Load prediction method and device, storage medium and electronic equipment

Abstract

The invention discloses a load prediction method and device, a storage medium and electronic equipment, and relates to the field of smart power grids and prediction analysis. The method comprises the following steps: acquiring a historical load sequence and first data of a first time period and second data of a second time period; decomposing the historical load sequence into N first components and M second components through target operation; dividing the N first components into Z sequences according to the numerical value change condition of each first component; adding the first data and the second data into each sequence of the Z sequences, and taking the sequence added with the first data and the second data as a target sequence to obtain Z target sequences; and determining a target load prediction result of the second time period according to the Z target sequences and the M second components. According to the invention, the technical problem of low accuracy of a load prediction result caused by difficulty in processing nonlinear and non-stationary data in a traditional prediction method is solved.

Description

Technical Field

[0001] The present application relates to the field of smart grid and prediction analysis, and specifically, to a load prediction method, device, storage medium and electronic device. Background Art

[0002] In the management and optimization of power systems, short-term power load forecasting technology plays a vital role, which directly affects the operating efficiency of the power grid, the rational allocation of power resources and the economic efficiency of the power market. Traditional forecasting methods, such as time series analysis, regression models and forecasting methods based on expert systems, can provide a certain degree of forecasting accuracy on data sets with high stability, but it is often difficult to achieve satisfactory forecasting results when faced with complex, variable, nonlinear and non-stationary power load data. This type of load data usually contains periodic fluctuations, sudden events and supply and demand patterns that change over time. These characteristics severely limit the forecasting ability of traditional models, resulting in a significant decrease in the accuracy of load forecasting results.

[0003] In the new power system, the access of distributed energy, the diversification of user needs and the dynamic nature of supply and demand balance make the nonlinearity and non-stationarity of power load data more significant. When dealing with such data, traditional forecasting methods face technical difficulties such as weak model generalization ability and inability to capture complex time and space characteristics. Especially when dealing with large-scale fluctuations, long-term trend changes and the impact of incidental events on load, the forecast accuracy is often difficult to meet the needs of efficient operation of the power system.

[0004] Traditionally, short-term power load forecasting methods rely on historical load data and various statistical analysis techniques, such as time series analysis and regression analysis. Although these methods can provide relatively accurate forecasting results under certain conditions, they are often limited in their ability to handle complex nonlinear relationships, and are particularly incapable of dealing with non-stationary and nonlinear power data. As the complexity of the power system increases, power load data presents higher-dimensional volatility, periodicity, and non-stationary characteristics, which pose challenges to traditional forecasting methods.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The present application provides a load forecasting method, device, storage medium and electronic device to at least solve the technical problem that traditional forecasting methods have difficulty in processing nonlinear and non-stationary data, resulting in low accuracy of load forecasting results.

[0007] According to one aspect of the present application, a load forecasting method is provided, comprising: obtaining a historical load sequence of a first time period, first data, and second data of a second time period, wherein the first data is used to characterize factors affecting the historical load sequence of the first time period, and the second data is used to characterize factors affecting the load sequence of the second time period, and the second time period is after the first time period; decomposing the historical load sequence into N first components and M second components through a target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1, and the target operation is used to decompose nonlinear and non-stationary data into modal components, the first component is used to characterize the dynamic characteristics of the load data within a target frequency range, and the second component is used to characterize the data in the load data whose frequency is less than a preset frequency; dividing the N first components into Z sequences according to the numerical change of each first component, wherein Z is an integer greater than 1, and each sequence includes a first component that meets a preset condition; adding the first data and the second data to each sequence of the Z sequences, and taking the sequence with the first data and the second data added as the target sequence to obtain Z target sequences; determining the target load forecasting result for the second time period according to the Z target sequences and the M second components.

[0008] Optionally, the historical load sequence is decomposed into N first components and M second components through a target operation, including: obtaining a white noise set; adding the white noise in the white noise set to the historical load sequence S times respectively to obtain S historical loads to be decomposed, where S is an integer greater than 1; decomposing the S historical loads to be decomposed through a target operation to obtain N first components and M second components.

[0009] Optionally, according to the numerical change of each first component, N first components are divided into Z sequences, including: performing a stationarity test on the N first components to obtain a test result, wherein the stationarity test is used to evaluate whether the statistical characteristics of the load data change over time; according to the test result, the N first components are divided into Z sequences, wherein the Z sequences include at least a fluctuation sequence, a periodic sequence, and a stationary sequence, wherein the fluctuation sequence includes a first component whose data change meets a first preset condition, the periodic sequence includes a first component whose data change meets a second preset condition, and the stationary sequence includes a first component whose data change meets a third preset condition, wherein the first preset condition indicates that the data change fluctuates beyond a preset threshold, the second preset condition indicates that the data changes periodically, and the third preset condition indicates that the statistical characteristics of the data remain unchanged.

[0010] Optionally, after dividing the N first components into Z sequences according to the test results, the load forecasting method also includes: performing data cleaning and preprocessing on the Z sequences and the M second components, wherein the data cleaning is used to process outliers in the Z sequences and the M second components, and the preprocessing is used to normalize and process missing values ​​on the Z sequences and the M second components.

[0011] Optionally, the target load forecast result for the second time period is determined based on Z target sequences and M second components, including: inputting the Z target sequences into Z first target forecasting models to obtain Z first initial load forecasting results, wherein the first target forecasting model is used to extract features from the target sequence, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling; adding the first data and the second data to each second component of the M second components, and using the second component to which the first data and the second data are added as the target component to obtain M target components; inputting the M target components into M second target forecasting models to obtain M second initial load forecasting results, wherein the second target forecasting model is used to extract features from the target component, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling; determining the target load forecast result for the second time period based on the Z first initial load forecasting results and the M second initial load forecasting results.

