Zone area node load prediction method and system based on cloud side-end collaborative architecture

Abstract

The invention relates to a transformer area node load prediction method and system based on a cloud side-end collaborative architecture. The method comprises the following steps: constructing the cloud side-end collaborative architecture; acquiring historical power load data, calculating key load parameters, and transmitting the key load parameters to a side end industrial computer for data preprocessing; constructing a historical power load data set, and storing the historical power load data set to a cloud database; using the training set to train and evaluate an LSTM prediction model based on the MindSpore cloud to obtain a trained LSTM prediction model; carrying out weight pruning and quantization to obtain an optimized LSTM prediction model; acquiring real-time power load data for load prediction to obtain a load prediction result, and transmitting the load prediction result to the cloud; the cloud performs error analysis on the load prediction result and the real-time power load data to obtain an error analysis result; and iteratively performing parameter tuning based on an error analysis result, updating to obtain an optimal LSTM load prediction model, and obtaining an optimal load prediction result.

Description

Technical Field

[0001] The present application relates to the technical field of power system substation area node load forecasting, and mainly relates to a method and system for substation area node load forecasting based on a cloud-edge-end collaborative architecture. Background Art

[0002] With the popularization of smart grid technology, substation area node load forecasting, as one of the key technologies in power demand side management, is playing an increasingly important role. Power node load forecasting not only provides a solid technical guarantee for the safe operation and optimal dispatching of the power grid, but also lays a data foundation for the energy conservation and consumption reduction of the power grid, the integration of distributed energy, and the formulation of demand response strategies.

[0003] However, with the continuous expansion of the power grid scale and the increasing diversification of user electricity consumption behaviors, a series of problems have gradually emerged when traditional load forecasting methods face the complex demands of large-scale distributed substation area nodes. On the one hand, the acquisition of power load data of each node in the substation area is relatively scattered, and the historical power load data collected by remote terminal units has inconsistent formats and uneven quality, which brings difficulties to subsequent analysis and utilization; on the other hand, the previous simple load forecasting models cannot effectively process complex and changeable power load data, and the forecasting accuracy is difficult to meet the growing power dispatching and management requirements. In addition, with the intelligent development of the power system, the requirements for the real-time performance and accuracy of load forecasting are constantly increasing, and the traditional centralized computing mode is difficult to handle the rapid processing of a large amount of data.

[0004] For example, the Chinese invention patent with the publication number of "CN118249313A" discloses a "Method and System for Real-Time Forecasting of Substation Area Load Based on Cloud-Edge Collaboration", specifically discloses that "cluster analysis is performed on the daily load curves of each distribution transformer in the substation area to obtain the daily load curve cluster centers of each distribution transformer; the daily load curve cluster centers of each distribution transformer are integrated to form a load curve data set, and re-cluster analysis is performed on the load curve data set to obtain the global daily load pattern; at least one load forecasting model corresponding to the global daily load pattern is constructed based on a multi-layer LSTM network structure; the real-time operation data of a certain distribution transformer in the future for a period of time is obtained, and the real-time operation data is input into a certain load forecasting model corresponding to the certain distribution transformer, and the certain load forecasting model outputs the predicted load of the certain distribution transformer", but this method does not optimize the constructed load forecasting model and perform feedback adjustment based on the forecasting results, and it is impossible to improve the model with the changes of time and data, and it is difficult to continuously ensure the forecasting accuracy of the model; in addition, this method only constructs a corresponding load forecasting model for the global daily load pattern, and may not consider the special situations and local characteristics of different substations or distribution transformers enough. When the actual operation situation of a certain distribution transformer is quite different from the global pattern, the accuracy of the forecasting result is likely to be affected. Summary of the Invention

[0005] To solve the above problems existing in the prior art, the present application provides a method and system for predicting the load of substation area nodes based on a cloud-edge-terminal collaborative architecture.

