Distributed meter identification method and system

Abstract

The invention relates to a distributed meter recognition method and system, and belongs to the field of substation automatic monitoring, and the method comprises the steps: collecting a meter image through a plurality of intelligent camera nodes, carrying out the local reading recognition of the meter image through a carried lightweight local model, and training the local model; each intelligent camera node is networked to form a hierarchical aggregation network, and the hierarchical aggregation network dynamically plans a multi-hop route according to the residual electric quantity of the intelligent camera nodes in the network and the link signal intensity, and dynamically determines a relay cluster head node; hierarchical aggregation of local models uploaded by the intelligent camera nodes and forwarding of identification results are carried out through the relay cluster head node, and the cluster models after hierarchical aggregation and the forwarded identification results are uploaded to a central server; and the central server aggregates the cluster models to update a global model, issues the global model to each intelligent camera node, and summarizes meter reading identification results. Compared with the prior art, the working efficiency and stability of the system in a complex environment can be improved.

Description

Technical Field

[0001] The present invention relates to the field of transformer substation automation monitoring, and in particular to a distributed meter identification method and system. Background Art

[0002] With the continuous advancement of smart grid construction, substations, as a vital component of the power system, are moving towards digitalization and intelligence. Traditional manual meter reading methods, characterized by low efficiency, prone to errors, and high costs, are no longer able to meet the needs of modern substations. Furthermore, as substations continue to expand in size, the workload of manual inspections has increased significantly, leading to rising labor costs.

[0003] Currently, substation meter monitoring relies primarily on manual inspections, which presents problems such as low efficiency and high safety risks. Operations and maintenance personnel must regularly read meters on-site, which is time-consuming and involves the risk of working in high-risk environments and manual recording errors. Prior art has proposed methods that combine artificial intelligence technology to achieve automated meter recognition. For example, CN115171091A discloses a meter recognition method for substation inspections. This meter recognition method includes the following steps: S1, collecting meter images; S2, detecting feature points in the meter image and matching them with a meter template image; S3, calculating the meter pointer direction; and S4, analyzing the pointer reading and outputting a structured output. This method utilizes an intelligent meter recognition approach that enables automatic and accurate meter reading identification during routine power grid inspections, allowing for easy and convenient recording, statistics, and data analysis. This approach not only reduces the operational and maintenance costs of manual meter reading but also effectively improves efficiency and reliability. Furthermore, it utilizes deep learning methods to construct a multi-layer perceptron and graph neural network model to extract the information expressed by the high-dimensional vectors of feature points. The introduction of self-attention and cross-attention mechanisms allows for better learning of the intrinsic and extrinsic connections and correspondences between feature points, enhancing their ability to express deeper information. While this method achieves high-precision automatic meter recognition, it relies on centralized image processing, requiring the transmission of large numbers of images to a central server for processing. This leads to high bandwidth pressure, poor real-time performance, and difficulty adapting a single image recognition model to complex and diverse environmental conditions (such as electromagnetic interference and varying lighting). Traditional automation solutions utilize 4G / Wi-Fi for network transmission, resulting in high energy consumption per node and the need for frequent battery replacement. Furthermore, communication stability is poor in the complex electromagnetic environment of substations. Although low-power wide area network technologies (such as LoRa) can support device networking, their one-way communication characteristics cannot meet the needs of distributed model collaboration.

[0004] In practical applications, to achieve remote monitoring, edge devices can be deployed to improve recognition efficiency. However, when centralized deep learning models are deployed on edge devices, they face bottlenecks such as insufficient computing resources and excessive power consumption. To address this, existing technologies have proposed lightweight models to address these issues. For example, CN119027866A discloses a pointer-type meter hand recognition method and system for edge computing devices. The method includes acquiring and preprocessing transmission line images; inputting the image data into a trained pointer-type meter recognition model to output a pointer recognition result; and verifying repeated verification instructions by sending the recognized pointer recognition result to a remote monitoring center. This method uses the pointer-type meter recognition model for pointer recognition. A target collaborative pruning strategy is used to minimize redundant parameters in the target detection model. The model parameters are then fine-tuned and trained to generate a recognition model suitable for edge computing. This approach reduces the resource overhead of the automatic pointer-type recognition model while ensuring model accuracy. However, existing technologies primarily focus on lightweighting the recognition model and do not consider optimizing communication resources between edge devices and central servers. Some existing technologies have proposed federated learning to solve this problem, but federated learning faces the challenge of unstable network topology in substation scenarios.

