Urban component management iot sensing terminal and edge computing gateway coordination system

Abstract

The application provides a city component management Internet of Things sensing terminal and edge computing gateway cooperative system, relates to the technical field of smart city Internet of Things and edge computing, and the method comprises the following steps: deploying an Internet of Things sensing terminal and an edge computing gateway with local intelligence, and constructing a dynamic federated cooperative network through a wireless self-organizing network; the terminal performs analysis and decision-making locally, and can perform peer task cooperation with the gateway and other terminals based on a negotiation mechanism; the gateway maintains a global resource view through a federated cooperative scheduling engine, and generates an optimal cooperative scheme by using an adaptive deep reinforcement learning algorithm, so that idle computing, storage and communication resource pools in the network are collectively used to execute complex tasks; the system supports cross-node privacy protection federated learning to continuously optimize the model. The application realizes ultra-low delay response of city management tasks and a significant reduction in network load, improves system resource utilization efficiency, reliability and cross-modal cooperation capability, and effectively guarantees data privacy and security.

Description

Technical Field

[0001] This invention relates to the field of smart city IoT and edge computing technology, and in particular to a collaborative system of IoT sensing terminals and edge computing gateways for urban component management. Background Technology

[0002] In the construction of smart cities, the intelligent management of urban components relies on the deployment of large-scale IoT sensing terminals. Currently, most IoT sensing terminals only possess basic data collection and forwarding functions, lacking local computing and independent decision-making capabilities. All sensing data must be uploaded to the cloud or edge gateway for centralized processing. This results in a large amount of raw, redundant sensing data continuously consuming network bandwidth, causing excessive load on the urban communication network. Furthermore, the remote transmission and centralized processing of data introduce significant time delays, hindering the rapid identification and response to abnormal states of urban components.

[0003] Existing sensor terminals and edge computing gateways generally adopt a unidirectional centralized control architecture. Terminals can only passively receive instructions from the gateway and cannot perform peer-to-peer task collaboration with the gateway and other terminals, nor can they achieve resource sharing and allocation. Idle computing, storage, and communication resources of each node in the network cannot be effectively integrated and utilized, resulting in serious resource waste. Moreover, a single edge gateway can easily become the performance bottleneck of the entire system. Once the gateway fails, the data analysis and processing tasks of surrounding terminals cannot be executed normally, resulting in poor system reliability and scalability.

[0004] In the multi-node collaborative process of urban component management, existing technologies lack effective data privacy protection mechanisms. To achieve multi-node model collaborative training and optimization, the local perception data of each node needs to be uploaded to the central node for unified processing. This process is prone to leakage of sensitive data, which cannot meet the privacy protection and compliance requirements of smart city data management and limits the improvement of multi-node collaborative intelligence.

[0005] Meanwhile, existing systems mostly employ fixed, preset strategies for task scheduling, allocating resources solely based on static network configurations. They cannot adaptively adjust to dynamic changes in network topology, node resource status, and wireless channel quality. When faced with complex collaborative tasks involving multiple modes and devices, such as road flooding monitoring and manhole cover anomaly handling, they cannot achieve efficient task decomposition and optimal resource allocation. This results in low task execution efficiency, uneven resource utilization, and an inability to meet the refined and intelligent management needs of urban components. Summary of the Invention

[0006] To address the technical problems in existing technologies, such as the lack of local computing capabilities in IoT sensor terminals leading to high data transmission bandwidth consumption and large response latency, the inability of sensor terminals and gateways to coordinate equitably and low resource utilization in a central control architecture, the lack of privacy protection mechanisms in multi-node collaboration which easily leads to data leakage, and the fixed task scheduling strategy which cannot adapt to dynamic network changes and results in low execution efficiency for complex tasks, this invention provides a collaborative system between IoT sensor terminals and edge computing gateways for urban component management.

[0007] The technical solution provided by this invention is as follows: The present invention provides a collaborative system for urban component management of IoT sensing terminals and edge computing gateways, comprising: Multiple IoT sensor terminals and at least one edge computing gateway; the IoT sensor terminals and the edge computing gateway are connected through a wireless ad hoc network to form a dynamic federated collaborative network; wherein, the IoT sensor terminals have local computing and decision-making capabilities, and can perform peer-to-peer task collaboration with the edge computing gateway and other IoT sensor terminals in the network based on a negotiation mechanism to jointly execute urban component management tasks. When any node in the dynamic federated collaborative network faces a complex task that it cannot complete independently, the node, as the task initiator, sends a collaborative request containing the task requirements to other nodes through the network. Each node that receives a collaboration request decides independently whether to participate and how to participate, based on its own resource status and task strategy. The federated coordination scheduling engine in the edge computing gateway maintains a global resource view of the dynamic federated coordination network and coordinates relevant nodes based on the global resource view. Nodes that receive coordination requests independently decide whether to participate and how to participate based on their own resource status and task strategy. The relevant nodes negotiate and communicate to jointly determine the task decomposition, allocation and execution process, and complete the complex task in a distributed collaborative manner.

