Power load prediction method and device, electronic equipment and medium

Abstract

Embodiments of the present application provide a power load prediction method and device, electronic equipment and medium, comprising: obtaining historical power load sequence data of a target power grid system, wherein the historical power load sequence data contains power load information and temperature information; extracting a spatial feature vector and a time feature vector, the spatial feature vector being a multi-channel vector obtained by spatial feature extraction on the historical power load sequence data, and the time feature vector being a long short-term memory feature vector of at least one channel obtained by time feature extraction on the historical power load sequence data; performing feature fusion on the spatial feature vector and the time feature vector to generate a feature fusion vector; and predicting power load information of a future time point of the target power grid system based on the feature fusion vector. Thus, the power load data with complex characteristics can be processed, and the future power load can be more accurately predicted to ensure the safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a power load prediction method, device, electronic equipment and medium. Background Art

[0002] Power load forecasting is the process of predicting future electricity demand using a variety of data and methods. Power load forecasting is crucial for power system planning, operation, and dispatch. Its results provide a crucial basis for future power system planning, including decisions on the installation, capacity, location, and timing of generators, as well as grid expansion, reconstruction, construction, and development.

[0003] Currently, power load forecasting is primarily achieved through time series models or deep learning models in the field of artificial intelligence. This is typically accomplished by analyzing historical power load data to identify patterns in its changes over time. Time series models, such as the Auto-Regressive Moving Average (ARMA) model, are then used to predict future loads. However, because load fluctuations are not only time-dependent but can also be affected by various factors, such as weather and economic conditions, time series models can only predict periodic characteristics and are unable to address load fluctuations caused by unexpected events, resulting in reduced forecast accuracy.

[0004] With the development of artificial intelligence technology, deep learning models have demonstrated strong nonlinear mapping and learning capabilities, enabling them to handle complex nonlinear relationships. Recurrent Neural Network (RNN) models have shown particular strength in the time series field. Among these, the Long Short-Term Memory (LSTM) model is able to focus on long-term dependencies within a sequence, memorizing and transmitting long-term information through a gating mechanism. This allows it to better handle long-term dependencies and effectively capture long-term trends in power load. However, the LSTM model updates its hidden state at each time step, and information transmission is a relatively smooth process. This can cause localized, sudden key information to be smoothed out, potentially causing the LSTM model to overlook some local key information, impacting prediction accuracy. Summary of the Invention

[0005] The present invention provides a power load prediction method, device, electronic device and medium, which are used to solve the defect of low power load prediction accuracy in the prior art, achieve more accurate power load prediction and improve prediction accuracy.

[0006] The present invention provides a method for predicting electric load, comprising:

[0007] Acquiring historical power load sequence data of a target power grid system, wherein the historical power load sequence data includes power load information and temperature information;

[0008] Extracting a spatial feature vector and a temporal feature vector of the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the temporal feature vector is a long short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data;

[0009] Performing feature fusion on the spatial feature vector and the temporal feature vector to generate a feature fusion vector;

[0010] The power load information of the target power grid system at a future time point is predicted based on the feature fusion vector.

[0011] In one possible implementation, the method further includes:

[0012] Based on a preset time period, historical power load data corresponding to a plurality of historical time points of the target power grid system are collected to obtain historical power load sequence data.

[0013] In one possible implementation, the method further includes:

[0014] Performing spatial feature extraction on the historical power load sequence data through a convolutional neural network to obtain a multi-channel spatial feature vector;

[0015] The historical power load sequence data is subjected to time feature extraction through a long short-term memory network to obtain a time feature vector of at least one channel, where the time feature vector is a long short-term memory feature vector.

[0016] In one possible implementation, the method further includes:

[0017] The convolutional neural network includes a plurality of convolution units arranged in series, and the convolution unit includes a first convolution layer, a second convolution layer and a third convolution layer;

[0018] When there is an i-2th initial eigenvalue, convolution processing is performed on the i-th initial eigenvalue, the i-1th initial eigenvalue, and the i-2th initial eigenvalue in the initial sequence data by the first convolution layer to generate an i-th first eigenvalue, the initial sequence data including a plurality of initial eigenvalues, and the initial sequence data input to the first convolution layer of the first convolution unit is the historical power load sequence data;

[0019] When the i-2th initial eigenvalue does not exist, performing convolution processing on the i-th initial eigenvalue and the i-1th initial eigenvalue in the initial sequence data through the first convolution layer to generate an i-th first eigenvalue, where i≥2;

[0020] Inputting the first sequence data including each first eigenvalue into the second convolutional layer;

[0021] When there is a j-4th first eigenvalue, convolution processing is performed on the j-th first eigenvalue, the j-2nd first eigenvalue, and the j-4th first eigenvalue in the first sequence data by the second convolution layer to generate a j-th second eigenvalue;

[0022] When there is no j-4th first eigenvalue, convolution processing is performed on the jth first eigenvalue and the j-2th first eigenvalue in the first sequence data by the second convolution layer to generate a jth second eigenvalue, where j≥4;

[0023] Inputting the second sequence data including each second eigenvalue into the third convolutional layer;

[0024] When there is a k-6th second eigenvalue, convolution processing is performed on the k-th second eigenvalue, the k-3th second eigenvalue, and the k-6th second eigenvalue in the second sequence data by the third convolution layer to generate a k-th third eigenvalue;

[0025] When there is no k-6th second eigenvalue, convolution processing is performed on the kth second eigenvalue and the k-3th second eigenvalue in the second sequence data by the third convolution layer to generate a kth third eigenvalue, thereby obtaining third sequence data of each third eigenvalue, where k≥7;

[0026] A multi-channel spatial feature vector is obtained based on the third sequence data output by the last convolution unit.

[0027] In one possible implementation, the method further includes:

[0028] The spatial feature vector and the temporal feature vector are concatenated along the channel dimension through a feature fusion network to generate a feature fusion vector.

