Power load prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a power load prediction method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining previous load time series data which are non-stationary time series data; variational mode decomposition is carried out on the prior load time sequence data, a plurality of target intrinsic load modes are obtained, and the target intrinsic load modes are modes related to temperature characteristics; a pre-trained load prediction model is called, the target intrinsic load mode is input into the load prediction model, a target load prediction value which is output by the load prediction model and matched with the preceding load time sequence data is obtained, the target load prediction value is used for representing a post-load, and the post-load time sequence data is used for representing the post-load time sequence data. The load prediction model is used for obtaining a target load prediction value according to the target intrinsic load mode. Accurate load prediction can be carried out on a non-stationary load time sequence, so that the accuracy of seasonal load prediction can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a power load forecasting method, device, electronic equipment and storage medium. Background Art

[0002] In the power system, the demand pattern of power load is a complex and changeable process, which is affected by many factors, among which seasonal changes and temperature fluctuations are particularly important factors. These seasonal changes and temperature fluctuations can cause significant shifts in the power load on the time axis, making the load time series curve present non-stationary characteristics.

[0003] It is known from related technologies that traditional load forecasting methods often predict stable load time series, but cannot accurately predict non-stationary load time series in practice.

[0004] Therefore, finding a power load forecasting method that can accurately forecast non-stationary load sequences has become a current research hotspot. Summary of the invention

[0005] The present invention provides a power load forecasting method, device, electronic equipment and storage medium, which can realize accurate load forecasting for non-stationary load time series, thereby improving the accuracy of seasonal load forecasting.

[0006] The present invention provides a method for predicting electric load, the method comprising: obtaining prior load time series data, wherein the prior load time series data is non-stationary time series data; performing variational modal decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calling a pre-trained load prediction model, and inputting the target intrinsic load mode into the load prediction model to obtain a target load prediction value output by the load prediction model and matching the prior load time series data, wherein the target load prediction value is used to characterize the subsequent load, and the load prediction model is used to obtain the target load prediction value according to the target intrinsic load mode.

[0007] According to an electric load forecasting method provided by the present invention, the target intrinsic load mode is input into the load forecasting model to obtain a target load forecast value output by the load forecasting model that matches the prior load time series data, specifically comprising: inputting a plurality of the target intrinsic load modes into the load forecasting model respectively to obtain a plurality of load forecast values ​​output by the load forecasting model that correspond to the target intrinsic load modes; and performing weighted reconstruction processing on the plurality of load prediction values ​​to obtain a target load forecast value that matches the prior load time series data.

[0008] According to an electric power load forecasting method provided by the present invention, before performing weighted reconstruction processing on multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data, the method includes: determining a weighting coefficient of the load prediction value based on the frequency characteristics of the target intrinsic load mode corresponding to the load prediction value; performing weighted reconstruction processing on multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data specifically includes: performing weighted reconstruction processing on multiple load prediction values ​​based on the weighting coefficient of the load prediction value to obtain a target load prediction value that matches the prior load time series data.

[0009] According to a power load forecasting method provided by the present invention, the prior load time series data is subjected to variational modal decomposition to obtain multiple target intrinsic load modes, which specifically includes: initializing a penalty parameter and the number of decomposition levels; based on the penalty parameter and the number of decomposition levels, iteratively optimizing through variational modal decomposition to decompose the prior load time series data into multiple initial intrinsic load modes, wherein each of the initial intrinsic load modes corresponds to a specific frequency range, and the specific frequency range corresponding to the initial intrinsic load mode reflects the load time series characteristics of the initial intrinsic load mode; and extracting multiple target intrinsic load modes from the multiple initial intrinsic load modes.

[0010] According to a power load forecasting method provided by the present invention, a plurality of target intrinsic load modes are extracted from a plurality of the initial intrinsic load modes, specifically comprising: performing correlation analysis on the plurality of the initial intrinsic load modes to obtain features corresponding to each of the initial intrinsic load modes, wherein the features include at least temperature features; and extracting a plurality of target intrinsic load modes from a plurality of the initial intrinsic load modes based on the features corresponding to each of the initial intrinsic load modes.