[0005] Therefore, there is an urgent need for a low-power, stable and scalable solution to realize the automatic identification and reporting of traditional meter readings in substations, providing strong technical support for the intelligent management of substations. Summary of the Invention

[0006] The purpose of this invention is to provide a distributed meter recognition method and system. By deploying intelligent camera nodes equipped with lightweight local models, meter detection and reading analysis are performed at the edge. A dynamic cluster network topology is constructed using a low-power Bluetooth network. Incorporating a hierarchical federated learning mechanism, relay cluster head nodes perform intermediate aggregation of local model parameters, reducing network load. A central server is responsible for global model aggregation updates and recognition result management. By employing a joint communication and computation optimization strategy, low-power automated recognition and analysis of traditional meter readings within substations is achieved, reducing substation maintenance costs.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] According to a first aspect of the present invention, a distributed meter identification method is provided, the method comprising the following steps:

[0009] Multiple smart camera nodes are used to collect meter images. The lightweight local models installed in the smart camera nodes are used to perform local reading recognition on the meter images. Local datasets are then built to train the local models.

[0010] The smart camera nodes are networked to form a hierarchical aggregation network. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, and dynamically determines the relay cluster head node;

[0011] The hierarchical aggregation network performs hierarchical aggregation of the local models uploaded by each smart camera node and forwards the recognition results through the relay cluster head node, and uploads the hierarchically aggregated cluster model and forwarded recognition results to the central server;

[0012] The central server aggregates the cluster models to update the global model, sends it to each smart camera node to update the local model, and summarizes the meter reading recognition results.

[0013] As a preferred technical solution, in the hierarchical aggregation network, the state S of node i at time t is i The definition is as follows:

[0014]

[0015] Among them, RSSI i and RSSI avg They represent the signal strength of node i and the global average signal strength, E i and E avg They represent the remaining power of node i and the global average remaining power respectively, and α is the scaling factor.

[0016] As a preferred technical solution, the hierarchical aggregation network preferentially selects nodes in good condition as relay cluster head nodes based on the dynamic routing mechanism, forwards data from other weaker nodes adjacent to it, forms a multi-cluster topology, and obtains multiple relay cluster head nodes and their corresponding clusters. The relay cluster head node performs weighted aggregation on the local models uploaded by multiple smart camera nodes in the cluster according to the amount of data held locally by each node to obtain a cluster model. For cluster c, its cluster model w c The aggregation method is:

[0017]

[0018] Among them, |D i | represents the data volume of the local dataset corresponding to the i-th local model in cluster c, M is the number of local models in cluster c, and w i is the i-th local model.

[0019] As a preferred technical solution, the central server receives cluster models uploaded by all hierarchical aggregation networks and updates the global model. For time t+1, the global model is updated as follows:

[0020]

[0021] Among them, w g represents the global model at time t, η is the learning rate, K is the number of relay cluster head nodes, represents the total amount of data in cluster c, w c It is a cluster model.

[0022] As a preferred technical solution, the lightweight local model carried in the smart camera node adopts different recognition algorithms according to the type of meter, wherein:

[0023] For pointer-type meters, the local model performs the following steps: performing target detection on the meter image to obtain the dial area, performing image segmentation on the dial area to extract the pointer area, and calculating the angular position of the pointer using a geometric analysis method. An angle-to-reading mapping relationship is established based on the dial's range and scale distribution, and the meter reading is determined based on the calculated angle and the angle-to-reading mapping relationship.

[0024] For digital meters, the local model performs the following steps: locating the digital area of the meter image, applying a character recognition model to the digital area, extracting digital information on the meter, and obtaining a meter reading.