[0029] In one possible implementation, the method further includes:

[0030] Performing feature extraction on the feature fusion vector in a time dimension to obtain an intermediate fusion vector;

[0031] Performing feature extraction on the intermediate fusion vector in the channel dimension to obtain a target fusion vector;

[0032] The power load information of the target power grid system at a future time point is predicted according to the target fusion vector.

[0033] In one possible implementation, the method further includes:

[0034] Transpose the feature fusion vector to obtain a first feature vector of C×T;

[0035] Perform feature extraction on the first eigenvector row by row to obtain a second eigenvector of C×T;

[0036] Transpose the second eigenvector to obtain a T×C third eigenvector, and generate the intermediate fusion vector based on the third eigenvector;

[0037] Performing feature extraction on the intermediate fusion vector row by row to generate a T×C fourth feature vector, and generating the target fusion vector based on the fourth feature vector;

[0038] The feature fusion vector size is T×C, and C = c1 + c2, c1 is the number of channels of the spatial feature vector, c1≥2, c2 is the number of channels of the temporal feature vector, c2≥1, and T is the number of eigenvalues ​​of the spatial feature vector and the temporal feature vector.

[0039] The present invention also provides a power load prediction device, comprising the following modules:

[0040] A data acquisition module is used to acquire historical power load sequence data of the target power grid system, wherein the historical power load sequence data includes power load information and temperature information;

[0041] a feature extraction module, configured to extract a spatial feature vector and a temporal feature vector of the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the temporal feature vector is a long short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data;

[0042] A feature fusion module, configured to fuse the spatial feature vector and the temporal feature vector to generate a feature fusion vector;

[0043] A prediction module is used to predict the power load information of the target power grid system at a future time point based on the feature fusion vector.

[0044] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-described power load prediction methods when executing the computer program.

[0045] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which implements any of the above-described power load prediction methods when executed by a processor.

[0046] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-described power load prediction methods.

[0047] The power load prediction method, device, electronic device and medium provided by the present invention obtain historical power load sequence data of the target power grid system, wherein the historical power load sequence data includes power load information and temperature information; extract the spatial feature vector and time feature vector of the historical power load sequence data, the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the time feature vector is a long-short-term memory feature vector of at least one channel obtained by performing time feature extraction on the historical power load sequence data; perform feature fusion on the spatial feature vector and the time feature vector to generate a feature fusion vector; and predict the power load information of the target power grid system at a future time point based on the feature fusion vector. Compared with the existing power load forecasting methods using time series models or deep learning models in the field of artificial intelligence, which are affected by multiple factors such as weather and economic conditions, or ignore some local key information, resulting in low prediction accuracy, this solution combines convolutional neural networks and long short-term memory networks to extract features of different dimensions respectively, and finally fuses the features of different dimensions to generate a feature fusion vector containing multi-dimensional features. Power load forecasting is performed based on the feature fusion vector, which can process power load data with complex characteristics and can more accurately predict future power loads, providing strong guarantees for the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 This is one of the flow charts of the power load forecasting method provided by the present invention.

[0050] Figure 2 This is the second flow chart of the power load forecasting method provided by the present invention.

[0051] Figure 3 It is a structural diagram of the load prediction model provided by the present invention.

[0052] Figure 4 It is a structural diagram of the convolutional neural network provided by the present invention.

[0053] Figure 5 It is a schematic diagram of the principle of the convolution unit provided by the present invention.

[0054] Figure 6 It is a structural diagram of the feature fusion network provided by the present invention.

[0055] Figure 7 It is a schematic diagram of the prediction layer provided by the present invention generating a target fusion vector.

[0056] Figure 8 It is a structural schematic diagram of the power load prediction device provided by the present invention.

[0057] Figure 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] To facilitate understanding of the embodiments of the present invention, specific embodiments will be further explained below with reference to the accompanying drawings. The embodiments do not limit the embodiments of the present invention.

[0060] Figure 1 This is one of the flow charts of the power load forecasting method provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0061] S11. Obtain historical power load sequence data of the target power grid system.

[0062] In an embodiment of the present invention, first, historical sequence data of power load of a target power grid system is acquired, where the historical sequence data includes historical characteristic values ​​at multiple historical time points, and the historical characteristic values ​​include power load values.

[0063] Specifically, based on a preset time period, historical power load data corresponding to multiple historical time points of the target power grid system may be collected to obtain historical power load sequence data, wherein the historical power load data includes power load information and temperature information.

[0064] S12. Extracting the spatial feature vector and the temporal feature vector of the historical power load sequence data.

[0065] The spatial features of the historical power load sequence data are extracted through a convolutional neural network to obtain a multi-channel spatial feature vector; the temporal features of the historical power load sequence data are extracted through a long short-term memory network to obtain a time feature vector of at least one channel, and the time feature vector is a long short-term memory feature vector.

[0066] Specifically, when there is an i-2th initial eigenvalue, the i-th initial eigenvalue, the i-1th initial eigenvalue and the i-2th initial eigenvalue in the initial sequence data are convolved through the first convolution layer to generate the i-th first eigenvalue, the initial sequence data includes multiple initial eigenvalues, and the initial sequence data of the first convolution layer input to the first convolution unit is the historical power load sequence data; when there is no i-2th initial eigenvalue, the i-th initial eigenvalue and the i-1th initial eigenvalue in the initial sequence data are convolved through the first convolution layer to generate the i-th first eigenvalue, where i≥2; the first sequence data containing each first eigenvalue is input to the second convolution layer; when there is a j-4th first eigenvalue, the j-th first eigenvalue, the j-2th first eigenvalue and the j-4th first eigenvalue in the first sequence data are convolved through the second convolution layer to generate the j-th first eigenvalue. second eigenvalue; when the j-4th first eigenvalue does not exist, the j-th first eigenvalue and the j-2th first eigenvalue in the first sequence data are convolved by the second convolution layer to generate the j-th second eigenvalue, where j≥4; the second sequence data containing each second eigenvalue is input into the third convolution layer; when the k-6th second eigenvalue exists, the k-th second eigenvalue, the k-3th second eigenvalue and the k-6th second eigenvalue in the second sequence data are convolved by the third convolution layer to generate the k-th third eigenvalue; when the k-6th second eigenvalue does not exist, the k-th second eigenvalue and the k-3th second eigenvalue in the second sequence data are convolved by the third convolution layer to generate the k-th third eigenvalue, and the third sequence data of each third eigenvalue is obtained, where k≥7; the multi-channel spatial eigenvector is obtained based on the third sequence data output by the last convolution unit.

