Power load prediction method and system based on generative artificial intelligence adversarial network

Abstract

The invention discloses a power load prediction method and system based on a generative artificial intelligence adversarial network, and relates to the technical field of power, and the method comprises the steps: cleaning collected historical power load data; completing model robustness of linear interpolation, designing a learning architecture for processing time series data, and performing data distribution by using a discriminator and adopting a conditional variation auto-encoder structure; and evaluating parameters of the generator and the discriminator to complete power load prediction of the adversarial network. The method meets the prediction requirements in different scenes, thereby providing powerful support for stable operation of a power system. In addition, the application of the technical scheme is helpful for an electric power company to make resource allocation in advance, energy waste is reduced, and economic benefits and social benefits are improved.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to an electric power load prediction method and system based on a generative artificial intelligence adversarial network. Background Art

[0002] At present, power load forecasting is an important basis for power system planning and operation, and is of great significance to the safe and economic operation of the power system. Traditional power load forecasting methods mainly include regression analysis, time series method, artificial neural network method, etc. These methods can meet the needs of power load forecasting to a certain extent, but the prediction accuracy is not high enough, and it is difficult to fully capture the nonlinear characteristics and time-varying characteristics of power load data, resulting in less than ideal prediction accuracy, especially for short-term loads and loads with large fluctuations.

[0003] In response to the problems existing in the existing power load forecasting technology, this patent proposes a power load forecasting model based on a generative artificial intelligence adversarial network, aiming to improve the accuracy of power load forecasting, especially for short-term loads and loads with large fluctuations, and enhance the generalization ability of the prediction model so that it can adapt to different load patterns and abnormal situations. Summary of the invention

[0004] In view of the problems existing in the existing power load prediction method and system based on generative artificial intelligence adversarial network, the present invention is proposed.

[0005] Therefore, in order to solve the problems existing in the existing power load forecasting technology, the present invention adopts a power load forecasting method based on a generative artificial intelligence adversarial network.

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

[0007] In the first aspect, an embodiment of the present invention provides a power load prediction method based on a generative artificial intelligence adversarial network, which includes collecting historical power load data, wherein the historical power load data includes power load values ​​in different time periods, temperature and humidity meteorological data, holiday information economic indicators, and distribution of electricity user types, and cleaning the collected historical power load data; using a time series interpolation method to perform linear interpolation on the cleaned historical power load data, standardizing the linear interpolation, completing the model robustness of the linear interpolation, designing a learning architecture for processing time series data, using a discriminator and adopting a conditional variational autoencoder structure to distribute data; training the distributed data, evaluating the generator and discriminator parameters, and completing the power load prediction of the adversarial network.

[0008] As a preferred solution of the power load prediction method based on the generative artificial intelligence adversarial network described in the present invention, wherein: the cleaning of the collected historical power load data includes processing outlier detection, using the Z-score method to detect outliers, the Z-score method includes an outlier detection technology, and identifies outliers based on the mean and standard deviation of the data. The calculation formula of the Z-score is as follows:

[0009]

[0010] Where X is the value of the data point; μ is the mean of the data set; σ is the standard deviation of the data set;

[0011] The data points with |Z|≥3 are regarded as potential outliers. When the distance between the data point and the mean exceeds 3 standard deviations, the outliers are subjected to local regression smoothing. The specific steps of the local regression smoothing are as follows:

[0012] For each data point, a local neighborhood is selected, and a weighted least squares regression is fitted within the neighborhood. The fitted regression model is used to predict the value of the current point, and the predicted value replaces the original outlier.

[0013] As a preferred solution of the power load forecasting method based on the generative artificial intelligence adversarial network of the present invention, wherein: the linear interpolation of the cleaned historical power load data using the time series interpolation method includes drawing a straight line between two known data points and using this straight line to estimate the missing value in the middle, and the linear interpolation steps are as follows:

[0014] When there are two known time points and corresponding data values: (t1, y1) and (t2, y2), estimate the value y at time t, where t1≤t≤t2;

[0015]

[0016] Perform minimum-maximum standardization on all types of data to unify the value range to the [0,1] interval. The minimum-maximum standardization formula is as follows:

[0017]

[0018] Where X is the value of the original data point, X min is the minimum value in the data set, X max is the maximum value in the data set, X norm is the value of the data point after normalization.

