Substation Intelligent Service Model Management Method and System Based on Internet of Things Management Platform

Abstract

The present invention discloses a management method and system for a substation intelligent service model based on an IoT management platform. The method includes transmitting a first-level substation intelligent inspection system model to a second level based on a first-level artificial intelligence "two libraries and one platform"; issuing the substation intelligent inspection system model to the IoT management platform based on a second-level artificial intelligence "two libraries and one platform"; the IoT management platform providing an HTTP Restful interface service to push messages to MQS; and providing an SDK access specification; based on the MQS messages, a third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the IoT management platform, deploys the model within the third-level substation intelligent inspection system, and processes local substation intelligent inspection tasks. The present invention realizes data sharing and reference of three-level model data, realizes collaborative learning and common iterative update of multi-level models, improves the accuracy and real-time iterative update of the models, and meets the continuously changing business requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of substations, and specifically relates to a management method and system for a substation intelligent service model based on an Internet of Things management platform. Background Art

[0002] Under the severe situation of the rapid growth of substation scale, the increasingly complex external environment, and the relatively insufficient personnel strength, the large-scale application of substation intelligent inspection can greatly improve the inspection efficiency and accuracy, reduce the labor cost and risk management cost, and further improve the operation efficiency and operation and maintenance quality of power grid companies.

[0003] Currently, the indicators such as the accuracy rate, recall rate, and precision of the current substation model are not high, and the generalization ability is low, resulting in a low practical level of the model and being unable to meet the application requirements of the substation intelligent inspection scenario. It is urgent to improve the model optimization system, realize the regular iteration of the algorithm model, improve the accuracy of the substation algorithm model, and quickly improve the practical level of the model. Summary of the Invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a management method and system for a substation intelligent service model based on an Internet of Things management platform. The technical solution is as follows:

[0005] In a first aspect, a management method for a substation intelligent service model based on an Internet of Things management platform is provided. The method includes the following steps:

[0006] Transmit the first-level substation intelligent inspection system model to the second level based on the first-level artificial intelligence "two libraries and one platform";

[0007] Based on the second-level artificial intelligence "two libraries and one platform", send the substation intelligent inspection system model to the Internet of Things management platform through the HTTP Restful protocol;

[0008] The Internet of Things management platform provides an HTTP Restful interface service to push messages to MQS; and provides an SDK access specification;

[0009] Based on the MQS message, the third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the Internet of Things management platform, deploys the model in the third-level substation intelligent inspection system, and processes the intelligent inspection tasks of the substation locally.

[0010] In some embodiments, the step that based on the MQS message, the third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the Internet of Things management platform includes:

[0011] Obtain the parameters of the latest version of the substation intelligent inspection system model sent down to the Internet of Things management platform at the first level. The parameters include multi-dimensional technical parameters, performance index parameters, and function parameters of the model.

[0012] Compare with the parameters of the substation intelligent inspection system model deployed internally in the substation intelligent inspection system at the current third level to determine whether it is necessary to pull the substation intelligent inspection system model from the Internet of Things management platform.

[0013] In some embodiments, after pulling the substation intelligent inspection system model from the Internet of Things management platform, test the performance index of the current model based on local sample data. If the performance index does not meet the standard, send a message to the second level, and the second level updates the substation intelligent inspection system model based on the collected local sample data at the third level.

[0014] In some embodiments, the second level updates the substation intelligent inspection system model based on the collected local sample data at the third level, including:

[0015] The second-level artificial intelligence "two libraries and one platform" collects the parameters of the substation intelligent inspection system models at each third level, local sample data, and the performance index values of the third-level substation intelligent inspection system models in the local sample data.

[0016] Based on the local sample data at the third level with similar performance index values of the third-level substation intelligent inspection system model or based on each third-level substation intelligent inspection system model and local sample data, exchange the local sample data and input it into the corresponding third-level substation intelligent inspection system model to determine the performance index values of each third-level substation intelligent inspection system model after the exchange.

[0017] In the case where the performance index value after the exchange is similar to the performance index value before the exchange, determine that the corresponding local sample data characteristics at the third level are similar. In the case where the performance index value after the exchange is different from the performance index value before the exchange, determine that the corresponding third-level sample data characteristics are not similar.

