Low-voltage prediction method and system for power distribution area, storage medium and equipment

Abstract

The embodiment of the invention discloses a low-voltage prediction method for a power distribution area, and the method comprises the steps: presetting a feature vector, and obtaining historical data containing the feature vector as historical area data; preprocessing and classifying data in the historical transformer area data to obtain first transformer area data and second transformer area data; respectively training a CNN-LSTM load prediction model and a GNN voltage prediction model according to the first transformer area data and the second transformer area data until corresponding model export conditions are met; determining a CNN-LSTM-GNN low-voltage prediction model by combining the CNN-LSTM load prediction model and the GNN voltage prediction model after the training is completed; and according to the CNN-LSTM-GNN low-voltage prediction model, determining a prediction result of a to-be-predicted power distribution area. According to the method, the prediction result for predicting the power distribution area is obtained through data preprocessing and classification and model selection and training, and an accurate prediction result is provided for voltage fluctuation caused by emergencies.

Description

Technical Field

[0001] The present invention relates to the technical field of low voltage prediction, and in particular to a low voltage prediction method, system, storage medium and device for a distribution station area. Background Art

[0002] In the power system, the distribution substation, as the last level of the power supply network, undertakes the important task of safely and stably distributing high-voltage electricity in the transmission network to end users. However, with the continuous expansion of the power grid and the increasing diversification of user needs, the problem of low-voltage management of the distribution network has gradually become prominent, especially in areas with large fluctuations in power demand.

[0003] In order to better meet the needs of power supply reliability, it is urgent to study and implement effective low-voltage control measures. With the development of intelligent power grids, the automation and information level of power systems are constantly improving, and the difficulty of obtaining node load and voltage data in substations is greatly reduced, which makes it possible to predict voltage based on operating data. By accurately predicting the changing trend of substation voltage, corresponding measures can be taken in advance, such as installing and operating mobile reactive power compensation facilities, to achieve optimal allocation of power resources and low-voltage control of multiple outgoing lines in the region.

[0004] Voltage forecasting technology can be divided into ultra-short-term, short-term, medium-term and long-term forecasts according to the time scale. Among them, ultra-short-term forecasting is mainly aimed at the real-time control needs of the power system, short-term forecasting is suitable for intraday scheduling and planning, while medium-term and long-term forecasts focus more on the long-term change trends and seasonal fluctuations of load characteristics. These forecast results are of great significance for formulating governance strategies and optimizing the configuration of reactive compensation devices. Summary of the invention

[0005] Based on this, it is necessary to propose a low voltage prediction method for distribution station areas to address the above problems.

[0006] A low voltage prediction method for a distribution station area, the method comprising the following steps:

[0007] Preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0008] Preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data;

[0009] According to the first substation data and the second substation data, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained respectively until the corresponding model export conditions are met;

[0010] Determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0011] The prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model.

[0012] In the above scheme, the preprocessing and classification of the data in the historical area data to obtain the first area data and the second area data specifically includes:

[0013] Searching the first substation area data and the second substation area data for data where the load active power is 0 for more than three consecutive points, and deleting the data;

[0014] Determine the missing data in the first station area data and the second station area data, and supplement them according to the linear interpolation method;

[0015] The historical active power data and voltage data in the first substation area data and the second substation area data are respectively subjected to minimum normalization processing and maximum normalization processing, and the data are linearly mapped to a fixed interval.

[0016] In the above scheme, the preprocessing and classification of the data in the historical area data to obtain the first area data and the second area data specifically includes:

[0017] Classify the data in the historical area data according to working days and holidays to obtain the first area data and the second area data;

[0018] The data of the second station area is divided into four categories according to the length of holidays and whether major events occur;

[0019] A training set and a test set are determined according to the first station area data and the second station area data.

[0020] In the above scheme, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the first substation data and the second substation data respectively until the corresponding model export conditions are met, specifically including:

[0021] According to the mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 Evaluate the training status of the CNN-LSTM load forecasting model and the GNN voltage forecasting model:

[0022]

[0023] Where n is the number of samples, yi is the i-th true value, is the ith predicted value, is the average of all true observations.