[0011] According to a method for power load forecasting provided by the present invention, the load forecasting model is pre-trained in the following manner: constructing a training data set, wherein the training data set includes a first training data set on a source domain and a second training data set on a target domain, wherein the first training data set and the second training data set are historical load time series data belonging to different seasons, and the number of the first training data sets is greater than the number of the second training data sets; jointly training a load forecasting model based on the first training data set and the second training data set to obtain an initially trained load forecasting model; based on a timing offset minimization algorithm, migrating model parameters of the initially trained load forecasting model to the target domain, and adjusting the model parameters in a gradient descent optimization manner based on a loss function to obtain adjusted model parameters, wherein the loss function is determined based on the first training data set on the source domain and the second training data set on the target domain; based on the adjusted model parameters, obtaining the load forecasting model so that the load forecasting model has generalization properties on the source domain and the target domain.

[0012] According to an electric load forecasting method provided by the present invention, after obtaining the load forecasting model based on the adjusted model parameters, the method further includes: calling a verification training set, wherein the verification training set is a verification training set on the target domain; verifying the load forecasting model based on the verification training set to obtain a verification result; if the verification result meets the preset requirements, using the load forecasting model as the final load forecasting model; if the verification result does not meet the preset requirements, repeatedly executing the steps of jointly training the load forecasting model based on the first training data set and the second training data set to obtain the initial trained load forecasting model to obtaining the load forecasting model based on the adjusted model parameters, until the verification result of the load forecasting model meets the preset requirements, and using the obtained load forecasting model as the final load forecasting model.

[0013] The present invention also provides an electric load prediction device, which includes: an acquisition module for acquiring prior load time series data, wherein the prior load time series data is non-stationary time series data; a decomposition module for performing variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; a prediction module for calling a pre-trained load prediction model and inputting the target intrinsic load mode into the load prediction model to obtain a target load prediction value output by the load prediction model that matches the prior load time series data, wherein the target load prediction value is used to characterize the subsequent load time series data, and the load prediction model is used to obtain the target load prediction value based on the target intrinsic load mode.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the power load forecasting method described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the power load forecasting method as described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned power load forecasting methods.

[0017] The present invention provides a method, device, electronic device and storage medium for predicting electric loads, which obtains prior load time series data, wherein the prior load time series data is non-stationary time series data; performs variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calls a pre-trained load prediction model, and inputs the target intrinsic load mode into the load prediction model, so as to obtain a target load prediction value output by the load prediction model that matches the prior load time series data. This achieves accurate load prediction for non-stationary load time series, thereby improving the accuracy of seasonal load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of the power load forecasting method provided by the present invention.

[0020] Figure 2 It is a flow chart of inputting a target intrinsic load mode into a load forecasting model to obtain a target load forecasting value output by the load forecasting model and matching the previous load time series data.

[0021] Figure 3 It is a flow chart of the invention for performing weighted reconstruction processing on multiple load prediction values ​​to obtain a target load prediction value that matches the previous load time series data.

[0022] Figure 4It is a flow chart of performing variational modal decomposition on prior load time series data to obtain multiple target intrinsic load modes provided by the present invention.

[0023] Figure 5 It is one of the flow charts of the pre-trained load prediction model provided by the present invention.

[0024] Figure 6 This is the second flow chart of the pre-trained load prediction model provided by the present invention.

[0025] Figure 7 It is a structural schematic diagram of the power load prediction device provided by the present invention.

[0026] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The power load forecasting method provided by the present invention is based on a combined forecasting model of variational mode decomposition (VMD) and a timing offset minimization algorithm. It strives to improve the accuracy of seasonal load forecasting by performing modal decomposition on the load sequence and applying the timing offset minimization algorithm for forecasting.

[0029] Figure 1 It is a flow chart of the power load forecasting method provided by the present invention.

[0030] The following will be combined Figure 1 The process of the power load forecasting method provided by the present invention is described.

[0031] In an exemplary embodiment of the present invention, Figure 1 It can be seen that the power load prediction method may include steps 110 to 130, and each step will be introduced below.

[0032] In step 110, previous load time series data is obtained, wherein the previous load time series data is non-stationary time series data.

[0033] In one embodiment, the previous load time series data may be obtained, wherein the previous load time series data may be considered as non-stationary time series data. That is, data in which the power load may shift on the time axis. In one example, the previous load time series data may be the load time series data from one day ago. In this embodiment, the previous duration of the previous load time series data is not specifically limited.

[0034] In step 120, variational modal decomposition is performed on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics.

[0035] In another embodiment, variational modal decomposition can be performed on the prior load time series data to decompose the prior load time series data into a number of intrinsic mode functions (IMFs) with different frequency characteristics, that is, to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to the temperature characteristic.