[0018] Based on the network parameters of the third-level substation intelligent inspection system model with the best performance index among multiple third-level substation intelligent inspection system models with similar local sample data characteristic parameters at the third level and the network parameters of the latest version of the substation intelligent inspection system model sent down to the Internet of Things management platform at the first level as a reference, and using the local sample data of multiple third levels with similar local sample data characteristic parameters at the third level as training samples together, train and update the network parameters of the third-level model, and update the substation intelligent inspection system model pulled by the Internet of Things management platform.

[0019] In some embodiments, multiple third-level local sample data with similar third-level local sample data characteristic parameters are jointly used as training samples, including: deleting duplicates;

[0020] Taking one local sample data as one data set, performing correlation analysis on multiple sample sets, increasing the information entropy of multiple sample sets by deleting related items, and re-obtaining multiple sample data for training the model.

[0021] In some embodiments, the substation intelligent service model management method based on the Internet of Things management platform further includes the following steps:

[0022] Based on the two-level artificial intelligence "two libraries and one platform", uploading each second-level model data to the first level to achieve unified management of the high-quality model library;

[0023] The first level updates its own model data based on the uploaded third-level model data of each level and the first-level original model data;

[0024] The first level updates its own model data based on the uploaded third-level model data of each level and the first-level original model data, including:

[0025] Based on the model data of each third level, testing the performance of the first-level sample data on each third-level model;

[0026] Statistical the first performance of the first-level sample data on each third-level model and the second performance of the first-level sample data on the first-level original model;

[0027] Based on the comparison between the first performance and the second performance, determine whether to update the first-level original model data. When it is determined that the first-level original model data needs to be updated, update its own model data based on the data of multiple third-level models whose first performance meets the preset conditions and the first-level original model data.

[0028] In some embodiments, based on the comparison between the first performance and the second performance, determining whether to update the first-level original model data includes:

[0029] Obtaining the generation time of each third-level model data uploaded to the first level at the third level;

[0030] Respectively obtaining the distance between the first performance and the second performance corresponding to each third-level model. The performance is characterized by multiple index dimensions. In the calculation of the distance, weights are obtained for each dimension data in the performance based on preset requirements, the optimized performance is obtained based on the weights, and the distance is calculated based on the optimized performance;

[0031] Determine the result of whether to update the first-level original model data based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data.

[0032] In some embodiments, the determining the result of whether to update the first-level original model data based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data includes:

[0033] Using the difference vector of the feature vectors of the first performance situation and the second performance situation corresponding to each third-level model as the distance, and based on the distance and the generation time of the model data of each third-level model to form input data, inputting the input data into a preset analysis model to obtain the result of whether to update the first-level original model data, specifically including:

[0034] Taking the first-level model as the central node and each third-level model as the node connected to the central node to form a model performance graph; the node data includes the performance situation of the model and the generation time of the model data; characterizing the data distribution of the distances corresponding to each third-level model based on the model performance graph;

[0035] Obtaining a node feature matrix and a node adjacency matrix based on the model performance graph, where the feature of each node in the node feature matrix is determined based on the performance situation of the node model and the generation time of the model data;

[0036] Inputting the model performance graph into a preset graph neural network to obtain the result of whether to update the first-level original model data.

[0037] In some embodiments, the determining the result of whether to update the first-level original model data based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data includes:

[0038] Obtain the distances corresponding to each third-level model and the generation time of the model data;

[0039] Determine the influence degree of the model parameters of the third-level model on the result of whether to update the first-level original model data based on the generation time of the third-level model data. The longer the generation time of the third-level model is from the current time, the smaller the influence degree;

[0040] Aggregate the distances corresponding to each third-level model based on the influence degree of the model parameters of the third-level model on the result of whether to update the first-level original model data. When the aggregation result is greater than a preset distance threshold, determine to update the first-level original model data.

[0041] Second aspect, a substation intelligent service model management system based on an Internet of Things management platform is provided. The system includes:

[0042] A first-level management unit, configured to transmit a first-level substation intelligent inspection system model to a second level based on a first-level artificial intelligence "two libraries and one platform";

[0043] A second-level management unit, configured to send the substation intelligent inspection system model to the Internet of Things management platform through the HTTP Restful protocol based on a second-level artificial intelligence "two libraries and one platform";

[0044] A transfer management unit, where the Internet of Things management platform provides an HTTP Restful interface service to push messages to MQS; and provides an SDK access specification;

[0045] A third-level management unit, based on the MQS message, the third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the Internet of Things management platform, and deploys the model in the third-level substation intelligent inspection system to process the local intelligent inspection tasks of the substation.