[0067] The time point eigenvalues ​​corresponding to each time step of the historical power load sequence data are predicted through a long short-term memory network to obtain the hidden state of the next time step; the eigenvectors of multiple time steps are determined based on the hidden state to obtain the time eigenvector of at least one channel.

[0068] S13: Fusing the spatial feature vector and the temporal feature vector to generate a feature fusion vector.

[0069] The spatial feature vector and the temporal feature vector are concatenated along the channel dimension through the feature fusion network to generate a feature fusion vector.

[0070] Specifically, the feature fusion network includes a fusion layer and a prediction layer. The fusion layer is used to fuse the spatial feature vector (a type of convolutional feature vector) and the temporal feature vector (a type of long short-term memory feature vector) in the channel dimension to generate a fusion feature vector. The fusion layer is used to fuse the convolutional feature vector and the long short-term memory feature vector in the channel dimension to generate a fusion feature vector.

[0071] S14. Predicting power load information of the target power grid system at a future time point based on the feature fusion vector.

[0072] Feature extraction is performed on the feature fusion vector in the time dimension to obtain an intermediate fusion vector; feature extraction is performed on the intermediate fusion vector in the channel dimension to obtain a target fusion vector; and the power load information of the target power grid system at a future time point is predicted based on the target fusion vector.

[0073] The power load prediction method provided by the present invention obtains historical power load sequence data of a target power grid system, wherein the historical power load sequence data includes power load information and temperature information; extracts spatial feature vectors and time feature vectors of the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the time feature vector is a long-short-term memory feature vector of at least one channel obtained by performing time feature extraction on the historical power load sequence data; performs feature fusion on the spatial feature vector and the time feature vector to generate a feature fusion vector; and predicts power load information of the target power grid system at a future time point based on the feature fusion vector. Compared with the existing power load forecasting methods using time series models or deep learning models in the field of artificial intelligence, which are affected by multiple factors such as weather and economic conditions, or ignore some local key information, resulting in low prediction accuracy, this method combines convolutional neural networks and long short-term memory networks to extract features of different dimensions respectively, and finally fuses the features of different dimensions to generate a feature fusion vector containing multi-dimensional features. Power load forecasting is performed based on the feature fusion vector, which can process power load data with complex characteristics and can more accurately predict future power loads, providing strong guarantees for the safe and stable operation of the power system.

[0074] Figure 2 This is the second flow chart of the power load forecasting method provided by the present invention, as shown in FIG. Figure 2 As shown, the method includes the following:

[0075] S21. Based on a preset time period, collect historical power load data corresponding to multiple historical time points of the target power grid system to obtain historical power load sequence data.

[0076] In the embodiment of the present invention, the historical power load data includes power load information and temperature information.

[0077] During the operation of power grid systems or power supply equipment such as generators, power load values ​​at corresponding time points can be collected at regular intervals. The power load values ​​are used to indicate the magnitude of the power load, which can specifically be load power. For example, if data is collected every 15 minutes, a total of 96 power load values ​​can be collected within a day (24 hours).

[0078] For each sampling time point, a historical characteristic value, including the corresponding power load value, can be determined. Combining the historical characteristic values ​​at multiple consecutive time points can form corresponding sequence data. Because this sequence data is derived from previously collected data, it is referred to as historical sequence data. Accordingly, the time point at which the power load value was collected is referred to as a historical time point.

[0079] Optionally, the historical characteristic value at each time point may further include: time information corresponding to the historical time point, and temperature information at the historical time point.

[0080] In this embodiment, power loads generally exhibit strong periodicity, and using time information as a historical feature value further facilitates the extraction of temporal features. Many power loads are also related to the actual ambient temperature. For example, power loads in winter and summer are generally higher than those in spring and autumn. Using temperature information as a historical feature value enables the load forecasting model to learn the correlation between temperature and power load, enabling more accurate load forecasting in situations such as sudden temperature changes.

[0081] It should be noted that, for each historical time point, its historical characteristic value may specifically include: the power load value, time information and temperature information at the historical time point.

[0082] S22. Perform spatial feature extraction on the historical power load sequence data through a convolutional neural network to obtain a multi-channel spatial feature vector.

[0083] S23. Perform time feature extraction on the historical power load sequence data through a long short-term memory network to obtain a time feature vector of at least one channel, where the time feature vector is a long short-term memory feature vector.

[0084] In an embodiment of the present invention, a power load prediction model is pre-constructed and trained; when load prediction is required based on historical sequence data, the historical sequence data is input into the power load prediction model, and after the model processes the historical sequence data, the power load value at a future time point after the historical sequence data can be output.

[0085] The future time point may be the next time point after the last historical time point in the historical sequence data. Furthermore, it is possible to predict the power load value at only one future time point, or to predict the power load values ​​at multiple future time points, which is not limited in this embodiment. During model training, for a piece of historical data, the predicted load value L1 is output after the training phase prediction, and the actual load value L2 in the future is used as the data label. A loss function is constructed based on L1 and L2 and minimized. The most commonly used loss function is to calculate the quadratic norm of L1-L2. Conventional methods such as gradient descent are generally used to train the model. When the loss function is minimized, a trained power load prediction model is obtained.