[0008] The IoT sensing terminal includes: The basic sensing module is used to collect status data of the target city components, and the status data includes at least one of vibration, image, temperature and humidity. The local intelligent processing module is used to run a lightweight artificial intelligence model to perform preliminary analysis and anomaly detection on the state data. The collaborative communication and resource management module is used to broadcast its own status and available computing resources, discover neighboring collaborative nodes, and process collaborative task requests from other nodes.

[0009] The lightweight AI model refers to a deep neural network model processed using model compression techniques, including but not limited to model pruning, weight quantization, low-rank decomposition, or knowledge distillation. These techniques significantly reduce the model's demand for terminal computing and storage resources while maintaining recognition accuracy, enabling it to run efficiently on resource-constrained embedded devices.

[0010] Furthermore, the edge computing gateway includes: The gateway core processing module is used for protocol conversion, data aggregation, and communication with the upper-level cloud platform; The federated collaborative scheduling engine is used to maintain a global resource view of the dynamic federated collaborative network, respond to task collaboration requests, generate distributed collaborative execution schemes based on preset algorithms, and coordinate relevant nodes to reach an execution consensus.

[0011] Furthermore, the system also includes a dynamic resource pooling virtual layer, which is composed of the federated collaborative scheduling engine and the collaborative communication and resource management modules of each IoT sensor terminal. This virtual layer is used to abstract the idle computing, storage and communication resources of all nodes in the network into a unified logical resource pool for dynamic invocation by the distributed collaborative execution scheme.

[0012] Furthermore, the preset algorithm on which the federated collaborative scheduling engine generates the distributed collaborative execution scheme is a model-assisted adaptive deep reinforcement learning algorithm. The algorithm aims to minimize the total task execution delay and the total system energy consumption. Its state space includes network topology, node resource status and channel quality, and its action space is a set of feasible task offloading and resource allocation strategies.

[0013] The optimization objective is a multi-objective optimization problem. In a specific implementation of this invention, the multi-objective problem is transformed into a single-objective problem for solution using a linear weighted summation method. Specifically, weight coefficients are assigned to the total delay and total energy consumption, and these weights are dynamically adjusted based on the real-time operating status of the system. In other embodiments, the Pareto optimization method can also be used to find a set of non-dominated solutions that satisfy the constraints, allowing the scheduling engine to select based on the current strategy.

[0014] Furthermore, the model-assisted adaptive deep reinforcement learning algorithm integrates a prediction network to predict short-term trends in network state and node load, thereby assisting the federated collaborative scheduling engine in making forward-looking decisions when generating collaborative schemes.

[0015] Furthermore, the collaborative communication and resource management module and the federated collaborative scheduling engine communicate through a standardized task description language for task publishing, capability querying, and solution description; this standardized task description language defines the input data, processing logic, resource requirements, and output format required for the computation task.

[0016] Furthermore, the dynamic federated collaborative network supports a privacy-preserving federated learning mechanism across nodes; each participating node uses local data to train a shared lightweight artificial intelligence model, and uploads encrypted model parameter updates to the edge computing gateway for secure aggregation, forming a global model update before distributing it to each node.

[0017] The encrypted model parameter updates are implemented using homomorphic encryption. Each terminal encrypts its local model gradient using a homomorphic encryption public key, generates ciphertext, and uploads it to the gateway. The gateway performs weighted average and other aggregation operations directly on the ciphertext without decrypting it; the aggregation result remains ciphertext. The gateway then decrypts the data using the federated learning private key to obtain the global model update. This mechanism ensures that no third party (including the gateway) can obtain the terminal's original gradient information during model aggregation, thus achieving data usability without visibility. Alternatively, as another privacy protection method, the terminal can introduce differential privacy noise before uploading parameters, sacrificing a small amount of model accuracy for stronger individual data privacy protection.

