Power load prediction method, device and equipment

Abstract

The invention relates to a power load prediction method, device and equipment, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining the power load data of a target region in a first time period; decomposing the power load data based on the plurality of decomposition parameters by adopting a moving average method to obtain a component set corresponding to each decomposition parameter; performing correlation calculation between each component set and the power load data to obtain a correlation coefficient corresponding to each component set, and based on the correlation coefficient corresponding to each component set, determining a target decomposition parameter meeting a condition from the plurality of decomposition parameters; and the target decomposition parameter is used as the hyper-parameter of the power load prediction model, the power load prediction model is trained by using the power load data, and the trained power load prediction model is used for power load prediction, so that the accuracy of power load prediction can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and particularly to a power load forecasting method, device, and equipment. Background Art

[0002] Power load forecasting is a key link in the operation and planning of power systems. Accurate power load forecasting helps optimize power generation plans, reduce costs, and improve power grid reliability.

[0003] In related technologies, deep learning technology is used to achieve power load forecasting. However, power load data has complex time series characteristics, including trends, seasonality, and randomness, etc. Traditional deep learning methods have certain limitations in dealing with these characteristics, resulting in relatively low accuracy of power load forecasting. Summary of the Invention

[0004] The present application provides a power load forecasting method, device, and equipment. The technical solution of the present application is as follows:

[0005] In a first aspect, the present application provides a power load forecasting method, and the method includes:

[0006] Obtain power load data of a target area within a first time period;

[0007] Using the moving average method, decompose the power load data based on multiple decomposition parameters to obtain a component set corresponding to each decomposition parameter, where the component set includes a trend component and a seasonal component;

[0008] Calculate the correlation between each component set and the power load data to obtain a correlation coefficient corresponding to each component set, where the correlation coefficient is used to reflect the matching degree between the component set and the power load data;

[0009] Based on the correlation coefficient corresponding to each component set, determine at least one target decomposition parameter that meets the conditions from the multiple decomposition parameters;

[0010] Use the at least one target decomposition parameter as hyperparameters of the power load forecasting model, and use the power load data to train the power load forecasting model to obtain a trained power load forecasting model, where the trained power load forecasting model is used to forecast the power load of the target area.

[0011] In some implementation manners, the step of using the moving average method to decompose the power load data based on multiple decomposition parameters to obtain a component set corresponding to each decomposition parameter includes:

[0012] After filling the power load data, based on the first decomposition parameter and formula (1), decompose the filled power load data into the trend component corresponding to the first decomposition parameter, where the first decomposition parameter refers to any one of the multiple decomposition parameters;

[0013] L t =Avgpool(Padding(y t )) (1)

[0014] where y t represents the power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the padding operation;

[0015] Based on formula (2), remove the trend component corresponding to the first decomposition parameter from the power load data to obtain the seasonal component corresponding to the first decomposition parameter;

[0016] M t =y t -L t (2)

[0017] where M t represents the seasonal component.

[0018] In some implementation manners, calculating the correlation coefficient corresponding to each component set between each component set and the power load data includes:

[0019] Obtain the first correlation coefficient between the trend component in the first component set and the power load data, and the second correlation coefficient between the seasonal component in the first component set and the power load data;

[0020] Take the weighted average of the first correlation coefficient and the second correlation coefficient as the correlation coefficient corresponding to the first component set;

[0021] where the first correlation coefficient and the second correlation coefficient are calculated based on formula (3):

[0022]

[0023] where r yx represents the correlation coefficient, X represents the seasonal component or the trend component, y represents the power load data, i represents any moment, and n represents the length of the time series.

[0024] In some implementation manners, training the power load prediction model using the power load data includes:

[0025] Input the power load data into the power load prediction model, and decompose the power load data through the at least one target decomposition parameter in the power load prediction model to obtain a set of target components, where the set of target components includes a target trend component and a target seasonal component;

[0026] Screen the set of target components through the variable selection network in the power load prediction model to obtain the screened set of target components;

[0027] Process the screened set of target components through the encoder-decoder network in the power load prediction model, output the power load prediction result of the target area in the second time period, calculate a loss value based on the power load prediction result and the real result of the target area in the second time period, and update the model parameters of the power prediction model based on the loss value. The second time period is after the first time period.

[0028] In some implementation manners, the power load data includes the power load values, weather data, and calendar data of the target area in the first time period.

