Internet of Things equipment intelligent control system and method based on multi-protocol fusion

Abstract

The invention discloses an Internet of Things equipment intelligent control system and method based on multi-protocol fusion, and relates to the technical field of Internet of Things equipment control, and the system comprises a multi-protocol adaptation module which can rapidly recognize and convert more than 10 protocols; the equipment management module is used for realizing million-level equipment authentication and full-life-cycle management; the decision module dynamically optimizes the strategy based on reinforcement learning; the data processing module supports real-time processing and analysis of high-concurrency data; the safety protection module adopts a zero-trust architecture to guarantee the safety of the system; in addition, the system is also provided with a self-adaptive load balancing and edge module, all the modules operate cooperatively, efficient access, intelligent management and control and safe operation of the equipment are realized, and the overall performance of the Internet of Things system is improved. Rapid access of multi-protocol equipment is achieved, the data processing capacity is remarkably improved, the resource utilization rate is effectively increased, energy consumption is reduced, the problems of interconnection, management, safety and the like of Internet of Things equipment are greatly solved, and system reliability and operation efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) device control technology, and in particular to an intelligent control system and method for IoT devices based on multi-protocol fusion. Background Technology

[0002] With the rapid development of IoT technology, a large number of devices are connecting to the network, and the communication protocols used by different devices vary significantly. Currently, common IoT protocols on the market include MQTT, CoAP, HTTP, and LwM2M, each with its own applicable scenarios and technical characteristics. For example, MQTT is suitable for message transmission in low-bandwidth, unstable network environments; CoAP is designed for constrained devices and low-power networks; and HTTP is widely used in traditional web applications. In addition, some manufacturers use custom protocols to meet specific needs. This coexistence of multiple protocols makes interoperability between IoT devices difficult. Traditional single-protocol control systems cannot directly connect to devices using different protocols, resulting in low device access efficiency and poor compatibility, severely restricting the large-scale development of IoT systems.

[0003] Existing IoT device control systems have significant shortcomings in data processing and device management. Faced with massive amounts of real-time data generated by devices, traditional systems lack efficient data processing capabilities, making it difficult to complete data cleaning, analysis, and storage in a short time. For example, when processing tens of thousands of data points per second, traditional systems may experience response times of several seconds or even data loss, failing to meet the demands of applications with high real-time requirements. Simultaneously, device management methods are outdated, device authentication processes are cumbersome and insecure, device status monitoring is untimely, and fault diagnosis and prediction capabilities are insufficient, hindering intelligent management throughout the entire device lifecycle. Once a device malfunctions, manual troubleshooting is often required, resulting in low maintenance efficiency and impacting the overall stability of the system.

[0004] Traditional IoT systems also face numerous challenges in the fields of intelligent decision-making and security. Most existing systems lack intelligent decision-making capabilities, failing to automatically adjust control strategies based on device operating status, network environment, and business needs, resulting in low equipment resource utilization and significant energy waste. Furthermore, IoT systems suffer from weak security defenses, making them vulnerable to network threats such as malicious access, data tampering, and DDoS attacks. Traditional security measures, such as firewalls and intrusion detection systems, are ill-equipped to handle new attack methods, and data encryption and access control mechanisms are inadequate, failing to guarantee the security of devices and data. This severely hinders the application and promotion of IoT technology in critical areas. Summary of the Invention

[0005] The present invention proposes an intelligent control system and method for IoT devices based on multi-protocol fusion to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent control system for IoT devices based on multi-protocol fusion, comprising the following modules: Multi-protocol adaptation module: It adopts a configurable protocol parsing engine, supports IoT protocols, and automatically identifies the communication protocol of the access device through a protocol feature library matching mechanism. The engine has a built-in protocol conversion algorithm to convert data of different protocols into standard JSON format, and a hot-swappable architecture for protocol adapters is designed for special protocols. Device Management Module: Constructs a digital twin model of devices to map the status of physical devices in real time, supports concurrent management of millions of devices, adopts a layered device authentication mechanism, including triple authentication of device ID, key and digital certificate, realizes device status caching through device shadow technology, designs a device lifecycle management mechanism, and automatically identifies the online, offline and fault status of devices and triggers corresponding processing procedures. Decision module: A decision engine based on deep reinforcement learning. The action space covers the device configuration parameter adjustment strategy. It adopts a priority experience replay mechanism, which increases the sampling probability of important experience by 2 times. Meta-learning technology is introduced to enable the model to adapt to new device types and environmental changes. Data processing module: Designed with a fusion architecture of real-time stream processing and batch processing, built-in time-series database, data compression ratio of up to 10:1, and an innovative anomaly detection formula. Where A is the comprehensive equipment anomaly index, and n is the number of monitoring indicators. Let be the probability of an anomaly for the i-th monitoring indicator. Using the weights, this formula is used to identify equipment anomalies, develop a data value density assessment model, and dynamically adjust storage strategies based on the degree of data's impact on decision-making. Security Protection Module: Constructs a security protection system, adopts a zero-trust architecture, requires dynamic authentication and authorization for all access, uses blockchain technology to ensure that device operation logs are tamper-proof, and deploys a lightweight honeypot system to automatically identify and trap malicious attacks.

