Data processing system

Abstract

The present application relates to a system that automatically and flexibly links, adjusts, a plurality of independent network systems of telephone, data, mobile, train control, passenger guidance, reservation, advertisement distribution, equipment monitoring, etc. operated by railway operators or their subsidiaries, affiliated companies, by using a comprehensive gateway function of generative AI. The interface information or quality of service requirements of each network system are automatically collected, and the optimal linkage path or QoS setting is generated and reconfigured by AI, so that the adjustment work that has been performed by humans or additional hardware in the past becomes unnecessary, and the business efficiency of the entire group, the improvement of service quality, the rapid development of new services, and the automation of fault response are achieved.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention addresses is that multiple railway operators and their subsidiaries or affiliates independently construct and operate various networks or systems, such as telephone, data, mobile, and train control systems, which lack interoperability. Information sharing and service coordination between operators or subsidiaries are extremely difficult. This hinders rapid and flexible integration of passenger services or implementation of MaaS services, as well as improving overall group business efficiency or creating new businesses. System interoperability incurs significant time and costs, and maintaining or optimizing service quality is also challenging. Furthermore, previous methods of achieving system interoperability through manual simulation adjustments or retrofit hardware have revealed problems such as operational burdens, lack of flexibility, scalability limitations, and delays in responding to failures. To solve these problems, the invention aims to automatically collect and analyze interface information and service quality requirements from various networks or systems, and achieve comprehensive adjustment and interoperability through AI-based agent-based software. This fundamentally eliminates these previous challenges and accelerates the digital transformation of the entire railway operator group. Summary of the Invention

[0005] The means to solve the problem involved in this invention is to construct a system for comprehensively linking multiple independent network systems operated by operators or their affiliated companies. This system includes: an information collection unit that automatically collects interface information and service quality requirements of each network system; a linkage adjustment unit that automatically generates linkage paths between network systems based on the collected interface information and service quality requirements, and determines the optimal service quality setting for each linkage path; and an AI control unit that uses generative AI to flexibly and promptly reconfigure linkage paths or service quality settings based on prompts or additional requirements, automatically realizing linkage between network systems. The AI ​​control unit operates in a cloud environment, comprehensively judging the physical or logical connection status of each network system, the service quality requirements of the usage scenario, and the linkage requirements between operators and affiliated companies, and automatically performing adjustments that were previously done manually or with additional hardware through software unification. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8 This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10 It represents an emotion graph that maps multiple emotions.

[0016] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) The embodiments for implementing this invention will be described in more detail below. The system of this invention utilizes the comprehensive gateway function of generative AI to automatically and flexibly link and adjust various network systems operated by railway operator groups and their subsidiaries and affiliated companies, including telephone networks, data communication networks, mobile communication networks, train control networks, passenger guidance systems, reservation systems, advertising systems, and equipment monitoring systems. The system consists of an information collection unit, a linkage adjustment unit, and an AI control unit, primarily implemented on the server side. However, by distributing some functions at the on-site terminal side, the overall flexibility and real-time responsiveness are improved. Specifically, this system employs a hybrid artificial intelligence architecture combining a large-scale language model and a graph neural network. This system acquires the topology information of each network system and the real-time traffic status as multidimensional tensors. The acquired multidimensional tensors are provided to the input layer of the graph neural network. This graph neural network extracts the spatial features of the connections between nodes and the communication delay. The extracted spatial feature vectors are projected into the latent space of the large-scale language model. This large-scale language model uses an attention mechanism to comprehensively analyze the application strategies and spatial feature vectors described in natural language. Based on the analysis results, this system calculates the optimal routing path and bandwidth allocation parameters. The system outputs the calculated parameters as structured data. This structured data is then sent to the control interfaces of each gateway device. The gateway device dynamically updates the packet forwarding rules based on the received structured data. Through this process, the system minimizes packet loss even during sudden surges in communication traffic. This system effectively reduces communication latency. It provides extremely high reliability in critical scenarios such as real-time data synchronization between railway operation management systems and passenger guidance systems. The system enables the main model with numerous weighted parameters configured on the server side to work collaboratively with the lightweight model configured on the field terminals. The system uses federated learning to reflect local traffic variation patterns collected from the field terminals in the learning of the main model. This system continuously improves the overall inference accuracy without compromising communication bandwidth. This system eliminates manual settings required by human network administrators. Through nonlinear optimization operations in high-dimensional space, this system achieves a level of communication efficiency unattainable by existing methods.

[0040] First, the information collection department automatically collects interface information and Quality of Service (QoS) requirements through agent programs or APIs installed in various network systems. For example, it obtains information from telephone networks such as line type (analog, digital, IP), communication protocols in use, bandwidth, latency tolerance, and backup line information in case of failure. From data communication networks, it obtains information such as Ethernet specifications, VLAN configuration, traffic, packet loss rate, encryption methods, and firewall settings. From mobile communication networks, it obtains information such as wireless specifications (LTE, 5G, etc.), base station information, terminal authentication methods, communication areas, handover methods, and QoS category identifiers. From train control networks, it obtains information such as control signal transmission methods, redundancy configuration, real-time requirements, failover procedures in case of failure, and security authentication methods. From passenger guidance systems, it obtains information such as data update frequency, real-time performance, external linkage API specifications, number of users, and peak traffic information. From reservation systems, it obtains information such as transaction processing capacity, database configuration, external linkage interfaces, recovery procedures in case of failure, and session management methods. From advertising publishing systems, it obtains information such as publishing data capacity, publishing timing, targeting information, external linkage APIs, and target terminal information. The data collection unit acquires information from the equipment monitoring system, including the types of sensor data, data collection cycles, anomaly detection algorithms, notification methods, and external API linkages. Specifically, it converts diverse interface information and service quality requirements from various network systems into a unified numerical vector that can be processed by a machine learning model. It then uses one-hot encoding to convert categorical variables such as analog or digital lines into multi-dimensional binary vectors. Next, it uses min-max scaling to normalize continuous variables such as bandwidth or latency tolerances to a range of zero to one. Finally, it reconstructs time-series data such as packet loss rate or traffic into two-dimensional time-series tensors using a sliding window process with defined time windows. Combining this preprocessed data, it generates a single high-dimensional feature vector. This high-dimensional feature vector is then input into an autoencoder for anomaly detection. The autoencoder compresses the input high-dimensional feature vector into low-dimensional latent variables. Finally, it reconstructs the original high-dimensional feature vector from the compressed latent variables. Finally, it calculates the reconstruction error between the input vector and the reconstructed vector. If the calculated reconstruction error exceeds a preset threshold, this information collection unit determines that an unknown fault or abnormal traffic fluctuation has occurred in the network system. This unit then sends the identified abnormal data, along with a priority flag, to the subsequent linkage adjustment unit. Through this process, the system can detect early signs of minor network anomalies. This system effectively prevents large-scale communication failures. This system is applicable to scenarios such as detecting a slight increase in delay in control signals in the train control network and immediately initiating preparations for switching to a backup line. This system is not limited to simple data collection.This system significantly improves the overall response speed and fault tolerance of the system through advanced data analysis during the collection phase.

[0041] This information collection is carried out on-site by terminals (e.g., gateway devices, IoT devices, field terminals, management PCs, etc.) located in various network systems. The collected information is then aggregated at the information collection unit on the server side. For example, it is envisioned that IoT terminals located at train depots periodically send redundant configurations or fault occurrence records of the train control network to the server, or that management terminals of passenger guidance systems automatically send real-time user counts or traffic information to the server via API. Specifically, this field terminal utilizes edge computing technology to autonomously perform a single processing of the collected raw data. The field terminal uses a built-in lightweight convolutional neural network to extract local features from the time-series waveforms of sensor data or traffic logs. The field terminal generates the extracted feature vector. The field terminal sends the generated feature vector to the information collection unit on the server side. Compared to existing methods of sending raw data as is, this field terminal significantly reduces communication bandwidth consumption. For example, in an IoT terminal at a train depot, this field terminal receives a high-frequency time-series array obtained from a motor vibration sensor as input data. The field terminal applies a Fast Fourier Transform to this time-series array, converting it into frequency domain spectral data. This field terminal inputs spectral data into a convolutional neural network. The network outputs an anomaly score representing the degree of deviation from the normal spectral pattern. Only when the anomaly score exceeds a predetermined threshold does the field terminal send detailed log data to the server. This process drastically reduces the overall network communication load. It effectively prevents server-side processing resource exhaustion. In large-scale device monitoring systems connecting tens of thousands of IoT devices, this system is an essential configuration for ensuring scalability. Even in the event of a temporary interruption in communication with the server, the field terminal autonomously continues anomaly detection using its built-in model. The field terminal temporarily stores the detection results in local memory. Once communication is restored, the field terminal sends the stored data to the server. Through this distributed processing architecture, the system builds a robust information collection foundation without single points of failure.

[0042] Based on detailed interface information and service quality requirements of each network system collected from the information collection department, the linkage adjustment department automatically determines which systems should be linked, and through what paths and methods, and generates linkage paths. For example, when linking the mobile network of railway operator A with the reservation system of railway operator B, it automatically selects the optimal communication path or protocol conversion method by comparing the API specifications or communication protocols, security requirements, and QoS requirements (e.g., real-time reflection of reservation information, session maintenance, whether there is encrypted communication, etc.). Furthermore, when multiple network systems need to be linked simultaneously, such as linking train operation information with the passenger guidance system in real time while also forwarding data to the advertising system, it automatically designs complex linkage paths. The linkage adjustment department also handles various linkage modes, such as linking the train control network with the equipment monitoring system to automatically send control signals upon anomaly detection, or linking the passenger guidance system with the advertising system to display advertisements in conjunction with guidance displays in specific areas during events. Specifically, this linkage adjustment department uses deep reinforcement learning algorithms to dynamically explore the optimal linkage paths on complex network topologies. This linkage adjustment unit represents the node information and link status of each network system as graph structure data. This graph structure data is input into a graph attention network. The graph attention network dynamically calculates the importance between neighboring nodes and outputs the latent performance vector of each node. This linkage adjustment unit uses the output latent performance vector as the state space of a reinforcement learning agent. This reinforcement learning agent selects the next hop node in a manner that maximizes the action value function based on a Markov decision process. The input data to this reinforcement learning agent includes the latent performance vector of the current node, the latent performance vector of the destination node, and the numerical vector of the required quality of service. The output data from this reinforcement learning agent is the action probability distribution for the selectable group of neighboring nodes. This linkage adjustment unit determines the optimal path based on this action probability distribution. This linkage adjustment unit calculates the communication latency, packet loss rate, and bandwidth utilization of the determined path as evaluation functions. This linkage adjustment unit feeds back the reward signal based on the evaluation results to the reinforcement learning agent. This reinforcement learning agent uses temporal difference learning to update the weight parameters of its internal neural network. Through this processing, the system reconstructs paths in real-time and adaptively in response to changes in network congestion or failure conditions. Compared to existing methods using static routing tables, this system achieves a significant leap in communication reach and throughput. This system is particularly effective in scenarios such as disasters where, even if some communication base stations fail, it can instantly identify usable alternative routes and maintain the linkage between train operation information and passenger guidance systems.

[0043] The processing of the linkage adjustment unit is mainly implemented on the server side, but protocol conversion modules or QoS control modules can also be configured on the field terminal side to implement real-time response or local optimization processing. For example, it can automatically switch the field gateway device to the backup line when the train control network fails, or automatically adjust the bandwidth allocation of the advertising terminal when the number of users increases sharply. Specifically, the protocol conversion module and QoS control module on the field terminal side perform inference processing that utilizes hardware accelerators to achieve ultra-low latency processing on the edge side. The field terminal extracts the header information and a portion of the payload of the received data packets as a tensor of byte arrays. The field terminal inputs the extracted byte array tensor into a quantized recurrent neural network. This recurrent neural network infers the application category and required real-time performance of the communication flow from the sequence pattern of the data packets. The output data from this recurrent neural network is a category label representing the priority of the communication flow and numerical data of the recommended bandwidth allocation. Based on the output category label, the field terminal dynamically allocates data packets to multiple priority queues. Based on the recommended bandwidth allocation, the field terminal adjusts the parameters of the token bucket algorithm in real time. This process eliminates the need for centralized server-side control, enabling millisecond-level real-time responses to traffic fluctuations on-site. The system maximizes end-to-end communication latency. Applications include train control networks, where, upon detecting data packets requiring extremely high real-time performance, such as emergency stop signals, it temporarily blocks non-priority traffic and immediately routes it to a dedicated backup line. The field terminal calculates the error between its local inference results and actual communication quality measurements. This error information is periodically sent to the server. The system uses this error information to relearn the master model on the server side. By distributing the weight parameters of the relearned model to the field terminals, the system continuously improves the inference accuracy at the edge.

[0044] The AI ​​control unit utilizes generative AI to flexibly and promptly reconfigure linkage paths or QoS settings generated by the linkage adjustment unit based on input prompts or additional requests. For example, in the event of timetable disruptions, it can temporarily improve the QoS of the linkage path between the passenger guidance system and the train control system, or automatically integrate the network system of a new subsidiary into the existing linkage path when one is added, enabling dynamic reconfiguration. Generative AI receives natural language-based prompts (e.g., "Add the monitoring system of the new engineering subsidiary to the data linkage," "Enhance the redundancy of the train control system," "Increase the bandwidth of the advertising system during events," etc.), analyzes them, and automatically optimizes the overall linkage configuration or QoS settings of the system. The AI ​​control unit is primarily implemented on the server side, but lightweight AI agents can also be configured on the terminal side for distributed, real-time on-site responses or localized optimization. For example, the AI ​​agent can autonomously execute judgments on-site, such as "Automatically switch to the backup path when the latency of the train control network exceeds a threshold." Specifically, this AI control unit uses a large language model based on a Transformer architecture specialized for natural language processing as its core. This AI control unit uses a word segmenter to convert the natural language prompts input by the system administrator into a sequence of integer tokens. The AI ​​control unit then converts the token sequence into multi-dimensional word embedding vectors. Positional encoding is added to the word embedding vectors and fed into the encoder layer of the Transformer. Input data to this AI control unit includes, for example, the vector representation of the prompt "enhance the redundancy of the train control system" and an adjacency matrix tensor representing the current network topology. This large language model uses a self-attention mechanism to calculate the semantic correlation between the intent within the prompt and the current state of the network. Output data from this large language model is structured JSON data containing a set of configuration commands for routers to be changed, a vector of updated QoS parameters, and a list of node IDs of applicable objects. The AI ​​control unit feeds the structured output data into a parser. The parser verifies the appropriateness of the configuration content and its consistency with security policies. The AI ​​control unit automatically publishes the verified configuration command set to the network devices as targets. Through this process, the system enables network configuration tasks requiring advanced expertise to be performed solely through intuitive natural language-based instructions. This system effectively eliminates human error caused by configuration mistakes, significantly reducing operational and management costs. It provides extremely high real-time responsiveness in scenarios such as sudden traffic surges accompanying unexpected events, where on-site operators can request increased bandwidth via voice input, and the system immediately adjusts network configurations. A lightweight AI agent configured on the local terminal maintains the inference capabilities extracted from a large language model on the server side using knowledge distillation techniques.This lightweight AI agent continues to make autonomous judgments even when communication is interrupted.

[0045] Furthermore, generative AI autonomously controls the entire system based on prompts or learning results in all scenarios, including network system configuration changes, new service additions, fault response, and operational optimization. For example, when a railway operator group adds a new subsidiary and introduces an equipment monitoring system, the information collection department automatically obtains the interface information or QoS requirements of the new system, the linkage adjustment department generates the optimal linkage path with the existing network system, and the AI ​​control department can instantly optimize the data forwarding path or QoS settings based on the prompt "aggregate all monitoring data to the head office system in real time." Additionally, for example, when the number of users in the passenger guidance system increases sharply, the AI ​​control department automatically enhances bandwidth allocation and maintains service quality by improving the QoS of the linkage path. Specifically, this generative AI uses retrieval-enhanced generation technology and links with a database of past configuration change history and fault response inference manuals. When this generative AI detects the addition of a new subsidiary to the network system, it vectorizes the text data of the interface specifications of the new system provided by the information collection department. This generative AI uses the vectorized text data as a query to retrieve similar past system integration cases from the vector database. This generative AI combines retrieved past case data, current network state tensors, and hints from system administrators to generate an expanded context. This expanded context is then fed into a large language model. The output data from this large language model consists of a step-by-step execution plan for securely integrating the new system into the existing network, definition files for required protocol conversion rules, and differential data for the optimal routing table. The AI ​​control unit updates the settings of each network device in stages according to the output execution plan. After each step is completed, the AI ​​control unit automatically performs communication tests to confirm that there is no packet loss or increased latency. Through this process, the system dynamically utilizes past insights to deduce the optimal integration steps, even for unknown system configurations. This system effectively minimizes downtime during system integration and ensures service continuity. In use cases such as quickly connecting the acquired company's proprietary equipment monitoring system to a group-wide integrated monitoring platform and automatically absorbing differences in data formats, this system achieves overwhelming business efficiency. This generative AI possesses a reinforcement learning mechanism that uses the success or failure of the execution result as a reward. This generative AI continuously fine-tunes its inference model to increase the probability of success in similar tasks in the future.