[0019] As a preferred solution of the power load prediction method based on the generative artificial intelligence adversarial network described in the present invention, wherein: the evaluation of the parameters of the generator and the discriminator includes evaluating the input layer and the multi-scale convolution block, the input layer includes receiving standardized multi-dimensional time series data, converting the features of different scales into the same range, the multi-scale convolution block includes using convolution kernels of different sizes to capture features of different time scales, using a one-dimensional convolution layer with convolution kernel sizes of 1, 3, 5, and 7 to capture features of different time scales, and using normalization and ReLU activation functions to splice features of different scales;

[0020] The formula of the ReLU function is as follows:

[0021] ReLU(X)=max(0,X)

[0022] Among them, when the input X is a positive number, the ReLU function outputs X, and when X is a negative number, the ReLU function outputs 0;

[0023] The multi-scale convolution block includes a residual connection, a dilated convolution layer, an attention mechanism, and an output layer;

[0024] The residual connection includes adding a residual connection between the input and output of the multi-scale convolution block, and solving the gradient vanishing problem by introducing a new network architecture unit - residual block. A residual connection is added between the input and output of the multi-scale convolution block, that is, the formula is expressed as:

[0025] x out =x in +F(x in )

[0026] Among them, x in is the input of the residual block, F(x in ) is a series of operations inside the residual block, x out is the output of the residual block.

[0027] As a preferred solution of the power load prediction method based on the generative artificial intelligence adversarial network of the present invention, wherein: the dilated convolution layer includes using a dilated convolution with a dilation rate increasing layer by layer, and the receptive field of the convolution kernel is expanded by introducing a hole in the convolution kernel to calculate the number of parameters;

[0028] The attention mechanism includes adding a self-attention layer between multi-scale convolution blocks to focus on important features. The self-attention mechanism calculates the correlation between each position in the sequence and all other positions. The formula for calculating the self-attention mechanism is:

[0029]

[0030] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key vector Used to scale dot products;

[0031] The output layer includes a fully connected layer, and outputs a predicted load value.

[0032] As a preferred solution of the power load prediction method based on the generative artificial intelligence adversarial network described in the present invention, the training of the distributed data includes collecting and calculating the power load data, and the specific calculation formula is:

[0033] L W =E x～P [f(x)]-E z～Q [f(G(z))]

[0034] Among them, f(x) is the discriminator, G(z) is the generator, and z is the noise sampled from the prior noise distribution.

[0035] As a preferred solution of the power load prediction method based on the generative artificial intelligence adversarial network of the present invention, wherein: the completion of the power load prediction of the adversarial network includes analyzing the model performance to determine whether the requirements are met;

[0036] When the hyperparameters are not adjusted according to the performance analysis results, the training cycle is repeated until the model performance is satisfactory and the training process is completed.

[0037] In a second aspect, an embodiment of the present invention provides a power load prediction system based on a generative artificial intelligence adversarial network, which includes:

[0038] A collection module collects historical power load data, including power load values ​​by time period, temperature and humidity meteorological data, holiday information economic indicators, and distribution of electricity users, and cleans the collected historical power load data;

[0039] Design a module that uses a time series interpolation method to perform linear interpolation on the cleaned historical power load data, standardizes the linear interpolation, completes the model robustness of the linear interpolation, designs a learning architecture for processing time series data, uses a discriminator and adopts a conditional variational autoencoder structure for data distribution;

[0040] The prediction module trains the distributed data, evaluates the generator and discriminator parameters, and completes the power load prediction of the adversarial network.

[0041] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, it implements any step of the above-mentioned power load prediction method based on generative artificial intelligence adversarial network.

