Power load prediction method, device, equipment and medium

Abstract

The invention discloses a power load prediction method, device and equipment and a medium, and relates to the technical field of power load prediction.The method comprises the steps that firstly, a historical original load sequence is obtained; then, performing signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal; performing modal decomposition on the high-frequency signal and the low-frequency signal by using an improved variational modal decomposition algorithm to obtain a residual signal and a plurality of intrinsic modal components; decomposing the residual signal by using a wavelet basis function to obtain a wavelet coefficient corresponding to each intrinsic mode component; for each intrinsic mode component, inputting the intrinsic mode component and the wavelet coefficient corresponding to the intrinsic mode component into a trained load prediction model to obtain a prediction result corresponding to each intrinsic mode component; and finally, a final load prediction result is determined according to the prediction results corresponding to all the intrinsic mode components, and the power load prediction precision is improved.

Description

Technical Field

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

[0002] With the global energy transition and the advancement of smart grid technology, power load forecasting, a core technology for power system planning and operation, is facing unprecedented challenges and opportunities. The proportion of renewable energy generation continues to increase, and the power system is characterized by the complex interaction of multiple actors: power generation, grid, load, and storage. Traditional load forecasting methods based on historical data are no longer able to adapt to the dynamic changes associated with renewable energy consumption and the large-scale integration of electric vehicles. There is an urgent need to develop a high-precision, multi-scenario load forecasting system.

[0003] At the planning level, accurate forecasting of load growth trends can optimize power generation layout and grid investment, avoiding duplication of construction or insufficient capacity. At the operational level, short-term load forecasts directly influence power generation scheduling and reserve capacity allocation, reducing spinning reserve costs. At the market level, medium- and long-term forecasts provide a basis for decision-making by power trading entities and enhance the scientific nature of bidding strategies. Especially in the context of frequent extreme weather, ultra-short-term load forecasts can provide early warning of grid overload risks, ensuring the safety of electricity for people's livelihoods.

[0004] However, the current research methods on load forecasting have poor accuracy in load forecasting. Summary of the Invention

[0005] The purpose of this application is to provide a power load forecasting method, device, equipment and medium, which can improve the accuracy of load forecasting.

[0006] To achieve the above objectives, this application provides the following solutions.

[0007] In a first aspect, the present application provides a method for power load forecasting, comprising the following steps.

[0008] Get the historical raw load series.

[0009] Signal separation is performed on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal.

[0010] The high-frequency signal and the low-frequency signal are modally decomposed using an improved variational modal decomposition algorithm to obtain a residual signal and several intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence.

[0011] The residual signal is decomposed using a wavelet basis function to obtain a wavelet coefficient corresponding to each of the eigenmode components.

[0012] The eigenmode components and the wavelet coefficients corresponding to the eigenmode components are input into a trained load prediction model to obtain prediction results corresponding to each of the eigenmode components.

[0013] The final load prediction result is determined according to the prediction results corresponding to all the eigenmodal components.

[0014] In a second aspect, the present application provides an electric load forecasting device for implementing the electric load forecasting method described in the first aspect, comprising the following modules.

[0015] The historical original load sequence acquisition module is used to obtain the historical original load sequence.

[0016] The signal separation module is used to perform signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal.

[0017] A modal decomposition module is used to use an improved variational modal decomposition algorithm to perform modal decomposition on the high-frequency signal and the low-frequency signal to obtain a residual signal and several intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence.

[0018] The residual signal decomposition module is used to decompose the residual signal using a wavelet basis function to obtain a wavelet coefficient corresponding to each of the eigenmode components.

[0019] The load prediction module is used to input the eigenmode components and the wavelet coefficients corresponding to the eigenmode components into a trained load prediction model to obtain prediction results corresponding to each of the eigenmode components.

[0020] The final load forecast result determination module is used to determine the final load forecast result according to the prediction results corresponding to all the eigenmode components.

[0021] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned power load forecasting method.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned power load forecasting method when executed by a processor.