[0024] In the above scheme, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the first substation data and the second substation data respectively until the corresponding model export conditions are met, specifically including:

[0025] Based on the CNN-LSTM load forecasting model, determine the load data corresponding to the test set of the first substation data and the second substation data, and compare the load data with the real data:

[0026] When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the training of the CNN-LSTM load forecasting model is completed and the model export conditions are met;

[0027] Based on the GNN voltage prediction model, determine the voltage data corresponding to the test set of the first substation data and the second substation data, and compare the voltage data with the real data:

[0028] When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the training of the GNN voltage prediction model is completed and the model derivation conditions are met.

[0029] In the above scheme, the prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model, specifically including:

[0030] The load prediction data predicted by the CNN-LSTM load prediction model is input into the GNN voltage prediction model to obtain the corresponding voltage vector, and the CNN-LSTM-GNN low voltage prediction model is determined;

[0031] The voltage vector includes voltages of several measuring points in the predicted substation area, and the voltage vector is denormalized to obtain a prediction result of the distribution substation area to be predicted.

[0032] In the above solution, the voltage vector is denormalized to obtain the real voltage data, and then the method further includes:

[0033] Analyzing the voltages of the plurality of measurement points to obtain low voltage behavior indicators of each substation;

[0034] According to the low voltage behavior index of each substation, each substation is sorted and the sorting result is output:

[0035]

[0036] Where N is the total number of voltage values ​​predicted for the substation in a fixed time period, Ui is the i-th voltage value predicted for the substation, is the indicator function, 15 is the resolution of the voltage prediction data, U is the rated value of the station area voltage, and ΔU is the dead zone voltage.

[0037] The present application also proposes a low voltage prediction system for a distribution station area, the system comprising: an area data acquisition unit, a data processing unit, a model building unit and a prediction unit;

[0038] The station area data acquisition unit is used to preset a feature vector and acquire historical data containing the feature vector as historical station area data;

[0039] The data processing unit is used to pre-process and classify the data in the historical station area data to obtain the first station area data and the second station area data;

[0040] The model building unit is used to train the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data, respectively, until the corresponding model export conditions are met; and determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0041] The prediction unit is used to determine the prediction result of the distribution station area to be predicted according to the CNN-LSTM-GNN low voltage prediction model.

[0042] The present application also proposes a readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the following steps:

[0043] Preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0044] Preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data;

[0045] According to the first substation data and the second substation data, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained respectively until the corresponding model export conditions are met;

[0046] Determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0047] The prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model.

[0048] The present application also proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the following steps of the computer program:

[0049] Preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0050] Preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data;

[0051] According to the first substation data and the second substation data, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained respectively until the corresponding model export conditions are met;

[0052] Determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0053] The prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model.

[0054] The use of the embodiments of the present invention has the following beneficial effects: presetting a feature vector, and obtaining historical data containing the feature vector as historical substation data; preprocessing and classifying the data in the historical substation data to obtain the first substation data and the second substation data; training the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data respectively until the corresponding model export conditions are met; determining the CNN-LSTM-GNN low voltage prediction model in combination with the trained CNN-LSTM load prediction model and the GNN voltage prediction model; determining the prediction result of the distribution substation to be predicted according to the CNN-LSTM-GNN low voltage prediction model. The present invention adopts a graph neural network (GNN) model to establish the relationship between voltage and load, which can efficiently capture the complex relationship between electrical nodes in the power grid, provide more accurate load input parameters for voltage prediction, and thus obtain more accurate voltage prediction results for evaluating the severity of low voltage in the substation in the future medium and long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. 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 creative work.