[0086] like Figure 3 As shown, the load prediction model includes a convolutional neural network 201, a long short-term memory network 202 and a feature fusion network 203.

[0087] The convolutional neural network 201 is used to extract spatial features from historical sequence data and determine convolution feature vectors for multiple channels. The long short-term memory network 202 is used to extract temporal features from historical sequence data and determine a long short-term memory feature vector for at least one channel. The feature fusion network 203 is used to fuse the convolution feature vectors and the long short-term memory feature vectors along the channel dimension to generate a fused feature vector. The fused feature vector is then used to predict the power load value at a future point in time.

[0088] In this embodiment, the historical sequence data is used as input data for the convolutional neural network 201 and the long short-term memory network 202, and features of different dimensions are extracted from the historical sequence data based on the convolutional neural network 201 and the long short-term memory network 202, thereby forming feature vectors of different dimensions.

[0089] Specifically, convolution processing is performed on the historical sequence data based on the convolutional neural network 201, thereby achieving feature extraction and determining the corresponding feature vector, i.e., the convolution feature vector. Generally, the convolutional neural network 201 is provided with multiple convolution layers, and through multi-layer convolution processing, it is possible to more comprehensively extract features from the historical sequence data.

[0090] The long short-term memory network 202 can adopt the existing network architecture to predict the characteristic values ​​of the corresponding historical time points at each time step for historical sequence data containing characteristic values ​​of multiple historical time points, and obtain the hidden state of the next time step. Based on these hidden states, the characteristic vectors of multiple time steps, namely the long short-term memory characteristic vectors, can be determined.

[0091] For example, the historical sequence data corresponds to N historical time points, that is, contains N historical feature values. After processing N time steps, the long short-term memory network 202 can obtain N hidden states, thereby generating an N-dimensional long short-term memory feature vector.

[0092] Since the LSTM network 202 primarily extracts features in the time dimension, the feature extraction process implemented by the LSTM network 202 is referred to as temporal feature extraction. The convolution operation in the convolutional neural network 201 is primarily used for static features in historical sequence data. For ease of distinction, the feature extraction process implemented by the convolutional neural network 201 is referred to as spatial feature extraction.

[0093] Furthermore, when performing spatial feature extraction based on convolutional neural network 201, feature extraction is generally performed based on multiple convolution kernels, thereby converting historical sequence data into multi-channel convolution feature vectors. When performing temporal feature extraction based on long short-term memory network 202, a long short-term memory feature vector with a corresponding number of channels can be generated based on the number of LSTM units in long short-term memory network 202. This long short-term memory feature vector is a feature vector with at least one channel.

[0094] For example, the number of channels of the convolution feature vector is c1, and the number of channels of the long short-term memory feature vector is c2, then c1 ≥ 2, c2 ≥ 1. Generally, the long short-term memory network 202 can be provided with multiple LSTM units, that is, c2 ≥ 2.

[0095] like Figure 3 As shown, the convolution feature vector and the long short-term memory feature vector determined by the convolutional neural network 201 and the long short-term memory network 202, respectively, can be fused based on the feature fusion network 203 to obtain a fused feature vector, i.e., a fused feature vector. The feature fusion network 203 fuses the two feature vectors (i.e., the convolution feature vector and the long short-term memory feature vector) along the channel dimension, for example, by concatenating the two feature vectors along the channel dimension, thereby achieving feature vector fusion without information loss.

[0096] After determining a feature fusion vector containing both temporal and spatial features, load prediction based on this feature fusion vector can more accurately determine the power load value at a future point in time. For example, load prediction can be performed based on this feature fusion vector using other LSTM network structures, or using a fully connected layer, although this embodiment is not limited thereto.

[0097] like Figure 4 As shown, the convolutional neural network 201 includes a plurality of convolution units 301 arranged in series; the convolution unit 301 includes a first convolution layer, a second convolution layer and a third convolution layer.

[0098] The first convolutional layer is configured to convolve the i-th initial eigenvalue, the i-1-th initial eigenvalue, and the i-2-th initial eigenvalue in the initial sequence data to generate the i-th first eigenvalue when the i-2-th initial eigenvalue exists; and to convolve the i-th initial eigenvalue and the i-1-th initial eigenvalue in the initial sequence data to generate the i-th first eigenvalue when the i-2-th initial eigenvalue does not exist, where i ≥ 2. The first sequence data containing each first eigenvalue is input into the second convolutional layer; wherein the initial sequence data includes multiple initial eigenvalues, and the initial sequence data input into the first convolution unit 301 is historical sequence data. That is, the initial eigenvalues ​​input into the first convolution unit 301 are the aforementioned historical eigenvalues.

[0099] The second convolutional layer is used to perform convolution processing on the jth first eigenvalue, j-2th first eigenvalue and j-4th first eigenvalue in the first sequence data to generate the jth second eigenvalue when the j-4th first eigenvalue exists; when the j-4th first eigenvalue does not exist, perform convolution processing on the jth first eigenvalue and j-2th first eigenvalue in the first sequence data to generate the jth second eigenvalue, where j≥4; and input the second sequence data containing each second eigenvalue into the third convolutional layer.

[0100] The third convolutional layer is used to perform convolution processing on the kth second eigenvalue, k-3th second eigenvalue and k-6th second eigenvalue in the second sequence data when there is a k-6th second eigenvalue, to generate the kth third eigenvalue; when there is no k-6th second eigenvalue, perform convolution processing on the kth second eigenvalue and k-3th second eigenvalue in the second sequence data to generate the kth third eigenvalue, and obtain third sequence data of each third eigenvalue, where k≥7; the third sequence data containing each third eigenvalue is the output data of the convolution unit 301.