[0025] As a preferred technical solution, the central server dynamically adjusts the operating parameters of the smart camera nodes based on monitoring data to optimize system performance and energy consumption. The monitoring data includes the network load and ambient lighting conditions of the smart camera nodes. The operating parameters include shooting frequency, exposure time and data transmission interval. When the ambient lighting conditions are weaker than the preset conditions, the shooting frequency of the smart camera nodes is reduced or the exposure time is increased. When the network load is greater than the preset threshold, the data transmission interval of the smart camera nodes is extended.

[0026] As a preferred technical solution, the central server performs version management on the global model, including a rollback mechanism and a grayscale release strategy. The rollback mechanism is specifically as follows: when the verification set performance of the central server degrades by more than a preset threshold for a preset number of consecutive rounds, it automatically rolls back to the historical optimal global model; the grayscale release strategy is specifically as follows: after the updated global model passes the grayscale release verification, it is first sent to a preset proportion of smart camera nodes for trial operation. If there is no abnormal confidence alarm within the preset time interval, the full version is pushed to all smart camera nodes.

[0027] As a preferred technical solution, the method further includes:

[0028] While using the local model to identify the meter reading image, the confidence of the recognition result is determined. If the confidence is greater than the preset threshold, the recognition result is forwarded to the central server through the hierarchical aggregation network. Otherwise, the meter image is compressed and forwarded to the central server through the hierarchical aggregation network. The central server calls the high-precision recognition model for recognition and submits it for manual review.

[0029] As a preferred technical solution, the method further includes:

[0030] When a new smart camera node joins, the hierarchical aggregation network automatically detects it and assigns it a unique network address;

[0031] The new smart camera node establishes a connection with the existing nodes through broadcast messages and downloads the necessary configuration files and initial models from the central server as the initial local model.

[0032] According to a second aspect of the present invention, a distributed meter identification system is provided for implementing the above method, the system comprising:

[0033] Smart camera nodes: Serving as front-end sensing and edge computing units, they are deployed in the substation meter monitoring area. They support wireless communication and are battery-powered. Equipped with high-resolution cameras and edge computing modules, they collect meter images, perform local reading recognition, and upload federated learning local models. They also have local storage capabilities to cache unsuccessfully transmitted data.

[0034] Hierarchical aggregation network: formed by a network of multiple smart camera nodes, which communicate with each other via the Bluetooth Low Energy Mesh network protocol. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, dynamically determines the relay cluster head node, and uses the relay cluster head node to hierarchically aggregate the local models uploaded by each smart camera node and forward the recognition results. The hierarchical aggregation cluster model and forwarded recognition results are uploaded to the central server;

[0035] Central server: used to summarize the reading recognition results forwarded by the hierarchical aggregation network, and execute the federated learning top-level aggregation algorithm to aggregate the cluster models to update the global model, and send the updated global model to each smart camera node to update the local model.

[0036] According to a third aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the method when executing the program.

[0037] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) The present invention improves the deployment and working range efficiency and accuracy of large-scale meter identification through automated meter identification and monitoring, reduces the need for manual operation, and reduces operation and maintenance costs.

[0040] (2) The present invention deploys federated learning and lightweight edge algorithm models, eliminating the need to transmit a large amount of image resources to a central server for processing. Most reading recognition can be completed locally, and images only need to be uploaded to a central server when the confidence level of the local recognition results does not meet the requirements. This significantly reduces the pressure on communication bandwidth, has good real-time performance, and meets the collaborative needs of distributed models.

[0041] (3) The present invention combines the low power consumption characteristics of the Bluetooth Mesh network with a hierarchical aggregation network with a dynamic routing mechanism. It dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, thereby reducing the network link load caused by model uploading and continuously optimizing the algorithm accuracy to overcome recognition errors in complex environments, ensuring that the system can operate reliably in complex environments with limited resources.

