Power load prediction method and system, and computer equipment

Abstract

The invention provides a power load prediction method and system, and computer equipment, and the method comprises the steps: obtaining historical meteorological data and historical power load data corresponding to a historical date for a target region; performing hybrid coding on the historical meteorological data and the historical electrical load data, and decoding a hybrid coding result to obtain meteorological-load data; performing feature extraction on the historical meteorological data, the historical electrical load data and the meteorological-load data to obtain target feature data; performing aggregation processing on the target feature data, and outputting fused feature data; and performing power load prediction on the fused feature data by using a pre-constructed deep learning model, and outputting a prediction result. According to the power load prediction method combining deep learning and meteorological load coding, the deep learning technology is introduced, the capturing capacity of the model for a complex data mode is enhanced, and the response speed and accuracy of the model for external environment change are improved through meteorological load coding.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a power load forecasting method and system, and computer equipment. Background Art

[0002] With the rapid development of the economy and society, people's demand for electricity is increasing day by day, and the volatility of power load values ​​is also increasing. The power system is a top priority in the field of infrastructure, playing a key supporting and promoting role in economic and social development. The fluctuation of power load affects the construction of power infrastructure. Therefore, accurate prediction of power load is crucial to the entire power system.

[0003] At present, the stability of the power system has been greatly improved compared with the past, but with the continuous increase in the global population, the rapid economic development and the increasingly severe global climate, we still need to use power resources more efficiently. Power load data is closely related to power demand, and the supply and demand relationship of the power system should maintain a dynamic balance, so that power companies can formulate correct operation plans and strategies to adapt to the power market demand.

[0004] In the prior art, power load forecasting mainly adopts time series forecasting, traditional machine learning forecasting, neural network forecasting and expert system forecasting. Although they can provide relatively accurate forecasting results in some cases, they are often difficult to adapt to the nonlinear and dynamic characteristics of power load data. Especially in the face of complex meteorological conditions and changing socio-economic activities, the forecasting accuracy and adaptability of these methods are often limited. In addition, existing forecasting models rarely fully consider the complex relationship between meteorological factors and power load, resulting in inaccurate forecasting results. Furthermore, with the continuous increase in data volume, traditional forecasting models may encounter problems of low computational efficiency when processing large-scale data sets. Summary of the invention

[0005] To this end, the present invention provides a power load forecasting method and system, and computer equipment, aiming to solve the technical problem in the prior art that power load forecasting cannot adapt to the increasing complexity and dynamic requirements of the power system.

[0006] To achieve the above objectives, the present invention adopts the following technical solutions:

[0007] According to a first aspect of the present invention, the present invention provides a method for predicting power load, the method comprising:

[0008] For the target area, obtain historical meteorological data and historical power load data corresponding to historical dates;

[0009] Performing mixed coding on the historical meteorological data and the historical electric load data, and decoding the mixed coding result to obtain meteorological-load data;

[0010] Extracting features from the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data;

[0011] Aggregating the target feature data and outputting fused feature data;

[0012] The pre-built deep learning model is used to predict the power load of the fused feature data and output the prediction result.

[0013] Optionally, before the mixed encoding of the historical meteorological data and the historical electric load data, the method further includes:

[0014] The historical meteorological data and the historical electric load data are interpolated based on the generative adversarial network technology to obtain complete historical meteorological data and historical electric load data.

[0015] Optionally, the historical meteorological data and the historical electric load data are mixedly encoded, and the mixed encoding result is decoded to obtain meteorological-load data, including:

[0016] Construct a recurrent encoder based on a bidirectional LSTM structure in the Keras framework;

[0017] Using the cyclic encoder to perform mixed encoding on the historical meteorological data and the historical electric load data to obtain meteorological-load characteristics;

[0018] Constructing a decoder based on a bidirectional LSTM structure; the decoder uses the final state of the cyclic encoder as an initial state;

[0019] The decoder is used to decode the weather-load characteristics to obtain weather-load data.