[0046] Furthermore, generative AI can learn from past usage data, failure history, and user behavior data to predict future traffic increases or failures, proactively proposing and automatically applying optimal linkage paths or QoS settings. For example, during periods of rapid user growth, such as the end of the year, the beginning of the new year, or long holidays, the AI ​​control unit automatically increases the bandwidth of the passenger guidance or reservation system, strengthens redundancy configurations for paths with high failure risk, and can also perform preventative applications. Specifically, this generative AI incorporates a time series prediction model that combines a Long Short-Term Memory network and an autoregressive Transformer. This generative AI acquires multivariate time series data from the data storage unit, including traffic, failure logs, and calendar information (holidays, event dates, etc.) from the past few years. This generative AI converts the acquired multivariate time series data into a three-dimensional tensor composed of feature vectors for each time step. This generative AI inputs this three-dimensional tensor into the time series prediction model. This time series prediction model extracts periodicity and trends from past patterns to calculate the predicted traffic values ​​for each network node within a specific future time period. The output data from this time series forecasting model consists of the probability distribution of predicted traffic at each node and an array of risk scores representing the probability of congestion. This generative AI identifies nodes whose output risk scores exceed a specified threshold. The generative AI then generates preventative QoS setting change plans for the identified nodes. For example, during peak periods at the end of the year and the beginning of the new year, predicting concentrated access to a passenger guidance system, the generative AI outputs a setting script that pre-emptively doubles the bandwidth allocated to that system. The AI ​​control unit displays the output setting script on the system administrator's dashboard and requests approval. Upon approval, the AI ​​control unit automatically executes the setting changes on the specified date and time. Through this process, the system achieves a paradigm shift from reactive network management to data-driven proactive network management. This system effectively prevents service disruptions caused by sudden increases in traffic and improves user experience. The system calculates the prediction error as a loss function and periodically updates the weights of the time series forecasting model using error backpropagation. The system keeps the prediction accuracy up-to-date.

[0047] Furthermore, for example, when a new sensor device is introduced by an engineering subsidiary on-site, the AI ​​agent on the terminal side automatically identifies the interface specifications or data collection cycle of the new device and reports it to the information collection department on the server side. The linkage adjustment department automatically generates a linkage path with the existing device monitoring system, and the AI ​​control department can optimize the data forwarding path or QoS settings based on the prompt "notify the head office system immediately when an anomaly is detected." Specifically, the AI ​​agent on the terminal side performs dynamic analysis of newly connected devices based on a zero-trust network access architecture. When the AI ​​agent detects that a new sensor device is physically or logically connected to the network, it captures the payload and header information of the initial communication data packet sent by the device. From the captured data packet data, the AI ​​agent extracts features such as the type of communication protocol, data packet size, transmission interval, and port number used. The AI ​​agent converts the extracted features into a multi-dimensional vector and inputs it into a pre-learned clustering algorithm. The output data from this clustering algorithm is a classification label representing the device category and numerical data of the inferred data transmission cycle. The AI ​​agent sends the output classification label and numerical data to the information collection department on the server side through a secure encrypted communication channel. Based on the received device information, the server-side linkage adjustment unit automatically constructs the optimal virtual private network tunnel from the sensor device to the head office system. The AI ​​control unit compares the natural language parsing results based on the input prompts with the path information of the constructed tunnel. The AI ​​control unit then applies rules assigning the highest priority QoS labels to alarm packets detected in anomaly detection to the intermediate router group along the path. Through this process, the system completely eliminates manual network configuration work in dynamic field environments where various IoT devices are frequently added and removed. This system achieves plug-and-play secure device integration. It provides extremely high flexibility in application scenarios, such as immediately linking a temporarily installed environmental sensor group at the construction site with the head office monitoring platform to begin real-time security management.

[0048] Thus, the system of this invention, by employing core functions of information collection, linkage adjustment, and AI control on the server side, and implementing information acquisition or protocol conversion from various network systems and on-site optimization processing by local AI agents on the terminal side, can automatically and flexibly achieve comprehensive linkage and adjustment of the entire group's network or system by utilizing generative AI. Generative AI autonomously controls the entire system based on prompts or learning results in all scenarios, including network system configuration changes, new service additions, fault response, and optimization, thus becoming an indispensable element in realizing this invention. Specifically, the overall architecture of this system is constructed as a distributed machine learning foundation for the collaborative operation of cloud server clusters and edge device clusters. On the server side, the system runs a large language model with billions of parameters and a complex graph neural network, responsible for the global optimization calculation of the entire system. On the terminal side, the system runs a dedicated inference neural network, lightweighted through model quantization and pruning techniques, responsible for local optimization processing in milliseconds. Between the server and terminal sides, the system uses a federated learning protocol to periodically synchronize only the difference of the model's weight parameters. This system reflects the inherent traffic characteristics of each site in the overall learning model without sending confidential or private data to the server. Through this process, the system achieves a high-dimensional balance between data privacy protection and improved machine learning model performance. The system effectively minimizes communication bandwidth consumption while continuously improving the overall inference accuracy. It uses a Bayesian neural network to quantify the uncertainty of inference results for generative AI. The system has a mechanism that automatically switches to fail-safe mode and applies safe default settings when the confidence level of the inference is low. This system prevents fatal network failures caused by AI misjudgments. In the highly security-critical field of railway infrastructure, this system provides a technological breakthrough for the practical application of generative AI. This system is not merely a tool for improving business efficiency, but a next-generation computing technology capable of enabling the autonomous evolution of the network infrastructure itself.

[0049] (System Composition) The system involved in this implementation includes an information collection unit, a linkage adjustment unit, an AI control unit, a protocol conversion unit, a QoS control unit, a field terminal linkage unit, a data accumulation unit, and a user interface unit. The information collection unit automatically collects interface information, service quality requirements, communication protocols, bandwidth, latency tolerance, redundancy configuration, fault occurrence history, number of users, traffic information, sensor data collection cycle, external linkage API specifications, encryption methods, security authentication methods, transaction processing capabilities, database configuration, peak traffic information, data release capacity, release timing, targeted information, anomaly detection algorithms, and notification methods from various network systems operated by railway operators and their subsidiaries and affiliated companies, including telephone networks, data communication networks, mobile communication networks, train control networks, passenger guidance systems, reservation systems, advertising systems, and equipment monitoring systems. For example, examples include IoT terminals located at train depots periodically sending redundant configurations or fault occurrence histories of the train control network to the server; passenger guidance system management terminals automatically sending real-time user counts or traffic information to the server via API; advertising system terminals sending target terminal information or timing of advertising to the server; and equipment monitoring system terminals sending sensor data collection cycles or anomaly detection algorithms to the server. Specifically, this information collection unit uses a distributed message queue system to receive massive amounts of telemetry data asynchronously sent from tens of thousands of heterogeneous devices without loss. This information collection unit performs a data pipeline that converts the received data in various formats into multidimensional tensors with a common pattern. For example, this information collection unit extracts Boolean arrays representing redundant configurations of the train control network, lists of timestamp-error code pairs representing fault occurrence histories, and integer values ​​representing real-time user counts of the passenger guidance system. This information collection unit combines these extracted data as feature vectors representing each network node. This information collection unit uses principal component analysis to reduce the dimensionality of the combined feature vectors. This information collection unit expands the dimensionality-reduced vectors together with adjacency matrices representing the connections between nodes as graph-structured data in memory. Input data to this unit includes, for example, JSON-formatted sensor data sent from IoT terminals or XML-formatted traffic reports sent from management terminals via API. Output data from this unit is a standardized and structured graph tensor, which is used as direct input to machine learning models in subsequent linkage adjustment and AI control units. Through this processing, the system transforms siloed data generated from various system clusters with different vendors or specifications into a unified mathematical representation. This system effectively provides a high-quality data foundation for subsequent AI models to accurately learn the overall system relationships.This system, for example, serves as the basis for enabling AI to pinpoint hidden causal relationships that are difficult for humans to detect, such as variations in the sensor data collection cycle of a device monitoring system and a decrease in the transaction processing capacity of an appointment system.

[0050] Based on detailed interface information and service quality requirements of each network system gathered from the information collection department, the linkage adjustment department automatically determines which systems should be linked, and through what paths and methods, and generates linkage paths. For example, when linking the mobile network of railway operator A with the reservation system of railway operator B, it automatically selects the optimal communication path or protocol conversion method by comparing the API specifications or communication protocols, security requirements, and QoS requirements of both. Furthermore, it can handle a variety of linkage modes, such as designing complex linkage paths that link train operation information with the passenger guidance system in real time while also forwarding data to the advertising system; linking the train control network with the equipment monitoring system to automatically send control signals upon anomaly detection; and linking the passenger guidance system with the advertising system to display advertisements in conjunction with guidance displays in specific areas during events. Specifically, this linkage adjustment department uses a multi-agent reinforcement learning algorithm to calculate the Pareto optimal routing topology that simultaneously satisfies multiple competing linkage requirements. This linkage adjustment department receives graph tensors output from the information collection department as input. This linkage adjustment department models each network system as an independent learning agent. Each learning agent selects its action with the objective function of minimizing communication latency and maximizing throughput within its system. This coordinated adjustment unit formulates the competition for communication bandwidth among agents as a non-cooperative game based on game theory. The coordinated adjustment unit updates the weights of each agent's neural network to converge to a Nash equilibrium. The input data to this coordinated adjustment unit consists of the QoS requirement vector for each system, the current bandwidth utilization tensor, and the constraint matrix of the security policy. The output data from this coordinated adjustment unit is an update matrix of the routing table, containing identifiers of the optimal virtual paths for each data flow, IDs of the applicable protocol conversion modules, and the values ​​of the bandwidth to be allocated. This coordinated adjustment unit sends the output update matrix to the software-defined network controller, dynamically rewriting the flow tables of the physical switches. Through this process, the system does not perform single-path optimization but globally optimizes the overall resource allocation of the system. This system effectively prevents load concentration on specific networks and maximizes overall communication efficiency. For example, in scenarios where real-time train operation information and the forwarding of large-capacity advertising video data share the same physical line, this system can prioritize the processing of information data packets while dynamically allocating idle bandwidth to advertising data, thereby maximizing bandwidth utilization efficiency.

[0051] The AI ​​control department utilizes generative AI to flexibly and promptly reconfigure the linkage paths or QoS settings generated by the linkage adjustment department based on prompts or additional requests. For example, it can temporarily improve the QoS of the linkage path between the passenger guidance system and the train control system when timetables are disrupted; automatically incorporate the network system into existing linkage paths when a new subsidiary is added; automatically enhance bandwidth allocation and improve the QoS of linkage paths when the number of users in the passenger guidance system increases sharply; and increase the bandwidth of the passenger guidance system or reservation system and strengthen redundancy configuration for paths with a high risk of failure during periods of rapid user growth, such as the end of the year, the beginning of the new year, or long holidays. These actions enable dynamic reconfiguration or preventative applications. Specific examples of prompts read by generative AI include: "Adding the monitoring system of the new engineering subsidiary to the data linkage," "Enhancing the redundancy of the train control system," "Increasing the bandwidth of the advertising system during events," "Automatically expanding bandwidth when the number of users in the passenger guidance system exceeds a threshold," "Immediately notifying the head office system when anomalies are detected in the equipment monitoring system," "Automatically adjusting the transaction processing capacity of the reservation system during peak hours," "Automatically switching to the backup path when the latency of the train control network exceeds a threshold," and "Sharing the target terminal information of the advertising system with the passenger guidance system in real time." Specifically, this AI control unit installs an advanced inference engine that integrates prompt engineering technology and a retrieval-enhanced generation pipeline. This AI control unit uses a semantic role labeling algorithm to decompose the input natural language prompts into slots such as "action," "target system," "condition," and "constraint." This AI control unit uses the decomposed slot information as a query to extract similar context from a vector database storing past usage history. This AI control unit combines the extracted context with the current network state tensor to dynamically construct input prompts for a large language model. Input data to this large-scale language model includes, for example, a sequence of text tags stating "automatically expand bandwidth when the number of users in the passenger guidance system exceeds a threshold," and a numerical vector representing the current bandwidth usage. Output data from this large-scale language model consists of a script containing the specific command-line interface to be executed, a list of IP addresses of the routers targeted for configuration changes, and structured JSON data with predicted values ​​for the expected QoS metrics after the changes. The AI ​​control unit pre-executes the output script in a sandbox environment to verify that no network loops or communication interruptions occur. Only scripts that have been successfully verified are applied to the production environment. Through this process, the system automatically generates secure and accurate network control commands from fuzzy instructions based on natural language. This system significantly reduces the burden on application administrators and brings downtime associated with configuration changes to near zero.For example, in scenarios where traffic surges dramatically at the end of the year and the beginning of the new year, and administrators simply input "automatically adjust the transaction processing capacity of the reservation system during peak periods," the system autonomously performs horizontal scaling of servers and changes to load balancer settings, providing unparalleled operational efficiency.

[0052] The protocol conversion unit automatically performs various protocol conversion processes when communication protocols differ between different network systems. For example, it converts analog signals from telephone networks into IP packets, Ethernet frames from data communication networks into wireless packets from mobile communication networks, dedicated signals from train control networks into API formats for passenger guidance systems, and sensor data from equipment monitoring systems into database formats for reservation systems. Specifically, this protocol conversion unit executes an advanced data processing pipeline that combines deep packet inspection technology and semantic conversion algorithms. It parses the payload of received data packets at the byte level, extracting the semantic structure of the data. For example, it uses a digital signal processor to convert analog voice signals received from a telephone network into pulse code modulation (PCM) data. This PCM data is then encapsulated into IP packets for the Real-Time Transport Protocol (RTP). The unit uses a predefined parser to parse sensor data in its unique binary format received from the equipment monitoring system. The parsed data is converted into a dictionary object consisting of key-value pairs. This dictionary object is then dynamically reconstructed into SQL insert statements for relational databases or JSON payloads for RESTful APIs, as required by the reservation system. The input data to this protocol conversion unit is a raw byte array tensor based on the protocol specifications of the sending source system. The output data from this protocol conversion unit is a formatted data stream that fully conforms to the protocol specifications of the target system. To prevent data loss or precision degradation during the conversion process, this protocol conversion unit maintains a data type mapping table in memory and performs strict type checking. Through this process, this system completely eliminates communication barriers between legacy systems and the latest cloud-native systems. This system achieves a significant leap in improving interoperability between systems. For example, this system can convert signals from train control networks on dedicated lines that have been in use for decades into push notification data for passenger guidance applications on the latest smartphones in real time, instantly delivering delayed information to passengers.

[0053] The QoS control department implements various QoS controls based on the service quality requirements of each linked path, including bandwidth allocation, latency control, packet priority setting, automatic switching of redundant paths, traffic shaping, application of encrypted communication, and automatic recovery in case of failure. For example, it performs detailed QoS controls such as increasing bandwidth during peak hours in the passenger guidance system, automatically switching to backup paths when the latency in the train control network exceeds a threshold, dynamically adjusting bandwidth to match the timing of advertising releases, and increasing priority for real-time data forwarding when anomalies are detected in the equipment monitoring system. Specifically, based on the inference results of a machine learning model, this QoS control department dynamically adjusts the parameters of the token bucket algorithm used for traffic shaping and the hierarchical priority queue in microseconds. This QoS control department extracts a 5-tuple containing the source IP, destination IP, port number, and protocol number from the packet headers of each data stream. This QoS control department inputs the extracted 5-tuple into a hash function to generate a flow identifier. This QoS control department calculates a moving average of the packet arrival rate and latency for each flow identifier to generate a state vector. This QoS control department inputs this state vector into a deep Q-network trained through reinforcement learning. The output data from this Deep Q network represents numerical data indicating the optimal token replenishment rate, bucket size, and queue priority category for that flow. This QoS control unit directly writes the output numerical data into the hardware registers of the network interface card, performing packet forwarding control at the hardware level. Through this processing, the system eliminates the overhead of software processing, achieving line-rate advanced QoS control. This system effectively ensures latency for mission-critical communications even under network congestion. For example, when a device monitoring system detects a major anomaly such as a fire, the system immediately assigns the highest priority queue to the alarm data flow, temporarily limiting the bandwidth of all other background communications, thereby delivering the alarm data to the head office system with a millisecond delay. In such use cases, this plays a decisive role in protecting lives and infrastructure security. In encrypted communication applications, this QoS control unit also dynamically adjusts the encryption algorithm strength based on the importance of the data packet, optimizing the trade-off between security and processing latency.