[0042] (4) The present invention effectively overcomes the problem of unstable network topology in federated learning in the prior art. The unstable network topology in the prior art is mainly manifested in problems such as communication interruption between node devices, high latency, frequent disconnection, and accidental mounting of old nodes. The present invention introduces the node status indicator S i , taking into account the signal strength (RSSI) and remaining power (E) of the node, the stability and availability of the node are dynamically evaluated through a weighted function to ensure that the selection of the network relay cluster head node has strong communication quality and endurance, thereby reducing problems such as communication interruption between node devices, high latency, frequent disconnection, and accidental mounting of old nodes caused by unstable network topology.

[0043] (5) The present invention supports the dynamic addition and configuration of new nodes, which can improve the scalability of the system.

[0044] (6) The central server of the present invention dynamically adjusts the working parameters of the camera nodes through monitoring data to cope with the complex working conditions of the substation, thereby enhancing the environmental adaptability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the system structure of the present invention;

[0046] Figure 2 is a flow chart of the method of the present invention;

[0047] Figure 3Schematic diagram of the layered polymerization process of the present invention;

[0048] Figure 4 Schematic diagram of the model updating process of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0050] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0051] Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by a person of ordinary skill in the technical field to which this application belongs. The words "one", "a", "the" and the like used in this application do not indicate a limit on quantity and may indicate the singular or plural. The terms "include", "comprise", "have" and any variations thereof used in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units that are inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The word "multiple" used in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0052] Example 1

[0053] This embodiment relates to the cross-technical field of substation automation monitoring and edge artificial intelligence, specifically to a distributed meter recognition system based on edge computing and federated learning, which combines target detection algorithms, Bluetooth low-power communication networks, and federated learning hierarchical aggregation methods to achieve automated reading recognition and continuous learning of large-scale meters in substations under complex environments. Figure 1 As shown, the distributed meter identification system includes:

[0054] (1) Smart camera node: As a front-end perception and edge computing unit, it is deployed in the meter monitoring area of the substation, supports wireless communication and is battery-powered. The smart camera node is equipped with a high-resolution camera and an edge computing module (i.e., a local model equipped with a lightweight YOLO image recognition algorithm), which is used to collect meter images at the edge and perform local reading recognition and federated learning local model upload. Each smart camera node has a local storage function for caching data that has not been successfully transmitted to ensure that data is not lost when the network is unstable or interrupted. In a preferred embodiment, each smart camera node preferentially uses the lightweight detection algorithm installed locally to identify the meter reading and gives a confidence level. If the confidence level is lower than the threshold, the image is compressed and transmitted to the central server, which uses a more accurate algorithm model to identify it and submit it for manual review.

[0055] (2) Hierarchical aggregation network: It is formed by a network of multiple smart camera nodes. The nodes communicate with each other through the low-power Bluetooth Mesh network protocol. It supports large-scale multi-hop communication and dynamic routing between nodes. It is a low-power, highly scalable wireless network architecture. Each node in the hierarchical aggregation network can serve as both a data transmitter and a relay node, forming a multi-hop ad hoc network structure. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, dynamically determines the relay cluster head node, avoids the single point overload problem caused by fixed routing, and performs hierarchical aggregation of the local models uploaded by each smart camera node and forwards the recognition results through the relay cluster head node. The hierarchical aggregated cluster model and forwarded recognition results are uploaded to the central server, further reducing the network load, thereby achieving low-power, scalable data transmission. That is, in order to reduce the communication load in the network, the local model uploaded by the smart camera node will first complete hierarchical aggregation at the relay cluster head according to the ad hoc network structure, and the aggregated cluster model will continue to be sent to the central server. In addition, when the node signal strength is weak or the network is congested, the dynamic routing selection mechanism will actively rebuild the link and select nodes with stronger signals and lower loads for data transmission.

[0056] (3) Central server: The control and management core of the entire system, used to aggregate the reading recognition results forwarded by the hierarchical aggregation network, and execute the federated learning top-level aggregation algorithm to aggregate the cluster model to update the global model, and send the updated global model to each smart camera node to update the local model, as well as storage and analysis. In addition, the central server is also responsible for managing the firmware upgrade of the smart camera node. The central server also has a built-in high-performance computing module that can process image data with low confidence and further optimize the recognition results through deep learning models. In a preferred embodiment, the central server can dynamically adjust the working parameters of the smart camera node based on the monitoring data to optimize the system power consumption performance.