[0056] in:

[0057] Figure 1 A schematic diagram of a low voltage prediction method for a distribution station area in one embodiment;

[0058] Figure 2 A schematic diagram of the structure of a CNN-LSTM model in one embodiment;

[0059] Figure 3 A schematic diagram of a GNN graph structure in one embodiment;

[0060] Figure 4 For Figure 3 Schematic diagram of the corresponding adjacency matrix. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] In the following description, a large number of specific details are given in order to provide a more thorough understanding of the present invention; however, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details; in other examples, in order to avoid confusion with the present invention, some technical features known in the art are not described, and it should be understood that the present invention can be implemented in different forms and should not be interpreted as limited to the embodiments presented here; on the contrary, providing these embodiments will make the disclosure thorough and complete and fully convey the scope of the present invention to those skilled in the art.

[0063] The purpose of the terms used herein is only to describe specific embodiments and is not intended to be limiting of the present invention. When used herein, the singular forms "a", "an" and "said / the" are also intended to include the plural forms, unless the context clearly indicates otherwise, and it should also be understood that the terms "consisting of" and / or "comprising", when used in this specification, determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0064] Highly unstable load characteristics will cause distribution substations to often face voltage abnormalities. Especially during high-load periods, low voltage events occur frequently, which will seriously threaten the normal operation of user equipment and the service quality of power supply companies. Traditional voltage forecasting methods are often difficult to accurately capture the complex changing characteristics of loads and external influencing factors such as weather, holidays, social events, etc. Therefore, it is necessary to combine advanced data processing technology and machine learning algorithms to build a more accurate and robust prediction model to improve the accuracy and reliability of voltage forecasting.

[0065] The present invention proposes a medium- and long-term prediction method for low voltage in a 10kV distribution station area based on a CNN-LSTM-GNN hybrid model, aiming to achieve accurate prediction of the voltage change trend in the distribution station area by comprehensively considering multiple influencing factors, and to provide a scientific basis for the configuration and operation of mobile reactive compensation equipment.

[0066] In order to thoroughly understand the present invention, a detailed structure will be presented in the following description to illustrate the technical solution proposed by the present invention; optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementation methods.

[0067] To facilitate understanding, the relevant terms involved in this application are first introduced below.

[0068] (1) CNN-LSTM: CNN-LSTM is a deep learning model that combines convolutional neural network (CNN) and long short-term memory network (LSTM). This model combines the advantages of CNN and LSTM and can process spatial features and time series data at the same time. Therefore, it has been widely used in many fields.

[0069] (2) GNN, which is used to process graph structured data. For example, a graph in graph theory consists of nodes (vertex) and edges (edge). Nodes can represent entities, and edges can represent the relationship between entities. GNN updates the representation of nodes through the message passing mechanism between nodes, thereby capturing the structural information of the graph and the dependency relationship between nodes.

[0070] like Figure 1 As shown, in one embodiment, a low voltage prediction method for a power distribution area is provided, and the low voltage prediction method for a power distribution area includes steps S101 to S105, which are described in detail as follows:

[0071] S101, preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0072] The preset feature vectors define the key information required for subsequent analysis. The preset feature vectors ensure the pertinence and relevance of the data, and can capture the key factors affecting low voltage prediction (such as load changes, weather conditions, grid structure, etc.). Acquiring historical data containing these feature vectors provides a rich data source for model training, which is the basis for subsequent analysis and prediction accuracy.

[0073] Preferably, the historical substation data includes historical load data, voltage data, date data, date category data, weather data, and major social events that have an impact on travel behavior:

[0074] Historical load data: should include the active power of each substation of the 10kV outgoing line at the current moment and the serial number of the substation;

[0075] Historical voltage data: should include the voltage value of each substation of the 10kV outgoing line at the current moment and the serial number of the substation;

[0076] Date: includes year, month, day, hour and minute, with a resolution of 15 minutes;

[0077] Date category data: 0 for working days, 1 for non-working days, and 2 for holidays;

[0078] Weather data: including temperature (degrees Celsius), humidity (relative humidity percentage);

[0079] Major social events that affect travel behavior: For example, whether the epidemic is under lockdown or not has a significant impact on tourism load. No major social events are represented by 0, and major social events are represented by 1;

[0080] The historical data of the characteristic vectors listed in S1 required for the 10kV outgoing line are obtained respectively, and the historical data of the past three years and the data before and after major social events (such as epidemics) are collected to cover different situations of the characteristic vectors. The time resolution of the data is 15 minutes.