[0054] The field terminal linkage unit collaborates with terminals, gateway devices, IoT devices, and management PCs deployed in various network systems to implement on-site information collection, protocol conversion, QoS control, AI-based intelligent agent-based on-site optimization processing, immediate response to faults, and handling of on-site prompts and inputs. For example, it enables various on-site linkages such as automatic switching to backup lines when the train control network fails, automatic bandwidth allocation adjustment by advertising terminals when user numbers surge, automatic interface specification identification and server reporting by equipment monitoring terminals when new sensor devices are added, and on-site operators conveying on-site requirements to the AI ​​control unit via voice input or prompts sent from tablet terminals. Specifically, this field terminal linkage unit deploys a containerized microservice architecture on edge devices to achieve autonomous data processing and asynchronous communication with the server. The unit uses an inference engine integrated into the on-site gateway device to analyze time-series data streaming from sensors in real time. Input data to the unit includes, for example, one-dimensional array data from vibration sensors sampled at 100 Hz. This one-dimensional array data is then fed into a one-dimensional convolutional neural network. The output data from this one-dimensional convolutional neural network is a floating-point number representing the degree of deviation from the normal state, indicating the anomaly score. When the anomaly score exceeds a threshold, the field terminal immediately rewrites its local routing table and performs a failover process that redirects control signals to a backup line. The field terminal accepts voice input from tablet terminals held by field operators. Using a built-in speech recognition model, it converts the speech waveform data into text prompts. The field terminal then sends the converted text prompts to the AI ​​control unit on the server side. Through this processing, the system is unaffected by communication delays or network outages to the cloud server, enabling real-time on-site response. The system leverages edge-side data filtering to significantly reduce the inflow of traffic to the core network. For example, in scenarios with unstable communication environments, such as tunnels, where the field terminal autonomously detects anomalies and only sends compressed alarm logs to the server upon communication restoration, the system provides extremely high reliability. The field terminal has the capability to receive differential data of model weight parameters periodically published from the server and perform hot updates to the edge-side inference engine without downtime.

[0055] The data accumulation department collects interface information, quality of service requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results from various network systems, accumulating them in a time-series format and utilizing them as learning data for generative AI. For example, it enables data-driven optimization applications such as learning past fault occurrence patterns to predict future faults, analyzing user behavior data to optimize bandwidth allocation during peak hours, parsing prompt input history to grasp on-site demand trends and generate automated rules, and accumulating AI-based optimization results to reflect in future decision-making. Specifically, this data accumulation department adopts a hybrid data lake architecture that combines a distributed relational database for storing structured data with a vector database for storing unstructured data and high-dimensional vectors. This data accumulation department rapidly writes millions of telemetry data points per second, communication logs, and fault events sent from the information collection department into the time-series database using timestamps as keys. This data accumulation unit uses a text embedding model to convert the text data of natural language prompts input to the AI ​​control unit, as well as the AI ​​outputs (setting scripts or routing change history), into multi-dimensional dense vectors. This data accumulation unit indexes and stores the converted dense vectors in a vector database. Input data to this data accumulation unit includes, for example, a time-series array of network device CPU utilization and a string indicating "expand bandwidth." Output data from this data accumulation unit consists of vector groups of similar past events related to specific fault patterns, extracted for relearning or retrieval enhancement of generative AI. This data accumulation unit periodically applies self-supervised learning algorithms to the massive accumulated data. This data accumulation unit automatically acquires unknown traffic variation patterns or features that may indicate faults hidden within the data. Through this processing, the system becomes more than just a log repository; it builds a knowledge base for the continuous evolution of the system as a whole. This system leverages past experience as a mathematical model for accumulation, dramatically improving the accuracy and speed of prediction and response to unknown future faults. This system serves as the basis for scenarios such as analyzing user behavior data from the past few years by multiplying it with meteorological data, predicting in advance the concentration of visits to the passenger guidance system under specific weather conditions, and enabling the AI ​​control department to propose preventative bandwidth expansion suggestions.

[0056] The user interface department provides GUIs or dashboards, voice input interfaces, and mobile terminal applications for system administrators or field operators to monitor system status, provide prompts, confirm and change linkage paths or QoS settings, provide instructions for handling faults, confirm AI-based optimization suggestions, and refer to the history of the data accumulation department. For example, it enables administrators to grasp the overall network linkage status or QoS settings at a glance on the dashboard, field operators to input prompts using natural language from tablet terminals, to transmit immediate response instructions to the AI ​​control department via voice input in the event of a fault, and for administrators to approve or reject AI-based optimization suggestions. Specifically, this user interface department is installed as a web-based single-page application with an advanced graphics rendering engine that provides real-time visualization of the backend system status. This user interface department applies dimensionality reduction algorithms such as t-distributed random neighborhood embedding (t-SNE) or Unified Manifold Approximation and Projection (UMAP) to the multidimensional network traffic data obtained from the data accumulation department. This user interface department maps the high-dimensional traffic status as a two-dimensional or three-dimensional scatter plot onto the screen. The input data to this user interface unit is a multidimensional tensor containing bandwidth utilization, latency, and packet loss rate for each node. The output data from this user interface unit is an interactive dashboard that visually highlights abnormal nodes in traffic clusters that deviate from the normal flow. When a system administrator clicks on a specific node on the screen, this user interface unit displays detailed QoS settings or past input history related to that node. This user interface unit provides a chatbot interface with a built-in natural language processing model. This user interface unit receives vague questions from administrators, such as "Where is the current bottleneck?" This user interface unit vectorizes this text and, in conjunction with the AI ​​control unit, generates specific node IDs and suggested solution text, which is then displayed on the screen. Through this processing, the system transforms the complex state of the massive network infrastructure into a format intuitively understandable to human cognitive abilities. This system significantly reduces the time required to identify the cause of a failure and supports rapid decision-making. For example, this system enables field operators to wear smart glasses and, through an augmented reality (AR) interface, overlay current traffic data or AI-based optimization suggestions onto their field of vision while performing maintenance tasks.

[0057] Thus, the system involved in this embodiment includes an information collection unit, a linkage adjustment unit, an AI control unit, a protocol conversion unit, a QoS control unit, a field terminal linkage unit, a data accumulation unit, and a user interface unit. These units work closely together, and through the application of generative AI, automatically and flexibly achieve comprehensive linkage, adjustment, and optimization of the railway operator group's overall network or system. Specifically, this system functions as a comprehensive computer architecture that tightly couples the aforementioned information collection unit, linkage adjustment unit, AI control unit, protocol conversion unit, QoS control unit, field terminal linkage unit, data accumulation unit, and user interface unit via a high-speed internal bus and message broker. This system sends and receives multidimensional tensor or structured JSON data asynchronously and in a streaming format between modules. This system places the inferential capabilities of generative AI at the core of the overall system arrangement. This system autonomously explores and applies optimal system configurations to address the nonlinear and dynamic environmental changes that previous rule-based network management systems could not handle. This system executes a series of processes end-to-end, from the input network state observation data to the output of optimal control commands, without human intervention. Through this process, the system maximizes the efficiency of communication resource utilization to mathematical limits within the massive and mission-critical infrastructure of railway operator groups. It achieves the technical effect of reducing recovery time to milliseconds and maximizing overall system availability. This system goes beyond simply automating existing operations; it enables the network infrastructure itself to adapt to its environment and grow into a self-organizing network through continuous self-evolution of machine learning models. The system provides ultimate scalability and flexibility, allowing AI to autonomously parse protocols and dynamically generate optimal linkage paths, even when new communication specifications or unknown IoT devices are introduced in the future, without requiring software modifications.

[0058] (Implementation steps) Step 1: Automatic Collection of Network System Information. Through agent programs or APIs installed in field terminals or gateway devices, IoT devices, and management PCs of various railway operators and their subsidiaries and affiliated companies operating telephone networks, data communication networks, mobile communication networks, train control networks, passenger guidance systems, reservation systems, advertising systems, equipment monitoring systems, etc., the system automatically collects interface information, service quality requirements, communication protocols, bandwidth, latency tolerance, redundancy configuration, fault occurrence history, number of users, traffic information, sensor data collection cycle, external linkage API specifications, encryption methods, security authentication methods, transaction processing capabilities, database configuration, peak traffic information, published data capacity, publication timing, targeted information, anomaly detection algorithms, notification methods, etc., and summarizes this information to the information collection unit on the server side. Specifically, in Step 1, this information collection unit activates lightweight data collection agents deployed in various field terminals and gateway devices to begin a distributed data capture process. This information collection unit directly accesses system calls provided by the operating system of each device or hardware performance counters. This data collection unit acquires low-level metrics such as CPU utilization, memory consumption, network interface transmitted / received bytes, and packet drop count at a resolution of microseconds. The input data to this unit is raw binary data streams asynchronously transmitted from various devices. Following a predefined protocol buffer pattern, this unit deserializes the received binary data streams. It then converts the deserialized data into structured time-series data points composed of timestamps, device IDs, and metric values. Within a defined time window, this unit aggregates these data points, generating a feature vector for each node. The output data from this unit is a massive state matrix tensor representing the current state of the entire system. To avoid straining network bandwidth during data collection, this unit employs an event-driven sending algorithm that only sends data when the data change on the agent side exceeds a specified threshold. Through this process, the system can aggregate massive amounts of telemetry data generated by tens of thousands of devices to the server side in real time without overloading the network. This system leverages the technology of continuously generating low-noise, high-quality input datasets to enable accurate inference by subsequent AI models. This system continuously executes step 1 in the background during system operation, capturing minute changes in the network's state without omission, thus establishing the basis for dynamic optimization control.

[0059] Step 2: Automatic parsing and generation of linkage paths and service quality requirements. Based on the detailed interface information and service quality requirements of each network system collected from the information collection department, the linkage adjustment department automatically determines which systems should be linked and how, and generates linkage paths. For example, it automatically designs various linkage modes, such as linkage between the mobile network and the reservation system, simultaneous linkage between the train operation information guidance system and the advertising system, linkage between the equipment monitoring system and the train control network in case of anomalies, and linkage between the guidance system and the advertising system during events. Specifically, this linkage adjustment department uses a graph neural network path exploration algorithm. This linkage adjustment department generates a high-dimensional graph tensor with each network system as nodes and physical and logical connection links as edges. This graph tensor maintains the processing capacity of each node or the bandwidth and latency values ​​of each edge as feature vectors. This linkage adjustment department inputs this graph tensor into the graph neural network. This graph neural network performs convolution operations on the input graph tensor. This graph neural network updates the hidden state vectors of each node. This graph neural network outputs a probability distribution matrix representing the optimal linkage path. This probability distribution matrix uses floating-point numbers to represent the probability that each edge is included in the linkage path. For example, as a path from the mobile network to the reservation system, this probability distribution matrix outputs the probability of an edge passing through a specific firewall as 0.95. Furthermore, as a path from train operation information to the passenger guidance system, this probability distribution matrix outputs the probability of an edge on a low-latency dedicated line as 0.98. This linkage adjustment unit compares each element of this probability distribution matrix with a specified threshold. This linkage adjustment unit connects edges exceeding the threshold to determine the specific linkage path. Through this process, the system eliminates the existing manual work of designing network configurations. This system completes the optimal path exploration in complex network topologies in milliseconds. This system minimizes communication latency, dramatically improving the data transmission efficiency of the entire system. This technology is extremely useful as the foundation for real-time data linkage between heterogeneous systems in smart cities.

[0060] Step 3: Automatic Setting of Protocol Conversion and QoS Control. Based on the linkage path generated by the linkage adjustment unit, the protocol conversion unit automatically implements the required protocol conversion processing between different network systems. The QoS control unit automatically sets the communication bandwidth allocation, latency control, packet priority setting, automatic switching of redundant paths, traffic shaping, application of encrypted communication, and automatic recovery in case of failure, according to the service quality requirements required for each linkage path. Specifically, this protocol conversion unit has a built-in protocol mapping engine that uses a natural language processing model. This protocol conversion unit performs syntax parsing on the data format definition files of the sending source system and the target system. This protocol conversion unit converts the parsing results into word embedding vectors. This protocol conversion unit inputs these word embedding vectors into a Transformer-based neural network. This neural network uses a self-attention mechanism to calculate the semantic correlation between fields. This neural network outputs a transformation matrix representing the optimal data mapping rule. For example, this transformation matrix outputs a rule that maps the timestamp field of the sending source to the occurrence time field of the target. In addition, this transformation matrix outputs a function that converts the status code of the sending source into the status flag of the target. This protocol conversion unit dynamically rewrites the packet header and payload according to this transformation matrix. Simultaneously, this QoS control unit dynamically allocates network resources using reinforcement learning algorithms. This QoS control unit obtains the current network traffic or packet loss rate as a state vector. This state vector is then input into a deep Q-network. The deep Q-network outputs the optimal bandwidth allocation for each linked path as an action vector. Based on this action vector, this QoS control unit updates the queuing settings of routers or switches. Through this process, the system completely prevents communication errors caused by protocol inconsistencies. The system autonomously controls traffic based on network congestion levels. This system improves the determinism of packet arrival, dramatically improving communication quality. This technology can be applied to seamless communication control between heterogeneous devices in industrial IoT environments.

[0061] Step 4: Dynamic Optimization and Reconfiguration Based on Generative AI. The AI ​​Control Department utilizes generative AI to flexibly and promptly reconfigure the linkage paths or QoS settings generated by the Linkage Adjustment Department based on prompts or additional requests. For example, it can temporarily improve the QoS between the passenger guidance system and the train control system when timetables are disrupted, automatically add the equipment monitoring system of a new subsidiary to the existing linkage path, enhance bandwidth allocation when the number of users in the passenger guidance system increases sharply, and strengthen bandwidth or redundancy configuration when users surge at the end of the year or the beginning of the new year, or during long holidays, thus achieving dynamic reconfiguration or preventative use. Specific examples of prompts that the generative AI can read include: "Add the monitoring system of a new engineering subsidiary to the data linkage," "Enhance the redundancy of the train control system," "Enhance the bandwidth of the advertising system when events are held," "Automatically expand bandwidth when the number of users in the passenger guidance system exceeds the threshold," "Immediately notify the head office system when the equipment monitoring system detects an anomaly," "Automatically adjust the transaction processing capacity of the reservation system during peak hours," "Automatically switch to the backup path when the latency of the train control network exceeds the threshold," and "Share the target terminal information of the advertising system with the passenger guidance system in real time." Specifically, this AI control unit possesses a dynamic reconfiguration engine centered around a large-scale language model. This AI control unit acquires natural language prompts input from the administrator. It then tokenizes these prompts, generating an input tensor. Next, it acquires current network topology information and resource usage status as structured data. This structured data is converted into a context vector. Finally, it combines the input tensor and context vector and inputs them into the large-scale language model. This large-scale language model consists of a Transformer architecture with billions of parameters. The large-scale language model parses the input information and outputs a specific sequence of commands for changing network settings. This command sequence is output as structured data in JSON format. For example, given input regarding enhancing the bandwidth of an advertising publishing system, the large-scale language model outputs a command to change the bandwidth limit of a specific virtual LAN from 500Mbps to 1Gbps. Similarly, given input regarding adding a monitoring system, the large-scale language model outputs a command to register a new group of IP addresses into the routing table. The AI ​​control unit verifies the syntactic validity of the output command sequence. This AI control unit sends verified command sequences to the control interfaces of each network device. This process eliminates the need for human administrators to manually input complex commands. The system instantly modifies network configurations based on intuitive natural language-based instructions. This system significantly shortens the lead time for operation and maintenance, eliminating human error. This technology is extremely effective for autonomous network operation and management in large-scale data centers.

[0062] Step 5: Linkage with Field Terminals and Field Optimization. The field terminal linkage unit links with terminals or gateways, IoT devices, management PCs, etc., installed in various network systems to implement field information collection, protocol conversion, QoS control, AI agent-based field optimization, immediate response to faults, and handling of prompts from the field. For example, it enables various field linkages such as automatic switching of field terminals to backup lines when train control network failures occur, automatic bandwidth allocation adjustment by advertising terminals when user numbers increase sharply, automatic identification of interface specifications and reporting to the server by equipment monitoring terminals when new sensor devices are added, and field operators sending prompts to the AI ​​control unit via voice input or tablet terminals to convey field requirements. Specifically, this field terminal linkage unit executes a lightweight inference model in an edge computing environment. This field terminal linkage unit runs on the microcontrollers assembled in each field terminal. This field terminal linkage unit acquires raw data from field sensors or network interfaces at a frequency of thousands of times per second. This field terminal linkage unit converts the acquired raw data into a one-dimensional time series tensor. The field terminal linkage unit inputs a one-dimensional time series tensor into a one-dimensional convolutional neural network on the edge side. This one-dimensional convolutional neural network extracts local features from the input data and outputs an anomaly detection score. This anomaly detection score takes continuous values ​​from 0 to 1. For example, using fluctuations in the arrival interval of data packets in the train control network as input, the one-dimensional convolutional neural network outputs an anomaly detection score of 0.92. Furthermore, using a sharp increase in the CPU usage of an advertising terminal as input, the one-dimensional convolutional neural network outputs an anomaly detection score of 0.85. When the anomaly detection score exceeds a preset threshold, the field terminal linkage unit immediately performs local failover processing. The field terminal linkage unit invalidates the routing table of the main line and enables the interface of the backup line. Through this processing, the system can make autonomous judgments on-site without waiting for communication delays with the central server. The system achieves recovery in milliseconds from the occurrence of a fault. The system maximizes the availability of the entire network. This technology is indispensable in the onboard networks of autonomous vehicles or the real-time control systems of industrial robots.