[0020] Optionally, the extracting features of the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data includes:

[0021] Standardizing the historical meteorological data, the historical electric load data and the meteorological-load data to obtain standard meteorological data, standard electric load data and standard meteorological-load data;

[0022] A Ridge regression fitting model is used, different regularization parameters are introduced, and feature weights of the standard meteorological data, the standard electric load data, and the standard meteorological-load data are adjusted to extract target feature data.

[0023] Optionally, the aggregating the target feature data to output fused feature data includes:

[0024] Step S1: The target feature data is used to form a feature data set S = {s 1 ,s 2 ,s 3};

[0025] Step S2: randomly selecting K different samples from the feature data set as initial clustering centers of K-means;

[0026] Step S3: traverse all samples, calculate the distance between each sample and each cluster center, and classify each sample into the class corresponding to the cluster center with the shortest distance;

[0027] Step S4: traverse all feature types and recalculate the cluster center of each feature type. If the cluster center changes, repeat steps S3 to S4 until the cluster center no longer changes, and output the clustering result as fused feature data.

[0028] Optionally, before using the pre-built deep learning model to perform power load forecasting on the fused feature data, the method further includes:

[0029] Using a particle swarm optimization algorithm to optimize the hyperparameters of the fusion feature data to obtain the optimal hyperparameters;

[0030] Build a load forecasting model based on the LightGBM algorithm;

[0031] The optimal hyperparameters are used as model inputs of the load forecasting model.

[0032] Optionally, the using a pre-built deep learning model to perform power load forecasting on the fused feature data includes:

[0033] estimating a constant value that minimizes a loss function of the load forecasting model, and initializing the load forecasting model using the constant value;

[0034] Performing model iteration to calculate the negative gradient value of the loss function of the current load forecasting model;

[0035] Fit a regression tree according to the negative gradient value to obtain a leaf node region corresponding to the number of iterations;

[0036] Using linear search to estimate the value of the leaf node region, minimizing the loss function, and obtaining a target load forecasting model;

[0037] The target load prediction model is used to perform power load prediction and output prediction results.

[0038] Optionally, the method further comprises:

[0039] Using residual analysis method to evaluate the prediction accuracy of the target load prediction model;

[0040] The target load prediction model is updated based on the prediction accuracy.

[0041] According to a second aspect of the present invention, the present invention provides a power load forecasting system, the system comprising:

[0042] A data acquisition module is used to acquire historical meteorological data and historical electric load data corresponding to historical dates for a target area;

[0043] A data processing module, used for performing mixed coding on the historical meteorological data and the historical electric load data, and decoding the mixed coding result to obtain meteorological-load data;

[0044] A feature extraction module, used for extracting features from the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data;

[0045] A feature fusion module, used to aggregate the target feature data and output fused feature data;

[0046] The model prediction module is used to use a pre-built deep learning model to perform power load prediction on the fused feature data and output a prediction result.

[0047] According to the third aspect of the present invention, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the power load forecasting method as described in any one of the first aspect of the present invention when executing the computer program.

[0048] The present invention adopts the above technical solution and has at least the following beneficial effects:

[0049] Through the scheme of the present invention, for the target area, the historical meteorological data and historical electric load data corresponding to the historical date are obtained; the historical meteorological data and the historical electric load data are mixed encoded, and the mixed encoding result is decoded to obtain meteorological-load data; the historical meteorological data, the historical electric load data and the meteorological-load data are feature extracted to obtain target feature data; the target feature data is aggregated and fused feature data is output; the fused feature data is predicted by using a pre-built deep learning model to predict the power load and output the prediction result. Therefore, a power load prediction method combining deep learning with meteorological load coding is proposed. By introducing deep learning technology, the model's ability to capture complex data patterns is enhanced, and through meteorological load coding, the model's response speed and accuracy to changes in the external environment are improved.