[0036] In step 130, a pre-trained load forecasting model is called, and the target intrinsic load mode is input into the load forecasting model to obtain a target load forecasting value output by the load forecasting model that matches the prior load time series data.

[0037] In another embodiment, a pre-trained load forecasting model can be called, wherein the load forecasting model can be used to obtain a target load forecast value based on the target intrinsic load modal prediction. During the application process, the target intrinsic load modality can be input into the load forecasting model, so that the target load forecast value output by the load forecasting model and matching the previous load time series data can be obtained. Among them, the target load forecast value can be used to characterize the subsequent load. In one example, the subsequent load can be the load for the next 4 hours. In this embodiment, the subsequent duration of the subsequent load is not specifically limited. In this embodiment, the accuracy of seasonal load forecasting can be improved by performing modal decomposition and prediction on the load sequence. In other words, the load time series is decomposed into multiple intrinsic mode functions through variational mode decomposition, and temperature-related features (corresponding to the target intrinsic load mode) are extracted in a targeted manner, thereby capturing the seasonal changes of the load more accurately.

[0038] The present invention provides a method for predicting electric loads, which obtains prior load time series data, wherein the prior load time series data is non-stationary time series data; performs variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calls a pre-trained load prediction model, and inputs the target intrinsic load mode into the load prediction model, so as to obtain a target load prediction value output by the load prediction model that matches the prior load time series data. This method realizes accurate load prediction for non-stationary load time series, thereby improving the accuracy of seasonal load prediction.

[0039] Figure 2 It is a flow chart of inputting a target intrinsic load mode into a load forecasting model to obtain a target load forecasting value output by the load forecasting model and matching the previous load time series data.

[0040] The following will be combined Figure 2 The process of inputting the target intrinsic load mode into the load forecasting model provided by the present invention to obtain the target load forecasting value output by the load forecasting model and matching the previous load time series data is described.

[0041] In an exemplary embodiment of the present invention, Figure 2 It can be seen that inputting the target intrinsic load mode into the load forecasting model to obtain the target load forecast value output by the load forecasting model that matches the previous load time series data may include steps 210 to 220, and each step will be introduced below.

[0042] In step 210, a plurality of target intrinsic load modes are respectively input into a load forecasting model, and a plurality of load forecasting values ​​corresponding to the target intrinsic load modes output by the load forecasting model are respectively obtained; In step 220, a plurality of load prediction values ​​are weighted and reconstructed to obtain a target load prediction value that matches the previous load time series data.

[0043] In one embodiment, each target intrinsic load mode can be input into the load prediction model respectively, and the load prediction values ​​corresponding to the target intrinsic load mode output by the load prediction model can be obtained respectively. Further, a weighted reconstruction process is performed on multiple load prediction values, so that the prediction value of the overall load can be obtained, that is, the target load prediction value matching the previous load time series data can be obtained. In this embodiment, the target load prediction value is comprehensively determined based on the load prediction values ​​corresponding to each target intrinsic load mode, so that the prediction accuracy and comprehensiveness of the target load prediction value can be improved.

[0044] Figure 3It is a flow chart of the invention for performing weighted reconstruction processing on multiple load prediction values ​​to obtain a target load prediction value that matches the previous load time series data.

[0045] The following will be combined Figure 3 The process of weighted reconstruction of multiple load forecast values ​​to obtain a target load forecast value that matches the previous load time series data is described.

[0046] In an exemplary embodiment of the present invention, Figure 3 It can be seen that performing weighted reconstruction processing on the multiple load prediction values ​​to obtain the target load prediction value matching the prior load time series data may include step 310 and step 320, and each step will be introduced below.

[0047] In step 310, a weighting coefficient of the load prediction value is determined based on the frequency characteristics of the target intrinsic load mode corresponding to the load prediction value; In step 320, based on the weighting coefficients of the load prediction values, a weighted reconstruction process is performed on the multiple load prediction values ​​to obtain a target load prediction value that matches the previous load time series data.

[0048] In one embodiment, the weighting coefficient of the load prediction value can be determined based on the frequency characteristics of the target intrinsic load mode corresponding to the load prediction value. It is understandable that the frequency characteristics or energy characteristics of the target intrinsic load mode can be used to determine which target intrinsic load modes have a greater impact on the load, and the weighting coefficient corresponding to the target intrinsic load mode with a greater impact on the load can be determined to be greater.