[0029] In a second aspect, the present application provides a power load prediction device, and the device includes:

[0030] An acquisition module, configured to acquire the power load data of a target area in a first time period;

[0031] A decomposition module, configured to use the moving average method to decompose the power load data based on multiple decomposition parameters to obtain a set of components corresponding to each decomposition parameter, where the set of components includes a trend component and a seasonal component;

[0032] A correlation calculation module, configured to calculate the correlation between each set of components and the power load data to obtain a correlation coefficient corresponding to each set of components, where the correlation coefficient is used to reflect the matching degree between the set of components and the power load data;

[0033] A determination module, configured to determine at least one target decomposition parameter that meets the conditions from the multiple decomposition parameters based on the correlation coefficient corresponding to each set of components;

[0034] A training module, configured to use the at least one target decomposition parameter as a hyperparameter of the power load prediction model, and use the power load data to train the power load prediction model to obtain a trained power load prediction model, where the trained power load prediction model is used to predict the power load of the target area.

[0035] In some implementations, the decomposition module is configured to:

[0036] After padding the power load data, based on the first decomposition parameter and formula (1), decompose the padded power load data into a trend component corresponding to the first decomposition parameter, where the first decomposition parameter refers to any one of the multiple decomposition parameters;

[0037] L t = Avgpool(Padding(y t )) (1)

[0038] where y t represents the power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the padding operation;

[0039] Based on formula (2), remove the trend component corresponding to the first decomposition parameter from the power load data to obtain the seasonal component corresponding to the first decomposition parameter;

[0040] M t = y t - L t (2)

[0041] where M t represents the seasonal component.

[0042] In some implementations, the correlation calculation module is configured to:

[0043] Obtain a first correlation coefficient between the trend component in the first component set and the power load data, and a second correlation coefficient between the seasonal component in the first component set and the power load data;

[0044] Use the weighted average of the first correlation coefficient and the second correlation coefficient as the correlation coefficient corresponding to the first component set;

[0045] where the first correlation coefficient and the second correlation coefficient are calculated based on formula (3):

[0046]

[0047] where r yX represents the correlation coefficient, X represents the seasonal component or the trend component, y represents the power load data, i represents any moment, and n represents the length of the time series.

[0048] In some implementations, the training module is configured to:

[0049] Input the power load data into the power load prediction model, and decompose the power load data through the at least one target decomposition parameter in the power load prediction model to obtain a target component set, where the target component set includes a target trend component and a target seasonal component;

[0050] Screen the target component set through the variable selection network in the power load prediction model to obtain the screened target component set;

[0051] Process the screened target component set through the encoding and decoding network in the power load prediction model, output the power load prediction result of the target area in the second time period, calculate a loss value based on the power load prediction result and the true result of the target area in the second time period, and update the model parameters of the power prediction model based on the loss value. The second time period is after the first time period.

[0052] In a third aspect, the present application provides an electronic device, which includes a processor and a memory. The memory stores program code, and the processor is used to execute the program code to implement the above-mentioned power load prediction method.

[0053] In a fourth aspect, the present application provides a computer-readable storage medium, which includes: when the program code in the computer-readable storage medium is executed by the processor of the electronic device, the electronic device can execute the above-mentioned power load prediction method.

[0054] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Description of the Drawings

[0055] Figure 1 is a schematic diagram of the implementation environment of a power load prediction method;

[0056] Figure 2 is a flowchart of a power load prediction method;

[0057] Figure 3 is a schematic diagram of the structure of a power load prediction device;

[0058] Figure 4 is a schematic diagram of the structure of an electronic device. Detailed Embodiments

[0059] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings.

[0060] The data involved in this application can be data authorized by the user or fully authorized by all parties.

[0061] Figure 1 It is a schematic diagram of the implementation environment of a power load forecasting method. Refer to Figure 1 , this implementation environment includes: this implementation environment includes: terminal 101 and server 102.

[0062] The terminal 101 can be at least one of devices such as a smart phone, a smart watch, a desktop computer, a laptop computer, a virtual reality terminal, an augmented reality terminal, a wireless terminal, and a laptop portable computer. The terminal 101 has a communication function and can access a wired network or a wireless network. The terminal 101 can generally refer to one of multiple terminals, and only the terminal 101 is used as an example in this embodiment. Those skilled in the art can know that the number of the above terminals can be more or less. Schematically, the terminal 101 can install and run an application program, and this application program is used to provide power load data for the power load forecasting model, and this is not limited.