[0046] The substation intelligent service model management method and system based on the Internet of Things management platform of the present invention have the following beneficial effects: The first-level model is sent to the second level based on two levels of artificial intelligence "two libraries and one platform", and the second level sends the model to the Internet of Things management platform, and the selected model data is pushed to the third-level substation intelligent inspection system to realize model synchronization, supporting the in-situ analysis and application of the algorithm model at the station end; The second-level model data is uploaded to the first level based on two levels of artificial intelligence "two libraries and one platform" to realize data sharing and reference of multiple third-level station-end model data and three different-level model data, realize collaborative learning and common iterative update of multi-level models, improve the accuracy and real-time iterative update of the model, and meet the changing business requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a schematic flowchart of a substation intelligent service model management method based on an Internet of Things management platform provided by an embodiment of the present application;

[0048] Figure 2 is a schematic flowchart of a method for the second level to update the substation intelligent inspection system model based on the collected sample data of the third level locally in an embodiment of the present application;

[0049] Figure 3 is a schematic flowchart of a process for the first level to update its own model in an embodiment of the present application;

[0050] Figure 4 is a schematic structural diagram of a substation intelligent service model management system based on an Internet of Things management platform provided by an embodiment of the present application. Specific Embodiments

[0051] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0052] See Figure 1 , a substation intelligent service model management method based on an Internet of Things management platform provided by an embodiment of the present application includes the following steps:

[0053] Step 1: Transmit the first-level substation intelligent inspection system model to the second level based on the first-level artificial intelligence "two libraries and one platform".

[0054] Step 2: Based on the second-level artificial intelligence "two libraries and one platform", send the substation intelligent inspection system model to the Internet of Things management platform through the HTTP Restful protocol.

[0055] Step 3: The Internet of Things management platform provides an HTTP Restful interface service to push messages to the MQS and provides an SDK access specification.

[0056] Step 4: Based on the MQS message, the third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the Internet of Things management platform and deploys the model in the third-level substation intelligent inspection system to process the local intelligent inspection tasks of the substation.

[0057] Specifically, in the embodiment of the present application, the first level is the headquarters artificial intelligence platform. The headquarters artificial intelligence platform includes a "sample library", a "model library", and an "operation platform". The "sample library" contains sample data related to substation intelligent inspection data uploaded by all second-level platforms within its jurisdiction. The "model library" includes at least one related model for implementing the substation intelligent inspection function.

[0058] The second level is the jurisdiction scope under the headquarters, and the third level is the jurisdiction scope under the second level. For example, the second level is the provincial-side artificial intelligence platform. The provincial-side artificial intelligence platform also includes a provincial "sample library", a "model library", and an "operation platform". The "sample library" contains sample data related to substation intelligent inspection data uploaded by all third-level system platforms within its jurisdiction. The "model library" includes at least one related model for implementing the substation intelligent inspection function.

[0059] The third level is the substation terminal. The substation intelligent inspection system at the substation terminal includes an inspection host and an intelligent analysis host. The inspection host is used to summarize and manage all the collected substation inspection data, including video data, audio data, text data, etc. The intelligent analysis host is used to analyze the inspection results by using at least one relevant model with the function of substation intelligent inspection. It can be understood that the third-level system platform in this application can also be other terminal system models for providing other business services at the terminal side.

[0060] In the embodiment of this application, based on the two-level artificial intelligence "two libraries and one platform", the headquarters distributes the model to the provincial side, and the provincial side distributes the model to the Internet of Things management platform, and pushes the selected model data to the substation intelligent inspection system to realize model synchronization, supporting the in-situ analysis and application of the algorithm model at the terminal side; based on the two-level artificial intelligence "two libraries and one platform", each provincial company uploads the model data to the headquarters to realize the unified management of the high-quality model library.

[0061] The provincial-side artificial intelligence "two libraries and one platform" provides functions for adding, deleting, and modifying the information of each substation intelligent inspection system to realize the management of substation information; supports distributing the model to the corresponding substation, and transmits the model information to the substation intelligent inspection system according to the managed substation address; supports monitoring the model distribution status and can view the model distribution status. On the side of the Internet of Things management platform, register the application information and create a Topic (message subscription topic), and bind the application to the Topic. The provincial-side artificial intelligence "two libraries and one platform" sends information such as the model name, model OSS address, model startup command, and model unique identification ID of the model to the Internet of Things management platform - MQS through the HTTP Restful method. After receiving the message, the Internet of Things management platform stores the message in the Topic bound to the application and sends a message to the substation intelligent inspection system.