[0050] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 A schematic diagram of a flow chart of a power load forecasting method provided by an embodiment of the present invention is shown;

[0053] Figure 2 A schematic diagram of a data interpolation process according to an embodiment of the present invention is shown;

[0054] Figure 3 A schematic diagram showing the structure of a power load forecasting system provided by an embodiment of the present invention is shown;

[0055] Figure 4 A schematic diagram showing the structure of a power load forecasting system provided by another embodiment of the present invention is shown;

[0056] Figure 5 A schematic diagram of the physical structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0057] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0058] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0059] The embodiment of the present invention provides a method for predicting power load. Figure 1 As shown, at least the following steps S101 to S105 may be included:

[0060] Step S101, for a target area, obtaining historical meteorological data and historical electric load data corresponding to historical dates.

[0061] In the embodiment of the present invention, daily dimension statistics are performed on the target area, and based on the acquisition of historical 96-point load data and meteorological data, the change pattern of load with the meteorological mailbox is studied.

[0062] Step S102, performing mixed coding on the historical meteorological data and the historical electric load data, and decoding the mixed coding result to obtain meteorological-load data.

[0063] It should be noted that before the historical meteorological data and the historical electric load data are mixed-encoded, the embodiment of the present invention also needs to pre-process the data collected in step S101, that is, to perform data interpolation based on the generative adversarial network technology on the historical meteorological data and the historical electric load data to obtain complete historical meteorological data and historical electric load data.

[0064] Specifically, a generative adversarial network (GAN) can be used to interpolate the collected data and evaluate the interpolation effect by calculating the mean square error to ensure the applicability of the data. Figure 2As shown in the figure, the generative adversarial network (GAN) model consists of two parts: the generator and the discriminator. The generator accepts input data containing missing values ​​and a mask matrix, and generates interpolated data through a convolutional neural network (CNN) to replace the missing data in the original data. The discriminator accepts the data and mask matrix generated by the generator. The discriminator uses a convolutional neural network (CNN) to determine whether the input data is generated or original data, and outputs the true or false probability. After data interpolation, the mean square error can be used to measure the accuracy of the interpolation model. The calculation formula for the mean square error is:

[0065]

[0066] Where n is the total number of interpolated data sets to be observed, y i is the ith observation, is the interpolation value of the ith observation. By adjusting the parameters until the accuracy of the interpolation model meets the requirements, the missing values ​​in the data set are interpolated using the interpolation model to obtain complete historical meteorological data and historical electricity load data.

[0067] Furthermore, a recurrent encoder based on a bidirectional LSTM (Bi-LSTM) structure can be constructed under the Keras framework; the historical meteorological data and historical electric load data can be mixedly encoded using the recurrent encoder to obtain meteorological-load characteristics; a decoder based on a bidirectional LSTM structure can be constructed; the decoder uses the final state of the recurrent encoder as the initial state; and the decoder is used to decode the meteorological-load characteristics to obtain meteorological-load data.

[0068] As a special recurrent neural network, LSTM solves the long-term dependency problem of recurrent neural networks by introducing input gates, forget gates, and output gates. By performing bidirectional processing on LSTM, a bidirectional LSTM is generated, which has two LSTM layers working at the same time at each time step, one for processing the forward sequence and the other for processing the reverse sequence. The forward layer is used to process data in the natural order of the time series, and the reverse layer reverses the time series and then processes the data in reverse order. The data after data interpolation is imported into the encoder. While encoding, it can also be used as training data to train the encoder. The decoder is also a bidirectional LSTM structure, but it uses the final state of the encoder as the initial state and generates an output sequence time by time step. The output of the encoder and the input of the decoder are connected, and the final output is generated through one or more fully connected layers.

[0069] Step S103, extracting features from historical meteorological data, historical electric load data and meteorological-load data to obtain target feature data.

[0070] Since there are a lot of data in the prediction process, the embodiment of the present invention uses Ridge regression to perform feature selection on the data. Specifically, the historical meteorological data, historical electric load data and meteorological-load data can be standardized first to obtain standard meteorological data, standard electric load data and standard meteorological-load data; using the Ridge regression fitting model, different regularization parameters (α) are introduced to adjust the feature weights of the standard meteorological data, standard electric load data and standard meteorological-load data, and the importance of the features is observed through the regression coefficient to extract the target feature data.