[0063] Step 6: Data Accumulation and AI Learning, Application Optimization. The Data Accumulation Department will collect interface information, service quality requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results from various network systems and accumulate them in a time series format. This data will be used as learning data for generative AI. The AI ​​Control Department will use this data to learn past fault occurrence patterns to predict future fault occurrences, analyze user behavior data to optimize bandwidth allocation during peak hours, analyze prompt input history to grasp the trends in on-site requirements and generate automated rules, and accumulate AI-based optimization results to reflect in future decision-making, thus achieving data-driven application optimization. Specifically, this Data Accumulation Department uses a distributed time-series database system to manage petabytes of data. This Data Accumulation Department will format the collected data into multidimensional vectors with timestamps. This Data Accumulation Department will write these multidimensional vectors into a column-oriented storage engine. This AI Control Department will read out the accumulated multidimensional vector group through batch processing. This AI Control Department will normalize the read data and construct a learning dataset for a Long Short-Term Memory (LSTM) network. This AI control unit uses the learning dataset to update the weight parameters of the Long Short-Term Memory (LSTM) network via error backpropagation. The LTM network learns time-series patterns of past traffic fluctuations or failures. It outputs a load prediction for each network link within a specific future time period. This load prediction is output as a numerical array in megabytes per second (MB / s). For example, the LTM network outputs a load prediction of 850 MB / s for the main backbone link at 8 AM tomorrow. Furthermore, it outputs a failure probability of 0.15 for a specific router. Based on this load prediction, the AI ​​control unit generates a bandwidth allocation schedule for the network in advance. Through this process, the system transitions from reactive network management to predictive proactive network management. The system optimizes resource allocation before congestion or failures occur. It maximizes network utilization efficiency, bringing downtime close to zero. This technology is highly effective in the automatic optimization of core networks for telecommunications operators or large-scale cloud infrastructure.

[0064] (specific use case) For example, in situations where multiple operators within a railway operator group operate independently, and their train control networks, passenger guidance systems, advertising systems, equipment monitoring systems, reservation systems, mobile communication networks, data communication networks, and telephone networks are all independently constructed, the system of this invention can be applied when real-time information linkage and service quality optimization between these systems are required. The information collection department automatically collects data from various network systems through agents or APIs. For example, it automatically collects information such as control signal transmission methods, redundancy configurations, real-time requirements, and fault occurrence history from the train control network; data update frequency, external linkage API specifications, number of users, and peak-hour traffic information from the passenger guidance system; data volume, timing, target terminal information, and targeting information from the advertising system; sensor data types, data collection cycles, anomaly detection algorithms, and notification methods from the equipment monitoring system; transaction processing capabilities, database configuration, and external linkage interfaces from the reservation system; wireless specifications, base station information, terminal authentication methods, and communication areas from the mobile communication network; Ethernet specifications, VLAN configuration, and traffic from the data communication network; and line type, communication protocol, and bandwidth from the telephone network. Specifically, this information collection department uses a multi-threaded parallel polling engine. Each independent system is allocated a dedicated data collection thread. The department periodically sends status acquisition API requests through each thread. This information collection unit asynchronously receives response data in JSON or XML format from various systems. It performs syntax parsing on the received response data to extract system-specific parameters. The extracted parameters are then normalized according to a common data pattern. The normalized data is stored as a feature vector in an in-memory database. This feature vector is represented as a multi-dimensional array of floating-point numbers. For example, the latency requirement obtained from the train control network is stored as a value in milliseconds in the first element of the feature vector. Furthermore, the idle capacity of base stations obtained from the mobile communication network is stored as a percentage in the second element of the feature vector. The number of transactions per second obtained from the reservation system is stored as an integer value in the third element of the feature vector. These feature vectors are then streamed to the subsequent linkage adjustment unit. Through this processing, the system collects status information in real-time and comprehensively from isolated heterogeneous system clusters. This system absorbs the differences in data formats between systems, providing a unified analytical foundation. This system significantly improves the overall visibility of large-scale infrastructure. This technology plays a key role in data integration within a consortium-based intelligent mobility infrastructure involving multiple companies.

[0065] The linkage adjustment unit, for example, links the passenger guidance system with the train control network to generate linkage paths that update passenger guidance displays in real time when train operating conditions change, and also links with the advertising system to generate linkage paths that publish advertisements linked to specific operating or event information to passenger guidance terminals. It also links the equipment monitoring system with the train control network to generate linkage paths that automatically send control signals to ensure train operation safety upon anomaly detection. Furthermore, it links the reservation system with the mobile communication network to generate linkage paths that notify users of their reservation information to mobile terminals in real time. For each of these linkage paths, the necessary protocol conversions or QoS controls are automatically set. For example, low-latency and high-reliability communication paths are automatically applied between the passenger guidance system and the train control network; bandwidth enhancement during peak hours is automatically applied between the advertising system and the passenger guidance system; instant notifications upon anomaly detection are automatically applied between the equipment monitoring system and the train control network; and secure data transmission is automatically applied between the reservation system and the mobile communication network. Specifically, this linkage adjustment unit uses a multi-agent reinforcement learning algorithm to simultaneously optimize multiple linkage paths. This linkage adjustment unit assigns the generation task of each linkage path to an independent learning agent. This linkage adjustment unit assigns bandwidth or latency constraints of the entire network as a shared environmental state to each agent. As actions of each agent, this linkage adjustment unit defines the selection of routing paths or the allocation of QoS categories between specific nodes. For the combination of actions selected by each agent, this linkage adjustment unit calculates a reward based on the overall system's communication efficiency or requirement fulfillment. This linkage adjustment unit updates the policy network of each agent in a way that maximizes this reward. This policy network outputs configuration parameters for the optimal linkage path. For example, for the linkage path between the passenger guidance system and the train control network, this policy network outputs a QoS setting with the highest priority, guaranteeing a latency of less than 1 millisecond. Furthermore, for the linkage path between the advertising system and the passenger guidance system, this policy network outputs a setting to allocate 10Gbps of bandwidth. For the linkage path between the reservation system and the mobile communication network, this policy network outputs a command to construct an encrypted tunnel. This linkage adjustment unit sends these output parameters to the controllers of each network device, reflecting the settings. Through this process, this system enables the simultaneous, conflict-free linkage of multiple systems with competing communication requirements. This system dynamically allocates limited network resources to the highest-value communication. This system fully automates the design and configuration of complex inter-system linkages. This technology is extremely important in integrated network environments that combine mission-critical communications and best-effort communications.

[0066] The AI ​​Control Department leverages generative AI, enabling it to read prompts such as: "Automatically expand bandwidth when the number of users in the passenger guidance system exceeds a threshold," "Add a new monitoring system from a construction subsidiary to the data linkage," "Automatically switch to a backup path when the latency in the train control network exceeds a threshold," "Share the target terminal information of the advertising system with the passenger guidance system in real time," "Automatically adjust the transaction processing capacity of the reservation system to match peak hours," "Immediately notify the head office system when the equipment monitoring system detects anomalies," "Enhance the redundancy of the train control system," and "Increase the bandwidth of the advertising system during events." Based on these instructions, the department flexibly and promptly reconfigures linkage paths and service quality settings. For example, during events, it increases the bandwidth of the linkage path between the passenger guidance system and the advertising system to cope with a sharp increase in the number of users, or automatically incorporates a new equipment monitoring system into the existing linkage path when one is added, and immediately switches to a backup path in case of a failure, achieving dynamic optimization. Specifically, this AI Control Department adopts a hybrid architecture that integrates a natural language understanding model and a rule generation engine. The department converts the input prompts into an abstract syntax tree through morphological and syntactic parsing. This AI control unit inputs the abstract syntax tree into the Transformer encoder to generate context embedding vectors. This AI control unit then inputs these embedding vectors into the decoder network. The decoder network generates executable scripts or configuration files that can be interpreted by network devices. The decoder network outputs logical structures containing conditional branches or loops. For example, the decoder network outputs the condition that the number of users exceeds a threshold as code for an event listener receiving a status monitoring trap. Furthermore, the decoder network outputs the action of expanding bandwidth as a script that calls the network controller's API to rewrite bandwidth limiting parameters. The decoder network outputs the instruction to switch to a backup path as a router configuration command that changes the priority value of the routing protocol. This AI control unit tests the generated scripts in a sandbox environment to verify that they do not adversely affect existing communication. The AI ​​control unit deploys the successfully verified scripts to the orchestrator in the production environment. Through this process, the system automatically converts vague human intentions into rigorous computer programs. The system performs dynamic reconfiguration of the network based on changing conditions without downtime. The system achieves a high degree of automation without relying on the programming skills of administrators. This technology serves as a core technology for realizing intent-based networks.

[0067] The field terminal linkage unit enables various field linkages. For example, when a fault occurs in the train control network, the field gateway device automatically switches to a backup line; the advertising terminal automatically adjusts bandwidth allocation when the number of users increases sharply; the equipment monitoring terminal automatically identifies interface specifications and reports to the server when new sensor devices are added; and field operators can convey field requirements to the AI ​​control unit via voice input or by sending prompts from tablet terminals. Specifically, this field terminal linkage unit operates as a containerized microservice deployed in each field terminal. It executes a daemon process for the hardware resources of the monitoring terminal. It captures interrupt signals from connected peripheral devices or sensors in real time. When a new sensor device is detected, it reads the descriptor information of the communication interface. It extracts the device identifier from this descriptor information. Based on the extracted identifier, it retrieves the local device catalog database and determines the appropriate communication protocol or data format. Finally, it formats the determined specification information into JSON-formatted metadata. For example, this field terminal unit, as a specification for a newly connected temperature sensor, generates metadata using a serial communication protocol and outputs a 16-bit integer value. Furthermore, this field terminal unit acquires voice input from field workers via microphone and converts it into text data using a local speech recognition model. This text data is then transmitted as a prompt to the central AI control unit via a secure communication path. Through this processing, the system immediately reflects physical changes on-site or worker requests throughout the system. This system enables automatic identification and configuration of plug-and-play devices. By autonomous edge processing based on on-site conditions, the system significantly reduces the load on the central server. This technology is highly practical for the efficient management and application of a widely distributed array of IoT devices.

[0068] The data accumulation department accumulates interface information, service quality requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results collected from various network systems in a time-series manner. The AI ​​control department uses this data to learn past fault occurrence patterns and predict future fault occurrences, analyzes user behavior data to optimize bandwidth allocation during peak hours, analyzes prompt input history to understand on-site requirements and generate automated rules, and accumulates AI-based optimization results to reflect in future decision-making, thus achieving data-driven application optimization. Specifically, this data accumulation department adopts a data lake architecture built on a distributed file system. This data accumulation department saves structured, semi-structured, and unstructured data in formats tailored to their respective characteristics. This data accumulation department compresses and saves communication history or traffic information as columnar files. This data accumulation department stores the logs of prompt input history or optimization results as indexes for search engines. This AI control department performs large-scale data mining on this data lake using a distributed processing framework. This AI control department uses the random forest algorithm to analyze the correlation between past fault occurrence history and traffic information. This AI control unit extracts combinations of specific feature patterns that serve as early warning signs of faults. The extracted patterns are then output as a decision tree model. For example, the AI ​​control unit outputs a prediction rule that a link disconnection will occur within ten minutes if a specific router's CPU utilization exceeds 80% and the packet loss rate exceeds 2% for five consecutive minutes. Furthermore, the AI ​​control unit applies a topic model to the input history prompts and clusters the tendency of frequently requested setting changes by field personnel. Frequently occurring request patterns are used as templates for generating automated scripts. Through this process, the system discovers hidden patterns that are unnoticed by humans from massive amounts of operational data. This system achieves preventative maintenance based on objective data, independent of rules of thumb. The system is capable of self-evolution and continuous improvement. This technology is an indispensable component of IT applications utilizing artificial intelligence across large-scale networks.

[0069] Thus, the system of this invention, by leveraging the automation of generative AI and close collaboration with field terminals, can achieve real-time and flexible linkage, adjustment, and optimization between multiple independent network systems. Specifically, this system seamlessly integrates the high inference capabilities of a centralized cloud-based large-scale language model with ultra-low latency autonomous control based on lightweight machine learning models distributed at the edge. The system possesses a stream processing engine that processes multi-dimensional telemetry data collected from various network nodes in real time. Based on the feature vectors extracted by this stream processing engine, the system dynamically recalculates the topology and resource allocation of the entire network. The system, via the network controller, applies the recalculated optimal configuration parameters to each physical device and virtual network function in milliseconds. The system continuously executes this series of feedback loops without human intervention. Through this processing, the system strictly maintains the service level agreement (SLA) even in the face of sudden surges in traffic or physical hardware failures. This system solves optimization problems in extremely dynamic and complex communication environments that were previously unmanageable in static network designs. This system dramatically reduces the operating costs of communication infrastructure while significantly improving the quality of service for end users. This technology can be widely used as a backbone control architecture in building next-generation communication networks or fully autonomous smart city infrastructure.

[0070] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0071] (Implementation methods for carrying out the present invention) The embodiments used to implement this invention will be described in more detail below. This invention utilizes a generative AI-powered integrated gateway to automatically and flexibly link and optimize multiple independent systems constituting urban infrastructure and resident services, enabling real-time and dynamic inter-system linkage in various urban application scenarios. Specifically, this integrated gateway functions as a distributed message broker, which accommodates data streams from millions of IoT devices or various server clusters widely distributed throughout the city via a single logical interface. This integrated gateway employs a publish-subscribe model to mediate asynchronous communication between systems. This integrated gateway incorporates a built-in memory stream processing engine for real-time parsing of the payload of received messages. Using this stream processing engine, the integrated gateway extracts metadata representing the source, data type, and urgency of each message. The integrated gateway converts the extracted metadata into a multidimensional tensor and inputs it into the generative AI's inference pipeline. Based on the input multidimensional tensor, the generative AI determines the optimal routing destination for the message and necessary data transformation processing. The generative AI outputs the decision result as an update command for the routing table. Through this process, the system dynamically resolves the complex dependencies between heterogeneous systems constituting urban infrastructure. Even in the event of a specific system failure, this system instantly establishes a path to an alternative system, preventing the shutdown of city functions. This system breaks down data silos, providing a foundation for the horizontal utilization of data across the entire city. This technology plays a crucial role in data-driven smart city operations and in enhancing resilience during large-scale disasters.