[0018] Furthermore, the system is used to achieve collaborative monitoring and early warning of urban road waterlogging; wherein, The first type of IoT sensing terminal is a water level sensor, used to collect data on water depth. The second type of IoT sensing terminal is an image acquisition device used to collect visual data of waterlogged areas; The edge computing gateway or the third type of IoT sensing terminal with redundant computing power is dynamically scheduled to fuse the water depth data and visual data, and to perform water accumulation range identification and risk assessment.

[0019] Furthermore, the system is used to achieve collaborative monitoring and coordinated response to the status of urban manhole covers; wherein, When an IoT sensor terminal deployed on a manhole cover detects an abnormal status locally, it issues a collaborative verification task to the dynamic federated collaborative network. The edge computing gateway schedules nearby camera devices for visual verification, and links with the municipal operation and maintenance terminal to generate a work order for handling. At the same time, it publishes warning information through nearby public information publishing terminals.

[0020] When an IoT sensor terminal deployed on a manhole cover detects an anomaly locally, it immediately generates a "collaborative verification task" request according to the standardized task description language and broadcasts it to the dynamic federated collaborative network. This task request explicitly defines: the task type as "image acquisition and verification," the required input data as "the precise latitude and longitude coordinates of the target manhole cover," the required processing logic as "acquiring and returning a real-time image containing the manhole cover," and the expected output format as "JPEG format image." This standardized description enables any node in the network with image acquisition capabilities (such as a nearby camera) to accurately understand the task requirements and participate in the response.

[0021] The beneficial effects of the technical solution provided by this invention include at least the following: (1) In this invention, by giving the IoT sensing terminal local computing and independent decision-making capabilities, the terminal can complete the preliminary analysis of status data and anomaly detection locally, and only transmit key collaborative requests, analysis results and feature data, which greatly reduces the network transmission volume of the original data, effectively reduces network bandwidth load and data transmission latency, realizes ultra-low latency response to abnormal status of urban components, and fully utilizes the local computing capabilities of the terminal, changing the traditional centralized data processing mode.

[0022] (2) In this invention, by constructing a wireless self-organizing dynamic federated collaborative network between the terminal and the edge computing gateway, peer-to-peer task collaboration between nodes is realized. Then, through the federated collaborative scheduling engine and the dynamic resource pooling virtual layer, the idle resources of all nodes in the network are abstracted into a unified logical resource pool, realizing the global integration and dynamic scheduling of resources. This allows complex tasks to be decomposed into sub-tasks that can be executed in parallel by multiple nodes, improving the overall resource utilization of the system. At the same time, it avoids the performance bottleneck of a single gateway. When a node fails, other nodes can take over the task, significantly enhancing the reliability and scalability of the system.

[0023] (3) In this invention, by introducing a cross-node privacy-preserving federated learning mechanism, each node can use local data to complete the local training of the model. Only the encrypted model parameters are updated and uploaded to the gateway for secure aggregation. There is no need to share the original perception data. While realizing the continuous optimization of the shared model and improving the overall intelligence level of the system, it effectively protects the privacy of the data source and meets the data compliance requirements of smart cities. At the same time, the model-assisted adaptive deep reinforcement learning algorithm is used for task scheduling. Combined with the prediction network to grasp the trend of network state changes, the forward-looking optimization of task unloading and resource allocation is realized. It can efficiently handle complex collaborative tasks across modalities and devices, and improve the refinement and intelligence level of urban component management. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of the overall architecture and federated collaborative network topology of the IoT sensing terminal and edge computing gateway collaborative system for urban component management provided in an embodiment of the present invention. Figure 2 This is a task coordination scheduling flowchart of an IoT sensing terminal and edge computing gateway collaborative system for urban component management provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating a scenario of collaborative monitoring and coordinated handling of manhole cover status in an IoT sensor terminal and edge computing gateway collaborative system for urban component management, as provided in an embodiment of the present invention. Detailed Implementation

[0026] This invention provides a collaborative system for urban component management, consisting of IoT sensing terminals and edge computing gateways, which may include: Multiple IoT sensor terminals and at least one edge computing gateway; the IoT sensor terminals and the edge computing gateway are connected through a wireless ad hoc network to form a dynamic federated collaborative network; among them, the IoT sensor terminals have local computing and decision-making capabilities, and can conduct peer-to-peer task collaboration with the edge computing gateway and other IoT sensor terminals in the network based on a negotiation mechanism to jointly perform urban component management tasks. When any node in the dynamic federated collaborative network faces a complex task that it cannot complete independently, the node, as the task initiator, sends a collaborative request containing the task requirements to other nodes through the network. Each node that receives a collaboration request decides independently whether to participate and how to participate, based on its own resource status and task strategy. The federated coordination scheduling engine in the edge computing gateway maintains a global resource view of the dynamic federated coordination network and coordinates relevant nodes based on the global resource view. Nodes that receive coordination requests independently decide whether to participate and how to participate based on their own resource status and task strategy. The relevant nodes negotiate and communicate to jointly determine the task decomposition, allocation and execution process, and complete the complex task in a distributed collaborative manner.