[0007] Furthermore, it also includes: an adaptive load balancing module: which monitors the load status of each node in the system in real time and uses a dynamic weighted round-robin algorithm to allocate tasks. The weight calculation formula is: ,in Let i be the weight of node i. For CPU utilization, For memory usage, For network bandwidth utilization, α=0.5, β=0.3, and γ=0.2 are weighting coefficients. When the system load exceeds the threshold, a horizontal scaling mechanism is automatically triggered. An intelligent task scheduling algorithm is designed to dynamically adjust the scheduling strategy based on task priority and resource demand characteristics.

[0008] Furthermore, it also includes: Edge module: deploying edge computing nodes close to the device to realize local data preprocessing and edge decision-making. It adopts a federated learning framework, with edge nodes training models locally and uploading gradients to the central server after each round of training. It protects data privacy through differential privacy protection technology. Edge nodes support offline decision-making and develop a dynamic resource allocation algorithm for edge nodes to automatically adjust the allocation of computing resources according to task load and importance.

[0009] Furthermore, the multi-protocol adaptation module also includes: using a finite state machine model to parse protocol messages, supporting custom protocol extensions, identifying protocol types through a protocol fingerprint matching algorithm, designing a protocol compatibility detection subunit, automatically detecting protocol version differences and generating adaptation strategies, implementing protocol data format conversion based on a rule engine, using a bidirectional conversion mapping table, developing a protocol conversion performance optimization subunit, and caching commonly used conversion rules and results.

[0010] Furthermore, the decision-making module also includes: real-time collection of device status, network quality, and service indicator data to construct a multi-dimensional environmental state vector. ,in Let i represent the i-th environmental parameter. Design a data quality assessment subunit to evaluate the completeness, accuracy, and timeliness of the collected data. The data quality scoring model is as follows: ,in Let i be the data quality indicator. To correspond to the weights, a control strategy is generated based on a deep Q-network. The learning efficiency is improved by using an experience replay buffer. A multi-objective optimization mechanism is introduced to simultaneously optimize decision accuracy, response speed, and energy consumption. The Pareto optimal algorithm is used to generate the non-dominated solution set.

[0011] Furthermore, the data processing module also includes: employing the Apache Flink stream processing engine to support millisecond-level data processing latency; implementing time-series data aggregation through a sliding window algorithm; developing a stream data compression subunit; performing real-time compression of stream data based on the Lempel-Ziv algorithm; constructing a device fault prediction model based on the XGBoost algorithm; using an association rule mining algorithm to discover potential relationships between devices; designing a knowledge graph construction subunit; and transforming the mining results into a knowledge graph.

[0012] Furthermore, the security protection module also includes: a role-based access control model that supports fine-grained permission control; attribute-based encryption technology for dynamic data permission management; a dynamic permission adjustment subunit that adjusts access permissions in real time based on user behavior and environmental risks; a machine learning-based anomaly detection model that identifies abnormal behavior through autoencoder reconstruction errors; a threat intelligence sharing platform that supports real-time updates of threat features; an attack tracing subunit that constructs an attack path graph based on a graph database.

[0013] Furthermore, this includes the following steps: Multi-protocol adaptation steps: Scan the communication characteristics of the access device, compare them with the protocol feature library to identify the protocol type, call the corresponding parser to parse the data according to the protocol type, convert the parsed data into a standard format, perform feature extraction and cluster analysis on unknown protocols, and generate protocol templates for subsequent identification. Device management steps: Assign a unique identifier to the access device, establish a digital twin model of the device, periodically collect device status data to update the model, verify the device identity using a hierarchical authentication mechanism, conduct a health assessment of the device, and calculate the device health index based on a multi-indicator fusion algorithm. ,in For the i-th health indicator, For the corresponding weight, the maintenance process is triggered when HI < threshold; Decision-making steps: Construct an environment state vector, input it into a deep reinforcement learning model to generate a control policy, execute the policy and collect feedback, the decision optimization cycle is <1s, the probability of sampling important experience is increased by 2 times, and transfer learning technology is used to accelerate the learning process in new scenarios by leveraging existing model knowledge. Data processing steps: Real-time processing of data reported by the device, cleaning, aggregation and analysis, storage of data using a time-series database with a compression ratio of 10:1, value assessment of historical data, and automatic archiving or deletion of low-value data based on the assessment results; Security protection steps: Monitor system access behavior, detect abnormal activities and trigger response mechanisms, use blockchain technology to record device operation logs, regularly perform security vulnerability scans and risk assessments, generate risk assessment reports and provide remediation suggestions.

[0014] Furthermore, it also includes: adaptive load balancing steps: real-time monitoring of the load status of each node in the system, dynamic adjustment of task allocation weights based on the load status, automatic triggering of the horizontal scaling mechanism when the system load exceeds the threshold, the load status monitoring cycle is 100ms, the node addition time is <30s, prediction of system load change trends, and advance deployment of resources based on time series analysis, with a resource deployment lead time of up to 5 minutes.

[0015] Furthermore, it also includes: edge steps: deploying lightweight models on edge nodes to achieve local data preprocessing and decision-making, synchronizing model parameters between edge nodes and the central server periodically, using a federated learning framework to protect data privacy, and automatically switching edge nodes to offline decision-making mode and caching decision data when the network is disconnected, and automatically synchronizing cached data after the network is restored.

