A training method and system for power grid alarm perception model based on information fusion

Abstract

The present invention discloses a power grid alarm perception model training method and system based on information fusion, which relates to the technical field of power system monitoring and analysis. The method comprises: based on the current round of power grid operation status data, perception results and user-side electricity consumption data and load characteristic results, respectively constructing and training a current load perception model and a current user load model; then performing a potential relationship mapping analysis on the operation and electricity consumption data to determine the load mutual influence distribution, and fusing the two models to obtain a fused load perception model; obtaining the next round of information according to the loss degree value of the model to continue training; if the preset conditions are met, using the fusion model of the last round as the final power grid alarm perception model for real-time load perception; the present invention takes into account both power grid operation and user electricity consumption data, can cope with the diversity and personalized needs of user loads, accurately predict loads, and improve the accuracy of power grid load alarm perception.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring and analysis, and in particular to a power grid alarm perception model training method and system based on information fusion. Background Art

[0002] In the operation and management of power systems, grid load alarm perception is an important link in ensuring the safe, stable and efficient operation of power systems.

[0003] Existing methods primarily rely on power grid load alarm perception models trained using artificial intelligence to detect power grid load alarms. These models can quickly and accurately pre-monitor and analyze power grid load data, and issue timely alarms when load anomalies occur. During training, existing power grid load alarm perception models primarily focus on the overall load data of the power grid, ignoring the diversity and personalized needs of user loads. This results in limitations in the existing models' ability to utilize information on the user load side. Because different users have significantly different electricity usage habits and load characteristics at different times, existing power grid load alarm perception models lack detailed analysis of the user load side, making it impossible to accurately predict and issue alarms based on the user's actual situation, resulting in low accuracy in power grid load alarm perception. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to improve the accuracy of power grid load alarm perception.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for training a power grid alarm perception model based on information fusion, comprising:

[0008] Based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side power consumption data and its corresponding load characteristic results, the current load perception model and the current user load model are constructed and trained respectively;

[0009] Based on the current round of operating status data and user-side power consumption data, potential relationship mapping analysis is performed to determine the load interaction distribution between the grid side and the user side;

[0010] Based on the load mutual influence distribution, the current load perception model and the current user load model are fused to obtain a fused load perception model;

[0011] Based on the model loss degree value of the fusion load perception model, the next round of grid load information and user-side load information is obtained, and the next round of training of the fusion load perception model is carried out;

[0012] If the preset training conditions are met, the fusion load perception model obtained in the last round of training is used as the final power grid alarm perception model, and the power grid is perceived in real time based on the power grid load alarm perception model.

[0013] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0014] Determining the load interaction distribution between the grid side and the user side includes:

[0015] Correlating the operating status data of the current round with the user-side electricity consumption data according to the time series to obtain a correlation vector at each time point in the time series;

[0016] Based on the coupling degree between the grid parameters and the user-side parameters in the correlation vector at each time point, the influence of the user-side parameters on the grid parameters, and the feedback degree of the changes in the grid parameters on the user-side parameters, the dynamic interaction intensity between the grid parameters and the user-side parameters in the correlation vector at each time point is determined;

[0017] Based on the dynamic interaction intensity between the grid parameters and the user side parameters at each time point, the load interaction distribution between the grid side and the user side in the time series is determined.

[0018] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0019] Determining the load interaction distribution between the grid side and the user side under the time series includes:

[0020] Determine the comprehensive change rate between the grid parameters and the user-side parameters based on the parameter change rate of the grid parameters at each time point and the parameter change rate of the user-side parameters at each time point;

[0021] Determine the time-accumulated interaction intensity between grid parameters and user-side parameters based on the comprehensive change rate and the dynamic interaction intensity at each time point;

[0022] With grid parameters as rows, user-side parameters as columns, and time-accumulated interaction intensity as matrix elements, a load interaction matrix between the grid side and the user side under time series is constructed, and the load interaction matrix is ​​determined as the load interaction distribution between the grid side and the user side under time series.

[0023] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0024] The fusion of the current load perception model and the current user load model to obtain the fused load perception model includes:

[0025] Based on the load mutual influence distribution, the first model feature in the current load perception model is mapped to the second model feature corresponding to the current user load model to obtain the first mapped feature of the current user load model; based on the load mutual influence distribution, the second model feature is mapped to the corresponding first model feature to obtain the second mapped feature of the current load perception model;

[0026] constructing an initial hidden layer based on the first mapped feature and the second mapped feature; constructing an initial output layer based on the first output feature of the current load perception model and the second output feature of the current user load model;

[0027] The initial hidden layer is updated based on the correlation between the first output feature and the second output feature, and the initial output layer is updated based on the feature change rate between the first output feature and the second output feature to obtain a fusion load perception model.