[0101] The numbers of eigenvalues ​​in the initial sequence data, the first sequence data, the second sequence data, and the third sequence data are respectively m, m-1, m-3, and m-6. That is, m is the number of initial eigenvalues ​​in the initial sequence data.

[0102] Figure 5 A schematic diagram of a convolution unit 301 is shown, as shown in FIG. Figure 5 As shown in Figure 1, the initial sequence data includes 10 initial eigenvalues ​​(i.e., m=10), namely a1~a10.

[0103] The first convolution layer performs convolution processing on three consecutive initial eigenvalues ​​(i.e., the i-th initial eigenvalue, the i-1th initial eigenvalue, and the i-2th initial eigenvalue) to obtain the corresponding first eigenvalue and generate the first sequence data. Figure 5 As shown, the first sequence data includes 9 first eigenvalues ​​(ie, m-1=9), namely b2~b10.

[0104] The second convolution layer performs convolution processing on the three first eigenvalues ​​(i.e., the jth first eigenvalue, the j-2th first eigenvalue, and the j-4th first eigenvalue) that are separated by one eigenvalue in sequence to obtain the corresponding second eigenvalue and generate the second sequence data. Figure 5 As shown, the second sequence data includes 7 second eigenvalues ​​(ie, m-3=7), namely d4~d10.

[0105] The third convolution layer performs convolution processing on three second eigenvalues ​​separated by 2 eigenvalues ​​(i.e., the kth second eigenvalue, the k-3th second eigenvalue, and the k-6th second eigenvalue) to obtain the corresponding third eigenvalue and generate the third sequence data. Figure 5 As shown, the third sequence data includes 4 third eigenvalues ​​(i.e., m-6=4), namely h7~h10.

[0106] In this embodiment, when the three convolutional layers (i.e., the first convolutional layer, the second convolutional layer, and the third convolutional layer) perform convolution processing, the input eigenvalues ​​have gradually increasing intervals, so that the final generated third eigenvalue is related to more initial eigenvalues; Figure 5 As shown, the third eigenvalue h10 is derived from the initial eigenvalues ​​a1-a10. Each initial eigenvalue corresponds to the power load value at a historical time point, facilitating load prediction based on data from more time points. Furthermore, the final third eigenvalue is only related to a portion of the initial sequence data. This means that convolution unit 301 primarily extracts local features from the initial sequence data, making it easier to learn local mutation information in the historical sequence data (the initial sequence data of the first convolution unit 301 is the historical sequence data), enabling more accurate load prediction.

[0107] Alternatively, as Figure 4 As shown, the convolutional neural network 201 further includes a screening layer 302 and a one-dimensional convolution layer 303.

[0108] The filtering layer 302 is used to remove the oldest 6n eigenvalues ​​from the historical sequence data to generate intermediate sequence data. The one-dimensional convolution layer 303 is used to perform one-dimensional convolution on the intermediate sequence data to generate one-dimensional sequence data containing multiple eigenvalues. The one-dimensional sequence data is then added to the output data of the last convolution unit 301 to generate a convolution feature vector.

[0109] like Figure 5 As shown, for each convolution unit 301, after processing by three convolution layers, the third sequence data has 6 fewer oldest eigenvalues ​​compared to the initial sequence data; and the convolutional neural network 201 has n convolution units 301, so after processing by each convolution unit 301, the 6 oldest eigenvalues ​​will be reduced. Therefore, after n convolution units 301 process the historical sequence data, the generated output data (i.e., the output data of the last convolution unit 301) has 6 fewer oldest eigenvalues.

[0110] In this embodiment, Figure 4 As shown, a residual structure is introduced into the convolutional neural network 201 to avoid problems such as depth disappearance. In the residual branch, a screening layer 302 is introduced to remove the oldest 6n eigenvalues ​​in the historical sequence data, so that the final output data of the residual branch is the same size as the output data of the last convolution unit 301. In addition, for the intermediate sequence data generated by the screening layer 302, a one-dimensional convolution process is performed on it based on the one-dimensional convolution layer 303 to generate one-dimensional sequence data containing multiple eigenvalues. Finally, the one-dimensional sequence data and the output data of the last convolution unit 301 are added to obtain the final output data of the convolutional neural network 201, that is, the convolution feature vector.

[0111] Optionally, the size of the convolution feature vector is T×c1, and the size of the long short-term memory feature vector is T×c2, where c1 is the number of channels of the convolution feature vector, c1 ≥ 2; c2 is the number of channels of the long short-term memory feature vector, c2 ≥ 1; and T is the number of eigenvalues ​​of the convolution feature vector and the long short-term memory feature vector.

[0112] In this embodiment, the number of eigenvalues ​​of the convolution feature vector and the long short-term memory feature vector is the same, which is T. For example, T can be the number of historical eigenvalues ​​in the historical sequence data. Or, if Figure 3 and Figure 4As shown, the generated convolution feature vector has 6n fewer eigenvalues ​​than the historical sequence data, so T is 6n less than the number of historical eigenvalues, or in other words, the number of historical eigenvalues ​​in the historical sequence data is T+6n.

[0113] The convolution feature vector and the long short-term memory feature vector are set to have the same number of eigenvalues ​​T to facilitate subsequent feature fusion in the channel dimension.

[0114] S24. Concatenate the spatial feature vector and the temporal feature vector along the channel dimension through a feature fusion network to generate a feature fusion vector.

[0115] S25. Perform feature extraction on the feature fusion vector in the time dimension to obtain an intermediate fusion vector.

[0116] S26. Perform feature extraction on the intermediate fusion vector in the channel dimension to obtain a target fusion vector.

[0117] S27. Predicting power load information of the target power grid system at a future time point according to the target fusion vector.