[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned power load prediction method based on a generative artificial intelligence adversarial network is implemented.

[0043] The beneficial effects of the present invention are as follows: the present invention significantly improves the accuracy and reliability of power load prediction by integrating a variety of advanced technologies, including time series interpolation, Z-score outlier detection, multi-scale convolution blocks, residual connections, dilated convolution layers and attention mechanisms, etc., and ensures the quality of input data and reduces the impact of noise on model training by deep cleaning and linear interpolation of historical data. Secondly, the conditional variational autoencoder structure is used for data distribution learning, and the model parameters are optimized in combination with the generative adversarial network, which not only enhances the robustness of the model, but also simulates complex data distribution and improves the accuracy of prediction. Furthermore, the introduction of multi-scale convolution blocks and attention mechanisms effectively captures the feature changes at different time scales, especially long-term dependencies, so that the model can better adapt to the dynamic characteristics of power loads. Finally, the scheme also has the ability to flexibly adjust hyperparameters, can optimize model performance according to actual needs, and meet the prediction requirements in different scenarios, thereby providing strong support for the stable operation of the power system. In addition, the application of this technical solution helps power companies to make resource allocation in advance, reduce energy waste, and improve economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0045] Figure 1 A specific flow chart of a method and system for predicting power load based on a generative artificial intelligence adversarial network provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0048] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0049] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0050] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0051] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0052] Example 1

[0053] Reference Figure 1, which is the first embodiment of the present invention, provides a method for predicting power load based on a generative artificial intelligence adversarial network, including:

[0054] S1: Collect historical power load data, which includes power load values ​​in different time periods, temperature and humidity meteorological data, holiday information economic indicators, and distribution of electricity user types, and clean the collected historical power load data.

[0055] Among them, cleaning the collected historical power load data includes processing outlier detection, using the Z-score method to detect outliers. The Z-score method includes outlier detection technology, which identifies outliers based on the mean and standard deviation of the data. The calculation formula of Z-score is as follows:

[0056]

[0057] Where X is the value of the data point; μ is the mean of the data set; σ is the standard deviation of the data set;

[0058] The data points with |Z|≥3 are regarded as potential outliers. When the data point is more than 3 standard deviations away from the mean, the outliers are subjected to local regression smoothing. The specific steps of local regression smoothing are as follows:

[0059] For each data point, a local neighborhood is selected, and a weighted least squares regression is fitted within the neighborhood. The fitted regression model is used to predict the value of the current point, and the predicted value replaces the original outlier.

[0060] S2: Use the time series interpolation method to perform linear interpolation on the cleaned historical power load data, standardize the linear interpolation, complete the model robustness of the linear interpolation, design a learning architecture for processing time series data, use the discriminator and adopt the conditional variational autoencoder structure for data distribution.

[0061] Among them, the linear interpolation of the cleaned historical power load data using the time series interpolation method includes drawing a straight line between two known data points and using this straight line to estimate the missing value in the middle. The linear interpolation steps are as follows:

[0062] When there are two known time points and corresponding data values: (t1, y1) and (t2, y2), estimate the value y at time t, where t1≤t≤t2;

[0063]

[0064] Perform minimum-maximum standardization on all types of data to unify the value range to the [0,1] interval. The minimum-maximum standardization formula is as follows:

[0065]

[0066] Where X is the value of the original data point, X min is the minimum value in the data set, X max is the maximum value in the data set, X norm is the value of the data point after normalization.

[0067] S3: Train the distributed data, evaluate the generator and discriminator parameters, and complete the power load prediction of the adversarial network.

[0068] Among them, the evaluation of the generator and discriminator parameters includes evaluating the input layer and the multi-scale convolution block. The input layer includes receiving standardized multi-dimensional time series data and converting the features of different scales into the same range. The multi-scale convolution block includes using convolution kernels of different sizes to capture the features of different time scales. The one-dimensional convolution layer with convolution kernel sizes of 1, 3, 5, and 7 is used to capture the features of different time scales. Normalization and ReLU activation functions are used to splice the features of different scales.