[0028] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0029] The updating of the initial hidden layer based on the correlation between the first output feature and the second output feature, and the updating of the initial output layer based on the feature change rate between the first output feature and the second output feature to obtain the fused load perception model include:

[0030] determining a feedback adjustment coefficient of the initial hidden layer based on a correlation between the first output feature and the second output feature and an output of a last hidden layer in the initial hidden layer;

[0031] Based on the second output feature and the feedback adjustment coefficient, each layer in the initial hidden layer is updated to obtain each optimized hidden layer;

[0032] The initial output layer is updated based on the feature change rate to obtain an updated output layer;

[0033] The fused load-aware model is obtained based on the optimized hidden layer and updated output layer of each layer.

[0034] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0035] The step of obtaining the next round of grid load information and user-side load information based on the model loss degree value of the fused load perception model and performing the next round of training on the fused load perception model includes:

[0036] Determine a first number of model loss values ​​whose values ​​are greater than or equal to a preset threshold;

[0037] A model loss degree value of the fused load-aware model is determined based on the first number and the second number of sample test data.

[0038] As a preferred solution for the training method of power grid alarm perception model based on information fusion, the following are the methods:

[0039] The real-time grid load perception of the grid according to the grid load alarm perception model includes:

[0040] Obtain the grid load information of the current grid and the user side load information of the user load end connected to the grid;

[0041] Inputting the current grid load information and user-side load information into the grid load alarm perception model, and obtaining the current grid load perception result output by the grid load alarm perception model;

[0042] If the grid load is greater than or equal to a preset load threshold, an overload alarm is issued.

[0043] In a second aspect, an embodiment of the present invention provides a power grid alarm perception model training system based on information fusion, comprising:

[0044] The model acquisition module is used to build and train the current load perception model and the current user load model based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side power consumption data and its corresponding load characteristic results;

[0045] The data mapping analysis module is used to perform potential relationship mapping analysis based on the operating status data and user-side power consumption data in the current round to determine the load interaction distribution between the grid side and the user side;

[0046] a model fusion module, configured to fuse the current load perception model and the current user load model based on the load mutual influence distribution to obtain a fused load perception model;

[0047] The model testing module is used to obtain the next round of grid load information and user-side load information based on the model loss degree value of the fusion load perception model, and conduct the next round of training for the fusion load perception model;

[0048] The cyclic training module is used to use the fusion load perception model obtained in the last round of training as the final power grid alarm perception model if the preset training conditions are met, and to perform real-time power grid load perception on the power grid based on the power grid load alarm perception model.

[0049] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0050] memory and processor;

[0051] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the power grid alarm perception model training method based on information fusion as described in any embodiment of the present invention.

[0052] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the information fusion-based power grid alarm perception model training method.

[0053] Beneficial effects of the present invention: The information fusion-based power grid alarm perception model training method provided by the present invention, during the model training process, according to the load mutual influence distribution between the power grid operation status data and the user-side electricity consumption data of the user load end, the front load perception model trained based on the power grid operation status data and the current user load model trained based on the user-side electricity consumption data of the user load end are integrated to obtain a power grid load alarm perception model that ultimately meets the preset training conditions. Therefore, the final power grid load alarm perception model not only takes into account the power grid operation status data, but also takes into account the user-side electricity consumption data of the user load end. Therefore, even in the face of the diversity and personalized needs of the load on the user load end side, as well as the differences in electricity consumption habits and load characteristics of different users at different times, accurate load forecast perception can be performed according to the actual situation of the user, thereby improving the accuracy of power grid load alarm perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0055] Figure 1 It is an overall flow chart of the power grid alarm perception model training method based on information fusion described in the present invention. DETAILED DESCRIPTION

[0056] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only 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 persons in this field without creative work should fall within the scope of protection of the present invention.

[0057] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0058] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0059] Example 1, with reference to Figure 1 , which is the first embodiment of the present invention, provides a power grid alarm perception model training method based on information fusion, comprising:

[0060] S1: Based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side electricity consumption data and its corresponding load characteristic results, the current load perception model and the current user load model are constructed and trained respectively.

[0061] S2: Based on the current round of operating status data and user-side power consumption data, potential relationship mapping analysis is performed to determine the load interaction distribution between the grid side and the user side;

[0062] S3: Based on the load mutual influence distribution, the current load perception model and the current user load model are integrated to obtain a fused load perception model;

[0063] S4: Based on the model loss degree value of the fusion load perception model, the next round of grid load information and user-side load information is obtained, and the next round of training of the fusion load perception model is performed;

[0064] S5: If the preset training conditions are met, the fusion load perception model obtained in the last round of training is used as the final power grid alarm perception model, and the power grid is perceived in real time based on the power grid load alarm perception model.