[0049] Furthermore, based on the weight coefficient of the load prediction value, a weighted reconstruction process can be performed on multiple load prediction values, so that a target load prediction value matching the previous load time series data can be obtained. In this embodiment, by performing a weighted reconstruction process on multiple load prediction values ​​to obtain a target load prediction value, the contribution of each load prediction value can be fully considered, and the target load prediction value can be determined according to its contribution, thereby improving the prediction accuracy of the target load prediction value.

[0050] Figure 4 It is a flow chart of performing variational modal decomposition on prior load time series data to obtain multiple target intrinsic load modes provided by the present invention.

[0051] The following will be combined Figure 4 The process of performing variational modal decomposition on prior load time series data to obtain multiple target intrinsic load modes is explained.

[0052] In an exemplary embodiment of the present invention, Figure 4It can be seen that performing variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes may include steps 410 to 430, and each step will be introduced below.

[0053] In step 410, the penalty parameter and the number of decomposition levels are initialized.

[0054] In step 420, based on the penalty parameter and the number of decomposition levels, variational modal decomposition is used for iterative optimization to decompose the prior load time series data into multiple initial intrinsic load modes, wherein each initial intrinsic load mode corresponds to a specific frequency range, and the specific frequency range corresponding to the initial intrinsic load mode reflects the load time series characteristics of the initial intrinsic load mode.

[0055] In one embodiment, the penalty parameter and the number of decomposition layers can be initialized, and appropriate penalty parameters and decomposition layers can be selected. Further, signal decomposition can be performed, that is, based on the penalty parameter and the number of decomposition layers, iterative optimization is performed through variational modal decomposition, and the prior load time series data is gradually decomposed into multiple initial intrinsic load modes, so that each initial intrinsic load mode corresponds to a specific frequency range. Among them, the specific frequency range corresponding to each initial intrinsic load mode reflects the load time series characteristics of the initial intrinsic load mode, that is, each mode reflects different frequency components in the load sequence, for example, low-frequency components can correspond to seasonal characteristics, and high-frequency components can correspond to random fluctuations and abnormal changes.

[0056] In yet another embodiment, the prior load timing data may be is decomposed into components, that is, the number of decomposition levels is , and ensure that the decomposed sequence (corresponding to each initial eigenload mode) is a modal component with a finite bandwidth and a center frequency, and the sum of the estimated bandwidths of each mode is minimized. The constraint is that the sum of the frequencies of all modes is equal to the frequency of the original signal (corresponding to the prior load time series data).

[0057] In another embodiment, the variational mode decomposition constrained variational model can be represented by formula (1): (1) in, is the modal function; is the modal center frequency; Indicates instantaneous pulse signal; represents the Hilbert transform kernel; represents the time derivative; represents the frequency shift factor; express The square of the norm.

[0058] In step 430 , a plurality of target eigenload modes are extracted from the plurality of initial eigenload modes.

[0059] In another embodiment, multiple target intrinsic load modes can be extracted from the multiple initial intrinsic load modes obtained by decomposition. It can be understood that the target intrinsic load modes can be used for subsequent load prediction.

[0060] In another exemplary embodiment of the present invention, continuing with the above Figure 4 The above embodiment is used as an example for explanation, wherein, among multiple initial intrinsic load modes, extracting multiple target intrinsic load modes (corresponding to step 430) can be implemented in the following manner: Performing correlation analysis on a plurality of initial intrinsic load modes to obtain features corresponding to each initial intrinsic load mode, wherein the features at least include temperature features; Based on the features corresponding to each initial intrinsic load mode, a plurality of target intrinsic load modes are extracted from the plurality of initial intrinsic load modes.

[0061] In one embodiment, the frequency characteristics and the correlation with the load of each initial intrinsic load mode can be analyzed based on the correlation analysis, that is, correlation analysis can be performed on multiple initial intrinsic load modes to obtain features corresponding to each initial intrinsic load mode, wherein the features at least include temperature features. It can be understood that the initial intrinsic load mode with temperature features is correlated with the load. Furthermore, based on the features corresponding to each initial intrinsic load mode, multiple target intrinsic load modes with temperature features can be extracted from multiple initial intrinsic load modes. It can be understood that the extracted target intrinsic load modes with temperature features can be used for subsequent load prediction.

[0062] Figure 5 It is one of the flow charts of the pre-trained load prediction model provided by the present invention.

[0063] The following will be combined Figure 5 The process of pre-training the load prediction model provided by the present invention is described.