[0057] Compared with existing technologies, this system achieves localized data processing through edge computing, continuously optimizes model performance through hierarchical federated learning, and combines Bluetooth Mesh networks to ensure low power consumption and high reliability, thereby improving the system's work efficiency and stability in complex environments, while significantly reducing the need for manual maintenance.

[0058] Example 2

[0059] This embodiment provides a distributed meter identification method based on embodiment 1, such as Figure 2 As shown, the method includes the following steps:

[0060] S1 uses multiple smart camera nodes to collect meter images, uses the lightweight local model installed in the smart camera nodes to perform local reading recognition on the meter images, and builds a local dataset to train the local model.

[0061] In this embodiment, the lightweight local model implemented in the smart camera node has been specifically optimized to efficiently identify meter readings using limited computing resources. This algorithm incorporates specialized feature extraction models for both analog and digital meters, leveraging image segmentation, edge detection, object detection, key point detection, and character recognition to rapidly identify meter readings.

[0062] In a preferred embodiment, different recognition algorithms are used according to different types of meters, wherein:

[0063] For pointer meters, the local model performs the following steps: performing target detection on the meter image to obtain the dial area, performing image segmentation on the dial area to extract the pointer area, and calculating the angular position of the pointer using geometric analysis methods. An angle-to-reading mapping relationship is established based on the dial's range and scale distribution, and the meter reading is determined based on the calculated angle and the angle-to-reading mapping relationship.

[0064] For digital meters, the local model performs the following steps: locate the digital area of the meter image, apply the character recognition model to the digital area, extract the digital information on the meter, and obtain the meter reading.

[0065] Each smart camera node i has a local dataset D i Train the local model w i , then its objective function F is:

[0066]

[0067] Among them, l(·) is the loss function, x and y are the local dataset D i The data and labels in .

[0068] While using the local model to identify the meter readings, the confidence level of the recognition results is determined. If the confidence level is greater than the preset threshold, the recognition results are forwarded to the central server through the hierarchical aggregation network. Otherwise, the meter image is compressed and forwarded to the central server through the hierarchical aggregation network. The central server calls a high-precision recognition model for identification and submits it for manual review to ensure the accuracy of the final data.

[0069] In S2, each smart camera node is networked to form a hierarchical aggregation network. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, and dynamically determines the relay cluster head node.

[0070] like Figure 3 As shown, this embodiment introduces federated learning to transfer model parameters between the edge and the central server. After the smart camera nodes complete local model training, they upload the model to the hierarchical aggregation network for cluster model aggregation. The hierarchical aggregation network then uploads the cluster model to the central server for global model aggregation. Finally, the central server redistributes the updated global model to the smart camera nodes to update the local models.

[0071] The hierarchical aggregation network supports a dynamic routing mechanism that dynamically establishes the network topology based on the load of smart camera nodes. Local models uploaded by smart camera nodes are hierarchically aggregated at cluster relay nodes based on the network topology. This prevents single-point overload and network congestion, optimizes network resource utilization, and prevents overload on certain links from impacting overall performance. This mechanism dynamically adjusts routing and network topology strategies by monitoring each node's resource utilization and network link load. If a smart camera node experiences high resource utilization or excessive link load, the dynamic routing mechanism redirects the link to a less loaded node, ensuring balanced resource utilization across the network.

[0072] Specifically, the dynamic routing mechanism is determined based on the state of the node, the state S of node i at time t i The definition is as follows:

[0073]

[0074] Among them, RSSI i and RSSI avg They represent the signal strength of node i and the global average signal strength, E i and E avg They represent the remaining power of node i and the global average remaining power respectively, and α is the scaling factor.

[0075] S3, the hierarchical aggregation network performs hierarchical aggregation of the local models uploaded by each smart camera node and forwards the recognition results through the relay cluster head node, and uploads the hierarchically aggregated cluster model and forwarded recognition results to the central server.