[0072] First, the information collection department automatically collects interface information or service quality requirements, communication protocols, bandwidth, latency tolerance, redundancy configuration, fault occurrence history, number of users, traffic information, sensor data collection cycle, external linkage API specifications, encryption methods, security authentication methods, transaction processing capabilities, database configuration, peak traffic information, data release capacity, release timing, targeted information, anomaly detection algorithms, and notification methods from various network systems that constitute the city, such as traffic control systems, power supply systems, disaster prevention and alarm systems, environmental sensor networks, information services for residents, water supply and drainage monitoring systems, medical institution networks, educational institution networks, logistics network systems, and administrative procedure systems.For example, traffic control systems collect signal control APIs, traffic volume data APIs, emergency contact APIs, signal status, traffic congestion information, automatic notifications in case of accidents, emergency vehicle priority control information, and traffic control information during events. Power supply systems collect supply and demand adjustment APIs, power monitoring data, fault notifications, power consumption, peak-hour power demand, distributed power generation operation status, power outage history, and priority supply information to critical infrastructure. Disaster warning systems collect earthquake warning APIs, evacuation guidance APIs, fire detection data, disaster history, tsunami warnings, debris flow warnings, shelter opening information, and resident safety information during disasters. Environmental sensor networks collect atmospheric data APIs, noise data, temperature and humidity data, PM2.5 concentration, anomaly detection information, pollen dispersal, ultraviolet radiation, and river water level data. Resident-oriented information services collect application notification APIs, administrative procedure APIs, event information, resident attribute information, number of users, emergency report release history, questionnaire results, and feedback from residents. Sewer and drainage monitoring systems collect... The system automatically collects data on water quality, flow rate, anomaly detection, emergency shut-off control API, leak detection, pump operation status, water purification plant operation data, and wastewater treatment plant load information. It also automatically collects data from medical institution networks, including patient information linkage API, emergency transport information, medical equipment operation status, medical system information during disasters, vaccine inventory information, infectious disease outbreak status, remote diagnosis and treatment data, and medical staff scheduling information. Furthermore, it automatically collects data from educational institution networks, including attendance management API, remote teaching data, emergency contact API, educational content release information, school activity information, safety confirmation information for children and students, guardian contact information, and learning progress data. Finally, it automatically collects data from logistics network systems, including delivery status API, inventory data, delivery route information, emergency delivery instructions, drone delivery status, warehouse temperature and humidity management data, cargo tracking information, and delivery vehicle location information. Finally, it automatically collects data from administrative procedures systems, including application acceptance API, resident ticket issuance data, special procedures information during disasters, personal identification number linkage information, tax processing status, subsidy application status, congestion information at administrative windows, and electronic signature data. Specifically, this information collection unit possesses a schema-on-read data integration engine for uniformly processing various data formats provided by different systems. This unit issues efficient data acquisition requests to endpoints of each system. The acquired raw data is temporarily stored in a memory buffer. Natural language processing techniques are used to automatically assign semantic tags to field names or metadata of each data point. The tagged data is then structured as a multi-dimensional knowledge graph. This knowledge graph uses each system as nodes and data dependencies or physical location relationships as edges.This information collection department assigns the collected parameters as attribute values ​​to each node of this knowledge graph. For example, for nodes in the traffic control system, the department assigns the current intersection congestion length as a floating-point number in meters. Furthermore, for nodes in the environmental sensor network, the department assigns the concentration of particulate matter as micrograms per cubic meter. For nodes in the medical institution network, the department assigns the number of available emergency beds as an integer value. This information collection department persists the constructed knowledge graph in a graph database, enabling it to respond to complex queries from subsequent coordination and adjustment departments. Through this process, the system integrates vast and diverse data scattered throughout the city into a semantically linked, massive dataset. This system eliminates data silos between systems, providing a real-time overview of the entire city's condition. The system fully automates the data collection and structuring process, eliminating manual work by data engineers. This technology is extremely effective as a foundational data collection mechanism for building digital twins of cities.

[0073] Based on detailed interface information and service quality requirements from various systems gathered by the Information Collection Department, the Coordination and Adjustment Department automatically determines which systems will be linked, and through what paths and methods, and generates linkage paths. For example, linking the traffic control system with the disaster warning system generates a linkage path that automatically controls traffic signals to ensure evacuation routes during an earthquake. Furthermore, linking resident-facing information services with the disaster warning system generates a linkage path that instantly disseminates evacuation instructions or disaster information to residents' smartphones or digital signage. Linking the power supply system with the medical institution network generates a linkage path that automatically prioritizes power supply to critical medical institutions during power outages. Linking the environmental sensor network with the traffic control system generates a linkage path that restricts traffic volume in specific areas when air pollution exceeds baseline values. Linking the sewer monitoring system with the disaster warning system generates a linkage path that automatically executes emergency sewer shut-off in conjunction with flood warnings during heavy rain. Linking the logistics network system with the traffic control system generates a linkage path that automatically ensures emergency material delivery routes during disasters. This system links educational institution networks with resident-oriented information services to generate a mechanism that instantly disseminates safety information for children and students to their guardians during disasters. It also links administrative procedures systems with resident-oriented information services to generate a mechanism that rapidly notifies residents of special procedures or support measures during disasters. Furthermore, for example, during city marathons, it links traffic control systems with resident-oriented information services to generate a mechanism that provides residents with real-time traffic control information or detour routes; during summer heatstroke warnings, it links environmental sensor networks with medical institution networks to generate a mechanism that strengthens emergency medical transport; and during infectious disease outbreaks, it links medical institution networks with educational institution networks to generate a mechanism that disseminates school closure information or remote learning information to children, students, and their guardians. This allows for flexible responses to various urban events and emergencies. Specifically, this linkage adjustment unit executes a hybrid optimization engine that combines constraint satisfaction problem solvers and evolutionary algorithms. This linkage adjustment unit receives a knowledge graph constructed by the information collection unit as input. This linkage adjustment unit formulates the linkage requirements between various systems into logical expressions. This linkage adjustment unit sets network bandwidth, latency, and security level as constraints. It uses an evolutionary algorithm to generate a candidate group of linkage paths that satisfy these constraints. The unit applies a fitness function to each candidate path and calculates a score. This fitness function comprehensively evaluates communication reliability, latency minimization, and resource consumption efficiency. The unit selects the linkage path with the highest score as the optimal solution. The selected linkage path is then output as a specific routing setting or a routing rule for the API gateway.For example, this linkage adjustment unit, acting as a linkage path between the traffic control system and the disaster warning system during an earthquake, outputs settings to establish a dedicated low-latency virtual private network tunnel. Furthermore, as a linkage path between the environmental sensor network and the traffic control system during air pollution, this linkage adjustment unit outputs settings for asynchronous communication via a message queuing system. As a linkage path between the environmental sensor network and the medical institution network during a heatstroke warning, this linkage adjustment unit outputs settings for data streams with high-priority QoS tags. Through this processing, the system derives the optimal linkage configuration in milliseconds from countless existing system combinations based on changes in urban conditions. This system mathematically solves complex constraints that human designers cannot fully consider. This system autonomously constructs the most effective inter-system linkages for urban activities or disasters. This technology is a core technology for realizing cross-domain service orchestration in smart cities.

[0074] The AI ​​control department utilizes generative AI to flexibly and promptly reconfigure the linkage paths or service quality settings generated by the linkage adjustment department based on prompts or additional requirements. For example, by providing generative AI with prompts such as "prioritizing the linkage between disaster warning systems and traffic control systems during a large-scale earthquake to ensure the immediate dissemination of evacuation information to residents' information services," "prioritizing notifications to residents' information services when an air pollution warning is issued, and linking with traffic control systems to restrict traffic in specific areas," "strengthening the linkage between disaster warning systems and residents' information services during power outages to provide real-time notifications of power supply status to critical infrastructure," "adding new logistics network points to the emergency delivery network," "linking emergency transport information from medical institution networks with residents' information services to notify local residents," "linking emergency contact APIs from educational institution networks with residents' information services to provide immediate updates to guardians," "strengthening the linkage between traffic control systems and residents' information services during city marathons to provide real-time notifications of traffic control information," and "strengthening the linkage between environmental sensor networks and medical institution networks when a heatstroke warning is issued to enhance the emergency transport system," the AI ​​can automatically optimize and reconfigure linkage paths or service quality settings based on these instructions. For example, maximizing the QoS of the linkage path between the disaster warning system and the traffic control system during an earthquake, enabling real-time dissemination of evacuation information to residents' information services, and strengthening the linkage of emergency transport information to medical institution networks can dynamically optimize the safety and service quality of the entire city. Specifically, this AI control unit is equipped with a large-scale language model fine-tuned to understand the city's context. This AI control unit receives natural language prompts input from city managers as text tensors. This AI control unit generates sensor data or traffic information representing the current city status as context tensors. This AI control unit combines this text tensor with this context tensor and feeds it to the input layer of the large-scale language model. This large-scale language model uses a self-attention mechanism to calculate the correlation between the intent within the prompt and the current city status. This large-scale language model generates specific control command sequences for each system of the city's infrastructure. These control commands are output as declarative configuration files. For example, in response to a prompt during a large-scale earthquake, this large-scale language model outputs a configuration file that prioritizes the traffic control system's signal control API to the highest level. Furthermore, this large-scale language model, in response to the issuance of heatstroke warnings, outputs a configuration file that changes the data transmission frequency of the environmental sensor network from a five-minute interval to a one-minute interval and doubles the data transmission bandwidth to the medical institution network. This AI control unit automatically applies the output configuration file to the edge routers or API gateways of each system via a configuration management tool. Through this processing, the system automatically translates the city manager's abstract policies into specific configuration values ​​for physical network devices or server clusters.This system eliminates the time available for manual setting changes during emergencies, enabling immediate response. It significantly enhances the system's adaptability to changing urban conditions. This technology provides an extremely innovative approach to the autonomous operation and management of software-defined urban infrastructure.

[0075] The protocol conversion unit automatically performs various protocol conversion processes when different systems use different communication protocols. For example, it converts the signal control API of a traffic control system to the evacuation instruction API of a disaster warning system, the supply and demand adjustment API of a power supply system to the priority power supply API of a medical institution network, the atmospheric data API of an environmental sensor network to the traffic volume control API of a traffic control system, the water quality data API of a sewer monitoring system to the flood warning API of a disaster warning system, and the delivery status API of a logistics network system to the emergency vehicle priority control API of a traffic control system. Specifically, this protocol conversion unit is equipped with a semantic data mapping engine using deep learning. The unit extracts the payload of the data packets received from the source system as a byte array. This byte array is input into a recurrent neural network to parse the structure and meaning of the data. The recurrent neural network extracts key-value pairs from the input data to generate an intermediate representation vector. This intermediate representation vector is then input into a decoder network based on the protocol specification of the destination system. The decoder network generates a new payload in a format that the destination system can interpret. For example, this decoder network takes binary signal status data received from a traffic control system as input and outputs evacuation instruction messages in JSON format for disaster warning systems. Additionally, it takes atmospheric data messages in message queuing protocol received from an environmental sensor network as input and outputs HTTP requests for traffic control systems. It also takes XML-formatted delivery status data received from a logistics network as input and outputs streaming data for traffic control systems. The protocol conversion unit assigns appropriate headers to the generated payload and sends it to the destination system. Through this processing, the system completely eliminates communication barriers between legacy systems and the latest cloud-native systems. This system reduces the cost of developing separate conversion programs for each system to zero. It prevents communication errors caused by inconsistent data formats, ensuring reliable data linkage. This technology is essential for ensuring interoperability in smart city environments where devices from multiple vendors coexist.

[0076] The QoS control unit implements various QoS controls based on the service quality requirements of each linked path, including bandwidth allocation, latency control, packet priority setting, automatic switching of redundant paths, traffic shaping, application of encrypted communication, and automatic recovery in case of failure. For example, during disasters, it maximizes the communication bandwidth between the disaster warning system and the traffic control system to deliver information service notifications to residents with minimal latency; automatically applies redundant paths in the linked path between the medical institution network and the power supply system; applies encrypted communication in the communication between the environmental sensor network and the traffic control system; applies packet priority setting in the communication between the logistics network system and the traffic control system; and applies traffic shaping in the communication between the educational institution network and information services for residents. Specifically, this QoS control unit performs dynamic traffic engineering that utilizes software-defined networking technology. The unit maintains the entire network topology and the current bandwidth used by each link as a state matrix. It obtains the service quality requirements of each linked path as constraint vectors. The unit then inputs this state matrix and constraint vectors into a linear programming solver. This linear programming solver calculates the bandwidth allocation for each flow that maximizes the overall network utility. Based on the calculated bandwidth allocation, the QoS control unit rewrites the flow tables of each switch using network control protocols. For example, for traffic from the disaster warning system during a disaster, the QoS control unit applies the highest priority flow rules to the switches. Furthermore, for communication between the healthcare network and the power supply system, the QoS control unit sets up a fast failover group that replicates and sends data packets to both the primary and backup paths. For communication from the environmental sensor network, the QoS control unit sends a command to the security gateway instructing the establishment of encrypted tunnels. For high-volume video publishing traffic from the educational institution network, the QoS control unit sets rate limits using a token bucket algorithm. Through this process, the system prioritizes the allocation of limited urban network resources to the most critical communications. The system completely prevents delays or packet loss of important data caused by network congestion. The system ensures the confidentiality and integrity of communications and protects urban infrastructure from cyberattacks. This technology is indispensable for achieving a highly reliable network foundation to support mission-critical urban services.

[0077] The field terminal linkage unit collaborates with terminals, gateway devices, IoT devices, and management PCs deployed in various systems to perform on-site information collection, protocol conversion, QoS control, AI-based agent-based on-site optimization, immediate response to faults, and reception of on-site prompts. For example, it enables diverse on-site linkages, such as automatically switching to backup paths when a traffic control system malfunctions, automatically adjusting bandwidth allocation when the number of users on a resident-facing information service terminal increases sharply, automatically identifying interface specifications and reporting to the server when new sensor devices are added to the environmental sensor network, allowing on-site personnel to convey on-site requirements to the AI ​​control unit via voice input or by sending prompts from tablet terminals, automatically linking emergency delivery instructions with the traffic control system at logistics network terminals, and instantly notifying resident-facing information services of emergency transport information from medical institution terminals. Specifically, this field terminal linkage unit manages a distributed edge node cluster based on a fog computing architecture. It executes lightweight container orchestration tools on each edge node. The unit dynamically deploys container images from the central server to each edge node to perform specific tasks. This edge node executes a local inference model within its deployed container. It processes data streams from connected sensors or devices in real time. For example, it takes video streams from cameras at intersections as input and uses an object detection model to detect accidents. Upon detecting an accident, it sends a command to switch the light to red directly to a nearby traffic signal controller, bypassing the central server. Furthermore, it functions as a gateway in medical facilities, calculating the estimated arrival time of ambulances. This calculated estimated arrival time is then forwarded directly to a local publishing server for resident-facing information services. It also acquires natural language text data input from tablets of field workers and extracts intent using a local natural language processing model. This generates control commands based on the extracted intent and applies them to the target devices. This process eliminates communication latency to the cloud, enabling ultra-low latency autonomous control on-site. Even in offline environments with WAN disconnections, the system maintains minimal urban functionality. This system significantly reduces the processing load and communication bandwidth consumption of the central server. This technology has a decisive advantage in the end-point control of urban infrastructure where real-time performance is extremely critical.

[0078] The Data Accumulation Department collects interface information, service quality requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results from various systems in a time-series manner. This enables data-driven application optimization. For example, the AI ​​Control Department uses this data to learn past fault occurrence patterns and predict future fault occurrences, analyzes user behavior data to optimize bandwidth allocation during peak hours, analyzes prompt input history to understand on-site requirements and generate automated rules, and accumulates AI-based optimization results to reflect in future decision-making. For instance, analyzing communication traffic history during past large-scale disasters allows for the preparation of optimal bandwidth allocation or linkage paths in advance for future disasters; analyzing resident feedback optimizes the timing or content of information dissemination; and analyzing logistics network delivery history automatically generates emergency delivery routes during disasters, contributing to the sophistication of urban operations. Specifically, this Data Accumulation Department constructs a distributed data warehouse capable of processing petabyte-scale data. This department acquires streaming data from various systems in real time, assigns timestamps and geospatial coordinates, and stores it as a multi-dimensional cube. This AI control unit applies a tensor decomposition algorithm to the multidimensional cube to extract latent feature vectors hidden in the data. The extracted feature vectors are then used as input to the environment state of a deep reinforcement learning model. This deep reinforcement learning model recreates past disaster-related traffic flow patterns in a simulation environment, learning the optimal network control strategy. As a learning outcome, the model outputs a predefined set of templates for specific disaster scenarios. For example, the deep reinforcement learning model outputs a matrix representing the bandwidth allocation ratio of each system during an earthquake of intensity six or higher. Furthermore, this AI control unit performs sentiment analysis on residents' feedback or application usage history using a natural language processing model. Based on the analysis results, the system outputs parameters determining the optimal time period or tone for information dissemination. The system inputs past GPS tracking data from logistics network points into a graph neural network, outputting a probability distribution of detour routes considering road blockage risks. Through this processing, the system crystallizes past experience data into the collective wisdom of the city. Even for unknown events, the system instantly derives the optimal response strategy from similar past patterns. This system evolves the use of urban infrastructure from passive response to proactive predictive control. This technology functions as the brain for realizing resilient and sustainable smart cities.