[0027] IoT sensing terminals are physically deployed on or near various urban components. Each IoT sensing terminal is an embedded device with independent computing capabilities, integrating a processor, memory, dedicated sensors, and wireless communication units. The key feature of this terminal is its local computing and decision-making capabilities, rather than being a simple data acquisition and forwarding node. It can perform pre-defined logical judgments or run lightweight analysis algorithms on the raw sensing data it collects, thereby initially forming local decision results with semantic information at the device level.

[0028] Edge computing gateways are deployed at network aggregation points in the areas where IoT sensor terminals are located. Their physical form can be a fixed cabinet device or a server based on an industrial computer. This gateway possesses significantly more computing and storage resources than a single IoT sensor terminal and is equipped with multiple network interfaces. One of the gateway's core functions is to act as a bridge between this local area and a remote cloud management platform, responsible for protocol conversion and data uplink. Simultaneously, its more critical role is as a high-level node in the collaborative network, participating in and coordinating distributed computing tasks within the local area.

[0029] IoT sensor terminals and edge computing gateways form a wireless self-organizing network via wireless communication modules. This network does not rely on pre-established centralized communication infrastructure. During system initialization, each node broadcasts discovery messages to probe for other neighboring nodes within its communication range and autonomously negotiates communication links and network routes. The resulting network topology is dynamic, adapting to changes in node additions, removals, movement, or wireless channel conditions, forming a resilient network foundation at the physical connection layer.

[0030] Based on the aforementioned wireless ad hoc network, IoT sensor terminals and edge computing gateways logically constitute a dynamic federated collaborative network. In this network, the relationships between participating nodes are equal. When a node faces a complex task that it cannot complete independently, it acts as the task initiator, sending a collaborative request containing the task requirements to other nodes via the network. Nodes receiving the request, including other IoT sensor terminals and edge computing gateways, independently decide whether and how to participate in the task based on their current resource status and task strategy. Subsequently, through multiple rounds of negotiation and communication, the relevant nodes jointly determine the task decomposition, allocation, and execution process, ultimately completing the overall task of city component management in a distributed collaborative manner.

[0031] like Figure 1As shown, this embodiment implements the above system through a collaborative network composed of multiple IoT sensor terminals and an edge computing gateway. The physical foundation of this network is a wireless ad hoc network, enabling direct and dynamic connections between terminals and the gateway. On top of this physical connection, a logical "dynamic federated collaborative network" is defined and constructed through software protocols. In this logical network, each IoT sensor terminal is no longer a passive data collection device, but an autonomous node with local intelligence and resource management capabilities. They are on equal footing with the edge computing gateway, which acts as the regional coordination center, and all declare their status, capabilities, and resources to the network through standardized interfaces. The core service in the edge computing gateway—the federated collaborative scheduling engine—is responsible for maintaining a global view of this dynamic network. The idle computing power, storage space, and communication bandwidth of all nodes (including the terminals and the gateway itself) are abstracted and aggregated into a unified, logical "dynamic resource pool." This architecture allows any complex task initiated by a node to transparently cross network boundaries, with the scheduling engine dynamically allocating the most suitable node combination from the resource pool based on the global state to collaboratively complete the task, thereby achieving distributed load balancing and global optimized resource utilization.

[0032] In one possible implementation, the IoT sensing terminal includes: The basic sensing module is used to collect status data of the target city components. The status data includes at least one of vibration, image, temperature and humidity. The local intelligent processing module is used to run a lightweight artificial intelligence model to perform preliminary analysis and anomaly detection on the status data. The collaborative communication and resource management module is used to broadcast its own status and available computing resources, discover neighboring collaborative nodes, and process collaborative task requests from other nodes.