[0006] The technical solution of the present application is as follows:

[0007] On the one hand, the present invention proposes a method for predicting the load of substation area nodes based on a cloud-edge-terminal collaborative architecture, and the method includes:

[0008] Construct a cloud-edge-terminal collaborative architecture, including a cloud, a remote terminal unit, and edge devices, where the cloud includes a cloud database and a MindSpore cloud, and the edge devices include a programmable logic controller and an edge industrial computer; obtain the historical power load data of the remote terminal unit in the substation area, and the programmable logic controller calculates the key load parameters based on the historical power load data; transmit the key load parameters to the edge industrial computer and perform data preprocessing; construct a historical power load data set based on the preprocessed key load parameters, divide the historical power load data set into a training set and a test set according to a preset ratio, and store them in the cloud database;

[0009] The MindSpore cloud extracts the training set from the cloud database, uses the training set to train a long short-term memory network (LSTM) prediction model based on the MindSpore cloud, and evaluates it based on the test set and preset performance evaluation indicators to obtain a trained LSTM prediction model; the cloud performs weight pruning and quantization on the trained LSTM prediction model to obtain an optimized LSTM prediction model, and transmits it to the edge industrial computer;

[0010] Obtain real-time power load data, and the edge industrial computer uses the optimized LSTM prediction model to predict the load of the real-time power load data to obtain a load prediction result, and transmit it to the cloud;

[0011] The cloud performs error analysis on the load prediction result and the real-time power load data to obtain an error analysis result; based on the error analysis result, iteratively perform parameter tuning to update and obtain an optimal LSTM load prediction model and obtain an optimal load prediction result.

[0012] Preferably, the programmable logic controller is used to calculate the key load parameters of each node in the substation area, including the power factor and the load rate, specifically:

[0013] Calculate the power factor, which is expressed by the formula:

[0014]

[0015] S = UI;

[0016] In the formula, PF represents the power factor; P represents the active power; S represents the apparent power; U represents the effective value of the power voltage; I represents the effective value of the power current; represents the power factor angle;

[0017] Calculate the load rate, which is expressed by the formula:

[0018]

[0019] In the formula, η represents the load rate; P a represents the actual power; P e represents the rated power.

[0020] Preferably, the data preprocessing includes data cleaning and data normalization, specifically:

[0021] The data cleaning includes processing missing values and outliers. Specifically, the processing of missing values is to delete the invalid data caused by the faults and anomalies of the substation area equipment and complete the filling by using the interpolation method; the processing of outliers is to set the normal data threshold, and determine the data smaller than the normal data threshold as abnormal data and delete it;

[0022] The data normalization is expressed by the formula;

[0023]

[0024] In the formula, represents the key load parameter after normalization; X max represents the maximum value of the key load parameter; X min represents the minimum value of the key load parameter.

[0025] Preferably, the training set is used to train the long short-term memory network LSTM prediction model based on the MindSpore cloud, specifically:

[0026] Convert the format of the training set into the Tensor format specified by the MindSpore cloud, initialize the parameter group of the LSTM prediction model, including the input feature dimension, the number of hidden layer units, the batch size, and the number of LSTM layer networks; and define the loss function and the optimizer;

[0027] The optimizer uses the model.trainable_params() method to obtain the parameters of the LSTM prediction model, and defines the learning rate hyperparameter to initialize the optimizer;

[0028] The loss function is defined as loss_fn = nn.MSELoss();

[0029] Each training includes a training phase and a testing phase. Specifically, the training phase is to find the optimal parameter group that minimizes the loss function; the testing phase is to evaluate based on the test set and the preset performance evaluation metrics to obtain the trained LSTM prediction model and the optimal parameter group.

[0030] The preset performance evaluation metrics include the mean absolute error MAE, the mean square error MSE, and the mean squared logarithmic error MSLE, which are expressed by the formula:

[0031]

[0032]

[0033] In the formula: y i represents the true value of the i-th historical power load data in the test set; represents the predicted value of the i-th historical power load data in the test set; N represents the number of historical power load data in the test set; i represents the index value of the i-th historical power load data.