[0064] In an exemplary embodiment of the present invention, Figure 5 It can be seen that the pre-training load prediction model may include steps 510 to 540, and each step will be introduced below.

[0065] In step 510, a training data set is constructed, wherein the training data set includes a first training data set on a source domain and a second training data set on a target domain, wherein the first training data set and the second training data set are historical load time series data belonging to different seasons, and the number of the first training data set is greater than the number of the second training data set.

[0066] In one embodiment, a training data set may be constructed, wherein the training data set may include a first training data set on a source domain and a second training data set on a target domain, wherein the first training data set and the second training data set are historical load time series data belonging to different seasons, and the number of first training data sets is greater than the number of second training data sets. For ease of explanation, the historical load time series data in summer is used as the first training data set on the source domain, and the historical load time series data in winter is used as the second training data set on the source domain. It should be noted that the load data on the source domain and the target domain have differences in statistical distribution, but there are still certain similarities in the overall trend.

[0067] In step 520, a load forecasting model is trained based on the first training data set and the second training data set to obtain an initially trained load forecasting model.

[0068] In step 530, based on the timing offset minimization algorithm, the model parameters of the initially trained load forecasting model are migrated to the target domain, and the model parameters are adjusted according to the gradient descent optimization method based on the loss function to obtain the adjusted model parameters, wherein the loss function is determined based on the first training data set on the source domain and the second training data set on the target domain.

[0069] In step 540, a load forecasting model is obtained based on the adjusted model parameters, so that the load forecasting model has generalization in the source domain and the target domain.

[0070] In one embodiment, the load forecasting model can be trained based on the first training data set and the second training data set to obtain the load forecasting model after initial training. Furthermore, the feature mapping relationship between seasons is learned based on a timing offset minimization algorithm, such as an Adversarial Transfer Learning (ATL) or an Adaptive Transfer Learning (Domain Adaptation) algorithm. In one example, a Generative Adversarial Network (GAN) or an AutoEncoder can also be used to reduce the distribution difference between the source domain and the target domain.

[0071] During the application process, the model parameters of the load forecasting model after initial training can be migrated to the target domain based on the timing offset minimization algorithm, and the model parameters can be adjusted according to the gradient descent optimization method based on the loss function to obtain the adjusted model parameters, wherein the loss function can be obtained based on the first training data set on the source domain and the second training data set on the target domain.

[0072] Furthermore, a load forecasting model is obtained based on the adjusted model parameters. Since the load forecasting model is trained based on data from the target domain and the source domain, it can be ensured that the load forecasting model has generalization properties in the source domain and the target domain.

[0073] In another embodiment, the timing offset minimization algorithm can learn the common knowledge shared by different time periods, that is, the common knowledge on the target domain and the source domain, by matching the distribution of different time periods according to a given time period. Therefore, the learned load forecasting model is expected to be well generalized on unseen test data. In one example, the loss function of the timing offset minimization algorithm used for prediction, that is, the loss function in adjusting the model parameters according to the gradient descent optimization method based on the loss function, can be expressed by formula (2): (2) in, represents the overall loss function of the prediction task (corresponding loss function); represents the number of divided domains. In this embodiment, the number of domains (including source domains and target domains) is 2; Indicates Datasets for domains; Indicates The number of samples in the dataset in each domain; Indicates In the domain The input features of samples; Indicates In the domain The true labels of samples; Represents the model prediction function (which can correspond to the load prediction model); is a loss function such as .

[0074] In another embodiment, the prediction result of the model (which may correspond to a load prediction model) is as follows: (3) in, For modal exist to Historical load values ​​for the time period; For modal exist to The load forecast value of the time period, where is the input sequence time length, is the time length of the output sequence; For modal exist Frequency characteristics of the moment; is the calculation function.

[0075] In this embodiment, the timing offset minimization algorithm is used to perform model parameter migration and adaptive training between seasons, which can effectively deal with the timing offset caused by the change of load curve with temperature, thereby improving the robustness and accuracy of the prediction model (corresponding to the load prediction model).

[0076] Figure 6 This is the second flow chart of the pre-trained load prediction model provided by the present invention.

[0077] The following will be combined Figure 6 Another process of pre-training a load forecasting model is described.

[0078] In an exemplary embodiment of the present invention, Figure 6 It can be seen that, for the process of pre-training the load forecasting model, after obtaining the load forecasting model based on the adjusted model parameters, the pre-training load forecasting model may further include steps 610 to 640, each of which will be described below: In step 610, a validation training set is called, wherein the validation training set is a validation training set on the target domain.