[0033] The hardware design of the IoT sensing terminal integrates three core functional modules. The basic sensing module directly connects to various dedicated sensor probes, such as vibration sensors, image sensors, or temperature and humidity sensors, responsible for continuously or on-demand collecting raw physical signals from target urban components and converting these analog or digital signals into a standardized data format that the system can process. The local intelligent processing module contains an embedded artificial intelligence computing unit that loads a lightweight artificial intelligence model file that has been pre-trained and compressed. This module receives data streams from the basic sensing module, performs model inference operations, and outputs preliminary analysis results of the data status and anomaly detection indicators. The collaborative communication and resource management module is embedded in the terminal's operating system. This module continuously monitors the terminal's processor load, remaining memory space, and battery level, and encapsulates this status information into specific resource broadcast messages that are periodically sent out. Simultaneously, this module also listens to network channels, receives and parses similar messages from other nodes, thereby maintaining a resource list of neighboring nodes. When receiving an external collaborative task request, this module calls its internal decision-making logic to evaluate the matching degree between the request content and local resources and generates a response message.

[0034] In one possible implementation, the edge computing gateway includes: The gateway core processing module is used for protocol conversion, data aggregation, and communication with the upper-level cloud platform; The federated collaborative scheduling engine is used to maintain a global resource view of the dynamic federated collaborative network, respond to task collaboration requests, generate distributed collaborative execution schemes based on preset algorithms, and coordinate relevant nodes to reach an execution consensus.

[0035] The edge computing gateway employs a multi-core processor architecture in hardware and is divided into two main functional parts in software. The gateway's core processing module runs the basic network protocol stack, handling heterogeneous communication protocols from IoT sensor terminals and converting them into protocols suitable for transmission back to the upper-level cloud platform. This module also temporarily aggregates and caches data reported by multiple terminals within the area. The federated collaborative scheduling engine is a core software service running on the gateway. This engine maintains a dynamically updated global resource view, which integrates the status reported by each IoT sensor terminal and the gateway's local resource situation. When the engine receives a task collaboration request, it runs a complex decision-making algorithm based on preset scheduling strategies and optimization objectives. This algorithm comprehensively considers multiple constraints such as task requirements, node computing power, network bandwidth, and link quality, generating one or more distributed collaborative schemes that decompose the total task and allocate it to different nodes for execution. Subsequently, the engine, acting as a coordinator, communicates and negotiates with all nodes involved in the scheme until a consensus is reached on the final execution plan.

[0036] In one possible implementation, the system also includes a dynamic resource pooling virtual layer, which is composed of a federated collaborative scheduling engine and collaborative communication and resource management modules of various IoT sensor terminals. This virtual layer is used to abstract the idle computing, storage and communication resources of all nodes in the network into a unified logical resource pool for dynamic invocation by the distributed collaborative execution scheme.

[0037] The system contains a logically dynamic resource pooling virtual layer. This virtual layer is not an independent physical device, but rather a logical construct of the federated collaborative scheduling engine within the edge computing gateway and the collaborative communication and resource management modules within each IoT sensor terminal, all connected through a unified interface specification and communication agreement. Each node's collaborative communication and resource management module quantifies and abstracts its idle computing cycles, available storage space, and remaining communication bandwidth resources into a unified resource descriptor. The federated collaborative scheduling engine continuously collects these resource descriptors from all nodes and integrates them into a global, logically unified resource pool. When the scheduling engine generates a distributed collaborative execution plan, each subtask in the plan dynamically requests and binds the required computing, storage, or communication resource quotas from this logical resource pool. After the task is completed, the relevant resource quotas are released and returned to the resource pool for use by other tasks.

[0038] In one possible implementation, the default algorithm on which the federated cooperative scheduling engine generates the distributed cooperative execution scheme is a model-assisted adaptive deep reinforcement learning algorithm. The algorithm aims to minimize the total task execution delay and the total system energy consumption. Its state space includes network topology, node resource status and channel quality, and its action space is a set of feasible task offloading and resource allocation strategies.

[0039] The pre-defined algorithm relied upon by the federated collaborative scheduling engine is a model-assisted adaptive deep reinforcement learning algorithm. Deployed on the gateway side, the algorithm's reinforcement learning agent's state space is meticulously designed to fully encompass the current dynamic federated collaborative network's topology, the real-time resource status of each node, and evaluation metrics for wireless channel quality. The action space is defined as a discrete or continuous set of all feasible task offloading decisions and resource allocation strategies. The algorithm's reward function aims for system-level optimization, specifically set as a weighted negative feedback of the total end-to-end latency of task execution and the total energy consumed by the entire collaborative network during execution. Through continuous interaction with the environment, the algorithm can autonomously learn and continuously optimize its task scheduling strategy to adapt to dynamic changes in network conditions and load.