[0034] Preferably, the cloud performs weight pruning and quantization on the trained LSTM prediction model, specifically:

[0035] Calculate the importance of each weight based on the mindspore.train.serialization module of the MindSpore cloud, and delete the weight connections of the LSTM prediction model whose importance approaches zero;

[0036] Convert the optimized LSTM prediction model into a low-precision format based on the mindspore.nn.Quantize module and the mindspore.nn.Quantization module of the MindSpore cloud.

[0037] Preferably, the cloud performs error analysis on the load prediction result and the real-time power load data to obtain the error analysis result, specifically:

[0038] Recalculate the current mean absolute error and mean square error, and calculate the coefficient of determination, which is expressed by the formula:

[0039]

[0040] In the formula, RES represents the residual sum of squares; TOT represents the total sum of squared deviations; represents the average value of the true values of the i-th historical power load data in the test set;

[0041] If the coefficient of determination is equal to the upper limit of a nearly preset value range, it indicates that the prediction effect of the LSTM prediction model is the best; otherwise, if the coefficient of determination is equal to the lower limit of the nearly preset value range, it indicates that the prediction effect of the LSTM prediction model is the worst.

[0042] Preferably, parameter tuning is iteratively performed based on the error analysis result, and the LSTM prediction model is retrained using the adjusted parameter group.

[0043] In the iterative training process, the error analysis result of each time is transmitted to the cloud for optimizing the LSTM prediction model until the preset performance evaluation index of the LSTM prediction model reaches the preset value or the preset maximum number of iterations is reached, obtaining the optimal LSTM load prediction model and the optimal load prediction result.

[0044] On the other hand, the present invention also proposes a substation node load prediction system based on a cloud-edge-end collaborative architecture. The system includes a data acquisition module, a model training module, a load prediction module, a model optimization module, and a result output module, where:

[0045] A cloud-edge-end collaborative architecture is constructed, including a cloud, a remote terminal unit, and an edge device. The cloud includes a cloud database and a MindSpore cloud, and the edge device includes a programmable logic controller and an edge industrial computer; historical power load data of the remote terminal unit in the substation area is acquired, and the programmable logic controller calculates key load parameters based on the historical power load data; the key load parameters are transmitted to the edge industrial computer and data preprocessing is performed; a historical power load data set is constructed based on the preprocessed key load parameters, and the historical power load data set is divided into a training set and a test set according to a preset ratio and stored in the cloud database.

[0046] The model training module is used for the MindSpore cloud to extract the training set from the cloud database, train a long short-term memory network (LSTM) prediction model based on the MindSpore cloud using the training set, and evaluate based on the test set and a preset performance evaluation index to obtain a trained LSTM prediction model; the cloud performs weight pruning and quantization on the trained LSTM prediction model to obtain an optimized LSTM prediction model and transmits it to the edge industrial computer.

[0047] The load prediction module is used to acquire real-time power load data, and the edge industrial computer uses the optimized LSTM prediction model to perform load prediction on the real-time power load data to obtain a load prediction result and transmits it to the cloud.

[0048] The model optimization module is used to perform error analysis on the load prediction result and the real-time power load data in the cloud to obtain an error analysis result; based on the error analysis result, parameter tuning is iteratively performed to update and obtain an optimal LSTM load prediction model, and an optimal load prediction result is obtained;

[0049] The result output module is used to display the optimal load prediction result.

[0050] On the other hand, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a method for predicting the load of a substation node based on a cloud-edge-end collaborative architecture as described in any embodiment of the present invention.

[0051] On the other hand, the present invention also proposes a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a method for predicting the load of a substation node based on a cloud-edge-end collaborative architecture as described in any embodiment of the present invention.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1) The present invention provides a method and system for predicting the load of a substation node based on a cloud-edge-end collaborative architecture. By using a programmable logic controller to calculate key load parameters and an edge industrial computer to perform data preprocessing, the quality of data processing is improved;

[0054] 2) The present invention provides a method and system for predicting the load of a substation node based on a cloud-edge-end collaborative architecture. The historical power load data set is divided into a training set and a test set according to a preset ratio, and MindSpore cloud is used for training and evaluation, which improves the scientific nature of model training and the accuracy of the model;