[0062] The substation intelligent inspection system is accessed through the SDK based on the JAVA language provided by the Internet of Things management platform, listens for messages on the same Topic as the artificial intelligence "two libraries and one platform" from the Internet of Things management platform - MQS. When the substation intelligent inspection system receives a message from the Internet of Things management platform, it can obtain the algorithm model information with the same structure sent by the artificial intelligence "two libraries and one platform" to the Internet of Things management platform in the message content.

[0063] In one implementation manner, in step 4 above, based on the MQS message, the substation intelligent inspection system at the third level pulls the substation intelligent inspection system model from the Internet of Things management platform, including the following steps:

[0064] Step 41, obtain the parameters of the latest version of the substation intelligent inspection system model issued by the first level to the Internet of Things management platform. The parameters include multi-dimensional technical parameters, performance index parameters, and function parameters of the model.

[0065] Step 42: Compare the parameters of the substation intelligent inspection system model deployed inside the current third-level substation intelligent inspection system, and determine whether it is necessary to pull the substation intelligent inspection system model from the IoT management platform. It can be understood that after the parameter comparison, if there are differences or the difference items meet the preset update conditions, it is necessary to pull the substation intelligent inspection system model from the IoT management platform to update its own model data.

[0066] In one implementation, in the above step 4, after pulling the substation intelligent inspection system model from the IoT management platform, it further includes:

[0067] Step 43: Test the performance indicators of the current model based on the local sample data. If the performance indicators do not meet the standards, send a message to the second level, and the second level updates the substation intelligent inspection system model based on the collected local sample data of the third level.

[0068] See Figure 2 In one implementation, in the above step 43, the second level updates the substation intelligent inspection system model based on the collected local sample data of the third level, including:

[0069] Step 431: The second-level artificial intelligence "two libraries and one platform" collects the parameters of the substation intelligent inspection system models of each third level, the local sample data, and the performance indicator values of the third-level substation intelligent inspection system models in the local sample data.

[0070] Step 432: Based on the local sample data of the third level with similar performance indicator values of the third-level substation intelligent inspection system model in the local sample data or based on each third-level substation intelligent inspection system model and the local sample data, exchange the local sample data and input it into the corresponding third-level substation intelligent inspection system model to determine the exchanged performance indicator values of each third-level substation intelligent inspection system model.

[0071] Step 433: When the exchanged performance indicator values are similar to the pre-exchange performance indicator values, determine that the corresponding local sample data features of the third level are similar; when the exchanged performance indicator values are different from the pre-exchange performance indicator values, determine that the corresponding third-level sample data features are not similar.

[0072] Step 434: Using the network parameters of the third-level substation intelligent patrol system model with the optimal performance index among multiple third-level substation intelligent patrol system models with similar third-level local sample data characteristic parameters and the network parameters of the latest version of the substation intelligent patrol system model sent from the first level to the IoT management platform as references, and using the multiple third-level local sample data with similar third-level local sample data characteristic parameters as training samples together, train and update the network parameters of the third-level model, and update the substation intelligent patrol system model pulled by the IoT management platform.

[0073] In the embodiment of the present application, by interleaving and swapping the sample sets and models, it is determined whether the characteristics of each third-level local sample data are similar, and the same model is trained jointly based on each third-level local sample data with similar sample characteristics, so as to realize the model training for a class of sample characteristics.

[0074] Based on the third-level local sample data with similar performance index values of the third-level substation intelligent patrol system model in the local sample data or based on each third-level substation intelligent patrol system model and local sample data, swap the local sample data and input it into the corresponding third-level substation intelligent patrol system model to determine the performance index values of each third-level substation intelligent patrol system model after the swap. Specifically, the station models and station local samples of multiple stations with similar performance on the local sample data of each station-end model at the current moment can be swapped. The swapping process is as follows: input the non-local sample data, that is, the sample data of multiple other station-ends, into the station-end model of this station-end, and analyze the performance of the sample data of other station-ends in the station-end model of this station-end. If the performance of the sample data of other station-ends in the station-end model of this station-end is the same as or similar to the performance of the local sample data of this station-end in the station-end model of this station-end, it is determined that the data characteristics of the sample data of other station-ends are similar to the local sample data of this station-end.