[0079] In step 620, the load forecasting model is verified based on the verification training set to obtain a verification result.

[0080] In one embodiment, a validation training set may be called, and the load forecasting model may be validated based on the validation training set to obtain a validation result. The validation training set is a validation training set on the target domain. The validation result may be understood as validating the generalization of the load forecasting model, that is, the prediction accuracy on the target domain.

[0081] In step 630, when the verification result meets the preset requirements, the load forecasting model is used as the final load forecasting model.

[0082] In step 640, when the verification result does not meet the preset requirements, the steps of training the load prediction model based on the first training data set and the second training data set to obtain the initial trained load prediction model are repeated, until the verification result of the obtained load prediction model meets the preset requirements, and the obtained load prediction model is used as the final load prediction model.

[0083] In another embodiment, when the verification result meets the preset requirements, that is, when the prediction accuracy of the load prediction model in the target domain meets the preset requirements, it means that the load prediction model has good generalization. Therefore, the load prediction model can be used as the final load prediction model and subsequent load prediction work can be performed.

[0084] In another embodiment, when the verification result does not meet the preset requirements, that is, when the prediction accuracy of the load prediction model in the target domain does not meet the preset requirements, it is necessary to train the load prediction model again, that is, adjust the model parameters. In the application process, the steps of training the load prediction model based on the first training data set and the second training data set to obtain the initial trained load prediction model (corresponding to step 520) to obtaining the load prediction model based on the adjusted model parameters (corresponding to step 540) can be repeated until the verification result of the obtained load prediction model meets the preset requirements, that is, when the prediction accuracy of the load prediction model in the target domain meets the preset requirements, the obtained load prediction model can be used as the final load prediction model, and subsequent load prediction work can be performed.

[0085] It can be understood that, according to the above analysis, the power load forecasting method provided by the present invention decomposes the load time series into multiple intrinsic mode functions (IMFs) through variational mode decomposition (VMD), and extracts temperature-related features in a targeted manner, thereby more accurately capturing the seasonal changes in the load and improving the accuracy of load forecasting. In addition, the use of the time series offset minimization algorithm to migrate model parameters and perform adaptive training between different seasons can effectively deal with the time series offset caused by the change in temperature of the load curve, thereby improving the robustness and accuracy of the forecasting model (load forecasting model).

[0086] The power load prediction device provided by the present invention is described below. The power load prediction device described below and the power load prediction method described above can be referenced to each other.

[0087] Figure 7 It is a structural schematic diagram of the power load prediction device provided by the present invention.

[0088] The following will be combined Figure 7 The structure of the power load prediction device provided by the present invention is described.

[0089] In an exemplary embodiment of the present invention, Figure 7 It can be seen that the power load prediction device may include an acquisition module 710, a decomposition module 720, and a prediction module 730, and the modules will be introduced separately below.

[0090] The acquisition module 710 may be configured to acquire previous load time series data, wherein the previous load time series data is non-stationary time series data; The decomposition module 720 may be configured to perform variational modal decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; The prediction module 730 can be configured to call a pre-trained load prediction model and input the target intrinsic load mode into the load prediction model to obtain a target load prediction value output by the load prediction model that matches the prior load time series data, wherein the target load prediction value is used to characterize the subsequent load time series data, and the load prediction model is used to obtain the target load prediction value based on the target intrinsic load mode.

[0091] In an exemplary embodiment of the present invention, the prediction module 730 may input the target intrinsic load mode into the load prediction model to obtain the target load prediction value output by the load prediction model and matching the prior load time series data in the following manner: Inputting a plurality of the target intrinsic load modes into the load prediction model respectively, and obtaining a plurality of load prediction values ​​corresponding to the target intrinsic load modes output by the load prediction model respectively; A weighted reconstruction process is performed on the multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data.

[0092] In an exemplary embodiment of the present invention, the prediction module 730 may also be configured to: Determining a weighting coefficient of the load prediction value based on a frequency characteristic of the target intrinsic load mode corresponding to the load prediction value; The prediction module 730 may implement weighted reconstruction processing on the multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data in the following manner: Based on the weighting coefficients of the load prediction values, a weighted reconstruction process is performed on the multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data.