[0055] 3) The present invention provides a method and system for predicting the load of a substation node based on a cloud-edge-end collaborative architecture. The edge industrial computer processes power load data in real time and makes predictions, and transmits the results to the cloud, which improves the timeliness of load prediction; the cloud performs error analysis on the load prediction result and real-time data, iteratively performs parameter tuning, and continuously updates the model, enhancing the self-optimization ability of the system and ensuring the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is the flowchart of the method of the embodiment of the present invention;

[0057] Figure 2 is the schematic diagram of the cloud-edge-end collaborative architecture of the embodiment of the present invention. DETAILED DESCRIPTION

[0058] The specific embodiments of the present invention will be described below to facilitate those skilled in the art of this technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of this technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions made using the concept of the present invention are within the scope of protection.

[0059] The present invention provides the following technical solution: A method and system for predicting the load of a substation node based on a cloud-edge-terminal collaborative architecture.

[0060] Embodiment 1

[0061] Specifically refer to Figure 1 , this embodiment provides a method for predicting the load of a substation node based on a cloud-edge-terminal collaborative architecture. The specific steps include:

[0062] S1. Please refer to Figure 2 , construct a cloud-edge-terminal collaborative architecture, including a cloud, a remote terminal unit, and an edge device. The cloud includes a cloud database and a MindSpore cloud, and the edge device includes a programmable logic controller and an edge industrial computer;

[0063] S2. Obtain the historical power load data of the remote terminal unit in the substation area. The programmable logic controller calculates key load parameters based on the historical power load data, including power factor and load rate;

[0064] Calculate the power factor, which is expressed by the formula:

[0065]

[0066] S = UI;

[0067] In the formula, PF represents the power factor; P represents the active power; S represents the apparent power; U represents the effective value of the power voltage; I represents the effective value of the power current; represents the power factor angle;

[0068] Calculate the load rate, which is expressed by the formula:

[0069]

[0070] In the formula, η represents the load rate; P a represents the actual power; P e represents the rated power;

[0071] S3. Transmit the key load parameters to the edge industrial computer and perform data preprocessing; the data preprocessing includes data cleaning and data normalization;

[0072] The data cleaning includes handling missing values and outliers. Specifically, handling missing values means deleting invalid data caused by faults and anomalies of substation area equipment and complementing them using the interpolation method; handling outliers specifically means setting a normal data threshold, and determining and deleting data smaller than the normal data threshold as abnormal data;

[0073] The data normalization is expressed by the formula;

[0074]

[0075] In the formula, represents the key load parameter after normalization; X max represents the maximum value of the key load parameter; X min represents the minimum value of the key load parameter;

[0076] S4. Construct a historical power load dataset based on the preprocessed key load parameters, divide the historical power load dataset into a training set and a test set according to a preset ratio, and store them in the cloud database;

[0077] S5. The MindSpore cloud extracts the training set from the cloud database and uses the training set to train the long short-term memory network (LSTM) prediction model based on the MindSpore cloud;

[0078] Convert the format of the training set into the Tensor format specified by the MindSpore cloud, initialize the parameter group of the LSTM prediction model, including the input feature dimension, the number of hidden layer units, the batch size, and the number of LSTM layer networks; and define the loss function and the optimizer;

[0079] The optimizer obtains the parameters of the LSTM prediction model using the model.trainable_params() method and defines the learning rate hyperparameter to initialize the optimizer;

[0080] The loss function is defined as loss_fn = nn.MSELoss();

[0081] Each training includes a training phase and a test phase. Specifically, the training phase is to find the optimal parameter group that minimizes the loss function; the test phase is to evaluate based on the test set and the preset performance evaluation metrics to obtain the trained LSTM prediction model and the optimal parameter group;

[0082] S6. And evaluate based on the test set and the preset performance evaluation metrics to obtain the trained LSTM prediction model;

[0083] The preset performance evaluation metrics include the Mean Absolute Error (MAE), the Mean Squared Error (MSE), and the Mean Squared Logarithmic Error (MSLE), which are expressed by the formulas as follows:

[0084]

[0085] In the formulas: y i represents the true value of the i-th historical power load data in the test set; represents the predicted value of the i-th historical power load data in the test set; N represents the number of historical power load data in the test set; i represents the index value of the i-th historical power load data;

[0086] S7. The cloud performs weight pruning and quantization on the trained LSTM prediction model, calculates the importance of each weight based on the mindspore.train.serialization module of the MindSpore cloud, and deletes the weight connections of the LSTM prediction model with the importance approaching zero, where the weights are specifically the input gate weight, the forget gate weight, and the output gate weight;

[0087] Based on the mindspore.nn.Quantize module and the mindspore.nn.Quantization module of the MindSpore cloud, convert the optimized LSTM prediction model into a low-precision format;

[0088] S8. Obtain the optimized LSTM prediction model and transmit it to the edge industrial computer;

[0089] S9. Obtain real-time power load data. The edge industrial computer uses the optimized LSTM prediction model to perform load prediction on the real-time power load data, obtains the load prediction result, and transmits it to the cloud;

[0090] S10. The cloud performs error analysis on the load prediction result and the real-time power load data to obtain the error analysis result;

[0091] Recalculate the current Mean Absolute Error and Mean Squared Error, and calculate the coefficient of determination, which is expressed by the formulas as follows:

[0092]

[0093] In the formulas, RES represents the residual sum of squares; TOT represents the total sum of squared deviations; represents the average value of the true values of the i-th historical power load data in the test set;

[0094] If the coefficient of determination is equal to the upper limit of the near preset value range, it indicates that the prediction effect of the LSTM prediction model is the best; otherwise, if the coefficient of determination is equal to the lower limit of the near preset value range, it indicates that the prediction effect of the LSTM prediction model is the worst;

[0095] S11. Iteratively perform parameter tuning based on the error analysis results, and retrain the LSTM prediction model using the adjusted parameter group;

[0096] In the iterative training process, transmit the error analysis results of each time to the cloud for optimizing the LSTM prediction model until the preset performance evaluation index of the LSTM prediction model reaches the preset value or reaches the preset maximum number of iterations, update to obtain the optimal LSTM load prediction model, and obtain the optimal load prediction result.

[0097] Embodiment 2

[0098] This embodiment provides a load prediction system for substation node based on cloud-edge-end collaborative architecture. The system includes a data acquisition module, a model training module, a load prediction module, a model optimization module, and a result output module, where:

[0099] The data acquisition module is used to construct a cloud-edge-end collaborative architecture, including a cloud, a remote terminal unit, and an edge device. The cloud includes a cloud database and MindSpore cloud. The edge device includes a programmable logic controller and an edge industrial computer; obtain the historical power load data of the remote terminal unit in the substation area. The programmable logic controller calculates key load parameters based on the historical power load data; transmit the key load parameters to the edge industrial computer and perform data preprocessing; construct a historical power load data set based on the preprocessed key load parameters, divide the historical power load data set into a training set and a test set according to a preset ratio, and store them in the cloud database

[0100] The model training module is used for the MindSpore cloud to extract the training set from the cloud database, use the training set to train the long short-term memory network (LSTM) prediction model based on the MindSpore cloud, and evaluate based on the test set and preset performance evaluation indicators to obtain the trained LSTM prediction model; the cloud performs weight pruning and quantization on the trained LSTM prediction model to obtain an optimized LSTM prediction model, and transmits it to the edge industrial computer;

[0101] The load prediction module is used to obtain real-time power load data. The edge industrial computer uses the optimized LSTM prediction model to perform load prediction on the real-time power load data, obtain a load prediction result, and transmit it to the cloud;

[0102] The model optimization module is used to perform error analysis on the load prediction result and the real-time power load data in the cloud to obtain an error analysis result; based on the error analysis result, iteratively perform parameter tuning, update to obtain the final LSTM prediction model, and obtain the optimal load prediction result;

[0103] The result output module is used to display the optimal load prediction result.