[0065] It should be noted that, through steps S1-S5, in the process of model training, the embodiment of the present invention integrates the front load perception model trained based on the operation status data of the power grid and the current user load model trained based on the user side electricity consumption data of the user load end according to the load mutual influence distribution between the operation status data of the power grid and the user side electricity consumption data of the user load end, so as to obtain the power grid load alarm perception model that finally meets the preset training conditions. Therefore, the final power grid load alarm perception model not only takes into account the operation status data of the power grid, but also takes into account the user side electricity consumption data of the user load end. Even in the face of the diversity and personalized needs of the load on the user load end side, as well as the differences in electricity consumption habits and load characteristics of different users at different times, accurate load forecast perception can be performed according to the actual situation of the user, thereby improving the accuracy of power grid load alarm perception.

[0066] Example 2, reference Figure 1 , which is an embodiment of the present invention, provides a power grid alarm perception model training method based on information fusion based on the previous embodiment, including:

[0067] In the embodiment of the present application, respectively constructing and training the current load perception model and the current user load model in step S1 includes:

[0068] It should be noted that each round in the embodiment of the present invention can be understood as a time sequence. For each round, the grid edge server needs to perform model training based on the grid load information in the current round, and the user load end needs to perform model training based on the user side load information in the current round.

[0069] Specifically, for each round, the grid edge server obtains the operating status data of the grid in the current round and its corresponding current grid perception results, where the operating status data includes the voltage parameters, current parameters, power factor parameters and frequency parameters of the grid.

[0070] The grid edge server performs model training based on the voltage parameters, current parameters, power factor parameters and frequency parameters of the grid in the current round and their corresponding current grid perception results to obtain the current load perception model in the current round.

[0071] Furthermore, for each round, the user-side electricity consumption data of the current round and its corresponding current load characteristic results are obtained through the user load end, wherein the user-side electricity consumption data includes the usage time, usage power, usage frequency and startup time interval of the user-side electrical appliances.

[0072] The user load side performs model training based on the usage time, power usage, usage frequency and startup time interval of the user-side electrical appliances in the current round and their corresponding current load characteristic results to obtain the current user load model in the current round.

[0073] It should be noted that, through this step, the power grid alarm perception system obtains the current load perception model of the power grid edge server in the current round, and the current user load model of the user load end in the current round.

[0074] In the embodiment of the present application, determining the load interaction distribution between the grid side and the user side in the above step S2 includes:

[0075] The power grid alarm perception system performs potential relationship mapping analysis on the operating status data and user-side electricity consumption data in the current round to obtain the load mutual influence distribution between the grid side and the user side.

[0076] Specifically, the operating status data of the current round and the user-side power consumption data are associated according to the time series to obtain the association vector at each time point in the time series.

[0077] Furthermore, each current round (time series) in the embodiment of the present invention ) indicates the operating status data under , the user side electricity consumption data is expressed as ,in, express The grid operation state vector at time , express Therefore, the power grid alarm perception system uses each time point (moment) as a unit to convert the time series At each time point The grid operation state vector under the time series is associated with the user side power consumption vector, and the associated vector at each time point in the time series is obtained. At each time point , the correlation vector The grid operation state vector and user-side power consumption vector It is obtained by splicing in sequence, which is specifically expressed as:

[0078] .

[0079] in, Represents the grid operation state vector The transpose of Indicates the user-side power consumption vector The transpose of express The transpose of .

[0080] Based on the coupling degree of the grid parameters and the user-side parameters in the correlation vector at each time point, the influence of the user-side parameters on the grid parameters, and the feedback degree of the change of the grid parameters on the user-side parameters, the dynamic interaction intensity between the grid parameters and the user-side parameters in the correlation vector at each time point is determined.

[0081] Specifically, the grid parameter in the embodiment of the present invention is the grid operation state vector of the grid in the correlation vector, and the user side parameter is the user side power consumption vector in the correlation vector. Therefore, the grid alarm perception system analyzes the mutual dependence degree of the grid parameter and the user side parameter in the correlation vector at each time point, and determines the coupling degree of the grid parameter and the user side parameter in the correlation vector at each time point. , the formula is as follows:

[0082] .

[0083] The power grid alarm perception system determines the influence of user-side parameters on power grid parameters by analyzing the relative proportion of power grid parameter changes caused by user-side parameter changes. , the specific formula is as follows:

[0084] .

[0085] in, represents the modulo operation, express The grid operation state vector at time The user-side electricity consumption vector at the moment.

[0086] The power grid alarm perception system determines the feedback degree of the power grid parameter change on the user side parameter by analyzing the relative proportion of the user side parameter change caused by the power grid parameter change. , the specific formula is as follows:

[0087] ,

[0088] The power grid alarm perception system is based on the coupling degree at each time point. , influence and feedback , determine the dynamic interaction intensity between grid parameters and user side parameters in the correlation vector at each time point , the specific formula is as follows:

[0089] ,

[0090] Based on the dynamic interaction intensity between the grid parameters and the user side parameters at each time point, the load interaction distribution between the grid side and the user side in the time series is determined.

[0091] Specifically, the power grid alarm perception system constructs a distribution matrix based on the dynamic interaction intensity between the grid parameters and the user-side parameters at each time point, and determines the load interaction distribution between the grid side and the user side in the time series.