[0071] Step S104, performing aggregation processing on the target feature data and outputting fused feature data;

[0072] The embodiment of the present invention aggregates the target feature data processed in step S105 to better capture patterns and trends in the data. This step uses the K-means algorithm, specifically steps S1 to S4 as follows:

[0073] Step S1: The target feature data form a feature data set S = {s 1 ,s 2 ,s 3};

[0074] Step S2: Randomly select K different samples in the feature data set as the initial clustering centers of K-means;

[0075] Step S3: traverse all samples, calculate the distance between each sample and each cluster center, and classify each sample into the class corresponding to the cluster center with the shortest distance;

[0076] Step S4: traverse all feature types and recalculate the cluster center of each feature type. If the cluster center changes, repeat steps S3 to S4 until the cluster center no longer changes, and output the clustering result as the fused feature data.

[0077] It should be noted that the embodiment of the present invention uses the Davidson-Borgin index to measure the clustering effect, denoted as I DBI , and remember I DBI The best number of clusters is K, I DBI The calculation formula is:

[0078]

[0079] Among them, H i is the i-th cluster center; ||·|| 2 represents the two-norm; is the average distance from the sample in the i-th class to the cluster center, and the calculation formula is:

[0080]

[0081] Among them, R i is the number of the i-th sample; G i，l is the lth sample in the i-th class.

[0082] Step S105, using the pre-built deep learning model to perform power load prediction on the fused feature data and output the prediction result.

[0083] In the embodiment of the present invention, a load forecasting model based on the LightGBM algorithm is selected. The LightGBM algorithm adopts the gradient boosting idea of ​​XGBoost and has a high model training speed and prediction accuracy. As a type of GBDT algorithm, the LightGBM algorithm is a decision tree algorithm constructed based on iteration. It forms a strong learner by combining weak learning. Given a weather-load feature data set D = {(x i ,y i ): i = 1, 2, ..., n, x i ∈R p ,y i ∈R}, where n is the number of samples and each sample has p features. By giving the loss function L(y, f(x)), the output regression tree The specific algorithm steps are as follows:

[0084] Estimate the constant value θ that minimizes the loss function of the load forecasting model, and use the constant value θ to initialize the load forecasting model, which is a tree with only one root node:

[0085]

[0086] Set m as the number of iterations, m = 1, 2, ..., M, and perform model iteration, including the following steps S1 to S4:

[0087] Step S1: For i=1, 2, ..., n, calculate the residual r im , that is, the negative gradient value of the loss function under the current model:

[0088]

[0089] Step S2: According to r im Fit a regression tree and get the leaf node region R of the mth tree mj , j = 1, 2, ..., J, where j represents the number of leaf nodes.

[0090] Step S3: For j=1, 2, ..., J, use linear search to estimate the value θ of the leaf node region mj , so that the loss function is minimized:

[0091]

[0092] Step S4: Update the regression tree: where I is the indicator function, when the regression tree determines x∈R mj Its value is 1 when , otherwise it is 0:

[0093]

[0094] It should be noted that in the embodiment of the present invention, in order to improve the prediction speed and enhance the robustness of the model, the particle swarm optimization algorithm is used to optimize the hyperparameters before building the LightGBM load forecasting model, and the obtained optimal hyperparameters are used as the model input of the load forecasting model. The algorithm simulates the foraging behavior of bird flocks and finds the optimal solution through collaboration and information sharing between individuals. The specific steps are as follows:

[0095] Define the parameters of the particle swarm optimization algorithm. Assume that the swarm size is N and the position vector of each particle is represented as X in D-dimensional space. i =(x i1 , x i2 , ..., x id , ..., x iD ), the velocity vector is represented by V i =(v i1 , v i2 , ..., v id , ..., v iD ), the optimal position of the individual particle is expressed as P i =(p i1 , p i2 , ..., p id , ..., p iD ), the optimal position of the group is denoted as P g =(p g1 , p g2 , ..., p gd , ..., p gD ). The optimal position of the group is the optimal position of all individuals, and the update formula of its speed and position is as follows:

[0096]

[0097] Among them, ω is the inertia weight; c 1 and c 2 are individual learning factors and social learning factors respectively; rand is a random number between 0 and 1. The position of the particle represents the value of the corresponding hyperparameter, and the speed controls the movement of the particle in the parameter space.