[0076] The hierarchical aggregation network prioritizes nodes in good condition as relay cluster head nodes based on the dynamic routing mechanism, forwards data from other weaker nodes adjacent to it, forms a multi-cluster topology, obtains K relay cluster head nodes, and then forms clusters under K relay cluster head nodes. Figure 4 As shown in Figure 1, the relay cluster head node performs weighted aggregation on the local models uploaded by the M smart camera nodes in the cluster according to the amount of data held locally by each node to obtain a cluster model. This is done before the global model is aggregated at the central server, forming a hierarchical aggregation structure. For cluster c, its cluster model w c The aggregation method is:

[0077]

[0078] Among them, |D i | represents the data volume of the local dataset corresponding to the i-th local model in cluster c, M is the number of local models in cluster c, and w i is the i-th local model.

[0079] The relay cluster head node transmits the cluster model w c In one embodiment, if the network is complex enough, there may be multiple layers of nested clusters. In this case, the current cluster model needs to be aggregated again at the next layer of relay cluster head nodes until it reaches the central server.

[0080] S4, the central server aggregates the cluster models to update the global model, sends it to each smart camera node to update the local model, and summarizes the meter reading recognition results.

[0081] like Figure 4As shown, the central server receives cluster models uploaded by all hierarchical aggregation networks and updates the global model.

[0082] For time t+1, the global model is updated as follows:

[0083]

[0084] Among them, w g represents the global model at time t, η is the learning rate, K is the number of relay cluster head nodes, represents the total amount of data in cluster c, w c It is a cluster model.

[0085] Each time the central server updates the global model, it generates a model file and pushes it to each smart camera node via a hierarchical aggregation network. Upon receiving the updated data, the smart camera node merges it with the corresponding parameters of its local model to generate an updated model. After aggregation is complete, the smart camera node performs inference testing on the updated model to ensure its performance and stability.

[0086] In a preferred embodiment, a central server dynamically adjusts the operating parameters of smart camera nodes based on monitoring data to optimize system performance and energy consumption. The monitoring data includes the network load and ambient lighting conditions of the smart camera nodes, and the operating parameters include capture frequency, exposure time, and data transmission interval. When ambient lighting conditions are weaker than preset conditions, the capture frequency of the smart camera nodes is reduced or the exposure time is increased to obtain clearer images. When the network load exceeds a preset threshold, the data transmission interval of the smart camera nodes is extended to reduce congestion in the hierarchical aggregation network.

[0087] In a preferred embodiment, the central server performs version management on the global model, including a rollback mechanism and a grayscale release strategy. The rollback mechanism is specifically as follows: when the performance of the central server's verification set degrades by more than a preset threshold for a preset number of consecutive rounds, it automatically rolls back to the historically optimal global model; the grayscale release strategy is specifically as follows: after the updated global model passes the grayscale release verification, it is first distributed to 10% of the smart camera nodes for trial operation. If there are no abnormal confidence alarms within 24 hours, the full version is pushed to all smart camera nodes.

[0088] In this embodiment, the method also includes the dynamic addition and configuration of new smart camera nodes, eliminating the need for manual intervention, improving system scalability and reducing maintenance costs. Specifically, when a new smart camera node is added, the hierarchical aggregation network automatically detects it and assigns it a unique network address. The new smart camera node establishes a connection with existing nodes via broadcast messages and downloads the necessary configuration files and initial models from a central server as the initial local model. This entire process requires no manual intervention, simplifying system maintenance and making it suitable for substation expansion or the addition of new equipment.

[0089] Example 3

[0090] This embodiment, based on the second embodiment, provides an application process of a distributed meter identification method, including the following steps:

[0091] Step 1: Deployment: Multiple smart camera nodes are deployed throughout the substation based on data collection requirements, ensuring coverage of all required traditional meters. Each node's camera is aimed at the target meter, adjusting the focus and angle to achieve a clear image. The node is equipped with a pre-trained YOLO algorithm model and the Bluetooth Mesh protocol stack.

[0092] Step 2: During the networking phase, after powering on, the smart camera node automatically broadcasts a network access request, prioritizing establishing a connection with the neighboring node with the best signal strength (RSSI > -70dBm). It then receives the network key and application key from the central server and constructs a neighbor node table containing signal strength (RSSI) and remaining battery power. This table is dynamically updated every 30 minutes to optimize communication paths.