[0092] Specifically, based on the parameter change rate of the grid parameter at each time point and the parameter change rate of the user-side parameter at each time point, the comprehensive change rate between the grid parameter and the user-side parameter is determined.

[0093] The power grid alarm perception system calculates the parameter change rate of the power grid parameters at each time point and the parameter change rate of the user side parameters at each time point.

[0094] The power grid alarm perception system determines the comprehensive change rate between the power grid parameters and the user-side parameters in the time series based on the parameter change rate of the power grid parameters at each time point and the parameter change rate of the user-side parameters at each time point. The specific formula is as follows:

[0095] .

[0096] in, Indicates the The grid parameters and The comprehensive change rate between the user-side parameters, express The grid parameters at the moment parameter values, express The grid parameters at the moment parameter values, express The user side parameters at the moment parameter values, express The user side parameters at the moment parameter values.

[0097] Based on the comprehensive change rate and the dynamic interaction intensity at each time point, the time-accumulated interaction intensity between the grid parameters and the user-side parameters is determined.

[0098] The power grid alarm perception system determines the time-accumulated interaction intensity between the power grid parameters and the user-side parameters in the time series based on the comprehensive change rate and the dynamic interaction intensity at each time point. The specific formula is as follows:

[0099] .

[0100] in, Indicates the The grid parameters and The time-accumulated interaction intensity between user-side parameters, represents the number of grid parameters, Indicates the number of user-side parameters.

[0101] With grid parameters as rows, user-side parameters as columns, and time-accumulated interaction intensity as matrix elements, the load interaction matrix between the grid side and the user side under time series is constructed, and the load interaction matrix is ​​determined as the load interaction distribution.

[0102] Specifically, the grid alarm perception system uses grid parameters as rows, user side parameters as columns, and time-accumulated interaction intensity as matrix elements. , construct the load interaction matrix between the grid side and the user side under time series , and the load interaction matrix is ​​determined as the load interaction distribution.

[0103] It should be noted that this step determines the load mutual influence distribution, and then the previous load perception model and the current user load model are integrated according to the load mutual influence distribution, so that the grid load alarm perception model obtained by the integration can comprehensively consider the operating status data of the grid and the user-side electricity consumption data of the user load end, accurately perceive the load of the grid, and improve the accuracy of the grid load alarm perception.

[0104] In the embodiment of the present application, in step S3, the current load perception model and the current user load model are merged to obtain a fused load perception model, which includes:

[0105] The power grid alarm perception system integrates the current load perception model and the current user load model according to the load mutual influence distribution, and obtains a fusion load perception model that integrates the operation status data of the integrated power grid and the user-side power consumption data of the user load end.

[0106] Specifically, based on the load mutual influence distribution, the first model feature in the current load perception model is mapped to the second model feature corresponding to the current user load model to obtain the first mapped feature of the current user load model, and based on the load mutual influence distribution, the second model feature is mapped to the corresponding first model feature to obtain the second mapped feature of the current load perception model.

[0107] Furthermore, the grid alarm perception system extracts the current load perception model The feature space and the current user load model The feature space , where for the current load-aware model , its feature space It is composed of the relevant features of the power grid operation status data, for the current user load model , its feature space It consists of features related to user-side electricity consumption data.

[0108] In order to integrate the current load perception model and the current user load model, it is necessary to map the feature spaces of the current load perception model and the current user load model. Therefore, the power grid alarm perception system is based on the load mutual influence distribution (load mutual influence matrix ) Determine the characteristic coefficient, where the characteristic coefficient represents the importance of the feature in the entire feature space. The specific formula is as follows:

[0109] .

[0110] .

[0111] in, Indicates the operating status data of the power grid The characteristic coefficient of a feature in the entire feature space is Indicates the first The characteristic coefficient of each feature in the feature map, represents the feature dimension, Indicates the operating status data of the power grid The first feature is related to the user-side electricity consumption data The degree of mutual influence between the characteristics Indicates the operating status data of the power grid Features variance, Indicates the operating status data of the power grid The first feature is related to the user-side electricity consumption data The degree of mutual influence between the characteristics Indicates the user side electricity consumption data The variance of a feature.

[0112] The power grid alarm perception system is based on the characteristic coefficient and characteristic coefficients , the current load-aware model The feature space The first model feature in is mapped to the current user load model The feature space The corresponding second model feature is used to obtain the first mapped feature of the current user load model , where the specific mapping formula is as follows:

[0113] ,

[0114] in, Represents the current load-aware model The feature space The The first model feature, Represents the current user load model The feature space The The second model feature.

[0115] Similarly, the power grid alarm perception system uses the characteristic coefficient and characteristic coefficients , the current user load model The feature space The second model feature in is mapped to the current load-aware model The feature space The corresponding first model feature obtains the second mapped feature of the current load sensing model ,in:

[0116] ,

[0117] An initial hidden layer is constructed based on the first mapped feature and the second mapped feature, and an initial output layer is constructed based on the first output feature of the current load perception model and the second output feature of the current user load model.