[0040] like Figure 2As shown, this embodiment further details the internal process of intelligent decision-making by the federated collaborative scheduling engine. When the engine receives a task request defined by a standardized task description language, a complete scheduling cycle is initiated. The core of the decision-making is a model-assisted adaptive deep reinforcement learning algorithm. This algorithm takes the current topology of the collaborative network, the real-time resource status of each node, and channel quality as state inputs. Its internally integrated prediction network first analyzes these state sequences to predict the network load and performance change trends within a short future window. Subsequently, the deep reinforcement learning agent integrates the current state and future predictions, with the optimization objective of minimizing the total task latency and total system energy consumption, and searches and evaluates in a vast action space (i.e., all possible combinations of task decomposition, unloading, and resource allocation strategies) to generate one or more candidate distributed collaborative execution schemes. After the scheme is generated, the scheduling engine does not force the command, but acts as a coordinator to send the scheme to the relevant nodes for negotiation to reach an execution consensus. At the same time, the engine logically binds the required computing, storage, and communication resource quotas to each subtask from the dynamic resource pool according to the scheme content. Once the negotiation is successful and the resources are ready, each node executes its assigned subtask in parallel. Once the task is completed, all occupied resources are released back to the resource pool, completing a full scheduling loop.

[0041] In one possible implementation, a model-assisted adaptive deep reinforcement learning algorithm is integrated with a prediction network to predict short-term trends in network state and node load, thereby assisting the federated collaborative scheduling engine in making forward-looking decisions when generating collaborative schemes.

[0042] The aforementioned deep reinforcement learning algorithm framework integrates a lightweight prediction network. This prediction network takes historical and current network state sequences and node load sequences as input, and its internal structure enables it to capture implicit patterns and short-term trends in this time-series data. The output of the prediction network is a prediction of network throughput fluctuations, node computational load changes, and link stability trends within a short future time window. These prediction results are provided in real-time to the decision-making module of the reinforcement learning agent. When generating a collaborative solution for the current moment, the agent not only considers the immediate state of the system but also proactively references the short-term future state provided by the prediction network, thereby making more robust and efficient scheduling decisions and avoiding solution failures or performance drops due to sudden environmental changes.

[0043] In one possible implementation, the collaborative communication and resource management module communicates with the federated collaborative scheduling engine through a standardized task description language for task publishing, capability querying, and solution description; this standardized task description language defines the input data, processing logic, resource requirements, and output format required for the computation task.

[0044] The collaborative communication and resource management module interacts with the federated collaborative scheduling engine using a standardized task description language. This language defines strict syntax and semantic specifications for formally describing a computational task. A complete task description language message must explicitly specify the data type, format, and source of the input data required by the task. The message must clearly define the core processing logic of the task or the identifier of the algorithm invoked. The message also needs to declare the minimum computing resource requirements, storage space requirements, and expected completion time for the task execution. Finally, the message needs to specify the data format and delivery method of the task output results. Through this language, the task initiator can accurately express its requirements, while the task executor or scheduler can unambiguously parse the requirements and assess whether it has the conditions to complete the task.

[0045] In one possible implementation, the dynamic federated collaborative network supports a privacy-preserving federated learning mechanism across nodes; each participating node uses local data to train a shared lightweight artificial intelligence model, and uploads encrypted model parameter updates to the edge computing gateway for secure aggregation, forming a global model update before distributing it to each node.

[0046] The dynamic federated collaborative network supports a privacy-preserving federated learning mechanism across nodes. When the network needs to jointly train or optimize a shared lightweight AI model, each participating IoT sensor terminal independently completes a round of model training locally using its own collected local data, generating incremental updates to the model parameters. To protect the privacy of the original data, the terminal performs homomorphic encryption or differential privacy processing on the parameter updates before sending them out locally. The encrypted parameter updates are securely uploaded to the edge computing gateway. The gateway's federated collaborative scheduling engine does not decrypt these updates but directly performs a secure aggregation operation on all received encrypted updates within the encrypted domain or a secure multi-party computation framework, generating an aggregated global model update. This global update is distributed back to all participating nodes, and each node merges it with its local model, thereby achieving collaborative model evolution and performance improvement without sharing any original data.