[0093] In an exemplary embodiment of the present invention, the decomposition module 720 may implement variational modal decomposition of the prior load time series data to obtain a plurality of target intrinsic load modes in the following manner: Initialize penalty parameters and number of decomposition layers; Based on the penalty parameter and the number of decomposition levels, the prior load time series data is decomposed into a plurality of initial intrinsic load modes by performing iterative optimization through variational mode decomposition, wherein each of the initial intrinsic load modes corresponds to a specific frequency range, and the specific frequency range corresponding to the initial intrinsic load mode reflects the load time series characteristics of the initial intrinsic load mode; A plurality of target intrinsic load modes are extracted from the plurality of initial intrinsic load modes.

[0094] In an exemplary embodiment of the present invention, the decomposition module 720 may be implemented in the following manner to extract a plurality of the target intrinsic load modes from the plurality of the initial intrinsic load modes: Performing correlation analysis on the multiple initial intrinsic load modes to obtain features corresponding to the initial intrinsic load modes, wherein the features at least include temperature features; Based on the features corresponding to each of the initial intrinsic load modes, a plurality of target intrinsic load modes are extracted from the plurality of initial intrinsic load modes.

[0095] In an exemplary embodiment of the present invention, the prediction module 730 may implement pre-training to obtain the load prediction model in the following manner: Constructing a training data set, wherein the training data set includes a first training data set on a source domain and a second training data set on a target domain, wherein the first training data set and the second training data set are historical load time series data belonging to different seasons, and the number of the first training data set is greater than the number of the second training data set; Training a load forecasting model based on the first training data set and the second training data set to obtain an initially trained load forecasting model; Based on a timing offset minimization algorithm, model parameters of the initially trained load forecasting model are migrated to the target domain, and the model parameters are adjusted according to a gradient descent optimization method based on a loss function to obtain adjusted model parameters, wherein the loss function is determined according to a first training data set on the source domain and a second training data set on the target domain; Based on the adjusted model parameters, the load forecasting model is obtained so that the load forecasting model has generalization in the source domain and the target domain.

[0096] In an exemplary embodiment of the present invention, the prediction module 730 may also be configured to: Calling a verification training set, wherein the verification training set is a verification training set on the target domain; Verifying the load forecasting model based on the verification training set to obtain a verification result; When the verification result meets the preset requirements, the load forecasting model is used as the final load forecasting model; When the verification result does not meet the preset requirements, the steps of training the load prediction model based on the first training data set and the second training data set to obtain the load prediction model after initial training are repeated, and the steps of obtaining the load prediction model based on the adjusted model parameters are repeated until the verification result of the load prediction model meets the preset requirements, and the load prediction model is used as the final load prediction model.

[0097] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the power load forecasting method, which includes: obtaining the prior load time series data, wherein the prior load time series data is non-stationary time series data; performing variational mode decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calling a pre-trained load forecasting model, and inputting the target intrinsic load mode into the load forecasting model to obtain the target load forecasting value output by the load forecasting model and matching the prior load time series data, wherein the target load forecasting value is used to characterize the subsequent load, and the load forecasting model is used to obtain the target load forecasting value according to the target intrinsic load mode.

[0098] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0099] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the power load forecasting method provided by the above methods, which method includes: obtaining prior load time series data, wherein the prior load time series data is non-stationary time series data; performing variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calling a pre-trained load forecasting model, and inputting the target intrinsic load mode into the load forecasting model to obtain a target load forecasting value output by the load forecasting model that matches the prior load time series data, wherein the target load forecasting value is used to characterize the subsequent load, and the load forecasting model is used to obtain the target load forecasting value based on the target intrinsic load mode.

[0100] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the power load forecasting method provided by the above-mentioned methods, the method comprising: obtaining prior load time series data, wherein the prior load time series data is non-stationary time series data; performing variational modal decomposition on the prior load time series data to obtain multiple target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; calling a pre-trained load forecasting model, and inputting the target intrinsic load mode into the load forecasting model to obtain a target load forecasting value output by the load forecasting model that matches the prior load time series data, wherein the target load forecasting value is used to characterize the subsequent load, and the load forecasting model is used to obtain the target load forecasting value based on the target intrinsic load mode.

[0101] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0102] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting power load, characterized in that: The method comprises: Acquire previous load time series data, wherein the previous load time series data is non-stationary time series data; Performing variational modal decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; Call a pre-trained load forecasting model, and input the target intrinsic load mode into the load forecasting model to obtain a target load forecasting value output by the load forecasting model that matches the prior load time series data, wherein the target load forecasting value is used to characterize the subsequent load, and the load forecasting model is used to obtain the target load forecasting value based on the target intrinsic load mode.