[0118] Specifically, the embodiment of the present invention uses the first mapped feature and the second mapped feature , and according to the current load perception model The first output feature and the current user load model The second output feature Building a fusion load-aware model The infrastructure includes the initial hidden layer and the initial output layer of the fusion load sensing model. Therefore, it can be understood that setting the fusion load sensing model The input layer is , input layer Integrating the first mapping features and the second mapped feature ,Right now Furthermore, the power grid alarm perception system is combined with the input layer and multi-layer nonlinear transformation to construct the initial hidden layer , where the initial hidden layer The first hidden layer in can be expressed as:

[0119] ,

[0120] in, represents the output of the first hidden layer, Indicates the number of input layer elements, represents the weight coefficient of the first hidden layer, Represents the input layer No. elements, represents the bias term of the first hidden layer.

[0121] Initial hidden layer The The hidden layer can be expressed as:

[0122] ,

[0123] in, Indicates the The output of the hidden layer, Indicates the The weight coefficients of the hidden layers, Indicates the The output of the hidden layer elements, Indicates the The bias term of the hidden layer.

[0124] The power grid alarm perception system is based on the current load perception model The first output feature and the current user load model The second output feature Constructing the initial output layer , where the initial output layer Expressed as:

[0125] .

[0126] in, Represents an activation function, such as a sigmoid function; Represents the last hidden layer The number of elements in the output of Indicates the preset weight coefficient, Represents the last hidden layer The output of elements.

[0127] The initial hidden layer is updated based on the correlation between the first output feature and the second output feature, and the initial output layer is updated based on the feature change rate between the first output feature and the second output feature to obtain a fusion load perception model.

[0128] Specifically, the grid alarm perception system calculates the current load perception model The first output feature and the current user load model The second output feature The correlation between , the specific formula is as follows:

[0129] ,

[0130] in, represents the output dimension, Represents the first output feature Middle The elements of Represents the first output feature The average value of Represents the second output feature Middle The elements of Represents the second output feature The average value of .

[0131] The power grid alarm perception system calculates the current load perception model The first output feature and the current user load model The second output feature The characteristic change rate between , the formula is as follows:

[0132] ,

[0133] in, represents the number of features or output elements, Indicates the operating status data of the power grid The change in the feature, Indicates the operating status data of the power grid Features, Indicates the first The change in the feature, Indicates the first Features, Indicates the first output feature of the current load sensing model The change in the element, Indicates the first output feature of the current load sensing model elements, Indicates the second output feature of the current user load model The change in the element, represents the second output feature of the current user load model, Indicates the output of the last hidden layer. The change in the element, Indicates the output of the last hidden layer. elements.

[0134] The power grid alarm perception system is based on the correlation Update the initial hidden layer and adjust the feature change rate The initial output layer is updated to obtain the final fusion load-aware model.

[0135] Specifically, the feedback adjustment coefficient of the initial hidden layer is determined based on the correlation degree and the output of the last hidden layer in the initial hidden layer.

[0136] Specifically, the power grid alarm perception system is based on the correlation The output of the last hidden layer in the initial hidden layer determines the feedback adjustment coefficient of the initial hidden layer , the specific formula is as follows:

[0137] ,

[0138] in, The number of elements representing the second output feature of the current user load model or the output of the last hidden layer, Indicates the second output feature of the current user load model The element is the output of the last hidden layer. The influence factor of each element, Indicates the second output feature of the current user load model The element is the output of the last hidden layer. element covariance.

[0139] Each layer in the initial hidden layer is updated based on the second output feature and the feedback adjustment coefficient to obtain each optimized hidden layer.

[0140] Specifically, the power grid alarm perception system updates each layer in the initial hidden layer according to the second output feature of the current user load model and the feedback adjustment coefficient to obtain each optimized hidden layer. The update formula of each layer in the initial hidden layer is as follows:

[0141] .

[0142] in, represents the output of the hidden layer after each layer is optimized, represents the initial hidden layer of each layer, The second output feature of the current user load model elements, represents the output of each initial hidden layer. elements.

[0143] The initial output layer is updated based on the feature change rate to obtain an updated output layer.

[0144] Specifically, the power grid alarm perception system updates the initial output layer according to the characteristic change rate and obtains the updated output layer , where the update formula for the initial output layer is as follows:

[0145] .

[0146] A fused load-aware model is obtained based on each optimized hidden layer and updated output layer.

[0147] Specifically, the power grid alarm perception system integrates each optimized hidden layer and updated output layer to obtain a fused load perception model.

[0148] It should be noted that this step integrates the power grid load alarm perception model, so that the final power grid load alarm perception model not only takes into account the operating status data of the power grid, but also takes into account the user-side electricity consumption data of the user load end. Therefore, even in the face of the diversity and personalized needs of the load on the user load end, as well as the differences in electricity consumption habits and load characteristics of different users at different times, accurate load forecast perception can be performed according to the actual situation of the user, thereby improving the accuracy of power grid load alarm perception.