[0016] Compared with existing technologies, the beneficial effects of this invention are: In terms of device access, the system supports more than 10 mainstream protocols and custom protocols, reducing the average device access time by more than 70%, enabling plug-and-play functionality for devices with different protocols, significantly improving system compatibility and scalability, and effectively solving the problem of interconnection and interoperability of IoT devices.

[0017] In terms of data processing and equipment management, the system can process 100,000 data entries per second, with a data query response time in milliseconds and an equipment status update latency of less than 50ms. Through equipment digital twin and health assessment technologies, it can monitor equipment operating status in real time, with a fault prediction accuracy rate of over 90%, and improve equipment maintenance efficiency by nearly 60%, achieving intelligent management of the entire equipment lifecycle.

[0018] The decision-making module, based on deep reinforcement learning, can automatically generate optimal control strategies based on more than 20 dimensions, including device performance and network quality. This improves resource utilization by over 30%, reduces energy consumption by about a quarter, and significantly enhances device operating efficiency. In terms of security, the system constructs a multi-layered protection system, improving attack interception rates and reducing false alarm rates. It employs a zero-trust architecture and blockchain technology to ensure the security of data transmission and storage. Furthermore, adaptive load balancing and edge intelligence technologies reduce system response time in high-concurrency scenarios and improve the accuracy of offline decision-making at edge nodes, comprehensively enhancing system stability and reliability and promoting the deep application of IoT technology in more fields. Attached Figure Description

[0019] Figure 1 This is a schematic block diagram of the intelligent control system for IoT devices based on multi-protocol fusion proposed in this invention; Figure 2 This is a schematic block diagram of the intelligent control method for IoT devices based on multi-protocol fusion proposed in this invention; Figure 3 A combination chart comparing the access efficiency of devices using different protocols; Figure 4 Line chart comparing the accuracy of device fault prediction using different algorithms; Figure 5 A comparison chart showing the changes in edge node resource utilization under different strategies as a function of tasks. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0023] Reference Figures 1 to 5 A smart control system for IoT devices based on multi-protocol fusion includes the following modules: Multi-protocol adaptation module: The core of the multi-protocol adaptation module is a configurable protocol parsing engine, employing a heterogeneous computing architecture combining FPGA and ARM processors. The FPGA leverages its parallel processing capabilities for rapid protocol message parsing, while the ARM processor handles overall scheduling and management. In the protocol parsing unit, a protocol feature library containing over 1000 protocol features is pre-built, covering information such as protocol header fields, data formats, and port numbers. When an IoT device connects, the system extracts the first 100 bytes of the device's communication data and compares it with the protocol features in the feature library using the Hamming distance algorithm, with a similarity threshold set to 90%, thus achieving rapid protocol type identification. For identified protocols, the corresponding parser is invoked for data parsing. Taking the MQTT protocol as an example, the parser extracts the fixed header, variable header, and payload of the message according to the MQTT protocol specification, parsing out information such as the topic, message type, and QoS level. For the CoAP protocol, the parser identifies the request method, response code, options, and payload data based on its message format. After parsing, the data is transmitted to the protocol conversion unit.

[0024] The protocol conversion unit implements data format conversion based on a rule engine, which uses the Rete algorithm for rule matching to improve matching efficiency. For converting HTTP protocol data to standard JSON format, conversion rules are pre-defined, such as mapping URL parameters and request body data in HTTP requests to corresponding fields of JSON objects. A bidirectional conversion mapping table ensures information integrity during the data conversion process. Data is validated and calculated before conversion, and validated again after conversion. If the validation is inconsistent, a re-conversion mechanism is automatically triggered.

[0025] For specific protocols, a hot-swappable architecture for protocol adapters is designed. The adapters employ standardized interfaces, including data, control, and power interfaces. When a new protocol needs to be added, developers write a protocol parsing plugin. The plugin adheres to unified development specifications and includes protocol parsing functions, data conversion functions, etc. After the plugin is deployed to the system, the system detects the plugin through the control interface, with a loading time of less than 500ms, enabling online addition and support of new protocols.

[0026] Device Management Module: The device management module relies on a high-performance server cluster to build a distributed storage system for storing digital twin models of devices. When a device is connected, the system assigns it a unique UUID as a device identifier, which remains unchanged throughout the device's lifecycle. Simultaneously, based on the device's type, function, and other information, a corresponding digital twin model is created in the system. Taking a smart meter as an example, the digital twin model includes the meter's real-time power consumption, voltage, current, power factor, and other operating parameters, as well as basic information such as device model, production date, and supplier.

[0027] The layered device authentication mechanism comprises three layers of verification. The first layer involves the device sending its device ID upon connection. The system verifies the ID's format and validity; if the ID does not conform to preset format rules, access is rejected. The second layer involves the device sending a key generated using the AES-256 encryption algorithm. The system verifies the key's correctness by decrypting it. The third layer involves the device presenting a digital certificate issued by an authoritative CA (Certificate Authority). The system verifies the certificate's validity, expiration date, and the integrity of the certificate chain. Only devices that pass all three layers of authentication are allowed to access the system.