[0118] like Figure 6 As shown, the feature fusion network 203 includes: a fusion layer 501 and a prediction layer 502.

[0119] The fusion layer 501 is used to fuse the convolution feature vector and the long short-term memory feature vector in the channel dimension to generate a feature fusion vector; the size of the feature fusion vector is T×C, and C = c1 + c2.

[0120] The prediction layer 502 is used to first extract features from the feature fusion vector in the time dimension to determine the intermediate fusion vector, and then extract features from the intermediate fusion vector in the channel dimension to determine the target fusion vector; the power load value at the future time point is predicted based on the target fusion vector.

[0121] In this embodiment, the fusion layer 501 concatenates the T×c1 convolution feature vector and the T×c2 LSTM feature vector along the channel dimension, generating a T×C feature vector (where C = c1 + c2), i.e., a fusion feature vector. Furthermore, based on the processing principle of depthwise separable convolution, this embodiment sequentially extracts features from the fusion feature vector along the time dimension and the channel dimension, ultimately obtaining a fusion vector that strongly integrates the convolution feature vector and the LSTM feature vector, i.e., the target fusion vector. Subsequent load forecasting based on this target fusion vector can relatively accurately determine the power load value at a future point in time.

[0122] Optionally, Figure 7Schematic diagram showing the generation of a target fusion vector by the prediction layer 502 , wherein the prediction layer 502 includes: a first transposition layer 5021 , a first feature extraction layer 5022 , a second transposition layer 5023 and a second feature extraction layer 5024 .

[0123] The first transposition layer 5021 is used to perform transposition processing on the feature fusion vector to obtain a C×T first feature vector.

[0124] The first feature extraction layer 5022 is used to perform feature extraction on the first feature vector row by row to obtain a C×T second feature vector.

[0125] The second transposition layer 5023 is used to perform a transposition process on the second eigenvector to obtain a T×C third eigenvector; and generate an intermediate fusion vector based on the third eigenvector.

[0126] The second feature extraction layer 5024 is used to perform feature extraction on the intermediate fusion vector by row to generate a T×C fourth feature vector; and generate a target fusion vector based on the fourth feature vector.

[0127] Among them, when the prediction layer 502 performs feature extraction, a residual structure can also be introduced; Figure 7 As shown, the intermediate fusion vector is obtained by adding the third eigenvector and the feature fusion vector; the target fusion vector is obtained by adding the fourth eigenvector and the intermediate fusion vector.

[0128] The feature fusion vector is the feature vector of T×C. For the convenience of description, different shadows are used to represent the feature values ​​at different time points, and dotted lines are used to divide different channels. Figure 7 Take T = 6 and C = 4 as an example. After the transposition process of the first transposition layer 5021, the T×C feature fusion vector is transposed into a C×T feature vector, that is, the first feature vector.

[0129] At this time, each row of the first eigenvector corresponds to the corresponding channels, and the first feature extraction layer 5022 performs feature extraction on the first eigenvector row by row, so as to extract the features on the time dimension corresponding to each channel, that is, to realize feature extraction on the time dimension, thereby obtaining the second eigenvector of C×T; then, the transposition processing is performed again based on the second transposition layer 5023, and finally the T×C feature vector is restored, that is, the third feature vector, which has the same size as the feature fusion vector.

[0130] Based on the first residual structure, the third eigenvector of T×C and the feature fusion vector of T×C are added to obtain the intermediate fusion vector that extracts the features in the time dimension.

[0131] At this point, each row of the intermediate fusion vector corresponds to a specific time point. The second feature extraction layer 5024 performs feature extraction on the intermediate fusion vector row by row, extracting the channel-dimensional features corresponding to each time point. This achieves feature extraction on the channel dimension, thereby obtaining a T×C fourth eigenvector. Finally, based on the first residual structure, the fourth eigenvector is added to the intermediate fusion vector to obtain the desired target fusion vector. Subsequently, the fully connected layer performs load prediction on the target fusion vector to determine the power load value at a future time point.

[0132] Among them, the first feature extraction layer 5022 and the second feature extraction layer 5024 can be implemented based on a feedforward neural network, and feature extraction is achieved through multiple hidden layers.

[0133] ;

[0134] .

[0135] in, is the first eigenvector, is the second eigenvector; 、 are the two weight matrices in the first feature extraction layer; 、 are the two bias matrices in the first feature extraction layer; is the intermediate fusion vector, is the fourth eigenvector, 、 are the two weight matrices in the second feature extraction layer; 、 are the two bias matrices in the second feature extraction layer; is the activation function, and .

[0136] In this embodiment, the first feature extraction layer 5022 and the second feature extraction layer 5024 are each provided with two hidden layers, and each hidden layer has a corresponding weight matrix and bias matrix. By training the feature fusion network 203, the weight matrix and bias matrix of each hidden layer can be determined. Specifically, the weight matrix of the first hidden layer in the first feature extraction layer can be determined. and the bias matrix , and the weight matrix of the second hidden layer and the bias matrix , similarly, the weight matrix of the first hidden layer in the second feature extraction layer can also be determined and the bias matrix , and the weight matrix of the second hidden layer and the bias matrix .

[0137] And, set the activation function between the two hidden layers , in order to improve the model's ability to predict nonlinear characteristics. Among them, some activation functions are not differentiable at certain positions, so this embodiment sets the activation function based on the probability density function of the normal distribution .

[0138] Specifically, the activation function for: , simplifying it, we can get: The activation function It is differentiable and has a certain degree of continuity.