[0047] In one possible implementation, the system is used to achieve collaborative monitoring and early warning of urban road waterlogging; wherein, The first type of IoT sensing terminal is a water level sensor, used to collect data on water depth. The second type of IoT sensing terminal is an image acquisition device used to collect visual data of waterlogged areas; Edge computing gateways or third-class IoT sensing terminals with redundant computing power are dynamically scheduled to fuse water depth data and visual data, and perform water accumulation range identification and risk assessment.

[0048] When applying the system to collaborative monitoring and early warning of urban road flooding, the specific deployment and workflow are as follows: The first type of IoT sensor terminal is a water level sensor, installed in low-lying areas of roads or drainage outlets, specifically designed to collect millimeter-precision data on water depth. The second type of IoT sensor terminal is a high-definition image acquisition device, deployed at nearby high points to obtain aerial visual data of the flooded area. When the water level sensor locally detects a depth exceeding a threshold, it issues a collaborative analysis task through a dynamic federated collaborative network. The federated collaborative scheduling engine then responds, dynamically scheduling either the edge computing gateway itself or a third type of IoT sensor terminal with redundant computing power as the task execution node. This scheduled node simultaneously receives water level data and visual data, runs an image recognition and data fusion algorithm to accurately identify the actual coverage area of ​​the flooding, and integrates historical rainfall and drainage data to complete the risk assessment and level determination of the flooding, ultimately generating structured early warning information for reporting.

[0049] In one possible implementation, the system is used to achieve collaborative monitoring and coordinated response to the status of urban manhole covers; wherein, When an IoT sensor terminal deployed on a manhole cover detects an abnormal status locally, it issues a collaborative verification task to the dynamic federated collaborative network. The edge computing gateway schedules nearby camera devices for visual verification and links with municipal operation and maintenance terminals to generate handling work orders. At the same time, it publishes warning information through nearby public information publishing terminals.

[0050] When applying the system to the collaborative monitoring and coordinated handling of urban manhole cover status, the specific deployment and workflow are as follows: IoT sensing terminals are directly installed inside the manhole cover. Their built-in vibration and tilt sensors are used to monitor the manhole cover's illegal opening, abnormal movement, or damage status in real time. When an abnormal condition is detected, the terminal immediately issues an urgent collaborative verification task request to the dynamic federated collaborative network. Upon receiving the request, the federated collaborative scheduling engine quickly retrieves and schedules the nearest camera device node within the task area from the resource view. The scheduled camera device adjusts its angle to capture images or videos of the problematic manhole cover and transmits the images back to the scheduling engine in real time for visual analysis and verification. After verification, the scheduling engine automatically sends a handling work order containing precise location and event description to the municipal maintenance personnel's mobile terminal or work order system. Simultaneously, the scheduling engine sends instructions to public information publishing terminals near the anomaly point, controlling them to display warning information to alert pedestrians and vehicles.

[0051] like Figure 3As shown, this embodiment uses urban manhole cover status management as an example to specifically demonstrate how the above system achieves cross-device collaboration and coordinated handling in a real-world scenario. The entire process begins with an intelligent sensing terminal deployed on the manhole cover. Once its built-in sensors detect abnormal vibration or tilt patterns in local analysis and determine them as potential safety events, the subsequent process is immediately triggered. This terminal, acting as the task initiator, publishes a formatted "collaborative verification task" request to the dynamic federated collaborative network. This request is captured by the federated collaborative scheduling engine within the region. After parsing the task requirement, the scheduling engine quickly retrieves and selects the optimal node to respond from the global resource view: it first schedules one or more camera device nodes closest to the event point, instructing them to adjust their angles to capture images or videos of the target manhole cover and transmit the streaming media data back in real time for analysis and verification. Simultaneously with visual verification confirming the anomaly, the scheduling engine automatically executes pre-set linkage logic, sending a structured handling instruction to the municipal department's operation and maintenance management terminal or work order system, containing the precise location, event type, and timestamp. Almost simultaneously, the dispatch engine also sent control commands to public information display terminals on roads surrounding the anomaly point, driving their displays to show prominent safety warnings and guide pedestrians and vehicles to avoid the area. This series of actions was automatically orchestrated and triggered by the dispatch engine in an instant, forming a complete closed-loop handling process from intelligent perception and multi-source collaborative verification to cross-departmental automated linkage.