[0079] The User Interface (UI) Department provides GUIs, dashboards, voice input interfaces, and mobile applications for system administrators or field personnel to monitor system status, provide prompts, confirm and change linkage paths or QoS settings, provide instructions for handling faults, confirm AI-based optimization proposals, and reference the data accumulation department's history. For example, it implements diverse user interfaces, such as administrators clearly viewing the city's linkage status or QoS settings on a dashboard, field personnel using tablets to input prompts in natural language, providing immediate instructions to the AI ​​control department via voice input in case of faults, administrators approving or rejecting AI-based optimization proposals, residents changing information dissemination settings via smartphone applications, and administrative officials instantly disseminating special procedures information during disasters. Specifically, this UI department generates a digital twin dashboard for the city using 3D graphics technology. It obtains real-time traffic information or device status from the data accumulation department. The UI department overlays the acquired data as heatmaps or particle animations onto a 3D city model. When an administrator clicks on a specific building or intersection model, the UI department displays detailed parameters of the relevant network devices. This UI department also provides a chatbot interface embedded with a natural language processing model. This chatbot interface receives text or voice input from the administrator. It converts the input voice into Mel-frequency cepstral coefficients and feeds them into a speech recognition neural network for textualization. The chatbot then sends the textualized prompts to the AI ​​control unit. The user interface unit uses the optimization proposals returned by the AI ​​control unit as a visual guide to the differences in the network topology before and after the changes. This user interface unit and its differences... Figure 1 Upon initial display, the system renders "Approve" and "Reject" execution buttons on the screen. When the administrator presses the "Approve" button, this user interface sends an API call to the backend system to configure the application. Through this process, the system presents the highly complex internal state of urban infrastructure to humans in an intuitive and understandable way. This system enables even users without highly specialized knowledge to make accurate decisions with the support of AI. This system facilitates seamless collaboration between humans and AI, maximizing system efficiency. This technology becomes the standard model for human-machine interfaces in next-generation urban operations centers.

[0080] Thus, the embodiments of this invention, by utilizing generative AI, automatically and flexibly achieve comprehensive linkage, adjustment, and optimization of various systems constituting a city. This enables real-time and dynamic fulfillment of diverse requirements for smart cities, such as efficient city operations, enhanced disaster response capabilities, improved resident services, reduced operating costs, rapid recovery from system failures, dynamic response to city events, resident-participatory information flow, and enhanced overall city security. Specifically, this system processes massive amounts of streaming data generated in real-time from millions of IoT sensors and edge devices distributed throughout the city in a central parallel computing cluster. The system maps the collected heterogeneous mixed data into a high-dimensional tensor space. It then applies a deep learning model with a self-attention mechanism to the mapped tensor data. This deep learning model extracts the city's current state vector from the input tensor data. Based on the extracted state vector, the system calculates the optimal control parameters for the city's infrastructure. The calculated control parameters are then sent to each subsystem with a delay of milliseconds. Through this series of processes, the system completely eliminates manual intervention by human operators. This system dramatically reduces initial disaster response time from tens of minutes to milliseconds by eliminating manual intervention. This increased processing speed directly minimizes the impact of disasters. Even under normal circumstances, the system predicts minute fluctuations in traffic volume or electricity demand. Based on these predictions, the system dynamically allocates resources, significantly reducing the city's overall energy consumption. Instead of using rule-based heuristics, the system employs nonlinear function approximations learned from vast amounts of past operational data. This nonlinear function approximation enables robust control even for unknown, complex events. In specific application scenarios within smart cities, such as traffic control, power grid control, and disaster prevention systems, the system models the interdependencies between systems as a graph structure. Using graph neural networks, the system quantitatively assesses the impact of a failure in one system on other systems. Based on the assessment results, the system derives optimal network topology reconfiguration steps to localize the impact. By automatically executing these derived steps, the system maximizes the availability of urban infrastructure. This system receives, as input data, a one-dimensional array representing the number of vehicles at each intersection and a numerical vector representing the electricity consumption of each area. As output data, the system outputs structured data containing numerical values ​​representing the duration of the green light at each traffic signal and the output voltage of each substation. This output data is then sent to subsequent control modules to directly control the operation of the physical infrastructure equipment.

[0081] (System Composition) The system involved in this implementation includes an information collection unit, a linkage adjustment unit, an AI control unit, a protocol conversion unit, a QoS control unit, a field terminal linkage unit, a data accumulation unit, and a user interface unit. The information collection unit automatically collects interface information or service quality requirements, communication protocols, bandwidth, latency tolerance, redundancy configuration, fault occurrence history, number of users, traffic information, sensor data collection cycle, external linkage API specifications, encryption methods, security authentication methods, transaction processing capabilities, database configuration, peak traffic information, data release capacity, release timing, targeted information, anomaly detection algorithms, and notification methods from various network systems constituting the city, such as traffic control systems, power supply systems, disaster prevention and alarm systems, environmental sensor networks, information services for residents, water supply and drainage monitoring systems, medical institution networks, educational institution networks, logistics network systems, and administrative procedure systems.For example, the system automatically collects data such as: signal control API, traffic volume data API, emergency communication API, signal status, traffic congestion information, automatic notifications in case of accidents, emergency vehicle priority control information, and traffic control information during events from traffic control systems; supply and demand adjustment API, power monitoring data, fault notifications, power consumption, peak power demand, distributed power generation operation status, power outage history, and priority supply information to critical infrastructure from power supply systems; earthquake alarm API, evacuation instruction API, fire detection data, disaster history, tsunami warnings, debris flow warnings, shelter opening information, and resident safety information during disasters from disaster prevention alarm systems; atmospheric data API, noise data, temperature and humidity data, PM2.5 concentration, anomaly detection information, pollen dispersal, ultraviolet radiation, and river water level data from environmental sensor networks; and application notification API, administrative procedure API, event information, resident attribute information, number of users, emergency bulletin release history, questionnaire survey results, and feedback from residents from information services for residents. The system collects data from various sources, including: water quality data, flow data, anomaly detection data, emergency shut-off control API, leak detection information, pump operation status, water purification plant operation data, and wastewater treatment plant load information from sewer monitoring systems; patient information linkage API, emergency transport information, medical equipment operation status, disaster-related medical system information, vaccine inventory information, infectious disease occurrence status, remote diagnosis and treatment data, and medical staff scheduling information from medical institution networks; attendance management API, remote teaching data, emergency contact API, educational content release information, school activity information, children's and students' safety confirmation information, guardian contact history, and learning progress data from educational institution networks; delivery status API, inventory data, delivery route information, emergency delivery instructions, drone delivery status, warehouse temperature and humidity management data, cargo tracking information, and delivery vehicle location information from logistics network systems; and application acceptance API, resident ticket issuance data, disaster-related special procedures information, personal identification number linkage information, tax processing status, subsidy application status, administrative window congestion information, and electronic signature data from administrative procedures systems. Specifically, this information collection unit includes a preprocessing module that converts raw data in various formats asynchronously sent from each subsystem into a unified multidimensional tensor format. This preprocessing module normalizes the received text data, numerical data, and time-series data respectively. This preprocessing module combines the normalized data to generate, for example, a 256-dimensional feature vector. This information collection unit stores the generated feature vector along with timestamps and system identifiers in a distributed in-memory database. During this data collection process, this information collection unit performs semantic parsing using natural language processing techniques. This information collection unit parses the specifications or metadata of each API to extract semantic relationships between data items.Based on the extracted correlations, this information collection unit automatically constructs an ontology dictionary. By referring to this ontology dictionary, it automatically maps the same concept with different names across different systems. For example, it treats the number of vehicles in a traffic control system and traffic flow in an environmental sensor network as the same variable. This automatic mapping completely eliminates the manual time required for data integration between systems. The unit applies anomaly detection algorithms to the collected data. Using an autoencoder-type neural network, it compresses the input feature vector into a low-dimensional space. It then restores the compressed data to its original dimensions. The unit calculates the reconstruction error between the input vector and the restored vector. If the reconstruction error exceeds a pre-set threshold, the unit marks the data as an outlier. It prioritizes forwarding data containing outliers to subsequent linkage adjustment units. This prioritized forwarding enables early detection and rapid response to anomalies in urban infrastructure. This information collection unit receives, as input data, data such as a one-dimensional numerical array representing traffic volume, categorical variables representing signal status, and multi-dimensional vectors representing weather information. This unit combines these input data to generate a 1024-dimensional comprehensive state tensor. This comprehensive state tensor is then provided as input to the subsequent AI model. By generating this comprehensive state tensor, the unit can process the entire city's condition as a single mathematical expression. Using this mathematical expression, the unit significantly improves computational efficiency in subsequent processing.

[0082] The linkage adjustment unit automatically determines which systems are to be linked through what paths and modes based on the detailed interface information and quality of service requirements of each system aggregated from the information collection unit, and generates a linkage path. For example, the traffic control system is linked with the disaster prevention alarm system to generate a linkage path that automatically controls traffic signals and ensures evacuation paths when an earthquake occurs. The resident-oriented information service is linked with the disaster prevention alarm system to generate a linkage path that instantly issues evacuation instructions or disaster information to residents' smartphones or digital signage. The power supply system is linked with the medical institution network to generate a linkage path that automatically controls priority power supply to important medical institutions during a power outage. The environmental sensor network is linked with the traffic control system to generate a linkage path that restricts traffic volume in a specific area when air pollution exceeds a reference value. The water supply and sewerage monitoring system is linked with the disaster prevention alarm system to generate a linkage path that automatically performs emergency shutdown of sewers in linkage with flood alarms during heavy rain. The logistics outlet system is linked with the traffic control system to generate a linkage path that automatically secures emergency material distribution routes during disasters. The educational institution network is linked with the resident-oriented information service to generate a linkage path that instantly publishes the safety status information of children and students during disasters to their guardians. The administrative procedure system is linked with the resident-oriented information service to generate a linkage path that promptly notifies residents of special procedure information or support measures during disasters. When a city marathon is held, the traffic control system is linked with the resident-oriented information service to generate a linkage path that notifies residents of traffic control information or detour routes in real time. When a summer heatstroke alarm is issued, the environmental sensor network is linked with the medical institution network to generate a linkage path that strengthens the emergency transport system. When an infectious disease is prevalent, the medical institution network is linked with the educational institution network to generate a linkage path that publishes school closure information or remote teaching information to children, students and guardians. Specifically, the linkage adjustment unit constructs a directed graph in the memory, where all systems in the city are taken as nodes and communicable paths between the systems are taken as edges. The linkage adjustment unit assigns a multi-dimensional weight vector with bandwidth, latency, packet loss rate and security level as elements to each edge. The linkage adjustment unit receives the comprehensive state tensor provided from the information collection unit as an input. The linkage adjustment unit uses a graph neural network to infer the optimal linkage path for the current city situation. The graph neural network includes a plurality of message passing layers. The graph neural network exchanges the feature quantity of each node with adjacent nodes to update the embedding vector of the node. The linkage adjustment unit uses the updated embedding vectors of the nodes to calculate the probability that an edge should be formed between any two nodes. The linkage adjustment unit determines node pairs for which the calculated probability exceeds a threshold. The linkage adjustment unit uses a constrained shortest path algorithm extended from Dijkstra's algorithm to explore the optimal physical and logical paths connecting the determined node pairs. The linkage adjustment unit determines the explored paths as linkage paths. In the generation processing of the linkage paths, the linkage adjustment unit also uses a reinforcement learning algorithm in combination.This linkage adjustment unit defines the degree to which the generated linkage path meets the quality of service requirements as a reward. In past simulation environments, this linkage adjustment unit allowed the agent to undergo millions of trial and error iterations. Through this learning process, this linkage adjustment unit optimizes the weights of a deep neural network that approximates the Q-function. Using the optimized model, this linkage adjustment unit can instantly derive the optimal linkage path even under unknown disaster conditions. As input data, this linkage adjustment unit receives, for example, identifiers of the starting node and the ending node, a value representing the maximum required latency, and a value representing the minimum required bandwidth. As output data, this linkage adjustment unit outputs structured data containing a list of intermediate nodes to be traversed, identifiers of the communication protocols to be applied in each interval, and specific values ​​of the bandwidth to be allocated. This linkage adjustment unit sends this output data to the subsequent protocol conversion unit and QoS control unit. Through this series of processes, this linkage adjustment unit fully automates the complex routing settings that were previously manually designed by network administrators. Through this automation, this linkage adjustment unit reduces the time required for network configuration changes from days to milliseconds. This linkage adjustment unit achieves the technical effect of improving the real-time responsiveness to dynamically changing urban environments by increasing the processing speed.

[0083] The AI ​​control department utilizes generative AI to flexibly and promptly reconfigure the linkage paths or service quality settings generated by the linkage adjustment department based on prompts or additional requirements. For example, by having generative AI read prompts such as "prioritizing the linkage between disaster warning systems and traffic control systems during a large-scale earthquake to ensure the immediate dissemination of evacuation information to residents' information services," "prioritizing notifications to residents' information services when an air pollution warning is issued, and linking with traffic control systems to restrict traffic in specific areas," "strengthening the linkage between disaster warning systems and residents' information services during power outages to provide real-time notifications of power supply status to critical infrastructure," "adding new logistics network points to the emergency delivery network," "linking emergency transport information from medical institution networks with residents' information services to notify local residents," "linking emergency contact APIs from educational institution networks with residents' information services to provide immediate updates to guardians," "strengthening the linkage between traffic control systems and residents' information services during city marathons to provide real-time notifications of traffic control information," and "strengthening the linkage between environmental sensor networks and medical institution networks to enhance the emergency transport system when a heatstroke warning is issued," the AI ​​automatically optimizes and reconfigures linkage paths or service quality settings based on these instructions. Specifically, this AI control unit is equipped with a large-scale language model with billions of parameters as its core. This large-scale language model employs a Transformer architecture. It features a structure with multiple layers of self-attention mechanisms. The AI ​​control unit uses a tokenizer to convert the input natural language prompts into a sequence of integers. This AI control unit maps the integer sequence to word embedding vectors in a high-dimensional space. The AI ​​control unit adds positional encoding to the word embedding vectors. This AI control unit inputs the positionally encoded vector column into the large-scale language model. This large-scale language model calculates the dependencies between the elements of the input vector column. This large-scale language model generates hidden state vectors that consider context. This AI control unit converts the generated hidden state vectors into a probability distribution of control commands via fully connected layers and a Softmax function. This AI control unit decodes the sequence of control commands with the highest likelihood from the probability distribution. As input data, the AI ​​control unit receives text strings such as "earthquake occurred," "highest priority," and "link traffic control with disaster prevention." As output data, the AI ​​control unit outputs structured data containing pairs of system identifiers that should be linked, values ​​for the QoS categories that should be assigned, and identifiers for the security policies that should be applied. The AI ​​control unit inputs the output structured data into the syntax parser. The syntax parser verifies whether the structured data meets the system constraints. If the constraints are met, the syntax parser converts the data into an executable configuration script. The AI ​​control unit then sends the generated configuration script to the management interfaces of each network device.This AI control unit automatically translates ambiguous human intentions into precise, computer-interpretable control commands through a series of processes. During this translation process, the AI ​​control unit references past operational history and system configuration diagrams as an external knowledge base. It uses retrieval-enhanced generation technology to automatically supplement any missing technical parameters in the prompts. This supplementary function enables even operators without specialized knowledge to safely and accurately execute complex network controls. The introduction of this natural language interface dramatically reduces operational errors during emergencies.

[0084] When communication protocols differ between different systems, the protocol conversion unit automatically performs various protocol conversion processes. For example, it converts the signal control API of a traffic control system to the evacuation instruction API of a disaster warning system, the supply and demand adjustment API of a power supply system to the priority power supply API of a medical institution network, the atmospheric data API of an environmental sensor network to the traffic volume control API of a traffic control system, the water quality data API of a sewer monitoring system to the flood warning API of a disaster warning system, and the delivery status API of a logistics network system to the emergency vehicle priority control API of a traffic control system. Specifically, this protocol conversion unit uses deep packet inspection technology to parse the header information and payload of received data packets. Based on the parsing results, the unit determines the data format of the sending source system. The unit temporarily converts the determined data format to a general data model within the system. The unit then converts the general data model back to the data format required by the destination system. During this conversion process, the unit executes a semantic mapping algorithm. This semantic mapping algorithm uses a natural language processing model to calculate the semantic similarity between the names of data items in the sending source system and the names of data items in the destination system. The unit automatically associates items with similarity exceeding a specified threshold. This protocol conversion unit, for example, maps latitude and longitude data from the source to geolocation data of the destination. This protocol conversion unit also performs data type conversions simultaneously. For example, it converts a string-type timestamp to an integer type representing the UNIX epoch time. As input data, this protocol conversion unit receives signal status data from a traffic control system, described in Extensible Markup Language (XML) format. As output data, this protocol conversion unit outputs control commands for disaster prevention and alarm systems, described in JSON format. This protocol conversion unit sends this output data as an HTTP request to the API endpoint of the destination system. This protocol conversion unit uses a dedicated hardware accelerator to perform the protocol conversion processing at line speed. This protocol conversion unit utilizes a packet processing pipeline mounted on a field-programmable gate array (FPGA). Through this hardware processing, this protocol conversion unit reduces processing latency by several orders of magnitude compared to software-based conversion processing. This ultra-low latency processing enables seamless system interoperability in application scenarios where real-time performance is extremely important, such as priority control of emergency vehicles or protection against transient voltage drops in the power grid. This protocol conversion unit facilitates system integration across the entire city by ensuring interoperability in this heterogeneous environment.