[0081] S102, preprocessing and classifying the data in the historical area data to obtain the first area data and the second area data;

[0082] Data preprocessing is a key step in improving data quality. It can remove noise, fill missing values, standardize data, etc., thereby reducing deviations and errors during model training. Data classification helps the model to more accurately capture data features under different conditions and improve the model's adaptability and generalization capabilities. The first substation data and the second substation data obtained through classification may represent different time periods, load levels or power grid states, which provides a variety of data samples for subsequent model training.

[0083] In some embodiments, preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data specifically includes:

[0084] In the data of the first and second substation areas, search for data where the load active power is 0 for more than three consecutive points, and delete the data;

[0085] Determine the missing data in the first station area data and the second station area data, and supplement them according to the linear interpolation method;

[0086] The historical active power data and voltage data in the first substation area data and the second substation area data are respectively subjected to minimum normalization processing and maximum normalization processing, and the data are linearly mapped to a fixed interval.

[0087] Specifically, missing data that are not measured in the data set and invalid data deleted in the first step need to be filled in, and linear interpolation is used to fill the missing values. The interpolation formula is as follows:

[0088] y=y1+(x-x1)*(y2-y1) / (x2-x1)

[0089] Among them, x1 and x2 are the horizontal coordinates of the correct measurement points at the same time before and after the missing data corresponding to the time point of the missing data, that is, the measurement time, y1 and y2 are the values ​​of the corresponding correctly measured points, x is the horizontal coordinate of the unknown point, that is, the sampling time corresponding to the data to be supplemented, and y is the measurement value of the measurement point to be supplemented at time x.

[0090] Normalize the data. Perform Min-Max Normalization on the historical active power data and voltage data, and linearly map the data to the [0,1] interval:

[0091]

[0092] Where x* is the value after normalization, and x is the value before normalization. When normalizing the active power data, x is each active power data; when normalizing the voltage data, x is each voltage data, and x maxis the maximum value of the sample data, x min is the minimum value of the sample data.

[0093] Preferably, the total sample is divided into two categories according to working days and holidays, among which the holiday category is further divided into four categories according to the length of the holiday and whether there are major events, namely, long holidays (>3 days) with major events, long holidays (>3 days) without major events, short holidays with major events, and short holidays without major events. The five data sets obtained by the division are divided into training sets and test sets according to the ratio of 9:1.

[0094] In some embodiments, preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data specifically includes:

[0095] Classify the data in the historical area data according to working days and holidays to obtain the first area data and the second area data;

[0096] The data from the second area are divided into four categories according to the duration of holidays and whether major events occur;

[0097] The training set and the test set are determined based on the data from the first station area and the data from the second station area.

[0098] Preferably, the CNN network includes a convolution layer, an activation function, a pooling layer, a fully connected layer, and a Dropout layer. The convolution kernel size of the convolution layer is 3, and the activation function is a ReLU function. The linear features of the data can be extracted by performing linear calculations of element-by-element multiplication and summation of the input data and the convolution kernel. The activation function increases nonlinearity, so that the network can adapt to complex load pattern changes; the pooling function of the pooling layer is Max Pooling, and the pooling window size is 2. The risk of overfitting is reduced and the robustness of the model is improved by reducing the data dimension; the input feature number of the fully connected layer corresponds to the output dimension of the LSTM hidden layer, which is 64, and the output feature number is 1. The obtained data is expanded and converted into a one-dimensional time series sequence and then input into the LSTM for load prediction; Dropout layer: the discard rate is set to 0.5.

[0099] LSTM is a long short-term memory neural network, and its structure is as follows Figure 2 As shown in Figure 1, the basic unit of the LSTM network includes a forget gate, an input gate, and an output gate.

[0100] Among them, X t For input. t is the output hidden state of the network.

[0101] Forget Gate t Calculated by the following formula:

[0102] f t =σ(W f ·[H t-1 ,X t ]+b f )

[0103] Among them, the vector [H t-1 ,X t ], H t-1 is the hidden state vector of the previous step. f is the weight. f is the bias. σ is the activation function.