2. The power load forecasting method according to claim 1, characterized in that: The inputting the target intrinsic load mode into the load forecasting model to obtain the target load forecasting value output by the load forecasting model and matching the prior load time series data specifically includes: Inputting a plurality of the target intrinsic load modes into the load prediction model respectively, and obtaining a plurality of load prediction values ​​corresponding to the target intrinsic load modes output by the load prediction model respectively; A weighted reconstruction process is performed on the multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data.

3. The power load forecasting method according to claim 2, characterized in that: Before performing weighted reconstruction processing on the plurality of load prediction values ​​to obtain a target load prediction value matching the prior load time series data, the method includes: Determining a weighting coefficient of the load prediction value based on a frequency characteristic of the target intrinsic load mode corresponding to the load prediction value; The weighted reconstruction of the plurality of load prediction values ​​to obtain a target load prediction value matching the prior load time series data specifically includes: Based on the weighting coefficients of the load prediction values, a weighted reconstruction process is performed on the multiple load prediction values ​​to obtain a target load prediction value that matches the prior load time series data.

4. The power load forecasting method according to claim 1 or 2, characterized in that: The performing variational modal decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes specifically includes: Initialize penalty parameters and number of decomposition layers; Based on the penalty parameter and the number of decomposition levels, the prior load time series data is decomposed into a plurality of initial intrinsic load modes by performing iterative optimization through variational mode decomposition, wherein each of the initial intrinsic load modes corresponds to a specific frequency range, and the specific frequency range corresponding to the initial intrinsic load mode reflects the load time series characteristics of the initial intrinsic load mode; A plurality of target intrinsic load modes are extracted from the plurality of initial intrinsic load modes.

5. The power load forecasting method according to claim 4, characterized in that: The extracting of a plurality of target intrinsic load modes from the plurality of initial intrinsic load modes specifically includes: Performing correlation analysis on the multiple initial intrinsic load modes to obtain features corresponding to the initial intrinsic load modes, wherein the features at least include temperature features; Based on the features corresponding to each of the initial intrinsic load modes, a plurality of target intrinsic load modes are extracted from the plurality of initial intrinsic load modes.

6. The power load forecasting method according to claim 1, characterized in that: The load forecasting model is pre-trained in the following way: Constructing a training data set, wherein the training data set includes a first training data set on a source domain and a second training data set on a target domain, wherein the first training data set and the second training data set are historical load time series data belonging to different seasons, and the number of the first training data set is greater than the number of the second training data set; Training a load forecasting model based on the first training data set and the second training data set to obtain an initially trained load forecasting model; Based on a timing offset minimization algorithm, model parameters of the initially trained load forecasting model are migrated to the target domain, and the model parameters are adjusted according to a gradient descent optimization method based on a loss function to obtain adjusted model parameters, wherein the loss function is determined according to a first training data set on the source domain and a second training data set on the target domain; Based on the adjusted model parameters, the load forecasting model is obtained so that the load forecasting model has generalization in the source domain and the target domain.

7. The power load forecasting method according to claim 6, characterized in that: After obtaining the load forecasting model based on the adjusted model parameters, the method further includes: Calling a verification training set, wherein the verification training set is a verification training set on the target domain; Verifying the load forecasting model based on the verification training set to obtain a verification result; When the verification result meets the preset requirements, the load forecasting model is used as the final load forecasting model; When the verification result does not meet the preset requirements, the steps of training the load prediction model based on the first training data set and the second training data set to obtain the load prediction model after initial training are repeated, and the steps of obtaining the load prediction model based on the adjusted model parameters are repeated until the verification result of the load prediction model meets the preset requirements, and the load prediction model is used as the final load prediction model.

8. A power load forecasting device, characterized in that: The device comprises: An acquisition module, used for acquiring previous load time series data, wherein the previous load time series data is non-stationary time series data; A decomposition module, used for performing variational modal decomposition on the prior load time series data to obtain a plurality of target intrinsic load modes, wherein the target intrinsic load mode is a mode related to temperature characteristics; A prediction module is used to call a pre-trained load prediction model and input the target intrinsic load mode into the load prediction model to obtain a target load prediction value output by the load prediction model that matches the prior load time series data, wherein the target load prediction value is used to characterize the subsequent load time series data, and the load prediction model is used to obtain the target load prediction value based on the target intrinsic load mode.

9. An electronic 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 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the power load forecasting method according to any one of claims 1 to 7 is implemented.

Images