[0075] Based on the third-level local sample data with similar performance index values of the third-level substation intelligent patrol system model in the local sample data or based on each third-level substation intelligent patrol system model and local sample data, swap the local sample data and input it into the corresponding third-level substation intelligent patrol system model to determine the performance index values of each third-level substation intelligent patrol system model after the swap. Specifically, the station models and station local samples of all station-ends can be directly swapped. The swapping process is as follows: input the non-local sample data, that is, the sample data of multiple other station-ends, into the station-end model of this station-end, and analyze the performance of the sample data of other station-ends in the station-end model of this station-end. If the performance of the sample data of other station-ends in the station-end model of this station-end is the same as or similar to the performance of the local sample data of this station-end in the station-end model of this station-end, it is determined that the data characteristics of the sample data of other station-ends are similar to the local sample data of this station-end.

[0076] In one embodiment, in step 434 above, multiple third-level local sample data with similar third-level local sample data characteristic parameters are jointly used as training samples, including:

[0077] Step 4341, delete duplicates;

[0078] Step 4342, take one local sample data as one data set, perform correlation analysis on multiple sample sets, increase the information entropy of multiple sample sets by deleting relevant items, and re-obtain multiple sample data for training the model. Among them, the correlation analysis of multiple sample sets can be implemented by methods such as the principal component analysis method and the Pearson correlation method.

[0079] In one embodiment, for the substation intelligent service model management method based on the Internet of Things management platform in the embodiments of the present application, the method further includes:

[0080] Step 5, based on the two-level artificial intelligence "two libraries and one platform", upload each second-level model data to the first level to achieve unified management of the high-quality model library;

[0081] Step 6, the first level updates its own model data based on the uploaded third-level model data of each level and the first-level original model data;

[0082] See Figure 3 , the first level updates its own model data based on the uploaded third-level model data of each level and the first-level original model data, including:

[0083] Step 61, based on the model data of each third level, test the performance of the first-level sample data on each third-level model;

[0084] Step 62, count the first performance of the first-level sample data on each third-level model and the second performance of the first-level sample data on the first-level original model;

[0085] Step 63, based on the comparison of the first performance and the second performance, determine whether to update the first-level original model data. When it is determined that the first-level original model data needs to be updated, update its own model data based on the data of multiple third-level models whose first performance meets the preset conditions combined with the first-level original model data.

[0086] In the embodiments of the present application, the headquarters system platform collects and manages all the station - side model data and the sample data collected at the station - side, and forms high - quality models and samples in the headquarters system platform. The headquarters system platform checks the update status of its own model in real time based on the continuously updated station - side model data, so as to achieve the sharing of model data between the headquarters system platform model and the station - side model and their common learning and updating, and continuously improve the intelligent service effect.

[0087] In one implementation, in step 63 above, based on the comparison between the first performance situation and the second performance situation, determining whether to update the first - level original model data includes:

[0088] Step 631, obtaining the generation time of each third - level model data uploaded to the first level at the third level;

[0089] Step 632, respectively obtaining the distances between the first performance situation and the second performance situation corresponding to each third - level model. The performance is characterized based on multiple index dimensions. In the calculation of the distance, weights are obtained for each dimension data in the performance based on preset requirements, and the optimized performance situation is obtained based on the weights. The distance is calculated based on the optimized performance situation;

[0090] Step 633, determining the result of whether to update the first - level original model data based on the data distribution of the distances corresponding to each third - level model and the generation time of the model data.

[0091] In the embodiments of the present application, based on the combination of the gap between the performance of each third - level model in the first - level sample data and the performance of the first - level model and the generation time of each third - level model data, it is determined whether to update the first - level original model data. It can be understood that the greater the gap between the performance of each third - level model in the first - level sample data and the performance of the first - level model, the greater the probability that the first - level original model data needs to be updated. If the generation time of the third - level model data is relatively far from the current time, the decision - making importance of this third - level model for whether to update the first - level model is relatively small. In addition, in the present application, different weight coefficients can also be set for different index dimensions of the model performance. For example, in some scenarios, more emphasis is placed on the prediction accuracy of the model, and in some scenarios, more emphasis is placed on the speed of the prediction result.

[0092] In one implementation, in step 633 above, based on the data distribution of the distances corresponding to each third - level model and the generation time of the model data, determining the result of whether to update the first - level original model data includes:

[0093] Taking the difference vector of the feature vectors of the first performance situation and the second performance situation corresponding to each third-level model as the distance, and based on the distance and the generation time of each third-level model's model data to form input data, inputting it into a preset analysis model to obtain the result of whether to update the first-level original model data, specifically including:

[0094] Step 63301: Taking the first-level model as the central node and each third-level model as the nodes connected to the central node to form a model performance graph; the node data includes the performance situation of the model and the generation time of the model data; characterizing the data distribution situation of the distance corresponding to each third-level model based on the model performance graph;

[0095] Step 63302: Based on the model performance graph, obtaining a node feature matrix and a node adjacency matrix, where the feature of each node in the node feature matrix is determined based on the performance situation of the node model and the model data generation time;

[0096] Step 63303: Inputting the model performance graph into a preset graph neural network to obtain the result of whether to update the first-level original model data.