[0104] Candidate status of the unit Calculated by the following formula:

[0105]

[0106] Among them, W c is the weight. c is the offset.

[0107] Input gate vector i t Calculated by the following formula:

[0108] i t =σ(W i ·[H t-1 ,X t ]+b i )

[0109] Among them, W i is the weight. i is the offset.

[0110] Current cell status C t Calculated by the following formula:

[0111]

[0112] Output gate o t Calculated by the following formula:

[0113] o t =σ(W o ·[H t-1 ,X t ]+b o )

[0114] Among them, W o is the weight. o is the offset.

[0115] The output hidden state H of the network t for:

[0116] H t =ot *tanh(C t )

[0117] The size of the hidden layer in the LSTM unit used in the present invention is 64 units, and the model training adopts a batch size of 256 and 200 training rounds.

[0118] like Figure 2 As shown in the figure, based on the above model, a CNN-LSTM load forecasting model is constructed, where the input layer Xi contains the feature vector at a certain moment, that is, the feature vector defined in step S1, X i The input dimension of is 5, namely load value, date, date category, weather, and major social events that affect travel behavior. n It is the time series of the input quantity; this forecast is a monthly forecast, and the output of the model is the load data of 96 points one month from the current moment.

[0119] S103, training the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data respectively until the corresponding model export conditions are met;

[0120] By using the first substation data and the second substation data to train the CNN-LSTM load forecasting model and the GNN voltage forecasting model respectively, we can make full use of the characteristics of different data sets to enable the model to learn richer information. The training process continuously adjusts the model parameters and optimizes the model structure until the model export conditions are met (such as the prediction error is less than a certain threshold, the training rounds reach the preset value, etc.), ensuring that the model has good prediction performance.

[0121] In some embodiments, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the first substation data and the second substation data respectively until the corresponding model derivation conditions are met, specifically including:

[0122] According to the mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 Evaluate the training of the CNN-LSTM load forecasting model and the GNN voltage forecasting model:

[0123]

[0124] Where n is the number of samples, y i is the ith true value, is the ith predicted value, is the average of all true observations.

[0125] The smaller the mean square error (MSE) and root mean square error (RMSE), the smaller the prediction error and the higher the accuracy of the model prediction; the smaller the mean absolute error (MAE), the smaller the error and the closer the prediction result is to the actual result; and the closer the coefficient of determination (R2) is to 1 or the larger it is, the stronger the explanatory power of the model, the smaller the prediction error and the more accurately the model can capture the variability of the data.

[0126] In some embodiments, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the first substation data and the second substation data respectively until the corresponding model derivation conditions are met, specifically including:

[0127] Based on the CNN-LSTM load forecasting model, determine the load data corresponding to the test set of the first substation data and the second substation data, and compare the load data with the real data:

[0128] When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the CNN-LSTM load forecasting model training is completed and the model export conditions are met;

[0129] Based on the GNN voltage prediction model, determine the voltage data corresponding to the test set of the first substation data and the second substation data, and compare the voltage data with the real data:

[0130] When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the GNN voltage prediction model training is completed and the model export conditions are met.

[0131] If the CNN-LSTM load prediction model and the GNN voltage prediction model do not meet the model export conditions after training, they will return to continue training.

[0132] For CNN-LSTM load forecasting model training: the training sets of the five data sets are converted into a time series format with 96 points in each group, where each point contains a total of five features: load value at that moment, date, date category, weather, and major social events that affect travel behavior. The time series is input for training to obtain the model parameters corresponding to each data set.

[0133] For GNN voltage prediction model training: convert the training set into a time series format of 96 points per group according to bus load, substation i load, and total load of all substations downstream of substation i, and input it into the corresponding nodes in the graph structure of step S3, where each point of bus load includes: bus load loadi, date; each point of substation i load includes: substation i load, date; each point of the total load of all substations downstream of substation i includes: load loadi-ds of all substations downstream of substation i, date, and train each substation to obtain a GNN model.