[0098] Step S5: After iterating q times, the final target load forecasting model is output:

[0099]

[0100] Furthermore, the target load forecasting model can be used to forecast the power load based on the meteorological data in the future period and output the forecast results.

[0101] Furthermore, the embodiment of the present invention can also use the residual analysis method to evaluate the prediction accuracy of the target load prediction model (model fit test); the target load prediction model is updated based on the prediction accuracy, and the residual analysis formula is as follows:

[0102]

[0103] Among them, yt is the observed value, is the model prediction value, and εt is the residual value.

[0104] Based on the residual analysis, the embodiment of the present invention can also evaluate the distribution of residuals by drawing a residual histogram to check whether the residual distribution satisfies the normal distribution. This step is very important for correlation testing. Based on the residual histogram, it can be checked whether the model fully captures the time series characteristics of the data, and based on the residual distribution test, it can be checked whether the residuals show the autocorrelation of the time series.

[0105] In addition, the embodiment of the present invention can also use the root mean square error to test the accuracy of the model prediction. Based on the residual analysis, the residual sum of squares is calculated as follows:

[0106]

[0107] The formula for calculating the root mean square error is as follows:

[0108]

[0109] Among them, S is the residual sum of squares, εt is the residual, and RMSE is the root mean square error.

[0110] It can be understood that the smaller the RMS error value, the smaller the difference between the predicted value and the observed value, indicating that the prediction accuracy of the target load prediction model is higher; the larger the RMS error, the greater the difference between the predicted value and the observed value, indicating that the prediction accuracy of the target load prediction model is lower. When the prediction accuracy of the target load prediction model does not meet the actual demand, the model is further iterated and updated to improve the prediction accuracy.

[0111] The embodiment of the present invention provides a method for predicting electric load, including: for a target area, obtaining historical meteorological data and historical electric load data corresponding to historical dates; performing mixed encoding on the historical meteorological data and historical electric load data, and decoding the mixed encoding result to obtain meteorological-load data; performing feature extraction on the historical meteorological data, historical electric load data and meteorological-load data to obtain target feature data; performing aggregation processing on the target feature data and outputting fused feature data; using a pre-built deep learning model to predict electric load on the fused feature data, and outputting the prediction result. Through the present invention, a method for predicting electric load by combining deep learning with meteorological load encoding is proposed, which introduces deep learning technology to enhance the model's ability to capture complex data patterns, and improves the model's response speed and accuracy to changes in the external environment through meteorological load encoding.

[0112] Further, as Figure 1 The specific implementation of the present invention provides a power load forecasting system, such as Figure 3 As shown, the system may include: a data acquisition module 310 , a data processing module 320 , a feature extraction module 330 , a feature fusion module 340 and a model prediction module 350 .

[0113] The data acquisition module 310 can be used to acquire historical meteorological data and historical electric load data corresponding to historical dates for a target area;

[0114] The data processing module 320 may be used to perform mixed coding on the historical meteorological data and the historical electric load data, and decode the mixed coding result to obtain meteorological-load data;

[0115] The feature extraction module 330 can be used to extract features from historical meteorological data, historical electric load data and meteorological-load data to obtain target feature data;

[0116] The feature fusion module 340 may be used to aggregate the target feature data and output fused feature data;

[0117] The model prediction module 350 can be used to use a pre-built deep learning model to perform power load prediction on the fused feature data and output the prediction results.

[0118] Alternatively, if Figure 4 As shown, another embodiment of the present invention provides a power load forecasting system, which also includes: a data interpolation module 360, a parameter optimization module 370 and a forecasting analysis module 380.

[0119] The data interpolation module 360 ​​can be used to perform data interpolation on the historical meteorological data and the historical electric load data based on the generative adversarial network technology before the mixed encoding of the historical meteorological data and the historical electric load data to obtain complete historical meteorological data and historical electric load data.