[0012] Optionally, determining a target load forecast result for a second time period based on Z first initial load forecast results and M second initial load forecast results includes: performing inverse processing of the normalization processing on the Z first initial load forecast results and the M second initial load forecast results to obtain L third initial load forecast results, wherein the sum of Z and M is equal to L; determining the weight of each third initial load forecast result in the L third initial load forecast results; and determining the target load forecast result for the second time period based on the L third initial load forecast results and the weights corresponding to the L third initial load forecast results.

[0013] Optionally, the first target prediction model and the second target prediction model are obtained by the following steps: obtaining a training data set, wherein the training data set includes N first components, M second components and first data of a first historical time period, and N first components, M second components and second data of a second historical time period, wherein the first historical time period is before the first time period and the second historical time period is after the first historical time period; using the N first components, first data of the first historical time period and the second data of the second historical time period as first training samples; using the N first components of the second historical time period as first training labels; iteratively training the first neural network according to the first training samples and the first training labels to obtain Z first target prediction models; using the M second components, first data of the first historical time period and the second data of the second historical time period as second training samples; using the M second components of the second historical time period as second training labels; iteratively training the second neural network according to the second training samples and the second training labels to obtain M second target prediction models.

[0014] According to another aspect of the present application, a load forecasting device is also provided, including: an acquisition unit, which acquires a historical load sequence, first data, and second data of a second time period in a first time period, wherein the first data is used to characterize factors that affect the historical load sequence in the first time period, and the second data is used to characterize factors that affect the load sequence in the second time period, and the second time period is after the first time period; a decomposition unit, which decomposes the historical load sequence into N first components and M second components through a target operation, wherein N is an integer greater than 1, M is an integer greater than or equal to 1, and the target operation is used to decompose nonlinear and non-stationary data into modal components , the first component is used to characterize the dynamic characteristics of the load data within the target frequency range, and the second component is used to characterize the data in the load data whose frequency is less than the preset frequency; a division unit, according to the numerical change of each first component, divides the N first components into Z sequences, where Z is an integer greater than 1, and each sequence includes a first component that meets the preset conditions; a processing unit, adds the first data and the second data to each sequence of the Z sequences, and uses the sequence with the first data and the second data as the target sequence to obtain Z target sequences; a determination unit, determines the target load prediction result of the second time period according to the Z target sequences and the M second components.

[0015] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, wherein when the computer program runs, the device where the computer-readable storage medium is located executes the above-mentioned load forecasting method.

[0016] According to another aspect of the present application, an electronic device is also provided, wherein the electronic device includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors execute the above-mentioned load forecasting method.

[0017] In the present application, firstly, a historical load sequence, first data of a first time period and second data of a second time period are obtained, wherein the first data is used to characterize factors affecting the historical load sequence of the first time period, and the second data is used to characterize factors affecting the load sequence of the second time period. The second time period is after the first time period. Then, the historical load sequence is decomposed into N first components and M second components through a target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1. The target operation is used to decompose nonlinear and non-stationary data into modal components. The first component is used to characterize the dynamic characteristics of the load data within a target frequency range, and the second component is used to characterize the data whose frequency in the load data is less than a preset frequency. Then, according to the numerical change of each first component, the N first components are divided into Z sequences, wherein Z is an integer greater than 1, and each sequence includes a first component that meets a preset condition, and the first data and the second data are added to each sequence of the Z sequences, and the sequence with the first data and the second data added is used as a target sequence to obtain Z target sequences. Finally, the target load prediction result of the second time period is determined according to the Z target sequences and the M second components. The purpose of predicting nonlinear and non-stationary loads is achieved, thereby realizing the technical effect of improving the accuracy of load prediction, and further solving the technical problem that traditional prediction methods are difficult to handle nonlinear and non-stationary data, resulting in low accuracy of load prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 is a flow chart of an optional load forecasting method according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of an optional CEEMDAN method processing according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of an optional TCN-BiGRU model processing according to an embodiment of the present application;

[0022] Figure 4is a schematic diagram of an optional load forecasting method according to an embodiment of the present application;

[0023] Figure 5 It is a schematic diagram of an optional load forecasting device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] It should also be noted that the information and data collected by this application are authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation portals for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0027] According to an embodiment of the present application, an embodiment of a load forecasting method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] It should be noted that an intelligent prediction system can be used as the execution subject of the load prediction method of the embodiment of the present application. It is understandable that the load prediction method provided in the embodiment of the present application can also be executed by other systems or devices, and the embodiment of the present application does not specifically limit this.

[0029] Figure 1 is a flow chart of an optional load forecasting method according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0030] Step S101, obtaining a historical load sequence and first data in a first time period and second data in a second time period.

[0031] In step S101, the first data is used to characterize factors affecting the historical load sequence in a first time period, and the second data is used to characterize factors affecting the load sequence in a second time period, where the second time period is after the first time period.

[0032] Optionally, the first data of the first time period includes but is not limited to weather conditions, holidays, working day / rest day identification, temperature, humidity, etc., which are external conditions or events that can affect the power load.

[0033] Optionally, the second data is used to characterize factors that affect the load series within the forecast period, similar to the first data, but more focused on specific conditions in the forecast period, such as weather forecast data for the next week.

[0034] Step S102: decompose the historical load sequence into N first components and M second components through target operation.

[0035] In step S102, N is an integer greater than 1, M is an integer greater than or equal to 1, and the target operation is used to decompose the nonlinear and non-stationary data into modal components. The first component is used to characterize the dynamic characteristics of the load data within the target frequency range, and the second component is used to characterize the data in the load data whose frequency is less than the preset frequency.

[0036] Optionally, the target operation is an operation performed by a CEEMDAN method (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise).