[0134] S104, determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0135] In this step, by combining the CNN-LSTM load forecasting model and the GNN voltage forecasting model to form a CNN-LSTM-GNN low voltage forecasting model, the advantages of the two models can be combined to improve the accuracy of the overall forecast. Model fusion may involve multiple strategies such as weight allocation and result fusion to ensure that the fused model can more comprehensively capture the spatiotemporal characteristics of the data and improve the stability of the forecast.

[0136] Specifically, a GNN voltage prediction model is constructed, including:

[0137] First, construct the graph structure: all substations are numbered 1 to N from the nearest to the farthest distance from the start of the 10kV line. For substation i, the input includes: bus load loadbus, substation i serial number, substation i load loadi, and the load loadi-ds of all substations downstream of substation i. j>i, the load loadi of the area i, the load loadbus of the 10kV bus, and the total load loadi-ds of all areas downstream of the area i are regarded as three nodes in the graph structure, and the adjacency matrix Aadj is constructed, as follows: Figure 3 , 4 shown.

[0138] Use graph convolution to generate new features: Use a graph neural network to perform a graph convolution operation on the graph structure to aggregate the information features of adjacent nodes by averaging.

[0139]

[0140] in, represents the information in node v after one graph convolution operation; xv represents the original information of node v before the graph convolution operation; f (1)represents the function of aggregating information; N(v) represents the set of all neighbor nodes of node v; W(1) and B(1) represent the parameters of the first layer in the graph convolutional network. Information aggregation function f (1) is a linear layer, then

[0141] Finally, the aggregated information is input into the fully connected layer, and the voltage characteristics of the target node are obtained through mapping in the fully connected layer, and each 96 voltage values ​​are output as a group of time series.

[0142] S105. Determine the prediction result of the distribution substation to be predicted according to the CNN-LSTM-GNN low voltage prediction model.

[0143] By applying the trained CNN-LSTM-GNN low voltage prediction model, low voltage prediction can be performed on the distribution substation area to be predicted, providing strong support for power grid operation and maintenance, equipment protection, user services, etc. The accuracy of the prediction results directly affects the effectiveness of subsequent decision-making and the stability of power grid operation.

[0144] In some embodiments, the prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model, specifically including:

[0145] The load forecast data predicted by the CNN-LSTM load forecast model is input into the GNN voltage forecast model to obtain the corresponding voltage vector, and the CNN-LSTM-GNN low voltage forecast model is determined;

[0146] The voltage vector includes the voltages of several measuring points in the predicted substation area. The voltage vector is denormalized to obtain the prediction result of the distribution substation area to be predicted.

[0147] Preferably, for any 10kV distribution substation, historical data containing the required feature vectors are obtained, and after preprocessing, the data is input into the low voltage prediction system to obtain the prediction result, the result is denormalized to obtain the real voltage data, and the low voltage behavior index of each substation is calculated based on the predicted voltage data. According to the low voltage behavior index of each substation, each substation is sorted, and the sorting result is output.

[0148] In some embodiments, for any 10 kV distribution station area, assuming that the current time is t, the detailed process of predicting the low voltage situation in the time period t1 to t2 (t1, t2>t) is as follows:

[0149] First, train the model: obtain historical data from the past three years and data before and after major social events (such as epidemics); then preprocess the data and divide the data set; after that, through training, obtain the low voltage prediction model of the current substation and the model parameters of different data sets.

[0150] Then make a prediction: classify each day from t1 to t2 according to step S3, select the model parameters of the corresponding class and load them into the CNN-LSTM prediction model to obtain the predicted load data of 96 points for each day from t1 to t2, input the predicted load data of 96 points for each substation for each day into the trained GNN model, and obtain the voltage vector of each substation for each day from t1 to t2. The voltage vector contains the voltage of 96 measurement points of a substation. The voltage vector is denormalized to obtain the real voltage data, and the substation voltage in the time period from t1 to t2 is analyzed to calculate the low voltage behavior index of each substation. According to the low voltage behavior index of each substation, each substation is sorted and the sorting result is output.