[0120] The parameter optimization module 370 can be used to optimize the hyperparameters of the fused feature data using a particle swarm optimization algorithm to obtain the best hyperparameters before using the pre-built deep learning model to predict the power load on the fused feature data;

[0121] Build a load forecasting model based on the LightGBM algorithm;

[0122] The optimal hyperparameters are used as model inputs for the load forecasting model.

[0123] The prediction analysis module 380 can be used to evaluate the prediction accuracy of the target load prediction model using a residual analysis method;

[0124] The load forecasting model is updated based on the forecast accuracy target.

[0125] Optionally, the data processing module 320 can also be used to construct a recurrent encoder based on a bidirectional LSTM structure under the Keras framework;

[0126] The historical meteorological data and the historical electric load data are mixedly encoded using a cyclic encoder to obtain the meteorological-load characteristics.

[0127] Construct a decoder based on a bidirectional LSTM structure; the decoder uses the final state of the recurrent encoder as the initial state;

[0128] Optionally, the feature extraction module 330 may also be used to perform standardization processing on historical meteorological data, historical electric load data and meteorological-load data to obtain standard meteorological data, standard electric load data and standard meteorological-load data;

[0129] The Ridge regression fitting model is used, different regularization parameters are introduced, and the feature weights of standard meteorological data, standard electric load data and standard meteorological-load data are adjusted to extract the target feature data.

[0130] Optionally, the feature fusion module 340 may also be used to perform the following steps:

[0131] Step S1: The target feature data form a feature data set S = {s 1 ,s 2 ,s 3};

[0132] Step S2: Randomly select K different samples in the feature data set as the initial clustering centers of K-means;

[0133] Step S3: traverse all samples, calculate the distance between each sample and each cluster center, and classify each sample into the class corresponding to the cluster center with the shortest distance;

[0134] Step S4: traverse all feature types and recalculate the cluster center of each feature type. If the cluster center changes, repeat steps S3 to S4 until the cluster center no longer changes, and output the clustering result as the fused feature data.

[0135] Optionally, the model prediction module 350 may also be used to estimate a constant value that minimizes the loss function of the load prediction model, and initialize the load prediction model using the constant value;

[0136] Perform model iteration and calculate the negative gradient value of the loss function of the current load forecasting model;

[0137] Fit a regression tree according to the negative gradient value and obtain the leaf node area corresponding to the number of iterations;

[0138] The value of the leaf node region is estimated by linear search to minimize the loss function and obtain the target load forecasting model.

[0139] Use the target load forecasting model to forecast power load and output the forecast results.

[0140] It should be noted that for other corresponding descriptions of the functional modules involved in the power load forecasting system provided in the embodiment of the present invention, reference can be made to Figure 1 The corresponding description of the method shown will not be repeated here.

[0141] Based on the above Figure 1 The method shown and Figure 4 The embodiment of the system shown in the figure, the embodiment of the present invention also provides a physical structure diagram of a computer device, such as Figure 5 As shown, the computer device may include a communication bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device, wherein each functional unit may communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the steps of the method for constructing a page navigation bar in the above embodiment.

[0142] Those skilled in the art can clearly understand that the specific working processes of the systems, devices, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and for the sake of brevity, they are not further described here.

[0143] In addition, the functional units in various embodiments of the present invention may be physically independent of each other, or two or more functional units may be integrated together, or all functional units may be integrated into one processing unit. The above integrated functional units may be implemented in the form of hardware, or in the form of software or firmware.

[0144] Those skilled in the art can understand that if the integrated functional unit is implemented in the form of software and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can essentially or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, which includes a number of instructions to enable a computing device (such as a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention when running the instructions. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes.

[0145] Alternatively, all or part of the steps of implementing the aforementioned method embodiments may be accomplished by hardware associated with program instructions (such as a computing device such as a personal computer, a server, or a network device), and the program instructions may be stored in a computer-readable storage medium. When the program instructions are executed by a processor of a computing device, the computing device executes all or part of the steps of the methods described in the embodiments of the present invention.

[0146] 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 within the spirit and principles of the present invention, the technical solutions described in the aforementioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate from the protection scope of the present invention.