[0028] Device shadowing technology reduces frequent interactions with physical devices by caching device status data. The system creates a device shadow object for each device, which stores the device's latest status data. When a device's status changes, the device first updates its local shadow object and then sends the changed data to the system. The system then updates the device shadow on the server, with a status update latency of less than 50ms. In cases of network interruptions, the system can retrieve the device status from the device shadow, maintaining normal system operation.

[0029] The equipment lifecycle management mechanism automatically identifies the online, offline, and fault states of equipment by monitoring its communication status and operating parameters. When an equipment reports no data within 5 minutes, it is considered offline. When equipment operating parameters exceed normal thresholds, such as a smart meter current exceeding 120% of its rated current for 10 minutes, it is considered faulty. For different states, the system triggers corresponding processing procedures. For example, when the equipment is offline, an alarm is sent to maintenance personnel; when the equipment is faulty, a maintenance work order is automatically generated and pushed to the maintenance personnel's mobile device.

[0030] Decision Module: The intelligent decision module leverages the powerful computing capabilities of a GPU cluster to support the training and inference of deep reinforcement learning models. The environmental perception unit collects real-time data across 20 dimensions, including device performance, network quality, and business metrics, through sensors and data acquisition interfaces deployed on various system nodes, constructing a multi-dimensional environmental state vector. To ensure data quality, a data quality assessment sub-unit is designed. The data quality scoring model is as follows: ,in Let i be the data quality metric, which includes data completeness, accuracy, and timeliness. The corresponding weights are set according to the importance of each indicator, such as data integrity weight 0.4, accuracy weight 0.3, and timeliness weight 0.3. When the data quality score Q < 80, the data repair or re-collection process is triggered.

[0031] The decision optimization unit employs a Deep Q-Network (DQN) to generate the optimal control strategy. The experience replay buffer capacity is set to 100,000 entries. When the buffer reaches 80% capacity, random sampling begins for model training at a sampling frequency of 10 times per second, with a policy update frequency of 10Hz. Meta-learning technology is introduced to quickly adjust model parameters in the early stages of model training using meta-knowledge gained from training in multiple similar scenarios. This allows the model to rapidly adapt to new device types and environmental changes, reducing the adaptation time to one-third of traditional methods. Simultaneously, a multi-objective optimization mechanism is introduced to optimize decision accuracy, response speed, and energy consumption. A Pareto optimality algorithm is used to generate a non-dominated solution set, and the merits of each solution are displayed on the system management interface, allowing decision-makers to flexibly select control strategies based on actual needs.

[0032] Data Processing Module: The data processing module adopts a distributed storage and computing architecture, with the real-time processing unit built on the Apache Flink stream processing engine. At the data ingestion end, a Kafka message queue is used for data buffering to ensure stable data ingestion. Flink jobs are divided according to data type and processing logic, such as device status data processing jobs and alarm data processing jobs. A sliding window algorithm is used to aggregate time-series data to calculate the average performance metrics of the device over a certain period. A stream data compression subunit is developed, using the Lempel-Ziv algorithm for real-time compression of stream data. At the data sending end, data is divided into 1KB blocks, and each block is compressed with a compression ratio of 5:1. At the receiving end, the decompression latency is <1ms, ensuring real-time data processing. Simultaneously, a checkpoint mechanism is used for fault tolerance, generating a checkpoint every minute. In the event of a system failure, the job can be recovered from the most recent checkpoint, with a recovery time of <30 seconds.

[0033] The analysis and mining unit stores historical data using the Hadoop Distributed File System (HDFS) and manages the data using a Hive data warehouse. A device fault prediction model is built based on the XGBoost algorithm. During the model training phase, one year's worth of operational data from the devices is collected as the training set, including both normal and fault data. Model parameters are adjusted through cross-validation to achieve a prediction accuracy >90%. An association rule mining algorithm is used to discover potential relationships between devices, with a minimum support of 10% and a minimum confidence of 80%, supporting the discovery of association rules between devices. Furthermore, an innovative anomaly detection formula is introduced. Where A is the comprehensive equipment anomaly index, n is the number of monitoring indicators, and P i Let E be the probability of an anomaly for the i-th monitoring indicator. i Let be the weight of the i-th indicator. This formula is used to identify equipment anomalies, and a data value density assessment model is developed to dynamically evaluate the data value based on its impact on decision-making.

[0034] The design incorporates a knowledge graph construction subunit, utilizing the Neo4j graph database to build the knowledge graph. Mined information such as device relationships and failure modes is transformed into nodes and edges within the knowledge graph, supporting complex queries and reasoning. For example, when a device malfunctions, the knowledge graph can be used to query other related devices to predict the potential affected area, with an average query response time of less than 200ms. Furthermore, the knowledge graph is regularly updated and optimized to ensure the accuracy and timeliness of the knowledge.

[0035] Security Protection Module: This module deploys dedicated security devices on the hardware side, such as intrusion detection equipment and firewalls. The access control unit uses a Role-Based Access Control (RBAC) model, pre-defining 50+ roles, including system administrators, device maintenance personnel, and data analysts. Each role corresponds to a different set of permissions; for example, system administrators have management permissions for all devices and data deletion permissions, while device maintenance personnel only have permissions to view device status and handle faults. Attribute-Based Encryption (ABE) technology is used to achieve dynamic data permission management. Encryption keys are generated based on user attributes, such as department, position, years of service, and data sensitivity level. Only users meeting the attribute criteria can decrypt the data.