[0151] In some embodiments, the voltage vector is denormalized to obtain the real voltage data, and then the method further includes:

[0152] Analyze the voltages at several measurement points to obtain low voltage behavior indicators for each substation;

[0153] According to the low voltage behavior index of each substation, sort each substation and output the sorting results:

[0154]

[0155] Where N is the total number of voltage values ​​predicted for the substation in a fixed time period, U i is the ith voltage value predicted for this area, is the indicator function, 15 is the resolution of the voltage prediction data, U is the rated value of the station area voltage, and ΔU is the dead zone voltage.

[0156] Among them, when u≤U-ΔU holds, the indicator function is 1; otherwise it is 0, 15 represents that the resolution of the voltage prediction data is 15 minutes, the unit of T is minutes, U is the rated value of the substation voltage, and ΔU is a very small value, representing the dead zone voltage.

[0157] In summary, the present application can predict the load and voltage of each substation every 15 minutes in the next month, rather than just predicting typical days. It can more effectively calculate the statistical characteristics of low voltage in the substation in the next month, and achieve the optimal decision on the installation location and capacity of the mobile reactive compensation device. It can avoid the complexity of the three-phase flow calculation of the distribution network, and the calculation errors caused by inaccurate parameters or structural changes due to operation. Through online learning of graph neural networks, the voltage of each substation is predicted by the substation load, and the prediction accuracy is higher. By retaining the characteristic data before and after major social events, the load forecast not only considers the type of load, climate, and regional industrial and agricultural structure, but also takes into account the impact of social events.

[0158] The present application also proposes a low voltage prediction system for a distribution station area, the system comprising: an area data acquisition unit, a data processing unit, a model building unit and a prediction unit;

[0159] A station area data acquisition unit, used for presetting a feature vector and acquiring historical data containing the feature vector as historical station area data;

[0160] A data processing unit, used for preprocessing and classifying the data in the historical area data, and obtaining the first area data and the second area data;

[0161] The model building unit is used to train the CNN-LSTM load prediction model and the GNN voltage prediction model according to the data of the first substation area and the data of the second substation area respectively until the corresponding model export conditions are met; and determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0162] The prediction unit is used to determine the prediction result of the distribution substation to be predicted according to the CNN-LSTM-GNN low voltage prediction model.

[0163] The present application also proposes a readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the following steps:

[0164] Preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0165] Preprocess and classify the data in the historical area data to obtain the first area data and the second area data;

[0166] The CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the data of the first substation area and the second substation area respectively until the corresponding model export conditions are met;

[0167] The CNN-LSTM-GNN low voltage prediction model is determined by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0168] The prediction results of the distribution substation to be predicted are determined based on the CNN-LSTM-GNN low voltage prediction model.

[0169] The present application also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to perform the following steps:

[0170] Preset a feature vector, and obtain historical data containing the feature vector as historical station area data;

[0171] Preprocess and classify the data in the historical area data to obtain the first area data and the second area data;

[0172] The CNN-LSTM load prediction model and the GNN voltage prediction model are trained according to the data of the first substation area and the second substation area respectively until the corresponding model export conditions are met;

[0173] The CNN-LSTM-GNN low voltage prediction model is determined by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model;

[0174] The prediction results of the distribution substation to be predicted are determined based on the CNN-LSTM-GNN low voltage prediction model.

[0175] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0176] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0177] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present application. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. The above disclosure is only a preferred embodiment of the present invention, and of course it cannot be used to limit the scope of the rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope covered by the present invention.

Claims

1. A low voltage prediction method for a distribution station area, characterized in that: The method comprises: Preset a feature vector, and obtain historical data containing the feature vector as historical station area data; Preprocessing and classifying the data in the historical station area data to obtain the first station area data and the second station area data; According to the first substation data and the second substation data, the CNN-LSTM load prediction model and the GNN voltage prediction model are trained respectively until the corresponding model export conditions are met; Determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model; The prediction result of the distribution station area to be predicted is determined according to the CNN-LSTM-GNN low voltage prediction model.