[0063] The server 102 can be an independent physical server, or a server cluster or a distributed file system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. In some embodiments, the server 102 is directly or indirectly connected to the terminal 101 through a wired or wireless communication method, and this disclosure embodiment does not limit this. Optionally, the number of the above servers 102 can be more or less, and this disclosure embodiment does not limit this. Of course, the server 102 can also include other functional servers to provide more comprehensive and diverse services.

[0064] In this application, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work.

[0065] Figure 2 It is a flowchart of a power load forecasting method. As Figure 2 shown, this method is executed by the server and includes the following steps 201 to step 205.

[0066] 201. The server obtains the power load data of the target area within the first time period.

[0067] Among them, the target area can be a factory, an enterprise, a city, or a designated area set artificially, and is not limited thereto. The first time period is, for example, one year, one month, or one week, and is not limited thereto. The power load data includes the power load value, weather data, and calendar data of the target area within the first time period. The power load value is, for example, a time series of power load values collected from the power distribution system of the target area.

[0068] 202. The server uses the moving average method to decompose the power load data based on multiple decomposition parameters, and obtains a set of components corresponding to each decomposition parameter. The set of components includes a trend component and a seasonal component.

[0069] Among them, the decomposition parameter is the length of the moving average kernel used in the moving average method. It should be understood that different time series have different seasonal and periodic variations, and the required decomposition parameters are different. For example, for hourly power load data, when the decomposition parameter is equal to 24, the daily cycle pattern can be captured, but it is not applicable to weekly load data. By setting appropriate decomposition parameters, the original power load data can be effectively decomposed into a trend component and a seasonal component, so that the decomposition result can more accurately reflect the data characteristics.

[0070] In this application, the server generates multiple decomposition parameters. For any one of the decomposition parameters, the moving average method is used to decompose the power load data to obtain a set of components corresponding to the decomposition parameter.

[0071] Exemplarily, taking any one of the decomposition parameters as the first decomposition parameter as an example, after the server fills the power load data, based on the first decomposition parameter and formula (1), the filled power load data is decomposed into a trend component corresponding to the first decomposition parameter;

[0072] L t = Avgpool(Padding(y t )) (1)

[0073] Among them, y t represents the power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the padding operation;

[0074] Based on formula (2), the trend component corresponding to the first decomposition parameter is removed from the power load data to obtain a seasonal component corresponding to the first decomposition parameter;

[0075] M t = y t - L t (2)

[0076] Among them, M tRepresents the seasonal component.

[0077] It should be understood that the decomposition parameter is a key parameter in the moving average calculation process and can determine the calculation method of the moving average function. When decomposing the trend component of the filled power load data, the moving average function determines the data range participating in the average calculation according to the decomposition parameter. Different decomposition parameters will produce different moving average results, thereby affecting the decomposed trend component and seasonal component. By changing the decomposition parameter, the smoothing degree of the moving average function for the data and the ability to capture periodic characteristics can be adjusted. For example, for time series with different periodic characteristics, appropriate decomposition parameters need to be selected to ensure that the moving average function can accurately extract the trend component, so that the decomposed components are more conducive to the training of subsequent power load forecasting.

[0078] 203. The server calculates the correlation between each component set and the power load data to obtain the correlation coefficient corresponding to each component set, and the correlation coefficient is used to reflect the matching degree between the component set and the power load data.

[0079] Among them, for any decomposition parameter, the server calculates the correlation between the component set corresponding to the decomposition parameter and the power load data to obtain the correlation coefficient corresponding to the component set, so as to reflect the matching degree between the component set and the power load data through the correlation coefficient.

[0080] Exemplarily, taking any component set as the first decomposition set as an example, the server obtains the first correlation coefficient between the trend component in the first component set and the power load data, and the second correlation coefficient between the seasonal component in the first component set and the power load data; takes the weighted average of the first correlation coefficient and the second correlation coefficient as the correlation coefficient corresponding to the first component set; among them, the first correlation coefficient and the second correlation coefficient are calculated based on formula (3):

[0081]

[0082] Among them, r yX Represents the correlation coefficient, X represents the seasonal component or trend component, y represents the power load data, i represents any moment, and n represents the length of the time series.