[0139] The power load prediction method provided by the embodiment of the present invention combines a convolutional neural network and a long short-term memory network to extract features in the spatial dimension and the temporal dimension respectively, and fuses the features of different dimensions, so as to generate a feature fusion vector containing multi-dimensional features. Through the three-layer convolution layer processing of the convolution unit, the local features in the historical sequence data can be extracted, the global feature information can be strengthened, and the long short-term memory feature vector extracted by the long short-term memory network can be combined to more comprehensively and accurately determine the characteristics of the historical sequence data, so as to be able to more accurately predict the power load. The feature fusion network extracts features in the time dimension and the channel dimension in turn, and can fuse the convolution feature vector and the long short-term memory feature vector in multiple dimensions. The fused target fusion vector can more comprehensively characterize the short-term local features and long-term features of the historical sequence data, thereby improving the accuracy of load prediction.

[0140] The power load prediction device provided by the present invention is described below. The power load prediction device described below and the power load prediction method described above can be referenced to each other.

[0141] Figure 8 This is a schematic diagram of the structure of the power load prediction device provided by the present invention, which specifically includes:

[0142] The data acquisition module 801 is used to acquire historical power load sequence data of the target power grid system, wherein the historical power load sequence data includes power load information and temperature information. Detailed descriptions refer to the corresponding descriptions of the above method embodiments, which will not be repeated here.

[0143] Feature extraction module 802 is configured to extract spatial feature vectors and temporal feature vectors from the historical power load sequence data. The spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data. The temporal feature vector is a long-short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data. For detailed descriptions, please refer to the corresponding descriptions of the above-mentioned method embodiments and will not be repeated here.

[0144] The feature fusion module 803 is configured to fuse the spatial feature vector and the temporal feature vector to generate a feature fusion vector. Detailed descriptions are given in the corresponding descriptions of the above method embodiments, which will not be repeated here.

[0145] The prediction module 804 is configured to predict the power load information of the target power grid system at a future time point based on the feature fusion vector. Detailed descriptions are provided in the corresponding descriptions of the above method embodiments, which will not be repeated here.

[0146] Figure 9 An example of a physical structure diagram of an electronic device is shown below. Figure 9 As shown, the electronic device may include: a processor 910, a communications interface 920, a memory 930, and a communications bus 940, wherein the processor 910, the communications interface 920, and the memory 930 communicate with each other via the communications bus 940. The processor 910 may call logic instructions in the memory 930 to execute a power load forecasting method, which includes: obtaining historical power load sequence data of a target power grid system, wherein the historical power load sequence data includes power load information and temperature information; extracting spatial feature vectors and temporal feature vectors from the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the temporal feature vector is a long short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data; performing feature fusion on the spatial feature vector and the temporal feature vector to generate a feature fusion vector; and predicting power load information of the target power grid system at a future time point based on the feature fusion vector.

[0147] Furthermore, the logic instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0148] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the power load prediction method provided by the above methods, which includes: obtaining historical power load series data of the target power grid system, wherein the historical power load series data includes power load information and temperature information; extracting the spatial feature vector and time feature vector of the historical power load series data, the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load series data, and the time feature vector is a long and short-term memory feature vector of at least one channel obtained by performing time feature extraction on the historical power load series data; performing feature fusion on the spatial feature vector and the time feature vector to generate a feature fusion vector; and predicting the power load information of the target power grid system at a future time point based on the feature fusion vector.

[0149] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the power load forecasting method provided by the above-mentioned methods, the method comprising: obtaining historical power load series data of the target power grid system, wherein the historical power load series data includes power load information and temperature information; extracting the spatial feature vector and time feature vector of the historical power load series data, the spatial feature vector being a multi-channel vector obtained by performing spatial feature extraction on the historical power load series data, and the time feature vector being a long-short-term memory feature vector of at least one channel obtained by performing time feature extraction on the historical power load series data; performing feature fusion on the spatial feature vector and the time feature vector to generate a feature fusion vector; and predicting the power load information of the target power grid system at a future time point based on the feature fusion vector.

[0150] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0151] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for predicting power load, characterized in that: include: Acquiring historical power load sequence data of a target power grid system, wherein the historical power load sequence data includes power load information and temperature information; Extracting a spatial feature vector and a temporal feature vector of the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the temporal feature vector is a long short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data; Extracting the spatial feature vector of the historical power load sequence data includes: Performing spatial feature extraction on the historical power load sequence data through a convolutional neural network to obtain a multi-channel spatial feature vector, wherein the convolutional neural network includes a plurality of convolution units arranged in series, and the convolution units include a first convolution layer, a second convolution layer, and a third convolution layer; The spatial feature extraction of the historical power load sequence data by a convolutional neural network to obtain a multi-channel spatial feature vector includes: When there is an i-2th initial eigenvalue, convolution processing is performed on the i-th initial eigenvalue, the i-1th initial eigenvalue, and the i-2th initial eigenvalue in the initial sequence data by the first convolution layer to generate an i-th first eigenvalue, the initial sequence data including a plurality of initial eigenvalues, and the initial sequence data input to the first convolution layer of the first convolution unit is the historical power load sequence data; When the i-2th initial eigenvalue does not exist, performing convolution processing on the i-th initial eigenvalue and the i-1th initial eigenvalue in the initial sequence data through the first convolution layer to generate an i-th first eigenvalue, where i≥2; Inputting the first sequence data including each first eigenvalue into the second convolutional layer; When there is a j-4th first eigenvalue, convolution processing is performed on the j-th first eigenvalue, the j-2nd first eigenvalue, and the j-4th first eigenvalue in the first sequence data by the second convolution layer to generate a j-th second eigenvalue; When there is no j-4th first eigenvalue, convolution processing is performed on the jth first eigenvalue and the j-2th first eigenvalue in the first sequence data by the second convolution layer to generate a jth second eigenvalue, where j≥4; Inputting the second sequence data including each second eigenvalue into the third convolutional layer; When there is a k-6th second eigenvalue, convolution processing is performed on the k-th second eigenvalue, the k-3th second eigenvalue, and the k-6th second eigenvalue in the second sequence data by the third convolution layer to generate a k-th third eigenvalue; When there is no k-6th second eigenvalue, convolution processing is performed on the kth second eigenvalue and the k-3th second eigenvalue in the second sequence data by the third convolution layer to generate a kth third eigenvalue, thereby obtaining third sequence data of each third eigenvalue, where k≥7; Obtaining a multi-channel spatial feature vector based on the third sequence data output by the last convolution unit; Performing feature fusion on the spatial feature vector and the temporal feature vector to generate a feature fusion vector; The power load information of the target power grid system at a future time point is predicted based on the feature fusion vector.