[0085] The QoS control department implements various measures based on the required quality of service for each linkage path, including bandwidth allocation, latency control, packet priority setting, automatic switching of redundant paths, traffic shaping, application of encrypted communication, and automatic recovery in case of failure. For example, it performs detailed QoS control, such as maximizing communication bandwidth between the disaster warning system and the traffic control system during disasters to deliver information service notifications to residents with minimal latency; automatically applying redundant paths in the linkage path between medical institution networks and power supply systems; applying encrypted communication in the communication between environmental sensor networks and traffic control systems; applying packet priority setting in the communication between logistics network systems and traffic control systems; and applying traffic shaping in the communication between educational institution networks and information services for residents. Specifically, this QoS control department uses software-defined networking technology to manage the data plane and control plane of network devices separately. The QoS control department executes a deep reinforcement learning model in the control plane. This deep reinforcement learning model takes the current network traffic status, the utilization rate of each link, and the queue length as inputs to the state space. This deep reinforcement learning model outputs an action space that includes bandwidth allocation for each flow, weighted priority queues, and changes to routing paths. The QoS control unit calculates a reward function based on observations of packet latency, throughput, and packet loss rate. This QoS control unit uses a near-end policy optimization algorithm to update the neural network weights to maximize this reward. As input data, this QoS control unit receives, for example, a time-series array representing the number of bytes sent and received for each port collected from each switch, and a numerical vector representing the latency of each flow. As output data, this QoS control unit outputs a set of control rules containing pairs of matching conditions and actions that should be written into the flow tables of each switch. This QoS control unit publishes the output control rule set to each network switch via the OpenFlow protocol. Through this dynamic rule update, this QoS control unit reallocates bandwidth in milliseconds in response to sharp increases in traffic. In traffic shaping, this QoS control unit implements a token bucket algorithm at the hardware level. This QoS control unit smooths bursts of low-priority traffic transmissions and effectively protects the bandwidth of high-priority traffic. In encrypted communication applications, this QoS control unit uses advanced encryption standard algorithms to encrypt the payload of data packets. By offloading the encryption process to a dedicated encryption coprocessor, this QoS control unit minimizes the increase in communication latency. Through these advanced traffic engineering techniques, this QoS control unit utilizes limited network resources with extremely high efficiency. This improved resource utilization efficiency results in the prevention of network congestion and ensures the reliable delivery of mission-critical communications.

[0086] The field terminal linkage unit collaborates with terminals, gateway devices, IoT devices, and management PCs deployed in various systems to perform on-site information collection, protocol conversion, QoS control, AI-based agent-based on-site optimization, immediate response to faults, and reception of on-site prompts. For example, when a traffic control system malfunctions, the field terminal automatically switches to a backup path; when the number of users in the resident information service terminal increases sharply, the environmental sensor network terminal automatically identifies the interface specifications and reports to the server when new sensor devices are added; field operators can convey on-site requirements to the AI ​​control unit via voice input or by sending prompts from tablet terminals; logistics network field terminals automatically link emergency delivery instructions with the traffic control system; and medical institution field terminals instantly notify resident information services of emergency transport information, achieving diverse on-site linkages. Specifically, this field terminal linkage unit adopts an edge computing architecture to distribute the load on the central server. A lightweight inference engine is deployed within each field terminal. This inference engine executes a deep neural network with compressed model size through quantization and pruning techniques. The field terminal linkage unit performs a single local analysis on the sensor data collected at the field terminals. Only if an anomaly is detected in the analysis result is the feature vector sent to the central server. This data filtering significantly reduces network bandwidth consumption. The field terminal linkage unit performs federated learning at the field terminals. Each field terminal updates its model weights using local data. Only the updated weight differences are sent to the central server. The central server aggregates the weight differences from each terminal and updates the global model. The updated global model is then redistributed to the field terminals. This distributed learning approach achieves high-precision model learning using data from the entire city while protecting privacy. Input data for the field terminal linkage unit includes, for example, image tensors from on-site cameras or speech waveform data from microphones. Output data includes an array of bounding box coordinates representing object recognition results within an image, or a text string representing speech recognition results. The local terminal linkage unit sends the output text string as a prompt to the AI ​​control unit on the central server. Even in the event of a communication interruption with the central server, the local terminal linkage unit continues autonomous control using its local inference engine. For example, the local terminal linkage unit enables traffic signal control terminals to revert to a built-in rule-based engine during communication failures, maintaining safe signal cycles. Through this autonomous distributed control, the local terminal linkage unit prevents system-wide downtime caused by single points of failure. This improved fault tolerance significantly enhances the resilience of urban infrastructure.

[0087] The data accumulation department accumulates interface information, service quality requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results collected from various systems in a time-series manner. This enables data-driven application optimization. For example, the AI ​​control department uses this data to learn past fault occurrence patterns and predict future fault occurrences, analyzes user behavior data to optimize bandwidth allocation during peak hours, analyzes prompt input history to understand on-site demand trends and generate automated rules, and accumulates AI-based optimization results to reflect in future decision-making. For instance, analyzing communication traffic history during past large-scale disasters allows for the preparation of optimal bandwidth allocation or linkage paths in advance for future disasters; analyzing resident feedback optimizes the timing or content of information dissemination; and analyzing logistics network delivery history automatically generates emergency delivery routes during disasters, contributing to more sophisticated urban operations. Specifically, this data accumulation department possesses a distributed time-series database system capable of high-speed reading and writing of petabyte-level data. The department writes received data into storage in a columnar format with timestamps as the primary key. During data writing, the department uses data compression algorithms to reduce data size. This data accumulation unit applies a prediction model using Long Short-Term Memory (LSTM) networks to the accumulated time-series data. This prediction model takes as input the trend of traffic or error rate over a certain past period. It outputs the probability of traffic or failure occurring at a specific future time step. As input data, this data accumulation unit receives, for example, a 1440-dimensional numerical vector representing the packet loss rate per minute over the past 24 hours. As output data, it outputs a continuous value from 0 to 1 representing the probability of failure occurring in the next hour. If the output probability exceeds a predetermined threshold, the data accumulation unit sends an alert indicating preventative resource allocation to the AI ​​control unit. This data accumulation unit applies the Latent Dirichlet Allocation method as a topic modeling algorithm to the accumulated input history. It categorizes operator input tendencies into multiple potential topics. This data accumulation unit generates macro scripts to automate operations related to frequently occurring topics. By registering the generated macro scripts into the system, this data accumulation unit reduces the execution time of routine tasks. This data accumulation unit continuously monitors the difference between the AI-based optimization results and the actual system performance. When discrepancies exceed a certain tolerance range, this data accumulation unit automatically triggers a model relearning process. Through this continuous learning cycle, it keeps the AI ​​model's inference accuracy up-to-date, adapting to annual changes in the urban environment or the addition of new systems. This data accumulation unit, through its data-driven approach, eliminates the need for manual adjustments based on rules of thumb. This data accumulation unit delivers the technical effect of maximizing the overall system's efficiency through objective data-based optimization.

[0088] The User Interface Department provides GUI or dashboards, voice input interfaces, and mobile terminal applications for system administrators or field operators to monitor system status, provide prompts, confirm and change linkage paths or QoS settings, provide instructions for handling faults, confirm AI-based optimization proposals, and reference the history of the data accumulation department. For example, administrators can easily grasp the linkage status or QoS settings of the entire city on the dashboard; field operators can input prompts using natural language from tablet terminals; in the event of a fault, they can transmit immediate instructions to the AI ​​control department via voice input; administrators can approve or reject AI-based optimization proposals; residents can change information dissemination reception settings through smartphone applications; and administrative officials can instantly disseminate special procedures information during disasters, thus realizing a variety of user interfaces. Specifically, this User Interface Department is installed as a single-page application running on a web browser. This User Interface Department uses the WebSocket protocol to establish a persistent connection with the central server. This User Interface Department pushes and publishes events of status changes occurring on the server side to the client side with a delay of milliseconds. This User Interface Department updates the Virtual Document Object Model based on the received event data. This user interface (UI) optimizes rendering performance by reflecting only the altered differences in the actual screen rendering. It visualizes the network topology of the entire city using a graph rendering algorithm based on a mechanics model. It dynamically changes the color or thickness of nodes and edges based on real-time traffic or error rates. In the voice input interface, it executes a hybrid acoustic model combining a Hidden Markov Model and a deep neural network. It converts the input speech waveform into acoustic feature vectors such as Mel-frequency cepstral coefficients. It decodes these acoustic feature vectors into text strings. Input data for this UI includes, for example, an array of speech waveforms with a 16kHz sampling rate. Output data includes structured data containing the recognized text strings and confidence scores for each word. If the confidence score is below a threshold, it displays a confirmation prompt to the user on the screen. When displaying AI-based optimization proposals, it visualizes the data or inference process that forms the basis of the proposal using interpretable AI technology. This user interface (UI) section uses heatmaps to illustrate, for example, the branching conditions of decision trees or the attention weights of neural networks. Through this visualization, the UI improves the reliability of human operators' judgments regarding AI, which can easily be perceived as a black box. This UI provides an intuitive and responsive operating environment that reduces the cognitive load on operators in emergencies and supports rapid and accurate decision-making.

[0089] Specific examples of the prompts required for generative AI to read when implementing this invention include: "Prioritizing the linkage between the disaster warning system and the traffic control system during a large-scale earthquake to ensure the immediate dissemination of evacuation information to residents' information services"; "Prioritizing notifications to residents' information services when issuing an air pollution warning, and linking with the traffic control system to restrict traffic in specific areas"; "Strengthening the linkage between the disaster warning system and residents' information services during power outages to provide real-time notifications of power supply status to critical infrastructure"; "Adding new logistics network points to the emergency delivery network"; "Linking emergency transport information from the medical institution network with residents' information services to notify local residents"; "Linking the emergency contact API of the educational institution network with residents' information services to provide immediate updates to guardians"; "Strengthening the linkage between the traffic control system and residents' information services during city marathons to provide real-time notifications of traffic control information"; and "Strengthening the linkage between the environmental sensor network and the medical institution network when issuing a heatstroke warning to enhance the emergency transport system." Specifically, this system uses a natural language processing pipeline to convert these natural language-based prompts into machine-readable intent representations. This system performs morphological and syntactic parsing on the input prompt text. It extracts entities and actions related to them from the text. Using slot-filling, the system maps the extracted entities and actions to blank spaces in a predefined command template. Input data includes, for example, a string like, "In the event of a large-scale earthquake, prioritize the linkage between the disaster warning system and the traffic control system to ensure the immediate dissemination of evacuation information to residents." Output data consists of a structured control object with the trigger condition set to a large-scale earthquake, the target systems set to the disaster warning system and the traffic control system, and the action set to maximize bandwidth. This control object is converted into an API call that the network controller can interpret. During this conversion, the system leverages the contextual learning capabilities of a large language model. Along with the prompt text, the system provides the model with past successful conversion examples as a few-sample demonstration. Through this prompt engineering approach, the system strictly controls the model's output format, minimizing the syntactic error rate. The generated control object undergoes a security check using static parsing tools. The system preemptively intercepts commands that might cause unexpected system shutdowns or privilege escalations. This system generates deterministic and secure system control commands from unstructured natural language data through a series of processing steps. This unconventional approach allows domain experts without programming knowledge to directly incorporate their expertise into system control. The system's abstract interface eliminates the human element in system operation and significantly reduces operational costs.

[0090] (Implementation steps) Step 1: Automatic Collection of Urban System Information. The information collection department automatically collects interface information, service quality requirements, communication protocols, bandwidth, latency tolerance, redundancy configuration, fault occurrence history, number of users, traffic information, sensor data collection cycle, external linkage API specifications, encryption methods, security authentication methods, transaction processing capabilities, database configuration, peak traffic information, published data capacity, publication timing, targeted information, anomaly detection algorithms, and notification methods from various network systems constituting the city, including traffic control systems, power supply systems, disaster warning systems, environmental sensor networks, resident information services, water supply and drainage monitoring systems, medical institution networks, educational institution networks, logistics network systems, and administrative procedure systems. This information is then aggregated in the server-side data storage department. Specifically, in this step, the information collection department performs asynchronous polling and event-driven subscription processing in parallel for each subsystem. The department uses a message queue telemetry transport protocol to receive lightweight sensor data published from millions of IoT devices via a proxy. Simultaneously, the department periodically calls APIs based on a representational state transition architecture to obtain static configuration information or batch processing results from each system. This data collection unit temporarily buffers the diverse data streams received into a message queue in memory. It asynchronously reads data from the message queue and performs data cleaning. This includes interpolating missing values, removing outliers, and synchronizing timestamps. The cleaned data is then converted into a system-wide common multidimensional tensor format. Input data includes temperature and traffic volume data acquired at different sampling rates. Output data is a feature matrix that has been resampled and normalized to a common time base. This feature matrix is ​​persisted to a distributed columnar database with high-speed write performance. The data collection pipeline is dynamically horizontally scaled using container orchestration tools. Data processing nodes are automatically added when traffic increases sharply and removed when load decreases. This auto-scaling mechanism keeps data collection latency in the millisecond range. The establishment of this high-throughput and low-latency data collection foundation brought about by this information collection department is directly related to the technical effects of improving the inference accuracy and ensuring real-time performance in subsequent AI processing.

[0091] Step 2: Automatic parsing and generation of linkage paths and service quality requirements. Based on detailed interface information and service quality requirements from each system collected by the information collection department, the linkage adjustment department automatically determines which systems should be linked, and in what ways, and generates linkage paths. For example, linkage between traffic control systems and disaster warning systems can be used for signal control during earthquakes; linkage between information services for residents and disaster warning systems can be used for the immediate dissemination of evacuation instructions; linkage between power supply systems and medical institution networks can be used for priority power supply during power outages; and linkage between environmental sensor networks and traffic control systems can be used for traffic restrictions during air pollution. This comprehensive and automatic design of linkage paths corresponds to a wide variety of urban application scenarios. Specifically, in this step, the linkage adjustment department uses the collected multidimensional tensor data as input and executes a constraint satisfaction problem solver. The linkage adjustment department treats all systems within the city as nodes and physical and logical connections as edges, unfolding a graph structure in memory. The linkage adjustment department assigns weights to each edge based on its current bandwidth, latency, and security level. The linkage adjustment department then formulates the required service quality requirements as constraints. This linkage adjustment unit sets constraints such as a latency of less than 10 milliseconds and a bandwidth of more than 1 Gbps. It uses mixed-integer linear programming to calculate the optimal path that minimizes the overall network load while satisfying these constraints. To reduce computational complexity, it employs the A* search algorithm as a heuristic. Input data includes the origin node, destination node, and a set of required QoS parameters. Output data includes a routing table containing the sequence of nodes to be traversed and the amount of resources to be allocated to each link. The unit utilizes parallel processing with multi-core CPUs to accelerate path computation. In the event of multiple simultaneous linkage requests, it schedules processing based on the priority of each request. For example, it prioritizes linkage requests from disaster alert systems involving human lives over those from routine environmental data collection. The unit pre-verifies the appropriateness of the generated linkage path using an internal simulator. Based on simulation results, it identifies bottlenecks that may lead to packet loss or congestion and fine-tunes the path. This coordinated adjustment unit, by applying advanced mathematical optimization techniques, instantly constructs complex and efficient network topologies that are beyond human intuition to design. This automated design capability dramatically improves the unit's adaptability to dynamic changes in urban infrastructure.

[0092] Step 3: Automatic Setting of Protocol Conversion and QoS Control. Based on the linkage path generated by the linkage adjustment unit, the protocol conversion unit automatically implements the required protocol conversion processing between different systems. The QoS control unit automatically sets the communication bandwidth allocation, latency control, packet priority setting, automatic switching of redundant paths, traffic shaping, application of encrypted communication, and automatic recovery in case of failure, according to the service quality requirements of each linkage path. Specifically, in this step, the protocol conversion unit dynamically deploys container images for protocol conversion in the virtual network functions configured in each node of the generated linkage path. Within the deployed containers, the protocol conversion unit performs payload serialization and deserialization processing. For example, the protocol conversion unit converts binary data into text data. The QoS control unit issues setting commands to the flow control subsystem running at the kernel level of the operating system. The QoS control unit uses a hierarchical token bucket algorithm to logically segment the transmission queues of the network interface cards. The QoS control unit sets parameters for guaranteed bandwidth and maximum bandwidth for each segmented queue. The QoS control unit marks the priority of packets throughout the network by rewriting the values ​​of code points contained in the IP header of the packets. This QoS control unit receives input data such as the identifier of the linked path, the source IP address, the destination IP address, and the required bandwidth. As output data, it outputs a low-level network control script. This QoS control unit automatically executes this control script on each router and switch via a secure shell. After applying the settings, this QoS control unit uses active measurement methods to test actual communication quality. It sends test packets to measure round-trip time or jitter. If the measurement results do not meet requirements, the QoS control unit automatically readjusts the settings through a feedback loop. Through this closed-loop control, this QoS control unit ensures stable communication quality unaffected by changes in the physical state of the network. This QoS control unit brings the technical effect of maximizing network utilization efficiency and preventing service interruptions caused by congestion through this software-defined, granular resource management.