[0083] 204. The server determines at least one target decomposition parameter that meets the conditions from multiple decomposition parameters based on the correlation coefficient corresponding to each component set.

[0084] Among them, the conditions can be set according to the requirements of the actual scenario. For example, the condition can be set to the maximum value, that is, the component parameters corresponding to the set of components with the largest correlation coefficient are used as the target decomposition parameters. Another example is to set the condition to the top 3, that is, the component parameters corresponding to the set of components with the top 3 correlation coefficients are used as the target decomposition parameters. This application does not limit this. It should be understood that the higher the correlation coefficient, the more accurately the trend component and seasonal component corresponding to the correlation coefficient can reflect the core characteristics and internal laws of the original time series.

[0085] 205. The server uses at least one target decomposition parameter as a hyperparameter of the power load forecasting model, and uses the power load data to train the power load forecasting model to obtain a trained power load forecasting model, and the trained power load forecasting model is used to forecast the power load of the target area.

[0086] Among them, the server configures at least one target decomposition parameter as a type of hyperparameter of the power load forecasting model, so that the power load forecasting model can decompose the power load data according to the target decomposition parameters. It should be understood that the hyperparameters of the power load forecasting model also include other types of parameters, such as the input sequence length, output sequence length, hidden layer size, and batch size, etc. This application does not limit the setting of these hyperparameters.

[0087] In this step, the server inputs the power load data into the power load forecasting model, decomposes the power load data through at least one target decomposition parameter in the power load forecasting model to obtain a target component set, and the target component set includes a target trend component and a target seasonal component; filters the target component set through the variable selection network in the power load forecasting model to obtain a filtered target component set; processes the filtered target component set through the encoder-decoder network in the power load forecasting model, outputs the power load forecasting result of the target area in the second time period, calculates the loss value based on the power load forecasting result and the real result of the target area in the second time period, and updates the model parameters of the power forecasting model based on the loss value. In addition, taking the first time period as January 2XXX as an example, the second time period can be February 2XXX.

[0088] Specifically, a gated residual network is used inside the variable selection network. Exemplarily, the variable selection network applies a gated residual network to each feature separately. Then, all the features are flattened and passed through a gated residual network, and then the Softmax function is used to generate feature weights. Finally, the outputs of each individual gated residual network are weighted and summed to obtain the output of the variable selection network, that is, the above-mentioned filtered set of target components is obtained. Among them, the gated residual network includes parts such as a fully connected layer, a non-linear activation function, a gating mechanism, and layer normalization, and has the ability to flexibly control the flow of information and can learn complex feature relationships.

[0089] By using the gated residual network in the variable selection network, the power load prediction model can more effectively select the most important features for prediction from the input features, and at the same time remove the noise inputs that may have a negative impact on performance, thereby improving the performance and interpretability of the model. Moreover, since the power load prediction model of the present application decomposes the power load data based on the target decomposition parameters, and the target decomposition parameters are determined based on the correlation coefficients, the set of target components obtained after decomposition can more accurately reflect the core features and internal laws of the original time series. When these highly correlated components are used as the input to the variable selection network, the model can obtain more valuable information. Among them, the accurate trend component can enable the model to grasp the overall change trend of the load over time, and the seasonal component can reflect the periodic fluctuations and help the model understand the impact of different seasons on the load, thereby improving the accuracy of the prediction. Based on this, during the training process of the model, it can better capture the long-term and short-term dependencies in the time series based on these highly correlated components, and then optimize its own parameters, improving the prediction performance of the model.

[0090] In summary, in the method provided in the embodiment of the present application, the power load data of the target area in the first time period is obtained; the moving average method is used to decompose the power load data based on multiple decomposition parameters to obtain a set of components corresponding to each decomposition parameter; the correlation between each set of components and the power load data is calculated to obtain the correlation coefficient corresponding to each set of components, and at least one target decomposition parameter that meets the conditions is determined from multiple decomposition parameters based on the correlation coefficient corresponding to each set of components; the at least one target decomposition parameter is used as the hyperparameter of the power load prediction model, and the power load prediction model is trained using the power load data to obtain the trained power load prediction model, and the trained power load prediction model is used for power load prediction, which can improve the accuracy of power load prediction.