[0097] The graph convolutional network (GCN) takes the node feature matrix H and the adjacency matrix A as inputs, and outputs high-dimensional node feature representations after multiple convolutions. Its inter-layer propagation can be expressed as:

[0098] ; where f() represents a non-linear mapping function, L is the number of hidden layers, represents the node feature matrix output by the -th hidden layer. If = 0, then = X, where X represents the input node feature matrix. , where is the adjacency matrix after adding self-connections, is the degree matrix of the adjacency matrix ;

[0099] For each layer of graph convolutional layer, its process is to multiply the adjacency matrix A and the node feature matrix and then perform a linear transformation using the trainable parameter matrix . That is, the inter-layer propagation can be written as:

[0100] , () represents the activation function.

[0101] In the embodiments of the present application, a neural network is used to mine the data distribution of the distances corresponding to each third-level model and the relationship between the generation time of the model data and whether to update the first-level original model data. Taking the first-level model as the central node and each third-level model as the nodes connected to the central node, a model performance graph is constructed. The node features of each node on the model performance graph include the performance data of the first-level model and each third-level model, and the corresponding model data generation time, and include the difference vector information of the feature vectors of the first performance situation and the second performance situation corresponding to each third-level model, that is, the distance information. Using a graph neural network for the model performance graph containing the distance information and the model data generation time data corresponding to the first performance situation and the second performance situation of each third-level model, through node update, the hidden spatio-temporal features of the input data are fully mined to obtain a decision result on whether to update the first-level original model data.

[0102] In one implementation manner, in step 633 above, based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data, the result of determining whether to update the first-level original model data includes:

[0103] Step 63311, obtain the distances and the model data generation times corresponding to each third-level model;

[0104] Based on the generation time of the third-level model data, determine the influence degree of the model parameters of the third-level model on the result of whether to update the first-level original model data. The longer the generation time of the third-level model is from the current time, the smaller the influence degree;

[0105] Step 63312, aggregate the distances corresponding to each third-level model based on the influence degree of the model parameters of the third-level model on the result of whether to update the first-level original model data. When the aggregation result is greater than a preset distance threshold, determine to update the first-level original model data.

[0106] In the embodiments of the present application, based on the generation time of the third-level model data, determine the influence degree of the model parameters of the third-level model on the result of whether to update the first-level original model data, and aggregate the distances corresponding to each third-level model based on the influence degree. The larger the aggregated distance is, the greater the difference between the model parameters of each station terminal and the first-level original model data is. At this time, it is necessary to update the first-level original model data.

[0107] See Figure 4 , based on the above method embodiments, a substation intelligent service model management system based on an Internet of Things management platform provided by the embodiments of the present application includes:

[0108] The first-level management unit is used to transmit the first-level substation intelligent inspection system model to the second level based on the first-level artificial intelligence "two libraries and one platform".

[0109] The second-level management unit is used to send the substation intelligent inspection system model to the IoT management platform through the HTTP Restful protocol based on the second-level artificial intelligence "two libraries and one platform".

[0110] The transfer management unit. The IoT management platform provides an HTTP Restful interface service to push messages to MQS; and provides an SDK access specification.

[0111] The third-level management unit. Based on the MQS message, the third-level substation intelligent inspection system pulls the substation intelligent inspection system model from the IoT management platform and deploys the model within the third-level substation intelligent inspection system to process the intelligent inspection tasks of the local substation.

[0112] It should be noted that: when the substation intelligent service model management system based on the IoT management platform provided in this embodiment conducts substation intelligent service model management, only the above-mentioned functional units are used for illustration. In actual applications, the above functions can be allocated to different functional units according to needs, that is, the internal structure of the device is divided into different functional units to complete all or part of the functions described above. In addition, the substation intelligent service model management system based on the IoT management platform provided in this embodiment and the method embodiment of the substation intelligent service model management method based on the IoT management platform provided in the above embodiment belong to the same concept. The specific implementation process can be seen in the method embodiment and will not be elaborated here.