[0104] Embodiment 3

[0105] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements a method for predicting the load of a substation node based on a cloud-edge-end collaborative architecture as described in any embodiment of the present invention.

[0106] Embodiment 4

[0107] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements a method for predicting the load of a substation node based on a cloud-edge-end collaborative architecture as described in any embodiment of the present invention.

[0108] It should be noted that the systems, electronic devices, and computer-readable storage media described in the present invention are all based on the same principle as the method described in Embodiment 1, and will not be elaborated here.

[0109] The above are only embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A method for predicting node load in a substation based on a cloud-edge-end collaborative architecture, characterized in that: The method comprises: Construct a cloud-edge collaborative architecture, including a cloud, a remote terminal unit, and an edge device, wherein the cloud includes a cloud database and a MindSpore cloud, and the edge device includes a programmable logic controller and an edge industrial computer; obtain historical power load data of the remote terminal unit in the substation area, and the programmable logic controller calculates key load parameters based on the historical power load data; transmit the key load parameters to the edge industrial computer and perform data preprocessing; construct a historical power load data set based on the preprocessed key load parameters, divide the historical power load data set into a training set and a test set according to a preset ratio, and store them in the cloud database; The MindSpore cloud extracts the training set from the cloud database, uses the training set to train the long short-term memory network LSTM prediction model based on the MindSpore cloud, and evaluates it based on the test set and preset performance evaluation indicators to obtain a trained LSTM prediction model; the cloud performs weight pruning and quantization on the trained LSTM prediction model to obtain an optimized LSTM prediction model, and transmits it to the edge industrial computer; Acquire real-time power load data. The edge industrial computer uses the optimized LSTM prediction model to perform load prediction on the real-time power load data, obtains load prediction results, and transmits them to the cloud; The cloud performs error analysis on the load forecast results and the real-time power load data to obtain error analysis results; iteratively optimizes parameters based on the error analysis results, updates the optimal LSTM load forecast model, and obtains the optimal load forecast results.

2. According to the method of claim 1, the node load prediction method based on the cloud-edge-end collaborative architecture is characterized in that: The programmable logic controller calculates key load parameters based on historical power load data, including power factor and load rate, specifically: The power factor is calculated as follows: S = UI; In the formula, PF represents power factor; P represents active power; S represents apparent power; U represents the effective value of power voltage; I represents the effective value of the power current; Indicates the power factor angle; The load rate is calculated using the formula: Where, η represents the load rate; P a Indicates actual power; P e Indicates rated power.

3. According to a method for predicting node load in a substation based on a cloud-edge-end collaborative architecture according to claim 1, it is characterized in that: The data preprocessing includes data cleaning and data normalization, specifically: The data cleaning includes processing missing values ​​and abnormal values, wherein the processing of missing values ​​specifically includes deleting invalid data caused by equipment failure and abnormality in the substation area, and using interpolation to complete the data; the processing of abnormal values ​​specifically includes setting a normal data threshold, determining data less than the normal data threshold as abnormal data and deleting it; The data is normalized and expressed as follows: In the formula, represents the normalized key load parameter; X max Indicates the maximum value of the key load parameter; X min Indicates the minimum value of the key load parameter.

4. According to the method of claim 1, the node load prediction method based on cloud-edge-end collaborative architecture is characterized in that: The training set is used to train the LSTM prediction model based on the MindSpore cloud. Specifically: Convert the format of the training set into the Tensor format specified by the MindSpore cloud, initialize the parameter group of the LSTM prediction model, including the input feature dimension, the number of hidden layer units, the batch size, and the number of LSTM layer network layers; and define the loss function and optimizer; The optimizer uses the model.trainable_params() method to obtain the parameters of the LSTM prediction model and defines the learning rate hyperparameter to initialize the optimizer; The loss function is defined as loss_fn = nn.MSELoss(); Each training includes a training phase and a testing phase, wherein the training phase is specifically to find the optimal parameter group that minimizes the loss function; the testing phase is specifically to evaluate based on the test set and the preset performance evaluation index to obtain the trained LSTM prediction model and the optimal parameter group; The preset performance evaluation indicators include mean absolute error MAE, mean square error MSE and mean square logarithmic error MSLE, which are expressed as follows: Where: y i represents the true value of the i-th historical power load data in the test set; represents the predicted value of the i-th historical power load data in the test set; N represents the number of historical power load data in the test set; i represents the index value of the i-th historical power load data.