Claims

1. A method for predicting power load, characterized in that: The method comprises: For the target area, obtain historical meteorological data and historical power load data corresponding to historical dates; Performing mixed coding on the historical meteorological data and the historical electric load data, and decoding the mixed coding result to obtain meteorological-load data; Extracting features from the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data; Aggregating the target feature data and outputting fused feature data; The pre-built deep learning model is used to predict the power load of the fused feature data and output the prediction result.

2. The method according to claim 1, characterized in that Before the mixed encoding of the historical meteorological data and the historical electric load data, the method further includes: The historical meteorological data and the historical electric load data are interpolated based on the generative adversarial network technology to obtain complete historical meteorological data and historical electric load data.

3. The method according to claim 1, characterized in that The mixing and encoding the historical meteorological data and the historical electric load data, and decoding the mixed encoding result to obtain meteorological-load data, includes: Construct a recurrent encoder based on a bidirectional LSTM structure in the Keras framework; Using the cyclic encoder to perform mixed encoding on the historical meteorological data and the historical electric load data to obtain meteorological-load characteristics; Constructing a decoder based on a bidirectional LSTM structure; the decoder uses the final state of the cyclic encoder as an initial state; The decoder is used to decode the weather-load characteristics to obtain weather-load data.

4. The method according to claim 1, characterized in that The extracting features of the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data includes: Standardizing the historical meteorological data, the historical electric load data and the meteorological-load data to obtain standard meteorological data, standard electric load data and standard meteorological-load data; A Ridge regression fitting model is used, different regularization parameters are introduced, and feature weights of the standard meteorological data, the standard electric load data, and the standard meteorological-load data are adjusted to extract target feature data.

5. The method according to claim 1, characterized in that The step of aggregating the target feature data and outputting fused feature data includes: Step S1: The target feature data is used to form a feature data set S = {s1, s2, s3}; Step S2: randomly selecting K different samples from the feature data set as initial clustering centers of K-means; Step S3: traverse all samples, calculate the distance between each sample and each cluster center, and classify each sample into the class corresponding to the cluster center with the shortest distance; Step S4: traverse all feature types and recalculate the cluster center of each feature type. If the cluster center changes, repeat steps S3 to S4 until the cluster center no longer changes, and output the clustering result as fused feature data.

6. The method according to claim 1, characterized in that Before using the pre-built deep learning model to perform power load prediction on the fused feature data, the method further includes: Using a particle swarm optimization algorithm to optimize the hyperparameters of the fusion feature data to obtain the optimal hyperparameters; Build a load forecasting model based on the LightGBM algorithm; The optimal hyperparameters are used as model inputs of the load forecasting model.

7. The method according to claim 6, characterized in that The using of a pre-built deep learning model to perform power load forecasting on the fused feature data includes: estimating a constant value that minimizes a loss function of the load forecasting model, and initializing the load forecasting model using the constant value; Performing model iteration to calculate the negative gradient value of the loss function of the current load forecasting model; Fit a regression tree according to the negative gradient value to obtain a leaf node region corresponding to the number of iterations; Using linear search to estimate the value of the leaf node region, minimizing the loss function, and obtaining a target load forecasting model; The target load prediction model is used to perform power load prediction and output prediction results.

8. The method according to claim 7, characterized in that The method further comprises: Using residual analysis method to evaluate the prediction accuracy of the target load prediction model; The target load prediction model is updated based on the prediction accuracy.

9. A power load forecasting system, characterized in that: The system comprises: A data acquisition module is used to acquire historical meteorological data and historical electric load data corresponding to historical dates for a target area; A data processing module, used for performing mixed coding on the historical meteorological data and the historical electric load data, and decoding the mixed coding result to obtain meteorological-load data; A feature extraction module, used for extracting features from the historical meteorological data, the historical electric load data and the meteorological-load data to obtain target feature data; A feature fusion module, used to aggregate the target feature data and output fused feature data; The model prediction module is used to use a pre-built deep learning model to perform power load prediction on the fused feature data and output a prediction result.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the power load forecasting method according to any one of claims 1 to 8 is implemented.

Images