[0113] The present invention is not limited to the above specific implementation manners. Those of ordinary skill in the art starting from the above concept and making various transformations without creative labor fall within the protection scope of the present invention.

Claims

1. A method for managing a substation intelligent service model based on an IoT management platform, characterized in that: The steps include: Based on the first-level artificial intelligence "two databases and one platform", the first-level substation intelligent inspection system model is transmitted to the second level; Based on the second-level artificial intelligence "two databases and one platform", the substation intelligent inspection system model is sent to the IoT management platform through the HTTP Restful protocol; The IoT management platform provides HTTP Restful interface service to push messages to MQS; And provide SDK access specifications; Based on MQS messages, the third-level substation intelligent patrol system pulls the substation intelligent patrol system model from the IoT management platform, deploys the model in the third-level substation intelligent patrol system, and processes the local intelligent patrol tasks of the substation; After the substation intelligent patrol system model is pulled from the IoT management platform, it also includes: testing the performance indicators of the current model based on local sample data, and sending a message to the second level if the performance indicators do not meet the standards, and the second level updates the substation intelligent patrol system model based on the collected third-level local sample data; The second level updates the substation intelligent patrol system model based on the collected third-level local sample data, including: the second-level artificial intelligence "two libraries and one platform" collects the parameters of each third-level substation intelligent patrol system model, local sample data and the performance index value of the third-level substation intelligent patrol system model in the local sample data; based on the third-level local sample data with similar performance index values ​​of the third-level substation intelligent patrol system model in the local sample data or based on each third-level substation intelligent patrol system model and local sample data, the local sample data is exchanged and input into the corresponding third-level substation intelligent patrol system model to determine the performance index value of each third-level substation intelligent patrol system model after exchange; the performance index value after exchange is compared with the performance index value before exchange. When the indicator values ​​are similar, it is determined that the corresponding third-level local sample data features are similar; when the performance indicator values ​​after exchange are different from the performance indicator values ​​before exchange, it is determined that the corresponding third-level sample data features are dissimilar; based on the network parameters of the third-level substation intelligent patrol system model with the best performance indicators among multiple third-level substation intelligent patrol system models with similar third-level local sample data feature parameters and the network parameters of the latest version of the substation intelligent patrol system model sent to the Internet of Things management platform from the first level as a reference, and using multiple third-level local sample data with similar third-level local sample data feature parameters as training samples, the network parameters of the third-level model are trained and updated, and the substation intelligent patrol system model pulled by the Internet of Things management platform is updated.

2. The method for managing a substation intelligent service model based on an IoT management platform according to claim 1 is characterized in that: Based on the MQS message, the third-level substation intelligent patrol system pulls the substation intelligent patrol system model from the IoT management platform, including: Obtain the parameters of the latest version of the substation intelligent patrol system model sent to the IoT management platform at the first level, including the multi-dimensional technical parameters, performance index parameters, and functional parameters of the model; Based on the comparison of the parameters of the substation intelligent patrol system model deployed inside the current third-level substation intelligent patrol system, it is determined whether it is necessary to pull the substation intelligent patrol system model to the Internet of Things management platform.

3. The method for managing a substation intelligent service model based on an IoT management platform according to claim 1, characterized in that: The method of using a plurality of third-level local sample data with similar third-level local sample data characteristic parameters as training samples includes: Remove duplicates; Take a local sample data as a data set, perform correlation analysis on multiple sample sets, increase the information entropy of multiple sample sets by deleting related items, and re-acquire multiple sample data for training the model.

4. The method for managing a substation intelligent service model based on an IoT management platform according to claim 1, characterized in that: The method further comprises: Based on the two-level artificial intelligence "two libraries and one platform", each second-level model data is uploaded to the first level to achieve unified management of high-quality model libraries; The first level updates its own model data based on the uploaded third-level model data and the first-level original model data; The first level updates its own model data based on the uploaded third level model data and the first level original model data, including: Based on the model data of each third level, test the performance of the first level sample data on each third level model; Count the first performance of the first-level sample data on each third-level model and the second performance of the first-level sample data on the first-level original model; Based on the comparison between the first performance and the second performance, determine whether to update the first-level original model data. When it is determined that the first-level original model data needs to be updated, the model data of the third-level model based on multiple first performances that meet the preset conditions is combined with the first-level original model data to update its own model data.