5. According to a method for predicting node load in a substation based on a cloud-edge-end collaborative architecture according to claim 1, it is characterized in that: The cloud performs weight pruning and quantization on the trained LSTM prediction model, specifically: The importance of each weight is calculated based on the mindspore.train.serialization module of the MindSpore cloud, and the weight connections of the LSTM prediction model whose importance approaches zero are deleted; The optimized LSTM prediction model is converted into a low-precision format based on the mindspore.nn.Quantize module and the mindspore.nn.Quantization module of the MindSpore cloud.

6. According to a method for predicting node load in a substation based on a cloud-edge-end collaborative architecture according to claim 1, it is characterized in that: The cloud performs error analysis on the load forecast result and the real-time power load data to obtain the error analysis result, which is specifically: Recalculate the current mean absolute error and mean square error, and calculate the coefficient of determination, expressed as: In the formula, RES represents the residual sum of squares; TOT represents the total deviation sum of squares; represents the average value of the true value of the i-th historical power load data in the test set; If the determination coefficient is equal to the upper limit of the preset value range, it means that the prediction effect of the LSTM prediction model is the best; otherwise, if the determination coefficient is equal to the lower limit of the preset value range, it means that the prediction effect of the LSTM prediction model is the worst.

7. According to a method for predicting node load in a substation based on a cloud-edge-end collaborative architecture according to claim 1, it is characterized in that: Iteratively perform parameter tuning based on the error analysis results, and retrain the LSTM prediction model using the adjusted parameter group; During the iterative training process, each error analysis result is transmitted to the cloud for LSTM prediction model optimization until the preset performance evaluation index of the LSTM prediction model reaches the preset value or reaches the preset maximum number of iterations, and the optimal LSTM load prediction model is obtained to obtain the optimal load prediction result.

8. A node load prediction system based on cloud-edge-end collaborative architecture, characterized in that: The system includes a data acquisition module, a model training module, a load forecasting module, a model optimization module and a result output module, wherein: The data acquisition module is used to build a cloud-edge collaborative architecture, including a cloud, a remote terminal unit and an edge device, wherein the cloud includes a cloud database and a MindSpore cloud, and the edge device includes a programmable logic controller and an edge industrial computer; historical power load data of the remote terminal unit in the substation area is obtained, and the programmable logic controller calculates key load parameters based on the historical power load data; the key load parameters are transmitted to the edge industrial computer and data preprocessing is performed; a historical power load data set is constructed based on the preprocessed key load parameters, and the historical power load data set is divided into a training set and a test set according to a preset ratio, and stored in the cloud database; The model training module is used for the MindSpore cloud to extract the training set from the cloud database, use the training set to train the long short-term memory network LSTM prediction model based on the MindSpore cloud, and evaluate it based on the test set and preset performance evaluation indicators to obtain the trained LSTM prediction model; the cloud performs weight pruning and quantization on the trained LSTM prediction model to obtain an optimized LSTM prediction model, and transmits it to the edge industrial computer; The load prediction module is used to obtain real-time power load data. The edge industrial computer uses the optimized LSTM prediction model to perform load prediction on the real-time power load data, obtains the load prediction result, and transmits it to the cloud; The model optimization module is used in the cloud to perform error analysis on the load forecast results and the real-time power load data to obtain error analysis results; iteratively perform parameter tuning based on the error analysis results, update the optimal LSTM load forecast model, and obtain the optimal load forecast result; The result output module is used to display the optimal load forecasting result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements a substation node load prediction method based on a cloud-edge-end collaborative architecture as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements a method for predicting substation node load based on a cloud-edge-end collaborative architecture as described in any one of claims 1 to 7.

Images