[0036] A dynamic permission adjustment subunit was developed to monitor user behavior and environmental risks in real time. User behavior monitoring includes operation frequency, operation type, and access time. Frequent access to sensitive data outside of working hours is considered abnormal behavior. Environmental risk assessment considers factors such as network attack trends and system vulnerabilities. When abnormal behavior or high-risk environments are detected, user access permissions are automatically adjusted. The threat detection unit uses a machine learning-based anomaly detection model trained with an autoencoder. Data such as network traffic and user operation logs during normal system operation are collected as a training set to train the autoencoder to learn normal patterns. During operation, real-time data is input into the autoencoder to calculate reconstruction errors. When the reconstruction error exceeds a preset threshold, it is judged as abnormal behavior. A threat intelligence sharing platform is built, collaborating with multiple well-known security vendors to obtain the latest threat characteristics in real time. When new threats are detected, the system's protection policies are automatically updated to ensure timely defense against attacks.

[0037] The attack attribution subunit is designed to construct an attack path graph based on a graph database (such as TigerGraph). When an attack is detected, relevant log information and network traffic data are collected, and the attack source is traced using graph analysis algorithms. Starting from the time the attack occurred, the propagation path and scope of the attack are analyzed step by step, with an attribution time of less than 10 seconds, providing strong support for the handling of security incidents.

[0038] This invention also includes: Adaptive Load Balancing Module: Deployed on the core network nodes of the system, the adaptive load balancing module monitors the load status of each node in real time via SNMP (Simple Network Management Protocol) and Prometheus monitoring system, with a monitoring cycle of 100ms. Monitoring metrics include CPU utilization, memory utilization, network bandwidth utilization, and task queue length. A dynamic weighted round-robin algorithm is used to allocate tasks, with the weight calculation formula as follows: ,in Let i be the weight of node i. For CPU utilization, For memory usage, For network bandwidth utilization, α=0.5, β=0.3, and γ=0.2 are weighting coefficients. Tasks are allocated to nodes using a weighted round-robin method based on their weights to ensure reasonable task allocation.

[0039] When the system load exceeds a threshold (80%), a horizontal scaling mechanism is automatically triggered. New node instances are quickly created on the cloud platform using container orchestration tools (such as Kubernetes), with new node addition time less than 30 seconds. An intelligent task scheduling algorithm is designed that, in addition to considering node load, dynamically adjusts the scheduling strategy based on task priority and resource requirements. For urgent tasks, priority is given to assigning them to nodes with lower load and matching resources; for CPU-intensive tasks, they are assigned to nodes with higher CPU performance, shortening the response time of critical tasks. Simultaneously, time series analysis algorithms (such as the ARIMA model) are used to predict system load trends over a 1-hour period. Based on the prediction results, resources are deployed in advance; for example, a certain number of node instances are pre-created before business peaks, providing a 5-minute lead time for resource deployment, effectively reducing the probability of system overload.

[0040] This invention also includes: Edge Module: The edge intelligence module deploys edge computing nodes close to IoT devices. These nodes utilize miniaturized, low-power hardware devices, such as Raspberry Pi 4B or NVIDIA Jetson Nano. A lightweight operating system and edge computing framework are installed on the edge nodes. A federated learning framework is used to enable collaborative learning between the edge nodes and the central server.

[0041] During the model training phase, edge nodes collect local device data for model training. After each training round, the model gradients are uploaded to the central server. To protect data privacy, differential privacy protection technology (ε=0.5, δ=10) is employed. -6 To ensure data privacy, noise is added before uploading gradients. The central server collects gradients from each edge node, aggregates and updates the global model, and then distributes the updated model to the edge nodes.

[0042] Edge nodes support offline decision-making, pre-storing trained models and related data. When the network is lost, the edge nodes automatically switch to offline decision-making mode and make decisions based on local data. Simultaneously, the edge nodes cache decision data; the cache size is dynamically adjusted based on device performance and data volume, typically set to 1GB. Once the network is restored, the cached data is automatically synchronized to the central server, with data verification and retransmission mechanisms ensuring a data synchronization success rate >99%.

[0043] Develop a dynamic resource allocation algorithm for edge nodes to monitor the real-time usage of CPU, memory, and storage resources. Automatically adjust computing resource allocation based on task load and importance, employing a genetic algorithm for optimized resource allocation. Allocate more computing resources to tasks with high real-time requirements; for non-real-time tasks, appropriately reduce resource allocation when resources are scarce, thereby improving resource utilization.

[0044] In this invention, the multi-protocol adaptation module further includes: parsing protocol messages using a finite state machine (FSM) model. It accurately extracts key information by transitioning between different states based on message characteristics, and also supports custom protocol extensions, allowing developers to define new states and rules as needed. A protocol fingerprint matching algorithm identifies protocol types, quickly scans messages to extract features, and compares them with a fingerprint database, with a matching time of less than 5ms. The protocol compatibility detection subunit automatically detects protocol version differences, analyzes the differences in syntax, semantics, and functionality between versions, and generates adaptation strategies.