2. The method according to claim 1, characterized in that The obtaining of historical power load sequence data of the target power grid system includes: Based on a preset time period, historical power load data corresponding to a plurality of historical time points of the target power grid system are collected to obtain historical power load sequence data.

3. The method according to claim 1, characterized in that Extracting the time feature vector of the historical power load sequence data includes: The historical power load sequence data is subjected to time feature extraction through a long short-term memory network to obtain a time feature vector of at least one channel, where the time feature vector is a long short-term memory feature vector.

4. The method according to claim 3, characterized in that The step of fusing the spatial feature vector and the temporal feature vector to generate a feature fusion vector includes: The spatial feature vector and the temporal feature vector are concatenated along the channel dimension through a feature fusion network to generate a feature fusion vector.

5. The method according to claim 1 or 4, characterized in that The predicting of power load information of the target power grid system at a future time point based on the feature fusion vector includes: Performing feature extraction on the feature fusion vector in a time dimension to obtain an intermediate fusion vector; Performing feature extraction on the intermediate fusion vector in the channel dimension to obtain a target fusion vector; The power load information of the target power grid system at a future time point is predicted according to the target fusion vector.

6. The method according to claim 5, characterized in that The step of performing feature extraction on the feature fusion vector in a time dimension to obtain an intermediate fusion vector; and performing feature extraction on the intermediate fusion vector in a channel dimension to obtain a target fusion vector includes: Transpose the feature fusion vector to obtain a first feature vector of C×T; Perform feature extraction on the first eigenvector row by row to obtain a second eigenvector of C×T; Transpose the second eigenvector to obtain a T×C third eigenvector, and generate the intermediate fusion vector based on the third eigenvector; Performing feature extraction on the intermediate fusion vector row by row to generate a T×C fourth feature vector, and generating the target fusion vector based on the fourth feature vector; The size of the feature fusion vector is T×C, and C=c1+c2, c1 is the number of channels of the spatial feature vector, c1≥2, c2 is the number of channels of the temporal feature vector, c2≥1, and T is the number of eigenvalues ​​of the spatial feature vector and the temporal feature vector.

7. A power load prediction device, characterized in that: include: A data acquisition module is used to acquire historical power load sequence data of the target power grid system, wherein the historical power load sequence data includes power load information and temperature information; A feature extraction module is configured to extract a spatial feature vector and a temporal feature vector of the historical power load sequence data, wherein the spatial feature vector is a multi-channel vector obtained by performing spatial feature extraction on the historical power load sequence data, and the temporal feature vector is a long short-term memory feature vector of at least one channel obtained by performing temporal feature extraction on the historical power load sequence data; extracting the spatial feature vector of the historical power load sequence data comprises: performing spatial feature extraction on the historical power load sequence data through a convolutional neural network to obtain a multi-channel spatial feature vector, wherein the convolutional neural network comprises a plurality of convolution units arranged in series, wherein the convolution unit comprises a first convolution layer, a second convolution layer, and a third convolution layer. Three convolution layers; the spatial feature extraction of the historical power load sequence data by the convolution neural network is performed to obtain a multi-channel spatial feature vector, including: when there is an i-2th initial eigenvalue, the i-1th initial eigenvalue and the i-2th initial eigenvalue in the initial sequence data are convolved by the first convolution layer to generate the i-th first eigenvalue, the initial sequence data includes multiple initial eigenvalues, and the initial sequence data of the first convolution layer input to the first convolution unit is the historical power load sequence data; when there is no i-2th initial eigenvalue, the i-th initial eigenvalue in the initial sequence data is convolved by the first convolution layer to generate the i-th first eigenvalue. and the i-1th initial eigenvalue to generate the i-th first eigenvalue, where i≥2; input the first sequence data containing each first eigenvalue into the second convolution layer; when there is the j-4th first eigenvalue, convolve the j-th first eigenvalue, the j-2th first eigenvalue and the j-4th first eigenvalue in the first sequence data through the second convolution layer to generate the j-th second eigenvalue; when there is no j-4th first eigenvalue, convolve the j-th first eigenvalue and the j-2th first eigenvalue in the first sequence data through the second convolution layer to generate the j-th second eigenvalue, where j≥4; input the first sequence data containing each second eigenvalue into the second convolution layer; when there is the j-4th first eigenvalue, convolve the j-th first eigenvalue and the j-2th first eigenvalue in the first sequence data through the second convolution layer to generate the j-th second eigenvalue, where j≥4; The second sequence data of the eigenvalue is input into the third convolution layer; when there is a k-6th second eigenvalue, the k-3th second eigenvalue and the k-6th second eigenvalue in the second sequence data are convolved by the third convolution layer to generate a kth third eigenvalue; when there is no k-6th second eigenvalue, the k-3th second eigenvalue in the second sequence data are convolved by the third convolution layer to generate a kth third eigenvalue, and third sequence data of each third eigenvalue is obtained, wherein k≥7; a multi-channel spatial eigenvector is obtained based on the third sequence data output by the last convolution unit; A feature fusion module, configured to fuse the spatial feature vector and the temporal feature vector to generate a feature fusion vector; A prediction module is used to predict the power load information of the target power grid system at a future time point based on the feature fusion vector.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the power load prediction method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power load prediction method according to any one of claims 1 to 6 is implemented.

Images