[0069] The formula of the ReLU function is as follows:

[0070] ReLU(X)=max(0,X)

[0071] Among them, when the input X is a positive number, the ReLU function outputs X, and when X is a negative number, the ReLU function outputs 0;

[0072] The multi-scale convolutional block includes residual connections, dilated convolutional layers, attention mechanisms, and output layers;

[0073] The residual connection includes adding a residual connection between the input and output of the multi-scale convolution block. The gradient vanishing problem is solved by introducing a new network architecture unit - the residual block. A residual connection is added between the input and output of the multi-scale convolution block, that is, the formula is expressed as:

[0074] x out =x in +F(x in )

[0075] Among them, x in is the input of the residual block, F(x in ) is a series of operations inside the residual block, x out is the output of the residual block.

[0076] Furthermore, the dilated convolution layer includes dilated convolutions with increasing dilation rates layer by layer, and the receptive field of the convolution kernel is expanded by introducing holes in the convolution kernel to calculate the number of parameters;

[0077] The attention mechanism consists of adding a self-attention layer between multi-scale convolutional blocks to focus on important features. The self-attention mechanism calculates the correlation between each position in the sequence and all other positions. The formula for calculating the self-attention mechanism is:

[0078]

[0079] Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key vector Used to scale dot products;

[0080] The output layer includes a fully connected layer and outputs the predicted load value.

[0081] Furthermore, training the distributed data includes collecting and calculating the power load data. The specific calculation formula is:

[0082] L W =E x～P [f(x)]-E z～Q [f(G(z))]

[0083] Among them, f(x) is the discriminator, G(z) is the generator, and z is the noise sampled from the prior noise distribution.

[0084] Completing the power load forecasting of the adversarial network includes analyzing the model performance to determine whether it meets the requirements;

[0085] When the hyperparameters are not adjusted according to the performance analysis results, the training cycle is repeated until the model performance is satisfactory and the training process is completed.

[0086] In a preferred embodiment, a power load prediction system based on a generative artificial intelligence adversarial network includes a collection module, which collects historical power load data, the historical power load data includes power load values ​​in different time periods, temperature and humidity meteorological data, holiday information economic indicators, and distribution of electricity user types, and cleans the collected historical power load data; a design module, which uses a time series interpolation method to perform linear interpolation on the cleaned historical power load data, standardizes the linear interpolation, completes the model robustness of the linear interpolation, designs a learning architecture for processing time series data, uses a discriminator and adopts a conditional variational autoencoder structure for data distribution; a prediction module, which trains the distributed data, evaluates the generator and discriminator parameters, and completes the power load prediction of the adversarial network.

[0087] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0088] In summary, the present invention significantly improves the accuracy and reliability of power load prediction by integrating a variety of advanced technologies, including time series interpolation, Z-score outlier detection, multi-scale convolution blocks, residual connections, dilated convolution layers and attention mechanisms. Through deep cleaning and linear interpolation of historical data, the quality of input data is ensured and the impact of noise on model training is reduced. Secondly, the conditional variational autoencoder structure is used for data distribution learning, and the model parameters are optimized in combination with the generative adversarial network, which not only enhances the robustness of the model, but also simulates complex data distribution and improves the accuracy of prediction. Furthermore, the introduction of multi-scale convolution blocks and attention mechanisms effectively captures the feature changes at different time scales, especially long-term dependencies, so that the model can better adapt to the dynamic characteristics of power loads. Finally, the scheme also has the ability to flexibly adjust hyperparameters, optimize model performance according to actual needs, and meet the prediction requirements in different scenarios, thereby providing strong support for the stable operation of the power system. In addition, the application of this technical solution helps power companies to make resource allocation in advance, reduce energy waste, and improve economic and social benefits.