[0149] In the embodiment of the present application, the next round of training of the fusion load perception model in step S4 includes:

[0150] The power grid alarm perception system obtains preset test data, wherein the preset test data includes test power grid load information and test user side load information.

[0151] The power grid alarm perception system obtains a model loss degree value of the fusion load perception model according to the test power grid load information and the test user side load information.

[0152] If the model loss level of the fused load perception model is determined to be less than or equal to a preset loss threshold, the grid alarm perception system determines the current fused load perception model as the final grid load alarm perception model, where the preset loss threshold is set based on actual conditions. If the model loss level of the fused load perception model is determined to be greater than the preset loss threshold, the grid alarm perception system obtains the next round of grid load information and user-side load information for the next round of training.

[0153] Specifically, the sample test data is input into the fusion load perception model to obtain the load prediction perception result of each sample test data output by the fusion load perception model.

[0154] The power grid alarm perception system obtains sample test data of any time series, wherein the sample test data includes sample test power grid load information of the power grid under any time series and sample test user-side load information of the user load end, wherein the sample test power grid load information includes operating status data and its corresponding power grid perception results, and the sample test user-side load information includes user-side electricity consumption data and its corresponding load characteristic results.

[0155] The power grid alarm perception system inputs sample test power grid load information and sample test user side load information into the fusion load perception model, and obtains the load prediction perception result of each sample test data output by the fusion load perception model.

[0156] The loss function based on the fusion load perception model combines the load prediction perception results and load perception label results of each sample test data to obtain the model loss value of each sample test data.

[0157] The power grid alarm perception system inputs the load forecast perception results and load perception label results of each sample test data into the loss function of the fused load perception model to obtain the model loss value of each sample test data. The specific formula of the loss function of the fused load perception model is as follows:

[0158] .

[0159] in, Indicates the The model loss value of the sample test data is Indicates the The load forecast perception results of sample test data, Indicates the Load-aware labeling results of the test data.

[0160] Determine a first number of model loss values ​​whose values ​​are greater than or equal to a preset threshold.

[0161] Based on the first quantity and the total number of sample test data, ie, the second quantity, a model loss degree value of the fused load-aware model is determined.

[0162] The power grid alarm perception system obtains a first number of model loss values ​​whose values ​​are greater than or equal to a preset threshold, wherein the preset threshold is set according to actual conditions, such as the preset threshold is 0.1, 0.05, etc.

[0163] The power grid alarm perception system determines the model loss degree value of the fusion load perception model based on the first quantity and the second quantity of sample test data, wherein the model loss degree value is the quantity ratio, therefore, the model loss degree value = first quantity / second quantity.

[0164] It should be noted that this step calculates the model loss degree value of the fusion load perception model, providing a data basis for subsequent cyclic training based on the model loss degree value, so that the prediction accuracy of the final power grid load alarm perception model is higher, thereby enabling more accurate load perception of the power grid and improving the accuracy of power grid load alarm perception.

[0165] In the embodiment of the present application, in step S5, the fusion load perception model obtained in the last round of training is used as the final power grid alarm perception model, which includes:

[0166] The power grid alarm perception system repeats steps S1 through S4 and determines whether a preset training condition is satisfied. In this embodiment of the present invention, the preset training condition is that the model loss value of the fused load perception model is less than or equal to a preset loss threshold. Therefore, if the preset training condition is determined to be satisfied, the power grid alarm perception system uses the fused load perception model obtained in the last round of training as the final power grid load alarm perception model.

[0167] In the embodiment of the present application, the real-time grid load sensing of the grid according to the grid load alarm sensing model in step S5 includes:

[0168] The power grid alarm perception system obtains the grid load information of the power grid at the current time and the user-side load information of the user load end connected to the power grid, that is, it obtains the voltage parameters, current parameters, power factor parameters and frequency parameters of the power grid at the current time, the usage time, power usage, usage frequency and startup time interval of the user-side electrical appliances at the current time.

[0169] The power grid alarm perception system inputs the current power grid load information and user-side load information into the power grid load alarm perception model, and the power grid load alarm perception model outputs the current power grid load perception result, wherein the power grid load perception result is the power grid load amount at the current time.

[0170] If the grid load sensing results determine that the grid load is greater than or equal to a preset load threshold, the grid alarm sensing system will issue a grid overload alarm. The preset load threshold is set based on actual conditions. If the grid load sensing results determine that the grid load is less than the preset load threshold, the grid alarm sensing system will determine that the grid is currently in a normal state.

[0171] Example 3. The above is a schematic scheme of the power grid alarm perception model training method based on information fusion in this embodiment. It should be noted that the technical scheme of the power grid alarm perception model training system based on information fusion and the technical scheme of the power grid alarm perception model training method based on information fusion are based on the same concept. For details not described in detail in the technical scheme of the power grid alarm perception model training system based on information fusion in this embodiment, please refer to the description of the technical scheme of the power grid alarm perception model training method based on information fusion.