5. The method for managing a substation intelligent service model based on an IoT management platform according to claim 4 is characterized in that: The determining whether to update the first-level original model data based on the comparison between the first performance performance and the second performance performance includes: Obtain the generation time of each third-level model data uploaded to the first level at the third level; Respectively obtain the distance between the first performance performance and the second performance performance corresponding to each third-level model, wherein the performance is characterized based on multiple indicator dimensions, and the calculation of the distance includes obtaining weights for each dimension data in the performance based on preset requirements, obtaining optimized performance performance based on the weights, and calculating the distance based on the optimized performance performance; Based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data, it is determined whether to update the result of the first-level original model data.

6. The method for managing a substation intelligent service model based on an IoT management platform according to claim 5 is characterized in that: The result of determining whether to update the first-level original model data based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data includes: Based on the difference vector of the feature vectors of the first performance expression and the second performance expression corresponding to each third-level model as the distance, based on the distance and the generation time of each third-level model data to form input data, input a preset analysis model, and obtain the result of whether to update the first-level original model data, specifically including: The first-level model is used as a central node, and each third-level model is used as a node connected to the central node to form a model performance graph; the node data includes the performance of the model and the generation time of the model data; based on the model performance graph, the data distribution of the distance corresponding to each third-level model is characterized; Acquire a node feature matrix and a node adjacency matrix based on the model performance graph, wherein the feature of each node in the node feature matrix is ​​determined based on the performance of the node model and the model data generation time; Based on the model performance graph, the preset graph neural network is input to obtain the result of whether the first-level original model data needs to be updated.

7. The method for managing a substation intelligent service model based on an IoT management platform according to claim 5 is characterized in that: The result of determining whether to update the first-level original model data based on the data distribution of the distances corresponding to each third-level model and the generation time of the model data includes: Obtaining the distance and model data generation time corresponding to each third-level model; Determine the influence of the model parameters of the third-level model on the result of whether to update the first-level original model data based on the generation time of the third-level model data, the longer the generation time of the third-level model is from the current time, the smaller the influence; Based on the influence of the model parameters of the third-level model on whether to update the first-level original model data, the distances corresponding to the third-level models are aggregated, and when the aggregation result is greater than a preset distance threshold, it is determined to update the first-level original model data.

8. A substation intelligent service model management system based on the Internet of Things management platform, characterized in that: include: The first-level management unit is used to transmit the first-level substation intelligent inspection system model to the second level based on the first-level artificial intelligence "two libraries and one platform"; The second-level management unit is used to send the substation intelligent patrol system model to the IoT management platform through the HTTP Restful protocol based on the second-level artificial intelligence "two libraries and one platform"; The transit management unit, the IoT management platform provides HTTP Restful interface service to push messages to MQS; And provide SDK access specifications; The third-level management unit, based on the MQS message, the third-level substation intelligent patrol system pulls the substation intelligent patrol system model from the IoT management platform, and deploys the model in the third-level substation intelligent patrol system to process the local intelligent patrol task of the substation. After pulling the substation intelligent patrol system model from the IoT management platform, it also includes: testing the performance indicators of the current model based on local sample data, and sending a message to the second level when the performance indicators do not meet the standards. The second level updates the substation intelligent patrol system model based on the collected third-level local sample data; The second level updates the substation intelligent patrol system model based on the collected third-level local sample data, including: the second-level artificial intelligence "two libraries and one platform" collects the parameters of each third-level substation intelligent patrol system model, local sample data and the performance index value of the third-level substation intelligent patrol system model in the local sample data; based on the third-level local sample data with similar performance index values ​​of the third-level substation intelligent patrol system model in the local sample data or based on each third-level substation intelligent patrol system model and local sample data, the local sample data is exchanged and input into the corresponding third-level substation intelligent patrol system model to determine the performance index value of each third-level substation intelligent patrol system model after exchange; the performance index value after exchange is compared with the performance index value before exchange. When the indicator values ​​are similar, it is determined that the corresponding third-level local sample data features are similar; when the performance indicator values ​​after exchange are different from the performance indicator values ​​before exchange, it is determined that the corresponding third-level sample data features are dissimilar; based on the network parameters of the third-level substation intelligent patrol system model with the best performance indicators among multiple third-level substation intelligent patrol system models with similar third-level local sample data feature parameters and the network parameters of the latest version of the substation intelligent patrol system model sent to the Internet of Things management platform from the first level as a reference, and using multiple third-level local sample data with similar third-level local sample data feature parameters as training samples, the network parameters of the third-level model are trained and updated, and the substation intelligent patrol system model pulled by the Internet of Things management platform is updated.

Images