2. The low voltage prediction method for a distribution station area according to claim 1, characterized in that: The preprocessing and classification of the data in the historical area data to obtain the first area data and the second area data specifically includes: Searching the first substation area data and the second substation area data for data where the load active power is 0 for more than three consecutive points, and deleting the data; Determine the missing data in the first station area data and the second station area data, and supplement them according to the linear interpolation method; The historical active power data and voltage data in the first substation area data and the second substation area data are respectively subjected to minimum normalization processing and maximum normalization processing, and the data are linearly mapped to a fixed interval.

3. The low voltage prediction method for a distribution station area according to claim 2, characterized in that: The preprocessing and classification of the data in the historical area data to obtain the first area data and the second area data specifically includes: Classify the data in the historical area data according to working days and holidays to obtain the first area data and the second area data; The data of the second station area is divided into four categories according to the duration of holidays and whether major events occur; A training set and a test set are determined according to the first station area data and the second station area data.

4. The low voltage prediction method for a distribution station area according to claim 3 is characterized in that: The training of the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data respectively until the corresponding model export conditions are met specifically includes: According to the mean square error MSE, root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 Evaluate the training status of the CNN-LSTM load forecasting model and the GNN voltage forecasting model: Where n is the number of samples, yi is the i-th true value, is the ith predicted value, is the average of all true observations.

5. The low voltage prediction method for distribution station area according to claim 4 is characterized in that: The training of the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data respectively until the corresponding model export conditions are met specifically includes: Based on the CNN-LSTM load forecasting model, determine the load data corresponding to the test set of the first substation data and the second substation data, and compare the load data with the real data: When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the training of the CNN-LSTM load forecasting model is completed and the model export conditions are met; Based on the GNN voltage prediction model, determine the voltage data corresponding to the test set of the first substation data and the second substation data, and compare the voltage data with the real data: When the output mean square error MSE < 0.1, root mean square error RMSE < 0.1, mean absolute error MAE < 0.1, and determination coefficient R2 > 0.8, the training of the GNN voltage prediction model is completed and the model derivation conditions are met.

6. The low voltage prediction method for distribution station area according to claim 5, characterized in that: The step of determining the prediction result of the distribution station area to be predicted according to the CNN-LSTM-GNN low voltage prediction model specifically includes: The load prediction data predicted by the CNN-LSTM load prediction model is input into the GNN voltage prediction model to obtain the corresponding voltage vector, and the CNN-LSTM-GNN low voltage prediction model is determined; The voltage vector includes voltages of several measuring points in the predicted substation area, and the voltage vector is denormalized to obtain a prediction result of the distribution substation area to be predicted.

7. The low voltage prediction method for a distribution station area according to claim 6, characterized in that: After denormalizing the voltage vector to obtain real voltage data, the method further includes: Analyzing the voltages of the plurality of measurement points to obtain low voltage behavior indicators of each substation; According to the low voltage behavior index of each substation, each substation is sorted and the sorting result is output: Where N is the total number of voltage values ​​predicted for the substation in a fixed time period, U i is the ith voltage value predicted for this area, is the indicator function, 15 is the resolution of the voltage prediction data, U is the rated value of the station area voltage, and ΔU is the dead zone voltage.

8. A low voltage prediction system for a distribution substation, characterized in that: The system comprises: a station area data acquisition unit, a data processing unit, a model building unit and a prediction unit; The station area data acquisition unit is used to preset a feature vector and acquire historical data containing the feature vector as historical station area data; The data processing unit is used to pre-process and classify the data in the historical station area data to obtain the first station area data and the second station area data; The model building unit is used to train the CNN-LSTM load prediction model and the GNN voltage prediction model according to the first substation data and the second substation data, respectively, until the corresponding model export conditions are met; and determine the CNN-LSTM-GNN low voltage prediction model by combining the trained CNN-LSTM load prediction model and the GNN voltage prediction model; The prediction unit is used to determine the prediction result of the distribution station area to be predicted according to the CNN-LSTM-GNN low voltage prediction model.

9. A readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor is caused to execute the steps of the method according to any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

Images