[0045] This invention includes the following steps: Multi-protocol adaptation steps: When an IoT device connects to the system, a communication feature scan is first performed at the device's access port. The system extracts the first 100 bytes of data sent by the device, including the packet header information and protocol identifier fields, and compares them with a pre-built protocol feature library. The protocol feature library uses a hash index structure to speed up the comparison. During the comparison process, the Hamming distance between the extracted data and the protocol features in the library is calculated. If the distance is less than a set threshold (90%), it is identified as the corresponding protocol type.

[0046] For each identified protocol, the corresponding parser is invoked to parse the data. Taking the CoAP protocol as an example, the parser, according to the CoAP protocol specification, parses the data packet into request method, response code, options, and payload data. After parsing, the data is transmitted to the protocol conversion unit.

[0047] The protocol conversion unit performs data format conversion based on a rule engine. The rule engine is pre-configured with over 100 predefined conversion rules. For converting HTTP protocol to standard JSON format, the rules define how URL parameters in the HTTP request are converted into key-value pairs of JSON objects, and the request body data is mapped according to the corresponding format. A bidirectional conversion mapping table ensures the accuracy of data conversion. CRC checks are performed on the data before conversion and again after conversion. If the checks are inconsistent, a re-conversion mechanism is automatically triggered, ensuring that the conversion latency is less than 10ms.

[0048] For unknown protocols, the system initiates a protocol analysis process. First, traffic capture of the unknown protocol data is performed for 5 minutes to obtain sufficient sample data. Then, machine learning clustering algorithms (such as DBSCAN) are used to extract features and perform cluster analysis on the data, classifying it into different categories based on data similarity. For each category, the data format, field meanings, and other characteristics are analyzed to generate a protocol template. The template includes information such as the protocol header format, data field definitions, and encoding methods. The template generation time is less than 30 seconds, and it is used for subsequent identification and processing of this type of protocol.

[0049] Device Management Steps: When a device is connected to the system, the system uses a random number generation algorithm to assign a globally unique UUID as the device identifier, ensuring the uniqueness and independence of the device's identity from the source. Simultaneously, based on the device type (e.g., sensor, actuator, smart terminal), and in conjunction with manufacturer information, a precise digital twin model of the device is constructed within the system. Taking a smart camera as an example, its digital twin model not only includes operational parameters such as real-time video stream address, resolution, frame rate, and pan-tilt control parameters, but also basic information such as device model, production date, and installation location, comprehensively simulating the device's characteristics.

[0050] For device access authentication, a layered authentication mechanism is adopted to ensure security. The first layer strictly verifies the device ID in a 16-digit alphanumeric format; if it does not conform, the connection is rejected. The second layer involves the device sending a key generated by the AES-256 encryption algorithm, which the system decrypts and verifies based on pre-stored information. The third layer verifies the validity of the digital certificate, its expiration time, and the integrity of the certificate chain, and verifies it online with the CA server. Only devices that pass the three-layer authentication can access the system. Every minute, the system actively polls (using Modbus or SNMP protocols) or the device actively reports its status data, updates the digital twin model in real time, strictly verifies the validity of the data, and re-collects abnormal data to ensure data accuracy and reliability.

[0051] Equipment health assessment is based on a multi-indicator fusion algorithm to calculate the equipment health index. When the equipment health index HI is less than 60, the equipment is deemed to be in poor health, triggering the maintenance process. The system automatically generates a maintenance work order, which includes basic equipment information, a description of the fault, maintenance suggestions, etc., and pushes it to the maintenance personnel's mobile device.

[0052] Decision-making steps: An environmental state vector is constructed, integrating multi-dimensional information such as equipment operating parameters, network load, and task priorities into a numerical vector, which serves as input to the deep reinforcement learning model. The model generates optimal control strategies, such as resource scheduling schemes or task allocation decisions, through multi-layer neural network computation. After strategy execution, the system collects feedback data such as reward signals and state changes in real time for model iterative optimization. To improve decision-making efficiency, the decision optimization cycle is set to less than 1 second to ensure real-time system response; simultaneously, a priority experience replay mechanism is adopted, increasing the sampling probability of important experiences by 2 times, accelerating the model's learning of key scenarios. Furthermore, transfer learning technology is introduced, using model parameters trained in similar scenarios such as industrial equipment monitoring as the initialization basis. In new business scenarios, only a few network layers need to be fine-tuned, improving learning efficiency by 40% and significantly shortening the deployment cycle for new scenarios.

[0053] Data processing steps: First, the data stream reported by the device in real time is processed in the entire chain. Invalid values ​​are filtered and the format is standardized through the rule engine to complete the data cleaning. Then, the time series data is aggregated based on the sliding window algorithm to calculate the mean, peak and other statistical quantities of the device operating parameters, so as to provide structured data for subsequent analysis.

[0054] The storage component employs a time-series database, utilizing efficient compression algorithms to achieve a data compression ratio of 10:1. Simultaneously, index optimization enables millisecond-level query responses, meeting the high-frequency data retrieval needs of real-time monitoring scenarios. For historical data, a value assessment model is constructed: a value score is generated based on dimensions such as data access frequency and business relevance. Data with scores below a threshold is automatically archived to low-cost storage media, while valueless data is deleted, ultimately reducing overall storage costs and achieving intelligent management of storage resources.