[0091] Figure 3 It is a schematic structural diagram of a power load prediction device. Refer to Figure 3, the device includes an acquisition module 301, a decomposition module 302, a correlation calculation module 303, a determination module 304, and a training module 305.

[0092] The acquisition module 301 is configured to acquire power load data of a target area within a first time period;

[0093] The decomposition module 302 is configured to decompose the power load data based on a plurality of decomposition parameters by using a moving average method to obtain a component set corresponding to each decomposition parameter, where the component set includes a trend component and a seasonal component;

[0094] The correlation calculation module 303 is configured to calculate a correlation between each component set and the power load data to obtain a correlation coefficient corresponding to each component set, where the correlation coefficient is used to reflect the matching degree between the component set and the power load data;

[0095] The determination module 304 is configured to determine at least one target decomposition parameter that meets the conditions from the plurality of decomposition parameters based on the correlation coefficient corresponding to each component set;

[0096] The training module 305 is configured to use the at least one target decomposition parameter as a hyperparameter of a power load prediction model, and use the power load data to train the power load prediction model to obtain a trained power load prediction model, where the trained power load prediction model is used to predict the power load of the target area.

[0097] In some implementation manners, the decomposition module 302 is configured to:

[0098] After padding the power load data, based on a first decomposition parameter and formula (1), decompose the padded power load data into a trend component corresponding to the first decomposition parameter, where the first decomposition parameter refers to any one of the plurality of decomposition parameters;

[0099] L t = Avgpool(Padding(y t )) (1)

[0100] where y t represents the power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the padding operation;

[0101] Based on formula (2), remove the trend component corresponding to the first decomposition parameter from the power load data to obtain the seasonal component corresponding to the first decomposition parameter;

[0102] Mt = y t - L t (2)

[0103] where M t represents the seasonal component.

[0104] In some implementations, the correlation calculation module 303 is configured to:[[]]

[0105] Obtain a first correlation coefficient between the trend component in the first component set and the power load data, and a second correlation coefficient between the seasonal component in the first component set and the power load data;

[0106] Take the weighted average of the first correlation coefficient and the second correlation coefficient as the correlation coefficient corresponding to the first component set;

[0107] where the first correlation coefficient and the second correlation coefficient are calculated based on formula (3):

[0108]

[0109] where r yX represents the correlation coefficient, X represents the seasonal component or the trend component, y represents the power load data, i represents any moment, and n represents the length of the time series.

[0110] In some implementations, the training module 305 is configured to:[[]]

[0111] Input the power load data into the power load prediction model, and decompose the power load data through the at least one target decomposition parameter in the power load prediction model to obtain a target component set, where the target component set includes a target trend component and a target seasonal component;

[0112] Filter the target component set through the variable selection network in the power load prediction model to obtain the filtered target component set;

[0113] Process the filtered target component set through the encoding and decoding network in the power load prediction model, output the power load prediction result of the target area in the second time period, calculate a loss value based on the power load prediction result and the real result of the target area in the second time period, and update the model parameters of the power prediction model based on the loss value, where the second time period is after the first time period.

[0114] Figure 4It is a schematic structural diagram of an electronic device. The electronic device can be configured as the above-mentioned server. Among them, the electronic device 400 can vary greatly due to configuration or performance differences, and may include one or more processors (Central Processing Units, CPU) 401 and one or more memories 402. Among them, at least one program code is stored in the one or more memories 402, and the at least one program code is loaded and executed by the one or more processors 401 to implement the process executed by the server in the power load forecasting method provided by each of the above method embodiments. Of course, the electronic device 400 may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The electronic device 400 may also include other components for implementing the functions of the device, which will not be elaborated here.

[0115] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of the present application are pointed out by the following claims.

[0116] It should be understood that the present application is not limited to the exact structure already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. A method for predicting power load, characterized in that: The method comprises: Obtaining power load data of a target area in a first time period; Decomposing the power load data based on multiple decomposition parameters by using a moving average method to obtain a component set corresponding to each decomposition parameter, wherein the component set includes a trend component and a seasonal component; Calculating the correlation between each component set and the power load data to obtain a correlation coefficient corresponding to each component set, wherein the correlation coefficient is used to reflect the matching degree between the component set and the power load data; Based on the correlation coefficient corresponding to each component set, determining at least one target decomposition parameter that meets the condition from the multiple decomposition parameters; The at least one target decomposition parameter is used as a hyperparameter of the power load prediction model, and the power load prediction model is trained using the power load data to obtain a trained power load prediction model, and the trained power load prediction model is used to predict the power load in the target area.