[0037] It should be noted that the CEEMDAN method is an advanced signal processing technology, mainly used for data analysis of nonlinear and non-stationary signals. CEEMDAN is an improved EMD (Empirical Mode Decomposition) method, which achieves a more stable and accurate decomposition of the signal by adding adaptive noise.

[0038] Optionally, the first component is an IMF component (Intrinsic Mode Function components), and the second component is a RES component (Residual component).

[0039] It should be noted that the IMF component and the RES component are signal components obtained by decomposing the original power load sequence through the CEEMDAN method. Each IMF represents an inherent oscillation mode in the original signal, which manifests itself as a set of local, natural frequency fluctuations in time. In the CEEMDAN decomposition, the original signal is decomposed step by step, and each level of decomposition obtains an IMF and a new signal to be decomposed. The RES component is the remaining signal part of the CEEMDAN decomposition process. It is the signal residue that still exists after all IMF decompositions and cannot be decomposed into IMFs. The RES component usually contains the long-term trend or low-frequency component of the signal, as well as possible noise or non-oscillatory parts. In power load forecasting, the RES component can be understood as those very slowly changing and stable trends in the load sequence, such as seasonal changes or long-term load growth trends.

[0040] Optionally, the intelligent prediction system uses the CEEMDAN method to preprocess the historical power load sequence and decompose the historical power load sequence into six groups of IMF components and one residual component RES. These IMF components represent signals of different frequency components in the original sequence, while RES contains residual information that cannot be decomposed into IMF.

[0041] Step S103: divide the N first components into Z sequences according to the change of the value of each first component.

[0042] In step S103, Z is an integer greater than 1, and each sequence includes a first component that meets a preset condition.

[0043] Optionally, the intelligent forecasting system divides the six groups of IMF components into three sequences according to the changes of each IMF component. The sequence referred to here is the result of classifying and merging each IMF component according to its characteristics (such as volatility, periodicity or stability). For example, the IMF components are classified into volatility, periodicity and stability, and each class forms a sequence.

[0044] Step S104: add the first data and the second data to each of the Z sequences, and use the sequence to which the first data and the second data are added as a target sequence, to obtain Z target sequences.

[0045] Optionally, the intelligent prediction system adds the first data and the second data to each of the three sequences, and uses the sequence to which the first data and the second data are added as the target sequence to obtain three target sequences, thereby forming a sequence that can be used for training and prediction of the prediction model. The target sequence contains complete information on the dynamic characteristics of the historical load and external factors affecting the load.

[0046] Step S105: determining a target load forecast result for a second time period according to the Z target sequences and the M second components.

[0047] Optionally, the intelligent prediction system performs prediction based on three target sequences and one RES component, and finally obtains a target load prediction result for the second time period.

[0048] Based on the contents of the above steps S101 to S105, it can be known that in the present application, firstly, the historical load sequence of the first time period, the first data and the second data of the second time period are obtained, wherein the first data is used to characterize the factors affecting the historical load sequence of the first time period, and the second data is used to characterize the factors affecting the load sequence of the second time period. The second time period is after the first time period. Then, the historical load sequence is decomposed into N first components and M second components through the target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1. The target operation is used to decompose the nonlinear and non-stationary data into modal components. The first component is used to characterize the dynamic characteristics of the load data within the target frequency range, and the second component is used to characterize the data whose frequency in the load data is less than the preset frequency. Then, according to the numerical change of each first component, the N first components are divided into Z sequences, wherein Z is an integer greater than 1, and each sequence includes a first component that meets the preset conditions, and the first data and the second data are added to each sequence of the Z sequences, and the sequence with the first data and the second data added is used as the target sequence to obtain Z target sequences. Finally, the target load prediction result of the second time period is determined according to the Z target sequences and the M second components. The purpose of predicting nonlinear and non-stationary loads is achieved, thereby realizing the technical effect of improving the accuracy of load prediction, and further solving the technical problem that traditional prediction methods are difficult to handle nonlinear and non-stationary data, resulting in low accuracy of load prediction results.

[0049] In an optional embodiment, the intelligent prediction system first obtains a white noise set, and then adds the white noise in the white noise set to the historical load sequence S times to obtain S historical loads to be decomposed, where S is an integer greater than 1. Finally, the S historical loads to be decomposed are decomposed through a target operation to obtain N first components and M second components.

[0050] Optionally, the intelligent prediction system first generates a white noise set. White noise is a random signal with the same energy density at all frequencies. It is used here to assist the decomposition of the signal and improve the stability and accuracy of the decomposition. The adaptive white noise is added to the historical load sequence S times, and white noise of different scales is added each time. Through the above steps, S historical load sequences to be decomposed are obtained. Each sequence contains the original signal and white noise of different scales. Then, the CEEMDAN method is used to decompose the S historical load sequences to be decomposed to obtain N first components (i.e., IMF components) and M second components (RES components). Among them, the RES component can be obtained by formula (1):

[0051]

[0052] Among them, G represents the photovoltaic power generation series, C represents the IMF subset, ∑IMF r Represents the sum of all IMFs in the IMF subset.

[0053] Optionally, CEEMDAN also uses an adaptive method to determine the frequency range of each IMF and the intensity of random noise, so that it can be adjusted according to the characteristics of the signal and better adapt to various different signals.

[0054] From the above content, it can be seen that the intelligent prediction system can ensure that each IMF component is as stable and independent as possible by adding white noise multiple times and performing CEEMDAN decomposition, which can improve the stability and accuracy of IMF, thereby improving the training effect and prediction accuracy of subsequent prediction models. The retention of the RES component ensures the accuracy of the long-term trend of the prediction results, making the short-term power load forecast more comprehensive and accurate. In this way, the intelligent prediction system can more finely analyze the dynamic characteristics of the historical power load series, laying a solid foundation for subsequent data prediction and analysis.