[0052] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A collaborative system for urban component management using IoT sensing terminals and edge computing gateways, characterized in that, include: Multiple IoT sensing terminals and at least one edge computing gateway; The IoT sensing terminal and the edge computing gateway are connected through a wireless self-organizing network to form a dynamic federated collaborative network; The IoT sensing terminal has local computing and decision-making capabilities, and can perform equal task collaboration with the edge computing gateway and other IoT sensing terminals in the network based on a negotiation mechanism to jointly execute urban component management tasks. When any node in the dynamic federated collaborative network faces a complex task that it cannot complete independently, the node, as the task initiator, sends a collaborative request containing the task requirements to other nodes through the network. Each node that receives a collaboration request decides independently whether to participate and how to participate, based on its own resource status and task strategy. The federated coordination scheduling engine in the edge computing gateway maintains a global resource view of the dynamic federated coordination network and coordinates relevant nodes based on the global resource view. Nodes that receive coordination requests independently decide whether to participate and how to participate based on their own resource status and task strategy. The relevant nodes negotiate and communicate to jointly determine the task decomposition, allocation and execution process, and complete the complex task in a distributed collaborative manner.

2. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 1, characterized in that, The IoT sensing terminal includes: The basic sensing module is used to collect status data of the target city components, and the status data includes at least one of vibration, image, temperature and humidity. The local intelligent processing module is used to run a lightweight artificial intelligence model to perform preliminary analysis and anomaly detection on the state data. The collaborative communication and resource management module is used to broadcast its own status and available computing resources, discover neighboring collaborative nodes, and process collaborative task requests from other nodes.

3. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 2, characterized in that, The edge computing gateway includes: The gateway core processing module is used for protocol conversion, data aggregation, and communication with the upper-level cloud platform; The federated collaborative scheduling engine is used to maintain a global resource view of the dynamic federated collaborative network, respond to task collaboration requests, generate distributed collaborative execution schemes based on preset algorithms, and coordinate relevant nodes to reach an execution consensus.

4. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 3, characterized in that, The system also includes a dynamic resource pooling virtual layer, which is composed of the federated collaborative scheduling engine and the collaborative communication and resource management modules of each IoT sensor terminal. This virtual layer is used to abstract the idle computing, storage and communication resources of all nodes in the network into a unified logical resource pool for dynamic invocation by the distributed collaborative execution scheme.

5. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 3, characterized in that, The default algorithm on which the federated collaborative scheduling engine generates the distributed collaborative execution scheme is a model-assisted adaptive deep reinforcement learning algorithm. The algorithm aims to minimize the total task execution delay and the total system energy consumption. Its state space includes network topology, node resource status and channel quality, and its action space is a set of feasible task offloading and resource allocation strategies.

6. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 5, characterized in that, The model-assisted adaptive deep reinforcement learning algorithm integrates a prediction network to predict short-term trends in network state and node load, thereby assisting the federated collaborative scheduling engine in making forward-looking decisions when generating collaborative schemes.

7. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 3, characterized in that, The collaborative communication and resource management module and the federated collaborative scheduling engine communicate through a standardized task description language for task publishing, capability querying, and solution description. This standardized task description language defines the input data, processing logic, resource requirements, and output format required for computational tasks.

8. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 1, characterized in that, The dynamic federated collaborative network supports a privacy-preserving federated learning mechanism across nodes; Each participating node uses local data to train a shared lightweight artificial intelligence model, and uploads encrypted model parameter updates to the edge computing gateway for secure aggregation, forming a global model update before distributing it to each node.

9. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 1, characterized in that, The system is used to achieve collaborative monitoring and early warning of waterlogging on urban roads; wherein... The first type of IoT sensing terminal is a water level sensor, used to collect data on water depth. The second type of IoT sensing terminal is an image acquisition device used to collect visual data of waterlogged areas; The edge computing gateway or the third type of IoT sensing terminal with redundant computing power is dynamically scheduled to fuse the water depth data and visual data, and to perform water accumulation range identification and risk assessment.

10. The collaborative system of IoT sensing terminals and edge computing gateways for urban component management according to claim 1, characterized in that, The system is used to achieve collaborative monitoring and coordinated response to the status of urban manhole covers; wherein... When an IoT sensor terminal deployed on a manhole cover detects an abnormal status locally, it issues a collaborative verification task to the dynamic federated collaborative network. The edge computing gateway schedules nearby camera devices for visual verification, and links with the municipal operation and maintenance terminal to generate a work order for handling. At the same time, it publishes warning information through nearby public information publishing terminals.

Images