2. The method according to claim 1, characterized in that The moving average method is used to decompose the power load data based on multiple decomposition parameters to obtain a component set corresponding to each decomposition parameter, including: After the power load data is filled, based on a first decomposition parameter and formula (1), the filled power load data is decomposed into trend components corresponding to the first decomposition parameter, wherein the first decomposition parameter refers to any one of the multiple decomposition parameters; L t =Aυgpool(Padding(y t )) (1) Among them, y t Indicates power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the filling operation; Based on formula (2), the trend component corresponding to the first decomposition parameter is removed from the power load data to obtain the seasonal component corresponding to the first decomposition parameter; M t =y t -L t (2) Among them, M t represents the seasonal component.

3. The method according to claim 1, characterized in that The performing of correlation calculation between each component set and the power load data to obtain a correlation coefficient corresponding to each component set includes: Obtaining a first correlation coefficient between a trend component in a first component set and the power load data, and a second correlation coefficient between a seasonal component in the first component set and the power load data; Taking a weighted average of the first correlation coefficient and the second correlation coefficient as the correlation coefficient corresponding to the first component set; The first correlation coefficient and the second correlation coefficient are calculated based on formula (3): Among them, r yX represents the correlation coefficient, X represents the seasonal component or trend component, y represents the power load data, i represents any time, and n represents the length of the time series.

4. The method according to claim 1, characterized in that: The using the power load data to train the power load prediction model includes: Inputting the power load data into the power load forecasting model, decomposing the power load data by using the at least one target decomposition parameter in the power load forecasting model to obtain a target component set, wherein the target component set includes a target trend component and a target seasonal component; The target component set is screened by a variable selection network in the power load forecasting model to obtain the screened target component set; The filtered target component set is processed through the encoding and decoding network in the power load forecasting model, and the power load forecast result of the target area in the second time period is output. The loss value is calculated based on the power load forecast result and the actual result of the target area in the second time period, and the model parameters of the power forecasting model are updated based on the loss value. The second time period is after the first time period.

5. The method according to claim 1, characterized in that The power load data includes a power load value, weather data, and calendar data of the target area during the first time period.

6. A power load forecasting device, characterized in that: The device comprises: An acquisition module, used for acquiring power load data of a target area in a first time period; A decomposition module, used to decompose the power load data based on a plurality of decomposition parameters by using a moving average method to obtain a component set corresponding to each decomposition parameter, wherein the component set includes a trend component and a seasonal component; A correlation calculation module, used to calculate the correlation between each component set and the power load data, and obtain a correlation coefficient corresponding to each component set, wherein the correlation coefficient is used to reflect the matching degree between the component set and the power load data; A determination module, configured to determine at least one target decomposition parameter that meets the condition from the multiple decomposition parameters based on the correlation coefficient corresponding to each component set; A training module is used to use the at least one target decomposition parameter as a hyperparameter of the power load forecasting model, and use the power load data to train the power load forecasting model to obtain a trained power load forecasting model, and the trained power load forecasting model is used to predict the power load of the target area.

7. The device according to claim 6, characterized in that The decomposition module is used to: After the power load data is filled, based on a first decomposition parameter and formula (1), the filled power load data is decomposed into trend components corresponding to the first decomposition parameter, wherein the first decomposition parameter refers to any one of the multiple decomposition parameters; L t =Aυgpool(Padding(y t )) (1) Among them, y t Indicates power load data, L t represents the trend component, AvgPool(.) represents the moving average function, and Padding(.) represents the filling operation; Based on formula (2), the trend component corresponding to the first decomposition parameter is removed from the power load data to obtain the seasonal component corresponding to the first decomposition parameter; M t =y t -L t (2) Among them, M t represents the seasonal component.

8. An electronic device, characterized in that: The electronic device includes a processor and a memory, the memory stores program code, and the processor is used to execute the program code to implement the power load forecasting method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: When the program code in the computer-readable storage medium is executed by a processor of an electronic device, the electronic device is enabled to execute the power load forecasting method according to any one of claims 1 to 5.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor of an electronic device, the electronic device is enabled to perform the power load forecasting method according to any one of claims 1 to 5.

Images