[0089] Example 2

[0090] Reference Figure 1 , which is the second embodiment of the present invention, provides a power load prediction method based on a generative artificial intelligence adversarial network. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0091] For the cleaned data, linear interpolation was used to fill missing values, and all features were normalized to the minimum and maximum values ​​to ensure that the numerical range was unified to the [0,1] interval. Subsequently, a deep neural network model including multi-scale convolutional blocks, residual connections, dilated convolutional layers and attention mechanisms was constructed using the designed time series learning architecture, namely the conditional variational autoencoder combined with the GAN structure. During the model training process, 70% of the data was used as a training set, 15% as a validation set, and the remaining 15% for testing. Through multiple iterative optimizations, the model finally achieved an accuracy of 94.7%, which is about 6 percentage points higher than the traditional long short-term memory network.

[0092] When evaluating the parameters of the generator and the discriminator, it was observed that as the number of training rounds increased, the loss function gradually converged, and the loss of the generator dropped from the initial 2.3 to below 0.4, while the accuracy of the discriminator stabilized at around 98.5%, indicating that the model successfully learned to simulate the distribution of real power load data. In addition, by introducing the self-attention mechanism, the model can more effectively capture dependencies in long time series, especially in predicting power demand during peak hours, with an average absolute error of only 2.1%, which is lower than similar methods. The experimental data of the present invention is shown in Table 1 below:

[0093] Table 1 Experimental data table of the present invention

[0094] Data Types Value / Details Experimental data time range 2018 to 2023 Total number of data points Approximately 262,800 hours of data points Number of outliers Over 1,500 potentially anomaly data points Data standardization scope [0,1] interval Training set ratio 70％ Validation set ratio 15％ Test set ratio 15％ Model Accuracy 94.7％ Compared with LSTM, About 6 percentage points Initial generator loss About 2.3 Final generator loss 0.4 or less Discriminator accuracy About 98.5% Mean Absolute Error (MAE) 2.1％

[0095] Table 1 summarizes the key data and results used in the experiment, which is convenient for quick reference and understanding of the main findings of the experiment. The comparison between the present invention and the prior art is shown in Table 2 below:

[0096] Table 2 Comparison between the present invention and the prior art

[0097]

[0098] Table 2 shows the main advantages of the present invention over the prior art.

[0099] 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 preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A power load forecasting method based on a generative artificial intelligence adversarial network, characterized by: include, Collecting historical power load data, the historical power load data includes power load values ​​by time period, temperature and humidity meteorological data, holiday information economic indicators and distribution of electricity users, and cleaning the collected historical power load data; The time series interpolation method is used to perform linear interpolation on the cleaned historical power load data, and the linear interpolation is standardized to complete the model robustness of the linear interpolation. A learning architecture for processing time series data is designed, and the discriminator is used and the conditional variational autoencoder structure is used for data distribution; The distributed data is trained, the generator and discriminator parameters are evaluated, and the power load prediction of the adversarial network is completed.

2. The power load prediction method based on generative artificial intelligence adversarial network according to claim 1, characterized in that: The cleaning of the collected historical power load data includes processing outlier detection, using a Z-score method to detect outliers, the Z-score method includes an outlier detection technology, identifying outliers based on the mean and standard deviation of the data, and the calculation formula of the Z-score is as follows: Where X is the value of the data point; μ is the mean of the data set; σ is the standard deviation of the data set; The data points with |Z|≥3 are regarded as potential outliers. When the distance between the data point and the mean exceeds 3 standard deviations, the outliers are subjected to local regression smoothing. The specific steps of the local regression smoothing are as follows: For each data point, a local neighborhood is selected, and a weighted least squares regression is fitted within the neighborhood. The fitted regression model is used to predict the value of the current point, and the predicted value replaces the original outlier.