[0055] In an optional embodiment, the intelligent prediction system performs a stationarity test on the N first components to obtain a test result, wherein the stationarity test is used to evaluate whether the statistical characteristics of the load data change over time, and then divides the N first components into Z sequences based on the test result, wherein the Z sequences include at least a fluctuation sequence, a periodic sequence, and a stationary sequence, wherein the fluctuation sequence includes a first component whose data change satisfies a first preset condition, the periodic sequence includes a first component whose data change satisfies a second preset condition, and the stationary sequence includes a first component whose data change satisfies a third preset condition, wherein the first preset condition indicates that the data change fluctuates beyond a preset threshold, the second preset condition indicates that the data changes periodically, and the third preset condition indicates that the statistical characteristics of the data remain unchanged.

[0056] Optionally, the intelligent prediction system first obtains N IMF components decomposed by the CEEMDAN method, and then performs a stationarity test on these components to evaluate whether their statistical characteristics change over time, thereby determining the dynamic properties of each component, and based on the test results, further classifies and reorganizes the N first components to form Z sequences with specific properties, which at least include volatility sequences, periodic sequences and stationary sequences.

[0057] Optionally, the intelligent prediction system combines IMF components with similar characteristics based on the stationarity test results so that subsequent models can learn and predict better.

[0058] Specifically, the first preset condition is used to identify IMF components whose fluctuations exceed the preset threshold. These components are classified as fluctuation sequences, which usually reflect the short-term and drastic characteristics of the load; the second preset condition selects IMF components that show periodic changes to form periodic sequences, which capture the regular fluctuations in the load, such as daily and weekly load cycles; the third preset condition is applicable to IMF components whose statistical characteristics remain stable, forming stationary sequences, which reflect the long-term trend and stability of the load. Through this classification, it can be ensured that the model can learn and predict load changes of different natures in a targeted manner, thereby improving the accuracy of the prediction.

[0059] Optionally, the CEEMDAN method processes the input time series data and returns its IMF and RES. When performing data analysis, the modal components are first numbered, and the stationarity test is performed on each group of sequences and the residual component RES. For non-stationary sequences, a direct merging and reorganization strategy is adopted, which not only helps to reduce the computational burden in the prediction process, but also ensures that sequences with similar stationary characteristics are reasonably classified, thereby preventing the prediction accuracy from being reduced due to the mixing of sequences with different stationarities. Then, the decomposed components are checked and analyzed, showing strong volatility, but with a small average amplitude, which is classified as a volatility class, showing obvious periodic changes, and sequences with certain regularity are classified as periodic classes, and sequences with relatively stable changes are classified as stationary classes. As for the residual component RES, its volatility is significant and needs to be considered separately. Through such classification processing, the characteristics of each sequence can be understood and analyzed more accurately, laying the foundation for subsequent data analysis and prediction.

[0060] Optionally, Figure 2 is a schematic diagram of an optional CEEMDAN method processing according to an embodiment of the present application, such as Figure 2 As shown in the figure, when decomposing the historical power load sequence, firstly, a CEEMDAN object is created for the historical power load sequence, then the characteristic data of the historical power load sequence is extracted, and the characteristic data is decomposed using the CEEMDAN algorithm to obtain the corresponding number of IMF components and RES components, and the IMF component images and RES component images are drawn. Then, through the stationarity test, the IMF components with similar characteristics are merged into a group of sequences, such as fluctuation sequences, periodic sequences and stationary sequences. Finally, these sequences are input into the prediction model as input features for prediction to obtain the prediction results.

[0061] From the above content, it can be seen that by classifying the IMF components into Z sequences, each sequence represents the specific dynamic characteristics of the load, and the intelligent prediction system can handle the complexity and nonlinearity of power load data more accurately. At the same time, by classifying and reorganizing the IMF components, it provides a basis for subsequent model training, which not only simplifies the input of the prediction model so that it can focus on predicting specific types of load changes, but also improves the generalization ability and prediction accuracy of the model, which has significant advantages for power load prediction, especially for processing nonlinear and non-stationary data.

[0062] In an optional embodiment, the intelligent prediction system performs data cleaning and preprocessing on the Z sequences and the M second components, wherein the data cleaning is used to process outliers in the Z sequences and the M second components, and the preprocessing is used to normalize and process missing values ​​on the Z sequences and the M second components.

[0063] Optionally, during the model training process, in order to accelerate training and improve model performance, the data is usually processed to a certain extent. The intelligent prediction system performs data cleaning and preprocessing on the Z sequences (including fluctuation sequences, periodic sequences, and stationary sequences) and M second components (RES components) that have been initially classified and decomposed. The data cleaning step aims to identify and process outliers in the Z sequences and M second components. These outliers may originate from measurement errors, omissions in data records, or other abnormal factors, posing a threat to the accuracy and stability of the prediction model. Preprocessing includes normalization and missing value processing, the purpose of which is to unify the data dimensions, eliminate the impact of magnitude differences on model training, and fill in the missing data to avoid information loss affecting the prediction results.

[0064] From the above content, it can be seen that the intelligent prediction system effectively improves the input quality of the prediction model through data cleaning and preprocessing, and reduces the impact of outliers and missing data on the prediction results. Normalization ensures that the model can treat all data points equally, regardless of their original magnitude. The processing of outliers and missing values ​​avoids prediction errors caused by noise and incompleteness in the data, providing a more reliable data foundation for subsequent prediction model training. Through these preprocessing steps, the intelligent prediction system can more accurately capture and predict the dynamic changes of power load.