[0055] Security Protection Steps: A real-time behavior monitoring module is deployed to continuously collect operation logs such as user logins, data access, and command calls. Anomaly detection algorithms are used to identify abnormal patterns, such as high-frequency access from unauthorized IPs and batch export of sensitive data. Once a threat is detected, an alarm is immediately triggered, and the firewall is activated to block suspicious connections. At the log storage level, blockchain technology is introduced to build a distributed operation log system. Critical operations such as device startup / shutdown and configuration changes are packaged into blocks in chronological order. A consensus algorithm ensures data immutability, and the block generation interval is controlled within 2 seconds, achieving real-time and reliable operation traceability. Simultaneously, an automated security inspection module regularly performs deep vulnerability scans on system components, assesses risk levels using the CVSS scoring system, and generates a visual report containing vulnerability details, impact scope, and patch solutions. Leveraging an intelligently matched vulnerability signature database, the high-risk vulnerability detection rate exceeds 95%, providing precise guidance for security hardening.

[0056] This invention also includes: Adaptive load balancing steps: Real-time monitoring of each node's CPU, memory, and network traffic load status at 100ms intervals; dynamic adjustment of task allocation weights based on load, decreasing the weight of high-load nodes and increasing the weight of low-load nodes to achieve balanced task distribution. When the overall system load exceeds a preset threshold, a horizontal scaling mechanism is automatically triggered, using containerization technology to quickly call images from the resource pool to start new nodes. Node joining time is controlled within 30 seconds, rapidly improving cluster processing capacity.

[0057] Meanwhile, historical load data is modeled based on time series analysis algorithms to predict system load change trends and deploy resources 5 minutes in advance, such as pre-allocating computing resources and adjusting network bandwidth. Through real-time monitoring, dynamic adjustment of task weights, automatic expansion, and advance resource deployment, the probability of system overload is effectively reduced, ensuring that the system can maintain stable and efficient operation under different load conditions, thereby improving overall performance and resource utilization.

[0058] This invention also includes: Edge Steps: A compressed, lightweight model is deployed on edge nodes, utilizing local computing power to perform real-time data filtering, feature extraction, and other preprocessing, and then making real-time decisions based on the local model. Edge nodes and the central server employ a federated learning framework to build a collaborative training mechanism: edge nodes periodically upload encrypted model gradient parameters, and the central server aggregates parameters from multiple sources before issuing an updated version. The entire synchronization process protects the privacy of the original data through differential privacy technology, and the model parameter synchronization latency is controlled within 500ms. For unstable network scenarios, an adaptive switching mechanism is designed: when a network disconnection is detected, the edge node automatically activates offline decision-making mode, continuously processing data based on the latest locally cached model, and temporarily storing the decision results and original data in a local database. After network recovery, the cached data is automatically synchronized through a breakpoint resume mechanism, and a consistent hashing algorithm ensures that data is not duplicated or lost.

[0059] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent control system for IoT devices based on multi-protocol fusion, characterized in that, Includes the following modules: Multi-protocol adaptation module: It adopts a configurable protocol parsing engine, supports IoT protocols, and automatically identifies the communication protocol of the access device through a protocol feature library matching mechanism. The engine has a built-in protocol conversion algorithm to convert data of different protocols into standard JSON format, and a hot-swappable architecture for protocol adapters is designed for special protocols. Device Management Module: Constructs a digital twin model of devices to map the status of physical devices in real time, supports concurrent management of millions of devices, adopts a layered device authentication mechanism, including triple authentication of device ID, key and digital certificate, realizes device status caching through device shadow technology, designs a device lifecycle management mechanism, and automatically identifies the online, offline and fault status of devices and triggers corresponding processing procedures. Decision module: A decision engine based on deep reinforcement learning. The action space covers the device configuration parameter adjustment strategy. It adopts a priority experience replay mechanism, which increases the sampling probability of important experience by 2 times. Meta-learning technology is introduced to enable the model to adapt to new device types and environmental changes. Data processing module: Designed with a fusion architecture of real-time stream processing and batch processing, built-in time-series database, data compression ratio of up to 10:1, and an innovative anomaly detection formula. Where A is the comprehensive equipment anomaly index, and n is the number of monitoring indicators. Let be the probability of an anomaly for the i-th monitoring indicator. Using the weights, this formula is used to identify equipment anomalies, develop a data value density assessment model, and dynamically adjust storage strategies based on the degree of data's impact on decision-making. Security Protection Module: Constructs a security protection system, adopts a zero-trust architecture, requires dynamic authentication and authorization for all access, uses blockchain technology to ensure that device operation logs are tamper-proof, and deploys a lightweight honeypot system to automatically identify and trap malicious attacks.

2. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, Also includes: Adaptive load balancing module: Monitors the load status of each node in the system in real time and uses a dynamic weighted round-robin algorithm to allocate tasks. The weight calculation formula is as follows: ,in Let i be the weight of node i. For CPU utilization, For memory usage, For network bandwidth utilization, α=0.5, β=0.3, and γ=0.2 are weighting coefficients. When the system load exceeds the threshold, a horizontal scaling mechanism is automatically triggered. An intelligent task scheduling algorithm is designed to dynamically adjust the scheduling strategy based on task priority and resource demand characteristics.

3. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, Also includes: Edge module: Edge computing nodes are deployed close to the device to realize local data preprocessing and edge decision-making. A federated learning framework is adopted, and the model is trained locally on the edge nodes. After each round of training, the gradient is uploaded to the central server. Differential privacy protection technology is used to protect data privacy. Edge nodes support offline decision-making. An edge node resource dynamic allocation algorithm is developed to automatically adjust the allocation of computing resources according to task load and importance.

4. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, The multi-protocol adaptation module also includes: using a finite state machine model to parse protocol messages, supporting custom protocol extensions, identifying protocol types through a protocol fingerprint matching algorithm, designing a protocol compatibility detection subunit, automatically detecting protocol version differences and generating adaptation strategies, implementing protocol data format conversion based on a rule engine, using a bidirectional conversion mapping table, developing a protocol conversion performance optimization subunit, and caching commonly used conversion rules and results.

5. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, The decision-making module also includes: real-time collection of device status, network quality, and service indicator data to construct a multi-dimensional environmental state vector. ,in Let i represent the i-th environmental parameter. Design a data quality assessment subunit to evaluate the completeness, accuracy, and timeliness of the collected data. The data quality scoring model is as follows: ,in Let i be the data quality indicator. To correspond to the weights, a control strategy is generated based on a deep Q-network. The learning efficiency is improved by using an experience replay buffer. A multi-objective optimization mechanism is introduced to simultaneously optimize decision accuracy, response speed, and energy consumption. The Pareto optimal algorithm is used to generate the non-dominated solution set.

6. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, The data processing module also includes: using the Apache Flink stream processing engine to support millisecond-level data processing latency, implementing time-series data aggregation through the sliding window algorithm, developing a stream data compression subunit, performing real-time compression of stream data based on the Lempel-Ziv algorithm, constructing a device fault prediction model based on the XGBoost algorithm, using an association rule mining algorithm to discover potential relationships between devices, designing a knowledge graph construction subunit, and transforming the mining results into a knowledge graph.

7. The intelligent control system for IoT devices based on multi-protocol fusion according to claim 1, characterized in that, The security protection module also includes: a role-based access control model that supports fine-grained permission control; attribute-based encryption technology for dynamic data permission management; a dynamic permission adjustment subunit that adjusts access permissions in real time based on user behavior and environmental risks; a machine learning-based anomaly detection model that identifies abnormal behavior through autoencoder reconstruction errors; a threat intelligence sharing platform that supports real-time updates of threat features; an attack tracing subunit that constructs an attack path graph based on a graph database.

8. A method for an intelligent control system for IoT devices based on multi-protocol fusion as described in any one of claims 1-7, characterized in that, Includes the following steps: Multi-protocol adaptation steps: Scan the communication characteristics of the access device, compare them with the protocol feature library to identify the protocol type, call the corresponding parser to parse the data according to the protocol type, convert the parsed data into a standard format, perform feature extraction and cluster analysis on unknown protocols, and generate protocol templates for subsequent identification. Device management steps: Assign a unique identifier to the access device, establish a digital twin model of the device, periodically collect device status data to update the model, verify the device identity using a hierarchical authentication mechanism, conduct a health assessment of the device, and calculate the device health index based on a multi-indicator fusion algorithm. ,in For the i-th health indicator, For the corresponding weight, the maintenance process is triggered when HI < threshold; Decision-making steps: Construct an environment state vector, input it into a deep reinforcement learning model to generate a control policy, execute the policy and collect feedback, the decision optimization cycle is <1s, the probability of sampling important experience is increased by 2 times, and transfer learning technology is used to accelerate the learning process in new scenarios by leveraging existing model knowledge. Data processing steps: Real-time processing of data reported by the device, cleaning, aggregation and analysis, storage of data using a time-series database with a compression ratio of 10:1, value assessment of historical data, and automatic archiving or deletion of low-value data based on the assessment results; Security protection steps: Monitor system access behavior, detect abnormal activities and trigger response mechanisms, use blockchain technology to record device operation logs, regularly perform security vulnerability scans and risk assessments, generate risk assessment reports and provide remediation suggestions.

9. The method for an intelligent control system for IoT devices based on multi-protocol fusion according to claim 8, characterized in that, Also includes: Adaptive load balancing steps: Real-time monitoring of the load status of each node in the system, dynamic adjustment of task allocation weights based on the load status, automatic triggering of the horizontal scaling mechanism when the system load exceeds the threshold, load status monitoring cycle of 100ms, node addition time <30s, prediction of system load change trends, and advance deployment of resources based on time series analysis, with a resource deployment lead time of up to 5 minutes.

10. The method for an intelligent control system for IoT devices based on multi-protocol fusion according to claim 8, characterized in that, It also includes: edge steps: deploying lightweight models on edge nodes to achieve local data preprocessing and decision-making, synchronizing model parameters between edge nodes and the central server periodically, using a federated learning framework to protect data privacy, and automatically switching edge nodes to offline decision-making mode and caching decision data when the network is disconnected, and automatically synchronizing cached data after the network is restored.

Images