[0093] Step 4: Dynamic Optimization and Reconfiguration Based on Generative AI. The AI ​​Control Department utilizes generative AI to flexibly and promptly reconfigure the linkage paths or service quality settings generated by the Linkage Adjustment Department based on prompts or additional requirements. For example, by having the generative AI read prompts such as "Prioritize the linkage between the disaster warning system and the traffic control system during a large-scale earthquake to ensure the immediate dissemination of evacuation information to residents' information services," "Prioritize notifications to residents' information services when an air pollution warning is issued, and link with the traffic control system to restrict traffic in specific areas," "Strengthen the linkage between the disaster warning system and residents' information services during a power outage to provide real-time notifications of power supply status to critical infrastructure," "Add new logistics network points to the emergency delivery network," "Link emergency transport information from the medical institution network with residents' information services to notify local residents," and "Link the emergency contact API of the educational institution network with residents' information services to immediately disseminate information to guardians," the AI ​​automatically optimizes and reconfigures linkage paths or service quality settings based on these instructions. Specifically, in this step, the AI ​​control unit executes an inference pipeline for a large language model on a parallel computing cluster utilizing GPUs. The AI ​​control unit inputs the natural language prompts along with the system's state vector into the model. The AI ​​control unit tokenizes the prompt text, converting it into a high-dimensional embedding vector. The AI ​​control unit uses an attention mechanism to calculate the correlation between important keywords within the prompts and the system state. Based on this calculation result, the AI ​​control unit progressively generates network configuration change steps. The AI ​​control unit receives input data such as texts like "earthquake occurred" and "highest priority," along with an adjacency matrix representing the current network topology. The AI ​​control unit outputs structured patch data containing lists of edges to be deleted, lists of edges to be added, and QoS parameters to be updated. The AI ​​control unit applies the output patch data as a transaction to the network controller. During application, the AI ​​control unit uses a two-phase commit protocol to ensure the atomicity of the configuration changes. In the event of an error during configuration changes, the AI ​​control unit automatically performs a rollback, restoring the system to a safe state. This AI control unit completes the dynamic reconfiguration process within seconds to tens of seconds. This high-speed reconfiguration enables the system to adapt instantly to rapid environmental changes, such as those occurring during disasters. To prevent model illusions, the AI ​​control unit uses a rule-based verification engine to cross-check the output configuration schemes. Configuration schemes that violate physical constraints or security policies are automatically discarded and regenerated. Through this combination of generative AI and rigorous verification logic, the AI ​​control unit achieves a high level of balance between flexibility and security.This AI control unit brings an unconventional processing method that achieves a technological effect of autonomous and adaptive system operation that surpasses the limits of manual settings by human operators.

[0094] Step 5: Linkage with Field Terminals and On-site Optimization. The field terminal linkage unit links with terminals, gateway devices, IoT devices, management PCs, etc., installed in various systems to implement on-site information collection, protocol conversion, QoS control, AI agent-based on-site optimization, immediate response to faults, and reception of on-site prompts. For example, when a traffic control system malfunctions, the field terminal automatically switches to a backup path; when the number of users in the information service terminal for residents increases sharply, it automatically adjusts bandwidth allocation; when new sensor devices are added, the environmental sensor network terminal automatically identifies the interface specifications and reports to the server; and field operators convey on-site requirements to the AI ​​control unit through voice input or by sending prompts from a tablet terminal, thus achieving diverse on-site linkages. Specifically, in this step, the field terminal linkage unit executes a microservice-based software stack on the edge device. The field terminal linkage unit activates the local AI agent deployed in each field terminal. The local AI agent processes the raw data acquired from the sensors connected to the terminal in real time. For example, the local AI agent executes a convolutional neural network to identify vehicle license plates from camera images. The local terminal linkage unit receives, for example, image tensors with a resolution of 1920×1080 as input data. As output data, it outputs a vector containing the identified vehicle category, location coordinates, and confidence score. Based on this inference, the local terminal linkage unit determines real-time control actions on-site. For example, upon detecting a specific vehicle, it sends a control signal to automatically open the gate. The local terminal linkage unit continuously monitors the quality of the communication link with the central server. If the packet loss rate exceeds a threshold, the local terminal linkage unit autonomously switches to fallback mode. In fallback mode, it stops sending low-importance data and only sends critical alarms via a low-bandwidth backup line. When receiving voice input from field operators, the local terminal linkage unit executes a lightweight speech recognition model at the edge. It inputs the recognized text into an intent extraction model, and executes locally executable commands immediately. The local terminal linkage unit only forwards complex requests requiring judgment from the central server to the server. This field terminal linkage unit, through collaborative processing between the edge and the cloud, minimizes the overall system response latency while optimizing network bandwidth consumption. The adoption of this distributed processing architecture eliminates the risk of single points of failure and significantly improves the robustness of urban infrastructure.

[0095] Step 6: Data Accumulation and AI Learning & Application Optimization. The Data Accumulation Department will accumulate interface information, service quality requirements, communication history, fault occurrence history, traffic information, user behavior data, prompt input history, and AI-based optimization results collected from various systems in a time-series manner. The AI ​​Control Department will use this data to learn past fault occurrence patterns and predict future fault occurrences, analyze user behavior data to optimize bandwidth allocation during peak hours, analyze prompt input history to understand on-site requirements and generate automated rules, and accumulate AI-based optimization results to reflect in future decision-making, thereby achieving data-driven application optimization. Specifically, in this step, the Data Accumulation Department will use the accumulated massive log data to perform an offline batch learning process. The Data Accumulation Department will construct a supervised learning dataset with system state vectors from past fault occurrences as positive examples and system state vectors from normal occurrences as negative examples. The Data Accumulation Department will use this dataset to train a classification model based on a gradient boosting decision tree algorithm. The Data Accumulation Department will use cross-validation to optimize the model's hyperparameters to prevent overfitting. This data accumulation unit receives input data, such as multidimensional vectors containing statistics on CPU usage, memory usage, and network traffic over the past week. As output data, it outputs scores representing the probability of each subsystem failing within the next 24 hours. The trained model is deployed to the production inference pipeline. It uses online learning techniques to continuously fine-tune the model's weights using real-time incoming data. It uses stochastic gradient descent to update parameters in a way that minimizes prediction error. Past optimization results are stored in the experience replay buffer of the reinforcement learning agent. By allowing the agent to repeatedly learn from past successes and failures, the accuracy of the policy function is improved. Natural language processing techniques are applied to the input history to extract co-occurrence networks of frequently occurring words. From the extracted networks, the unit automatically identifies potential operator needs or system application challenges. Based on these insights, it generates new automation scripts or monitoring rules. This data accumulation department achieves the technical effect of system autonomous evolution and increased efficiency over time through this data-driven continuous improvement cycle.

[0096] (specific use case) For example, in the event of a large-scale earthquake in a city, the system of this invention first automatically collects earthquake occurrence information and evacuation instruction data from the disaster warning system. Simultaneously, it obtains signal control APIs, emergency vehicle priority control information, and traffic congestion information from the traffic control system, and application notification APIs, resident attribute information, and emergency report release history from the resident-oriented information service. Based on this information, the linkage adjustment unit generates an emergency signal control linkage path between the disaster warning system and the traffic control system, automatically ensuring evacuation routes and prioritizing emergency vehicle passage. Furthermore, it generates an instant evacuation instruction release linkage path between the disaster warning system and the resident-oriented information service, releasing evacuation information to residents' smartphones or digital signage in real time. Additionally, it obtains power outage information or priority power supply information for critical infrastructure from the power supply system, and links with the medical institution network to generate a path that automatically controls power supply to critical medical institutions during power outages. It also obtains secondary disaster risks caused by the earthquake (e.g., gas leak or fire detection data) from the environmental sensor network, and, as needed, links with the disaster warning system or the resident-oriented information service to issue additional alarms or evacuation instructions. The protocol conversion department automatically enables data linkage between different systems by converting the evacuation instruction API of the disaster warning system into the signal control API of the traffic control system, and further into the application notification API for information services to residents. The QoS control department ensures the accuracy and timeliness of information transmission by maximizing communication bandwidth, minimizing latency, and automatically applying redundant configuration paths for disaster-related linkage paths. The field terminal linkage department links with on-site traffic signal control terminals, evacuation management terminals, and power monitoring terminals of medical institutions, receiving real-time responses or prompts from the field. The data accumulation department accumulates the entire communication history, fault occurrence history, resident evacuation action data, and prompt input history from the field in a time sequence during the earthquake. The AI ​​control department learns from this data to prepare optimal linkage paths or bandwidth allocations for future disasters. The user interface department allows administrators to clearly view the linkage status or QoS settings of the entire city on the dashboard and input additional prompts to the AI ​​control department as needed. Specifically, in this large-scale earthquake scenario, the system runs a composite event processing engine based on an event-driven architecture. The system analyzes the initial micro-seismic data sent from seismographs in milliseconds. This system receives input data, such as time-series waveform tensors from a 3-axis accelerometer. As output data, it outputs structured data containing the predicted epicenter, magnitude, and arrival time in each region. Using this output data as a trigger, the system executes a pre-compiled emergency response workflow in memory in real time. The system multicasts control commands to the traffic control system, switching all traffic lights to red. Simultaneously, the system tracks the GPS coordinates of emergency vehicles in real time.This system employs a dynamic routing algorithm that switches traffic lights to green only in the direction of emergency vehicles. Within the power supply system, it analyzes electricity consumption data collected from smart meters for each household. The system designates the outage area as polygonal data. It then uses spatial calculations to compare the designated polygonal data with the location coordinates of medical institutions. The system identifies critical medical institutions within the outage area and automatically generates power transfer routes from nearby distributed power sources or batteries using a graph exploration algorithm. In its protocol conversion section, the system converts disaster warning system messages into binary control data packets that can be interpreted by the traffic control system at line speed. In its QoS control section, the system assigns emergency communication tags to data packets and routes them to the absolute priority queue of the network switch. This system completes this series of processes within seconds of an earthquake without human intervention. This extreme real-time processing directly minimizes the spread of disaster and maximizes the efficiency of lifesaving efforts. As a post-disaster processing measure, the system utilizes clustering algorithms to analyze accumulated evacuation tracking data. This system identifies bottlenecks in actual evacuation routes and utilizes them for automatic updates to the next disaster response plan.

[0097] Specific examples of prompts that generative AI can read include: "Prioritizing the linkage between disaster warning systems and traffic control systems during a large-scale earthquake to ensure the immediate dissemination of evacuation information to residents' information services"; "Strengthening the linkage between medical institution networks and power supply systems during power outages to provide real-time notifications of power supply status to critical medical institutions"; "Strengthening the linkage between disaster warning systems and residents' information services when environmental sensor networks detect a fire to issue immediate alerts to residents in the affected area"; "Strengthening the linkage between residents' information services and administrative procedures systems when establishing shelters to provide residents with shelter information and special procedure guidance"; and "Strengthening the linkage between traffic control systems and logistics network systems during traffic congestion to automatically reconfigure emergency delivery routes." Specifically, to process these prompts, this system utilizes a finely tuned, domain-specific large-scale language model. The system uses a byte-pair encoding algorithm to segment the input prompts into sub-word tags. The system embeds the segmented tag sequence into, for example, a continuous vector space of 4096 dimensions. In the Transformer's encoder layer, the system uses a multi-head attention mechanism to extract the contextual dependencies between tags. This system receives input data, such as text stating, "When traffic congestion occurs, strengthen the linkage between the traffic control system and the logistics network system, and automatically reconfigure emergency delivery routes." Output data includes a structured abstract syntax tree containing the URLs of the API endpoints to be executed, the HTTP methods, and the payloads. The system traverses this abstract syntax tree to generate specific network control commands corresponding to each node. The system compiles the generated command set into an execution plan represented as a directed acyclic graph. The system executes the tasks in the execution plan in parallel while satisfying dependencies. During the prompt parsing process, the system applies reinforcement learning techniques based on human feedback to continuously improve the model. When an operator corrects the generated control commands, the system feeds the correction back to the model as a reward signal. Through this feedback loop, the system enables the model to learn city-specific operational rules or tacit knowledge. By providing this highly abstract interface based on natural language, this system achieves the technical effect of eliminating human error and dramatically improving operational efficiency in the complex system management of smart cities. This system, through its unconventional processing method, can respond flexibly and rapidly to unknown complex disasters that traditional, fixed, rule-based systems cannot handle.

[0098] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0099] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0100] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0101] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0102] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0103] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0104] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0105] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0106] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0107] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0108] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0109] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0110] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0111] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0112] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0113] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0114] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0115] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0116] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0117] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0118] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0119] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0120] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0121] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0122] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0123] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0124] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0125] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0126] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0127] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0128] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0129] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0130] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0131] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0132] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0133] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0134] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0135] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0136] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0137] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0138] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0139] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0140] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0141] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0142] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0143] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0144] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0145] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0146] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0147] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0148] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0149] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0150] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0151] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0152] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0153] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0154] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0155] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0156] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0157] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The system determines the user's emotions. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0158] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0159] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0160] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0161] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0162] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0163] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, sentiment values ​​in nearby configurations are similar to each other. Figure 10 Examples show that multiple emotions such as "peace of mind", "stability", and "reassurance" have similar emotional values.

[0164] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0165] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0166] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0167] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0168] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0169] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0170] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0171] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0172] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0173] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0174] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0175] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0176] In addition, the following notes further explain the above explanation.

[0177] (Note 1) A system for comprehensively linking and optimizing multiple independent systems constituting urban infrastructure and resident services is characterized by comprising: an information collection unit that automatically collects interface information and service quality requirements of each system; a linkage adjustment unit that automatically generates linkage paths between systems based on the collected interface information and service quality requirements, and determines the optimal service quality setting for each linkage path; and an AI control unit that uses generative AI to flexibly and promptly reconfigure linkage paths or service quality settings based on prompts or additional requirements, automatically realizing linkage between systems. The AI ​​control unit comprehensively judges the physical or logical connection status of various systems constituting the city, such as urban operation status or emergency response during disasters, optimization of resident services, energy management, traffic control, environmental monitoring, disaster prevention alarms, administrative services, medical networks, education networks, and logistics networks, and, based on the service quality requirements of the usage scenario and the linkage requirements between systems, realizes through software unification and automatic adjustment operations that were previously performed manually or by adding hardware.

[0178] (Note 2) The system described in Appendix 1 is characterized in that the information collection unit monitors the interface information and service quality requirements of each system in real time; the linkage adjustment unit dynamically optimizes the linkage path and service quality settings between systems based on the monitoring results; and the AI ​​control unit can instantly change the linkage configuration or service quality settings based on natural language prompts input based on generative AI.

[0179] (Note 3) As described in Appendix 1, the system is characterized in that the AI ​​control unit learns the interface specifications and service quality requirements of each system; the linkage adjustment unit automatically determines the optimal linkage path and service quality settings between multiple systems based on the inference results of generative AI; and, in the event of a new system or a configuration change of an existing system, the AI ​​control unit also autonomously reconfigures the linkage path and service quality settings.

Claims

1. A system for comprehensively linking multiple independent network systems operated by operators or their affiliated companies, characterized in that, include: The Information Collection Department collects interface information and service quality requirements from various network systems. The linkage adjustment department automatically generates linkage paths between various network systems based on the collected interface information and service quality requirements, and determines the optimal service quality setting for each linkage path. as well as The AI ​​control unit uses generative AI to flexibly and promptly reconfigure linkage paths or quality of service settings based on prompts or additional requests, automatically enabling linkage between various network systems. The AI ​​control unit operates in a cloud environment, comprehensively judging the physical or logical connection status of each network system, the service quality requirements of the usage scenario, and the linkage requirements between operators and affiliated companies. It realizes the adjustment work that was previously done manually or by adding hardware through software unification and automatic.

2. The system as described in claim 1, characterized in that, The information collection unit monitors the interface information and service quality requirements of each network system in real time. The linkage adjustment unit dynamically optimizes the linkage paths and quality of service settings between network systems based on monitoring results. The AI ​​control unit can instantly change linkage configurations or service quality settings based on natural language prompts input using generative AI.

3. The system as described in claim 1, characterized in that, The AI ​​control unit learns the interface specifications and service quality requirements of each network system. The linkage adjustment unit automatically determines the optimal linkage path and service quality settings between multiple network systems based on the inference results of generative AI. Furthermore, in the event of a new network system or a configuration change in an existing system, the AI ​​control unit also autonomously reconfigures the linkage path and service quality settings.

Images