[0065] In an optional embodiment, the intelligent prediction system inputs Z target sequences into Z first target prediction models to obtain Z first initial load forecast results, wherein the first target prediction model is used to extract features from the target sequence, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling, then add the first data and the second data to each second component of the M second components, and use the second component to which the first data and the second data are added as the target component to obtain M target components, and then input the M target components into M second target prediction models to obtain M second initial load forecast results, wherein the second target prediction model is used to extract features from the target component, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling, and finally determine the target load forecast result for the second time period based on the Z first initial load forecast results and the M second initial load forecast results.

[0066] Optionally, the prediction model is a TCN-BiGRU model (Temporal Convolutional Network and Bidirectional Gated Recurrent Unit Combined Model, a one-dimensional causal convolutional neural network and a bidirectional gated recurrent unit combined model), which combines a one-dimensional causal convolutional neural network (Temporal Convolutional Network, referred to as TCN) and a bidirectional gated recurrent unit (Bidirectional Gated Recurrent Unit, referred to as BiGRU), using TCN to extract the temporal features of past data, extended convolution to more effectively extract temporal features, and residual connections to ensure the temporal properties of the model and accelerate convergence. BiGRU is a variant of a recursive neural network with strong nonlinear adaptability, making the feature extraction capability of the model more efficient. The combination of these two technologies can provide similar prediction effects and faster convergence to recursive neural networks.

[0067] Optionally, Figure 3 is a schematic diagram of an optional TCN-BiGRU model processing according to an embodiment of the present application, such as Figure 3 As shown in the figure, the first step is to input continuous short-term time series data, such as power load data samples, and non-time series data, such as daily date types; the second step is to input data into the time convolutional network layer for feature extraction. The time convolutional network layer is composed of stacked time blocks. Each convolution unit uses the ReLU function (Rectified Linear Unit function) as the activation function, normalizes the weight of the convolution kernel, and sets the size of the convolution kernel to 3, that is, slides 3 time steps on the input data each time to extract features. The output dimension of each TCN layer is [32, 64, 128]. Here, the number of output dimension channels means the number and complexity of features learned by each layer, see formula (2):

[0068]

[0069] Among them, input represents the input dimension, padding represents the padding size, which is used to keep the output feature map size consistent with the input feature map size, k represents the convolution kernel size, and S represents the stride during the convolution operation.

[0070] Generally, as the network depth increases, the number of output channels will gradually increase in order to extract more abstract and complex features. Here, [32, 64, 128] means that the TCN network contains three convolutional layers, which convert the input data into feature maps with 32, 64, and 128 channels respectively. The size of is 0.1, which means that in each training iteration, 10% of the neurons will be randomly discarded with a probability of 0.1 and will not participate in the calculation of the current layer. The remaining elements remain unchanged. The dimensions of the input and output of the residual unit are different. In order to unify the dimensions of the input and output data, it is necessary to add a one-dimensional expanded causal convolution layer in the residual mapping. In the third step, the features processed by TCN are further extracted through the bidirectional gated recurrent unit layer. The bidirectional GRU module contains forward and backward GRU modules. In the stacked GRU structure, the output of each GRU layer will be used as the input of the next layer. By stacking multiple GRU layers, the representation ability and learning ability of the model can be increased, and the modeling effect of the model on sequence data can be improved. The number of stacking times of GRU in this embodiment is 2, that is, the input of each time step will be processed by two GRU layers. The input layer (Input Layer) is responsible for receiving the feature vector or feature sequence of the input sequence. The input of each time step is represented by , which converts the features processed by TCN into a format suitable for GRU model processing. The hidden layer (Hidden The output layer is composed of multiple forward and backward GRU units, which are responsible for processing the forward and backward information of the input sequence. The dimension of the hidden layer determines the number and complexity of the features that the model can learn, and also represents the performance of the model. See formula (3). In this embodiment, the dimension of the hidden layer in each GRU layer is set to 64, that is, the dimension of the hidden state vector of each time step is 64. The output layer is a fully connected layer, which is responsible for receiving information from the forward and backward hidden layers and combining this information into the final output.

[0071] h=f(W·x+b) (3)

[0072] Where W x represents the matrix product of the weight matrix and the input vector, b is the bias vector, f is the activation function, and h is the output vector of the hidden layer.

[0073] The fourth step is to map the output of GRU through the fully connected layer to obtain the final prediction result. Finally, the prediction result is taken as the final output result according to the dimension of time step and the output of the last time step is returned.

[0074] Optionally, the intelligent prediction system inputs the merged sequences into the TCN-BiGRU model for prediction, defines respective models for different decomposition modes, and adds the prediction results of each mode as the final prediction result.

[0075] Optionally, the intelligent prediction system predicts the Z target sequences (i.e., fluctuation sequence, periodic sequence, and stable sequence) and M target components (the sequence after the RES component is combined with the historical and predicted influencing factors) that have been preprocessed and classified through a multi-model integrated prediction strategy, in order to more accurately capture the dynamic characteristics of the power load. First, the Z target sequences of each specific attribute are input into the corresponding Z first target prediction models. The model extracts features based on the inherent attributes of the sequence and establishes a time series model for load prediction, thereby obtaining Z first initial load prediction results, which correspond to the predicted values ​​of fluctuation, periodic, and stable load changes, respectively. Then, for the M second components (RES components), the intelligent prediction system merges the first data (historical influencing factors) and the second data (forecast period influencing factors) into each component to create M target components. These target components contain historical load trends and factors that may affect load changes during the forecast period. Subsequently, the M target components are input into the M second target prediction models, and the fused characteristics are deeply learned through the model, key time series features are extracted, and load forecasting is performed, and finally M second initial load forecasting results are obtained. These results capture the long-term trend forecast of the load. Finally, the intelligent forecasting system combines the Z first initial load forecasting results and the M second initial load forecasting results, and determines the target load forecasting results for the second time period through certain integration strategies or algorithms, such as weighted average, statistical fusion, etc. This multi-model parallel prediction and result integration strategy makes full use of the characteristics of different sequences and components, and improves the comprehensiveness and accuracy of the prediction.