[0172] This embodiment further provides a system for a power grid alarm perception model training method based on information fusion, including:

[0173] The model acquisition module is used to build and train the current load perception model and the current user load model based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side power consumption data and its corresponding load characteristic results;

[0174] The data mapping analysis module is used to perform potential relationship mapping analysis based on the operating status data and user-side power consumption data in the current round to determine the load interaction distribution between the grid side and the user side;

[0175] a model fusion module, configured to fuse the current load perception model and the current user load model based on the load mutual influence distribution to obtain a fused load perception model;

[0176] The model testing module is used to obtain the next round of grid load information and user-side load information based on the model loss degree value of the fusion load perception model, and conduct the next round of training for the fusion load perception model;

[0177] The cyclic training module is used to use the fusion load perception model obtained in the last round of training as the final power grid alarm perception model if the preset training conditions are met, and to perform real-time power grid load perception on the power grid based on the power grid load alarm perception model.

[0178] This embodiment further provides a computing device applicable to a power grid alarm perception model training method based on information fusion, including:

[0179] Memory and processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the power grid alarm perception model training method based on information fusion as proposed in the above embodiment.

[0180] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the power grid alarm perception model training method based on information fusion as proposed in the above embodiment.

[0181] The storage medium proposed in this embodiment and the power grid alarm perception model training method based on information fusion proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0182] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A power grid alarm perception model training method based on information fusion, characterized in that: include: Based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side power consumption data and its corresponding load characteristic results, the current load perception model and the current user load model are constructed and trained respectively; Specifically, for each round, the grid edge server obtains the grid operation status data of the current round and its corresponding current grid perception results, where the operation status data includes the voltage parameters, current parameters, power factor parameters, and frequency parameters of the grid; The grid edge server performs model training based on the voltage parameters, current parameters, power factor parameters, and frequency parameters of the grid in the current round and their corresponding current grid perception results to obtain the current load perception model in the current round; Furthermore, for each round, the user-side electricity consumption data of the current round and its corresponding current load characteristic results are obtained through the user load terminal, wherein the user-side electricity consumption data includes the usage time, power usage, usage frequency and startup time interval of the user-side electrical appliances; The user load side performs model training based on the usage time, power usage, usage frequency, and startup time interval of the user-side electrical appliances in the current round and their corresponding current load characteristics to obtain the current user load model in the current round; Based on the current round of operating status data and user-side power consumption data, potential relationship mapping analysis is performed to determine the load interaction distribution between the grid side and the user side; Based on the load mutual influence distribution, the current load perception model and the current user load model are fused to obtain a fused load perception model; Based on the model loss degree value of the fusion load perception model, the next round of grid load information and user-side load information is obtained, and the next round of training of the fusion load perception model is carried out; If the preset training conditions are met, the fusion load perception model obtained at the end of the training is used as the final grid alarm perception model, and the grid load is perceived in real time based on the grid load alarm perception model; Determining the load interaction distribution between the grid side and the user side includes: Correlating the operating status data of the current round with the user-side electricity consumption data according to the time series to obtain a correlation vector at each time point in the time series; Based on the coupling degree between the grid parameters and the user-side parameters in the correlation vector at each time point, the influence of the user-side parameters on the grid parameters, and the feedback degree of the changes in the grid parameters on the user-side parameters, the dynamic interaction intensity between the grid parameters and the user-side parameters in the correlation vector at each time point is determined; Based on the dynamic interaction intensity between grid parameters and user-side parameters at each time point, the load interaction distribution between the grid side and the user side in the time series is determined; Determining the load interaction distribution between the grid side and the user side under the time series includes: Determine the comprehensive change rate between the grid parameters and the user-side parameters based on the parameter change rate of the grid parameters at each time point and the parameter change rate of the user-side parameters at each time point; Determine the time-accumulated interaction intensity between grid parameters and user-side parameters based on the comprehensive change rate and the dynamic interaction intensity at each time point; With grid parameters as rows, user-side parameters as columns, and time-accumulated interaction intensity as matrix elements, a load interaction matrix between the grid side and the user side under time series is constructed, and the load interaction matrix is ​​determined as the load interaction distribution between the grid side and the user side under time series.

2. The method for training a power grid alarm perception model based on information fusion according to claim 1, characterized in that: The fusion of the current load perception model and the current user load model to obtain the fused load perception model includes: Based on the load mutual influence distribution, the first model feature in the current load perception model is mapped to the second model feature corresponding to the current user load model to obtain the first mapped feature of the current user load model; based on the load mutual influence distribution, the second model feature is mapped to the corresponding first model feature to obtain the second mapped feature of the current load perception model; constructing an initial hidden layer based on the first mapped feature and the second mapped feature; constructing an initial output layer based on the first output feature of the current load perception model and the second output feature of the current user load model; The initial hidden layer is updated based on the correlation between the first output feature and the second output feature, and the initial output layer is updated based on the feature change rate between the first output feature and the second output feature to obtain a fusion load perception model.