[0093] Step 3: The smart camera node captures meter images at set intervals and performs preliminary image preprocessing, including real-time correction of barrel distortion caused by wide-angle lenses and implementing fill-light strategies for low-light scenarios common in substations. The smart camera node then identifies the meter reading using its local model and checks the confidence level. If the recognition fails or the confidence level falls below a set threshold, the original image is compressed and uploaded to a central server via a hierarchical aggregation network for further processing.

[0094] Step 4: The smart camera node uses the newly acquired images and labels during operation as a local dataset to further train the built-in local model. The local model of each node is uploaded to the hierarchical aggregation network according to the preset reporting cycle.

[0095] Step 5: Based on the current status of each smart camera node and the network status (such as signal strength and network load), the hierarchical aggregation network plans the optimal multi-hop topology for transmission to the central server based on the dynamic routing algorithm and forms a relay cluster head node in the network.

[0096] Step 6: The local models uploaded by each smart camera node are transmitted via the routes planned by the hierarchical aggregation network. The cluster head relay node receives the local models relayed there and performs aggregation. The hierarchically aggregated cluster models are then sent to the central server for reception and global model aggregation.

[0097] Step 7: The central server receives the cluster model and performs global model aggregation. The central server's global model version management includes a rollback mechanism and a grayscale release strategy. If the server-side validation set drops below a preset threshold for three consecutive rounds, the system automatically rolls back to the historically optimal model (retaining the most recent five versions). The new model passes grayscale release verification: it is first distributed to 10% of nodes for a trial run. If there are no abnormal confidence alerts within 24 hours, the full model update is pushed.

[0098] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0099] Many components in a device are connected to the I / O interface, including: input units, such as a keyboard and mouse; output units, such as various types of displays and speakers; storage units, such as magnetic disks and optical disks; and communication units, such as network cards, modems, and wireless communication transceivers. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks.

[0100] The processing unit performs the various methods and processes described above, such as methods S1 to S4. For example, in some embodiments, methods S1 to S4 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S4 described above can be performed. Alternatively, in other embodiments, the CPU can be configured to execute methods S1 to S4 by any other appropriate means (for example, by means of firmware).

[0101] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0102] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0103] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A distributed meter identification method, characterized in that: The method comprises the following steps: Multiple smart camera nodes are used to collect meter images. The lightweight local models installed in the smart camera nodes are used to perform local reading recognition on the meter images. Local datasets are then built to train the local models. The smart camera nodes are networked to form a hierarchical aggregation network. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, and dynamically determines the relay cluster head node; The hierarchical aggregation network performs hierarchical aggregation of the local models uploaded by each smart camera node and forwards the recognition results through the relay cluster head node, and uploads the hierarchically aggregated cluster model and forwarded recognition results to the central server; The central server aggregates the cluster models to update the global model, sends it to each smart camera node to update the local model, and summarizes the meter reading recognition results.

2. A distributed meter identification method according to claim 1, characterized in that: In the hierarchical aggregation network, the state S of node i at time t is i The definition is as follows: Among them, RSSI i and RSSI avg They represent the signal strength of node i and the global average signal strength, E i and E avg They represent the remaining power of node i and the global average remaining power respectively, and α is the scaling factor.

3. A distributed meter identification method according to claim 1, characterized in that: The hierarchical aggregation network preferentially selects nodes in good condition as relay cluster head nodes based on the dynamic routing mechanism, forwards the data of other weaker nodes adjacent to it, forms a multi-cluster topology, and obtains multiple relay cluster head nodes and their corresponding clusters. The relay cluster head node performs weighted aggregation on the local models uploaded by multiple smart camera nodes in the cluster according to the amount of data held locally by each node to obtain a cluster model. For cluster c, its cluster model w c The aggregation method is: Among them, |D i | represents the data volume of the local dataset corresponding to the i-th local model in cluster c, M is the number of local models in cluster c, and w i is the i-th local model.