[0076] From the above content, it can be seen that through multi-model prediction and result integration strategies, the intelligent prediction system can more effectively process complex and changeable power load data, ensuring that each type of load change can be fully considered and accurately predicted. Finally, a comprehensive prediction that combines short-term fluctuations and long-term trends is obtained through result integration. The time features of past data are extracted through parallel splicing TCN, the extended convolution is used to more effectively extract the time features, and the residual connection is used to ensure the time attributes of the model and accelerate the convergence speed. At the same time, BiGRU is used to solve the problem of nonlinear time series data processing, making the feature extraction capability of the model more efficient.

[0077] In an optional embodiment, the intelligent prediction system performs inverse processing of the normalization process on the Z first initial load forecast results and the M second initial load forecast results to obtain L third initial load forecast results, where the sum of Z and M is equal to L, determines the weight of each third initial load forecast result in the L third initial load forecast results, and then determines the target load forecast result for the second time period based on the L third initial load forecast results and the weights corresponding to the L third initial load forecast results.

[0078] Optionally, during the model training phase, the intelligent prediction system normalizes the original load sequence, that is, converts the data into a unified range (such as between 0 and 1) to improve the efficiency of model training and prediction performance. This is because the model is easily affected by large numerical features when processing data of different scales, resulting in an unstable training process or inaccurate prediction results. Therefore, after the training phase is completed, when the model outputs prediction results, these results are based on normalized data. Therefore, in order to make the prediction results meaningful and match the dimensions of the original data, it is necessary to perform denormalization. After denormalization, L third initial load prediction results are obtained. Next, the system determines the weight of each third initial load prediction result. By combining the L third initial load prediction results with their corresponding weights, the intelligent prediction system can integrate the predictions of all models to obtain a more accurate and comprehensive target load forecast result for the second time period.

[0079] Optionally, Figure 4 is a schematic diagram of an optional load forecasting method according to an embodiment of the present application, such as Figure 4 As shown in the figure, historical power load data and related factors affecting the load data, such as date type, are obtained, and then the acquired data are processed for outliers and missing values. Then, the historical load data is decomposed using the CEEMDAN method to obtain IMF components and RES components. After that, the components are tested for stationarity and components with similar characteristics are merged to obtain three sequences, namely, periodic sequence, stationary sequence and fluctuating sequence. The three sequences are normalized to improve the efficiency of model training and prediction performance. Each sequence is input into the corresponding TCN-BiGRU model for load prediction to obtain three initial prediction results. Since the data is normalized before being input into the model, the initial prediction results need to be denormalized after obtaining the initial prediction results, and then the weights corresponding to each initial prediction result are obtained. Finally, the final prediction results are obtained according to the initial prediction results of the three modes and the corresponding weights.

[0080] From the above content, it can be seen that the intelligent prediction system effectively integrates the prediction results of different models through inverse normalization and weighted average strategy, overcomes the limitations of single model prediction, and provides a more comprehensive and accurate method for short-term prediction of power load. This method fully considers the dynamic characteristics of various power loads and uses the dynamic adjustment of model weights to optimize the prediction results based on real-time data and model performance to improve the accuracy and practicality of the prediction.

[0081] In an optional embodiment, the intelligent prediction system first obtains a training data set, wherein the training data set includes N first components, M second components and first data of a first historical time period, and N first components, M second components and second data of a second historical time period, wherein the first historical time period is before the first time period, and the second historical time period is after the first historical time period, then the N first components, first data of the first historical time period and the second data of the second historical time period are used as first training samples, and the N first components of the second historical time period are used as first training labels, and then the first neural network is iteratively trained according to the first training samples and the first training labels to obtain Z first target prediction models, and then the M second components, first data of the first historical time period and the second data of the second historical time period are used as second training samples, and the M second components of the second historical time period are used as second training labels, and finally the second neural network is iteratively trained according to the second training samples and the second training labels to obtain M second target prediction models.

[0082] Optionally, the intelligent prediction system collects data from two historical time periods: the first historical time period is located before the second historical time period, and includes N first components (corresponding to the IMF components after CEEMDAN decomposition) and M second components (RES components), as well as first data related to these components (historical influencing factors). Similarly, the data set of the second historical time period also includes N first components, M second components and second data (forecast period influencing factors).

[0083] Optionally, during the model training phase, the intelligent prediction system combines the N first components and the first data of the first historical time period and the second data of the second historical time period as the first training sample, and the N first components of the second historical time period as the first training labels to guide the learning of the first neural network. Through repeated iterations, the first training samples are input into the neural network, and the network weights are adjusted to minimize the error between the prediction results and the first training labels, and finally Z first target prediction models are obtained, which are specifically optimized for the prediction of the IMF component.

[0084] Optionally, the intelligent prediction system adopts a similar method to combine the M second components of the first historical time period, the first data, and the second data of the second historical time period as the second training sample; and the M second components of the second historical time period are used as the second training labels. Through the same iterative training process, the system optimizes the second neural network to predict the RES component, and finally obtains M second target prediction models.

[0085] From the above content, it can be seen that the intelligent prediction system can train the decomposed power load sequence in a targeted manner and build a high-precision model specifically for the prediction of IMF and RES components, thereby effectively improving the accuracy of power load prediction. This method makes full use of the characteristics of historical data, ensures the effectiveness and purpose of model training through the precise matching of the first and second training samples with the labels, and can learn the law of load changes from historical data, especially for fluctuation sequences, periodic sequences and stationary sequences, as well as RES components, and can capture more complex characteristics and time dependencies, thereby improving the accuracy of prediction.