3. The method for training a power grid alarm perception model based on information fusion according to claim 2, wherein: The updating of the initial hidden layer based on the correlation between the first output feature and the second output feature, and the updating of the initial output layer based on the feature change rate between the first output feature and the second output feature to obtain the fused load perception model include: determining a feedback adjustment coefficient of the initial hidden layer based on a correlation between the first output feature and the second output feature and an output of a last hidden layer in the initial hidden layer; Based on the second output feature and the feedback adjustment coefficient, each layer in the initial hidden layer is updated to obtain each optimized hidden layer; The initial output layer is updated based on the feature change rate to obtain an updated output layer; The fused load-aware model is obtained based on the optimized hidden layer and updated output layer of each layer.

4. The method for training a power grid alarm perception model based on information fusion according to claim 3, wherein: The step of obtaining the next round of grid load information and user-side load information based on the model loss degree value of the fused load perception model and performing the next round of training on the fused load perception model includes: Determine a first number of model loss values ​​whose values ​​are greater than or equal to a preset threshold; A model loss degree value of the fused load-aware model is determined based on the first number and the second number of sample test data.

5. The method for training a power grid alarm perception model based on information fusion according to claim 4, characterized in that: The real-time grid load perception of the grid according to the grid load alarm perception model includes: Obtain the grid load information of the current grid and the user side load information of the user load end connected to the grid; Inputting the current grid load information and user-side load information into the grid load alarm perception model, and obtaining the current grid load perception result output by the grid load alarm perception model; If the grid load is greater than or equal to a preset load threshold, an overload alarm is issued.

6. A power grid alarm perception model training system based on information fusion, applying the method according to any one of claims 1 to 5, characterized in that: include: The model acquisition module is used to build and train the current load perception model and the current user load model based on the grid operation status data and its corresponding grid perception results in the current round, as well as the user-side power consumption data and its corresponding load characteristic results; Specifically, for each round, the grid edge server obtains the grid operation status data of the current round and its corresponding current grid perception results, where the operation status data includes the voltage parameters, current parameters, power factor parameters, and frequency parameters of the grid; The grid edge server performs model training based on the voltage parameters, current parameters, power factor parameters, and frequency parameters of the grid in the current round and their corresponding current grid perception results to obtain the current load perception model in the current round; Furthermore, for each round, the user-side electricity consumption data of the current round and its corresponding current load characteristic results are obtained through the user load terminal, wherein the user-side electricity consumption data includes the usage time, power usage, usage frequency and startup time interval of the user-side electrical appliances; The user load side performs model training based on the usage time, power usage, usage frequency, and startup time interval of the user-side electrical appliances in the current round and their corresponding current load characteristics to obtain the current user load model in the current round; The data mapping analysis module is used to perform potential relationship mapping analysis based on the operating status data and user-side power consumption data in the current round to determine the load interaction distribution between the grid side and the user side; a model fusion module, configured to fuse the current load perception model and the current user load model based on the load mutual influence distribution to obtain a fused load perception model; The model testing module is used to obtain the next round of grid load information and user-side load information based on the model loss degree value of the fusion load perception model, and conduct the next round of training for the fusion load perception model; A cyclic training module is used to use the fused load perception model obtained at the end of the last round of training as the final power grid alarm perception model if the preset training conditions are met, and to perform real-time power grid load perception on the power grid based on the power grid load alarm perception model; Determining the load interaction distribution between the grid side and the user side includes: Correlating the operating status data of the current round with the user-side electricity consumption data according to the time series to obtain a correlation vector at each time point in the time series; Based on the coupling degree between the grid parameters and the user-side parameters in the correlation vector at each time point, the influence of the user-side parameters on the grid parameters, and the feedback degree of the changes in the grid parameters on the user-side parameters, the dynamic interaction intensity between the grid parameters and the user-side parameters in the correlation vector at each time point is determined; Based on the dynamic interaction intensity between grid parameters and user-side parameters at each time point, the load interaction distribution between the grid side and the user side in the time series is determined; Determining the load interaction distribution between the grid side and the user side under the time series includes: Determine the comprehensive change rate between the grid parameters and the user-side parameters based on the parameter change rate of the grid parameters at each time point and the parameter change rate of the user-side parameters at each time point; Determine the time-accumulated interaction intensity between grid parameters and user-side parameters based on the comprehensive change rate and the dynamic interaction intensity at each time point; With grid parameters as rows, user-side parameters as columns, and time-accumulated interaction intensity as matrix elements, a load interaction matrix between the grid side and the user side under time series is constructed, and the load interaction matrix is ​​determined as the load interaction distribution between the grid side and the user side under time series.

7. A computing device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 5.

Images