4. A distributed meter identification method according to claim 1, characterized in that: The central server receives cluster models uploaded by all hierarchical aggregation networks and updates the global model. For time t+1, the global model is updated as follows: Among them, w g represents the global model at time t, η is the learning rate, K is the number of relay cluster head nodes, represents the total amount of data in cluster c, w c It is a cluster model.

5. A distributed meter identification method according to claim 1, characterized in that: The lightweight local model carried in the smart camera node adopts different recognition algorithms according to the type of meter. For pointer-type meters, the local model performs the following steps: performing target detection on the meter image to obtain the dial area, performing image segmentation on the dial area to extract the pointer area, and calculating the angular position of the pointer using a geometric analysis method. An angle-to-reading mapping relationship is established based on the dial's range and scale distribution, and the meter reading is determined based on the calculated angle and the angle-to-reading mapping relationship. For digital meters, the local model performs the following steps: locating the digital area of the meter image, applying a character recognition model to the digital area, extracting digital information on the meter, and obtaining a meter reading.

6. A distributed meter identification method according to claim 1, characterized in that: The central server dynamically adjusts the operating parameters of the smart camera nodes based on monitoring data to optimize system performance and energy consumption. The monitoring data includes the network load and ambient lighting conditions of the smart camera nodes. The operating parameters include shooting frequency, exposure time, and data transmission interval. When the ambient lighting conditions are weaker than preset conditions, the shooting frequency of the smart camera nodes is reduced or the exposure time is increased. When the network load is greater than a preset threshold, the data transmission interval of the smart camera nodes is extended.

7. A distributed meter identification method according to claim 1, characterized in that: The central server performs version management on the global model, including a rollback mechanism and a grayscale release strategy. The rollback mechanism is specifically as follows: when the performance of the central server's verification set degrades by more than a preset threshold for a preset number of consecutive rounds, it automatically rolls back to the historical optimal global model; the grayscale release strategy is specifically as follows: after the updated global model passes the grayscale release verification, it is first sent to a preset proportion of smart camera nodes for trial operation. If there is no abnormal confidence alarm within a preset time interval, the full version is pushed to all smart camera nodes.

8. A distributed meter identification method according to claim 1, characterized in that: The method further comprises: While using the local model to identify the meter reading image, the confidence of the recognition result is determined. If the confidence is greater than the preset threshold, the recognition result is forwarded to the central server through the hierarchical aggregation network. Otherwise, the meter image is compressed and forwarded to the central server through the hierarchical aggregation network. The central server calls the high-precision recognition model for recognition and submits it for manual review.

9. A distributed meter identification method according to claim 1, characterized in that: The method further comprises: When a new smart camera node joins, the hierarchical aggregation network automatically detects it and assigns it a unique network address; The new smart camera node establishes a connection with the existing nodes through broadcast messages and downloads the necessary configuration files and initial models from the central server as the initial local model.

10. A distributed meter identification system, characterized in that: For implementing the method according to any one of claims 1 to 9, the system comprises: Smart camera nodes: Serving as front-end sensing and edge computing units, they are deployed in the substation meter monitoring area. They support wireless communication and are battery-powered. Equipped with high-resolution cameras and edge computing modules, they collect meter images, perform local reading recognition, and upload federated learning local models. They also have local storage capabilities to cache unsuccessfully transmitted data. Hierarchical aggregation network: formed by a network of multiple smart camera nodes, which communicate with each other via the Bluetooth Low Energy Mesh network protocol. The hierarchical aggregation network dynamically plans multi-hop routing based on the remaining power of the smart camera nodes in the network and the link signal strength, dynamically determines the relay cluster head node, and uses the relay cluster head node to hierarchically aggregate the local models uploaded by each smart camera node and forward the recognition results. The hierarchical aggregation cluster model and forwarded recognition results are uploaded to the central server; Central server: used to summarize the reading recognition results forwarded by the hierarchical aggregation network, and execute the federated learning top-level aggregation algorithm to aggregate the cluster models to update the global model, and send the updated global model to each smart camera node to update the local model.

Images