[0086] According to another aspect of the present application, a load prediction device is also provided. Figure 5 is a schematic diagram of an optional load forecasting device according to an embodiment of the present application, such as Figure 5 As shown, the load prediction device includes: an acquisition unit 501, a decomposition unit 502, a division unit 503, a processing unit 504, and a determination unit 505.

[0087] Optionally, an acquisition unit 501 is used to acquire a historical load sequence, first data, and second data of a second time period in a first time period, wherein the first data is used to characterize factors affecting the historical load sequence in the first time period, and the second data is used to characterize factors affecting the load sequence in the second time period, and the second time period is after the first time period; a decomposition unit 502 is used to decompose the historical load sequence into N first components and M second components through a target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1, and the target operation is used to decompose nonlinear and non-stationary data into modal components, and the first component is used to characterize the load data in Dynamic characteristics within the target frequency range, the second component is used to characterize the data in the load data whose frequency is less than the preset frequency; the division unit 503 is used to divide the N first components into Z sequences according to the numerical change of each first component, where Z is an integer greater than 1, and each sequence includes the first component that meets the preset conditions; the processing unit 504 is used to add the first data and the second data to each sequence of the Z sequences, and use the sequence with the first data and the second data as the target sequence to obtain Z target sequences; the determination unit 505 is used to determine the target load prediction result for the second time period according to the Z target sequences and the M second components.

[0088] Optionally, the decomposition unit 502 includes: a first acquisition subunit, a first processing subunit, and a first decomposition subunit. The first acquisition subunit is used to acquire a white noise set; the first processing subunit is used to add the white noise in the white noise set to the historical load sequence S times to obtain S historical loads to be decomposed, where S is an integer greater than 1; the first decomposition subunit is used to decompose the S historical loads to be decomposed through a target operation to obtain N first components and M second components.

[0089] Optionally, the division unit 503 includes: a first inspection subunit and a first division subunit. The first inspection subunit is used to perform a stationarity test on the N first components to obtain a test result, wherein the stationarity test is used to evaluate whether the statistical characteristics of the load data change over time; the first division subunit is used to divide the N first components into Z sequences according to the test result, wherein the Z sequences at least include a fluctuation sequence, a periodic sequence, and a stationary sequence, wherein the fluctuation sequence includes a first component whose data change condition satisfies a first preset condition, the periodic sequence includes a first component whose data change condition satisfies a second preset condition, and the stationary sequence includes a first component whose data change condition satisfies a third preset condition, wherein the first preset condition indicates that the data change fluctuates beyond a preset threshold, the second preset condition indicates that the data changes periodically, and the third preset condition indicates that the statistical characteristics of the data remain unchanged.

[0090] Optionally, the division unit 503 includes: a second processing subunit, used to perform data cleaning and preprocessing on the Z sequences and the M second components, wherein the data cleaning is used to process outliers in the Z sequences and the M second components, and the preprocessing is used to normalize and process missing values ​​on the Z sequences and the M second components.

[0091] Optionally, the determination unit 505 includes: a third processing subunit, a fourth processing subunit, a fifth processing subunit and a first determination subunit. The third processing subunit is used to input Z target sequences into Z first target prediction models to obtain Z first initial load prediction results, wherein the first target prediction model is used to extract features from the target sequence, perform time series modeling based on the extracted features, and perform load prediction based on the time series modeling; the fourth processing subunit is used to add the first data and the second data to each second component of the M second components, and use the second component to which the first data and the second data are added as the target component to obtain M target components; the fifth processing subunit is used to input the M target components into the M second target prediction models to obtain M second initial load prediction results, wherein the second target prediction model is used to extract features from the target component, perform time series modeling based on the extracted features, and perform load prediction based on the time series modeling; the first determination subunit is used to determine the target load prediction result for the second time period based on the Z first initial load prediction results and the M second initial load prediction results.

[0092] Optionally, the first determination subunit includes: a first processing module, a first determination module and a second determination module. The first processing module is used to perform inverse processing of normalization processing on Z first initial load forecast results and M second initial load forecast results to obtain L third initial load forecast results, wherein the sum of Z and M is equal to L; the first determination module is used to determine the weight of each third initial load forecast result in the L third initial load forecast results; the second determination module is used to determine the target load forecast result of the second time period according to the L third initial load forecast results and the weights corresponding to the L third initial load forecast results.

[0093] Optionally, the load prediction device further includes: a first acquisition unit, a first processing unit, a second processing unit, a first training unit, a third processing unit, a fourth processing unit, and a second training unit. Among them, the first acquisition unit is used to acquire a training data set, wherein the training data set includes N first components, M second components and first data of a first historical time period, and N first components, M second components and second data of a second historical time period, wherein the first historical time period is before the first time period, and the second historical time period is after the first historical time period; the first processing unit is used to use the N first components, first data of the first historical time period and the second data of the second historical time period as the first training sample; the second processing unit is used to use the N first components of the second historical time period as the first training label; the first training unit is used to iteratively train the first neural network according to the first training sample and the first training label to obtain Z first target prediction models; the third processing unit is used to use the M second components, first data of the first historical time period and the second data of the second historical time period as the second training sample; the fourth processing unit is used to use the M second components of the second historical time period as the second training label; the second training unit is used to iteratively train the second neural network according to the second training sample and the second training label to obtain M second target prediction models.

[0094] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the above-mentioned load forecasting method.

[0095] According to another aspect of the present application, an electronic device is also provided, wherein the electronic device includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned load forecasting method.