3. The power load prediction method based on generative artificial intelligence adversarial network according to claim 2, characterized in that: The linear interpolation of the cleaned historical power load data using the time series interpolation method includes drawing a straight line between two known data points and using the straight line to estimate the missing value in the middle. The linear interpolation steps are as follows: When there are two known time points and corresponding data values: (t1, y1) and (t2, y2), estimate the value y at time t, where t1≤t≤t2; Perform minimum-maximum standardization on all types of data to unify the value range to the [0,1] interval. The minimum-maximum standardization formula is as follows: Where X is the value of the original data point, X min is the minimum value in the data set, X max is the maximum value in the data set, X norm is the value of the data point after normalization.

4. The power load prediction method based on generative artificial intelligence adversarial network according to claim 3, characterized in that: The evaluation of the parameters of the generator and the discriminator includes evaluating the input layer and the multi-scale convolution block, wherein the input layer includes receiving standardized multi-dimensional time series data and converting features of different scales into the same range, and the multi-scale convolution block includes using convolution kernels of different sizes to capture features of different time scales, using a one-dimensional convolution layer with convolution kernel sizes of 1, 3, 5, and 7 to capture features of different time scales, and using normalization and ReLU activation functions to splice features of different scales; The formula of the ReLU function is as follows: ReLU(X)=max(0,X) Among them, when the input X is a positive number, the ReLU function outputs X, and when X is a negative number, the ReLU function outputs 0; The multi-scale convolution block includes a residual connection, a dilated convolution layer, an attention mechanism, and an output layer; The residual connection includes adding a residual connection between the input and output of the multi-scale convolution block, and solving the gradient vanishing problem by introducing a new network architecture unit - residual block. A residual connection is added between the input and output of the multi-scale convolution block, that is, the formula is expressed as: x out =x in +F(x in ) Among them, x in is the input of the residual block, F(x in ) is a series of operations inside the residual block, x out is the output of the residual block.

5. The power load prediction method based on generative artificial intelligence adversarial network according to claim 4, characterized in that: The dilated convolution layer includes dilated convolution with a dilated rate increasing layer by layer, and the receptive field of the convolution kernel is expanded by introducing a hole in the convolution kernel to calculate the number of parameters; The attention mechanism includes adding a self-attention layer between multi-scale convolution blocks to focus on important features. The self-attention mechanism calculates the correlation between each position in the sequence and all other positions. The formula for calculating the self-attention mechanism is: Where Q is the query matrix, K is the key matrix, V is the value matrix, and d k is the dimension of the key vector Used to scale dot products; The output layer includes a fully connected layer, and outputs a predicted load value.

6. The power load prediction method based on generative artificial intelligence adversarial network according to claim 5, characterized in that: The training of the distributed data includes collecting and calculating the power load data, and the specific calculation formula is: L W =E x～P [f(x)]-E z～Q [f(G(z))] Among them, f(x) is the discriminator, G(z) is the generator, and z is the noise sampled from the prior noise distribution.

7. The power load prediction method based on generative artificial intelligence adversarial network according to claim 6, characterized in that: The completing the power load prediction of the adversarial network includes analyzing the model performance to determine whether the requirements are met; When the hyperparameters are not adjusted according to the performance analysis results, the training cycle is repeated until the model performance is satisfactory and the training process is completed.

8. A power load forecasting system based on a generative artificial intelligence adversarial network, based on the power load forecasting method based on a generative artificial intelligence adversarial network according to any one of claims 1 to 7, characterized in that: include, A collection module collects historical power load data, including power load values ​​by time period, temperature and humidity meteorological data, holiday information economic indicators, and distribution of electricity users, and cleans the collected historical power load data; Design a module that uses a time series interpolation method to perform linear interpolation on the cleaned historical power load data, standardizes the linear interpolation, completes the model robustness of the linear interpolation, designs a learning architecture for processing time series data, uses a discriminator and adopts a conditional variational autoencoder structure for data distribution; The prediction module trains the distributed data, evaluates the generator and discriminator parameters, and completes the power load prediction of the adversarial network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the power load prediction method based on the generative artificial intelligence adversarial network described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the power load prediction method based on a generative artificial intelligence adversarial network as described in any one of claims 1 to 7 are implemented.

Images