[0096] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0097] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0099] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0100] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0101] If the integrated unit is implemented in the form of a software functional unit 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 application, in essence, or the part that contributes to the prior art or all or 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 application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0102] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A load forecasting method, characterized in that: include: Acquire a historical load sequence in a first time period, first data, and second data in a second time period, wherein the first data is used to characterize factors that affect the historical load sequence in the first time period, and the second data is used to characterize factors that affect the load sequence in the second time period, and the second time period is after the first time period; Decomposing the historical load sequence into N first components and M second components through a target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1, and the target operation is used to decompose nonlinear and non-stationary data into modal components, wherein the first component is used to characterize the dynamic characteristics of the load data within a target frequency range, and the second component is used to characterize the data in the load data whose frequency is less than a preset frequency; Divide the N first components into Z sequences according to the change in the value of each first component, where Z is an integer greater than 1, and each of the sequences includes the first components that meet the preset conditions; Adding the first data and the second data to each of the Z sequences, and taking the sequence added with the first data and the second data as a target sequence, to obtain Z target sequences; A target load forecast result for a second time period is determined according to the Z target sequences and the M second components.

2. The load forecasting method according to claim 1, characterized in that: Decomposing the historical load sequence into N first components and M second components through a target operation includes: Get the white noise set; Adding the white noise in the white noise set to the historical load sequence S times respectively to obtain S historical loads to be decomposed, where S is an integer greater than 1; The S historical loads to be decomposed are decomposed through the target operation to obtain N first components and M second components.

3. The load forecasting method according to claim 1, characterized in that: According to the change of the value of each of the first components, the N first components are divided into Z sequences, including: Performing a stationarity test on the N first components to obtain a test result, wherein the stationarity test is used to evaluate whether the statistical characteristics of the load data change over time; According to the test result, the N first components are divided into Z sequences, wherein the Z sequences include at least a fluctuation sequence, a periodic sequence, and a stationary sequence, wherein the fluctuation sequence includes the first component whose data change satisfies a first preset condition, the periodic sequence includes the first component whose data change satisfies a second preset condition, and the stationary sequence includes the first component whose data change satisfies a third preset condition, wherein the first preset condition indicates that the data change fluctuates beyond a preset threshold, the second preset condition indicates that the data changes periodically, and the third preset condition indicates that the statistical characteristics of the data remain unchanged.

4. The load forecasting method according to claim 3, characterized in that: After dividing the N first components into Z sequences according to the test result, the load forecasting method further includes: Data cleaning and preprocessing are performed on the Z sequences and the M second components, wherein the data cleaning is used to process outliers in the Z sequences and the M second components, and the preprocessing is used to normalize and process missing values ​​on the Z sequences and the M second components.

5. The load forecasting method according to claim 1, characterized in that: Determining a target load forecast result for a second time period according to the Z target sequences and the M second components includes: Inputting Z target sequences into Z first target prediction models to obtain Z first initial load prediction results, wherein the first target prediction model is used to extract features from the target sequence, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling; Add the first data and the second data to each second component of the M second components, and use the second component added with the first data and the second data as a target component to obtain M target components; Input the M target components into M second target prediction models to obtain M second initial load forecast results, wherein the second target prediction model is used to extract features of the target components, perform time series modeling based on the extracted features, and perform load forecasting based on the time series modeling; determine the target load forecast result for the second time period based on the Z first initial load forecast results and the M second initial load forecast results.

6. The load forecasting method according to claim 5, characterized in that: Determining a target load forecast result for the second time period according to the Z first initial load forecast results and the M second initial load forecast results includes: Performing an inverse process of the normalization process on the Z first initial load forecast results and the M second initial load forecast results to obtain L third initial load forecast results, wherein the sum of Z and M is equal to L; Determining a weight of each of the L third initial load forecast results; The target load forecast result for the second time period is determined according to the L third initial load forecast results and the weights corresponding to the L third initial load forecast results.

7. The load forecasting method according to claim 5, characterized in that: The first target prediction model and the second target prediction model are obtained by the following steps: Acquire a training data set, wherein the training data set includes N first components, M second components, and first data of a first historical time period, and N first components, M second components, and second data of a second historical time period, wherein the first historical time period is before the first time period, and the second historical time period is after the first historical time period; Using the N first components and the first data of the first historical time period and the second data of the second historical time period as first training samples; Using the N first components in the second historical time period as first training labels; Iteratively training the first neural network according to the first training samples and the first training labels to obtain Z first target prediction models; Using the M second components of the first historical time period, the first data, and the second data of the second historical time period as second training samples; Using the M second components of the second historical time period as second training labels; The second neural network is iteratively trained according to the second training samples and the second training labels to obtain M second target prediction models.

8. A load forecasting device, characterized in that: include: an acquisition unit, acquiring a historical load sequence in a first time period, first data, and second data in a second time period, wherein the first data is used to characterize factors that have an impact on the historical load sequence in the first time period, and the second data is used to characterize factors that have an impact on the load sequence in the second time period, and the second time period is after the first time period; A decomposition unit, decomposing the historical load sequence into N first components and M second components through a target operation, wherein N is an integer greater than 1, and M is an integer greater than or equal to 1, and the target operation is used to decompose nonlinear and non-stationary data into modal components, wherein the first component is used to characterize the dynamic characteristics of the load data within a target frequency range, and the second component is used to characterize the data in the load data whose frequency is less than a preset frequency; a dividing unit, dividing the N first components into Z sequences according to the numerical change of each of the first components, wherein Z is an integer greater than 1, and each of the sequences includes the first components that meet a preset condition; a processing unit, adding the first data and the second data to each of the Z sequences, and using the sequence to which the first data and the second data are added as a target sequence, to obtain Z target sequences; A determination unit determines a target load forecast result for a second time period according to the Z target sequences and the M second components.

9. A computer-readable storage medium, characterized in that: A computer program is stored in the computer-readable storage medium, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the load forecasting method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the load forecasting method described in any one of claims 1 to 7.

Images