[0023] According to the specific embodiments provided in the present application, the present application discloses the following technical effects: the present application provides a method, apparatus, equipment and medium for power load forecasting, which uses an improved variational modal decomposition algorithm to perform modal decomposition on high-frequency signals and low-frequency signals, thereby solving the problem of physical feature coupling; by performing signal separation on the historical original load sequence, high-frequency signals and low-frequency signals are obtained, and the improved variational modal decomposition algorithm is used to perform modal decomposition on the high-frequency signals and low-frequency signals to obtain residual signals and several intrinsic modal components, and the residual signals are further decomposed using wavelet basis functions, that is, through a multi-level decomposition method, the problem of insufficient capture of the time series nonlinear characteristics of the input features by traditional load forecasting methods is solved, and the problems of modal aliasing and non-stationary residuals are effectively solved, which can reduce feature extraction errors and thereby improve the accuracy of power load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 This is an application environment diagram of a power load forecasting method in one embodiment of the present application; Figure 2 A flowchart of a power load forecasting method provided in one embodiment of the present application; Figure 3 A schematic diagram of a specific process of a power load forecasting method provided in one embodiment of the present application; Figure 4 A schematic diagram of the functional modules of a power load forecasting device provided in one embodiment of the present application; Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0027] At present, there are four main research methods for load forecasting: (1) Traditional time series models, such as ARIMA and seasonal decomposition methods, but they rely too much on the statistical laws of historical data and have poor adaptability to nonlinear and non-stationary load fluctuations; (2) Machine learning models, such as SVM and LSTM, use data-driven learning features, but suffer from dimensionality disaster and overfitting problems, and ignore the physical characteristics of the load waveform; (3) Signal decomposition methods, such as wavelet decomposition and empirical mode decomposition, improve prediction accuracy by decomposing signals into different modal components, but they have problems such as inability to adapt to signal characteristics and modal aliasing problems that lead to blurred physical meanings of components; (4) Feature extraction methods, existing methods mostly use single features (such as mean and variance), but fail to fully explore the time domain, frequency domain, and morphological characteristics of the waveform.

[0028] Therefore, this application addresses the problem that traditional load forecasting methods cannot adequately capture the temporal nonlinear characteristics of input features. Through a three-level decomposition structure of high-pass and low-pass filtering algorithms, an improved variational mode decomposition algorithm, and a wavelet decomposition algorithm, based on long short-term memory networks and temporal convolutional network models, combined with an attention mechanism to dynamically adjust weights for load types such as high-frequency mutations and low-frequency trends, a power load forecasting technology based on multi-level decomposition and dynamic fusion is proposed.

[0029] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0030] The power load forecasting method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, integrated on the server 104, or placed on the cloud or other servers. The terminal 102 can send the historical original load sequence to the server 104. After the server 104 receives the historical original load sequence, the server 104 performs signal separation on the historical original load sequence to obtain high-frequency signals and low-frequency signals. The high-frequency signal and the low-frequency signal are modally decomposed using the improved variational mode decomposition algorithm to obtain residual signals and several intrinsic mode components. The residual signal is decomposed using the wavelet basis function to obtain the wavelet coefficients corresponding to each intrinsic mode component. For each intrinsic mode component, the intrinsic mode component and the wavelet coefficients corresponding to the intrinsic mode component are input into the trained load forecasting model to obtain the prediction results corresponding to each intrinsic mode component. The final load forecast result is determined based on the prediction results corresponding to all intrinsic mode components. The server 104 can feed back the final load forecast result obtained for the historical original load sequence to the terminal 102. In addition, in some embodiments, the power load forecasting method can also be implemented independently by the server 104 or the terminal 102. For example, the terminal 102 can directly perform power load forecasting for the historical original load sequence, or the server 104 can obtain the historical original load sequence from a data storage system and perform power load forecasting for the historical original load sequence.

[0031] Terminal 102 may include, but is not limited to, various desktop computers, laptops, smartphones, tablet computers, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Server 104 may be implemented as a standalone server or a server cluster consisting of multiple servers, or may be a cloud server.

[0032] In an exemplary embodiment, Figure 2 and Figure 3 As shown, a method for predicting power load is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used for explanation, including the following steps 201 to 206.

[0033] Step 201: Obtain historical original load sequence.

[0034] Step 202: performing signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal.

[0035] Step 203: Using an improved variational modal decomposition algorithm, perform modal decomposition on the high-frequency signal and the low-frequency signal to obtain a residual signal and several intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence.

[0036] Step 204: Decompose the residual signal using wavelet basis functions to obtain wavelet coefficients corresponding to each of the intrinsic mode components.

[0037] Step 205 : Input the eigenmode components and the wavelet coefficients corresponding to the eigenmode components into a trained load prediction model to obtain prediction results corresponding to the eigenmode components.

[0038] Step 206: Determine a final load prediction result based on the prediction results corresponding to all the intrinsic mode components.

[0039] By implementing the above steps 201 to 206, high-frequency and low-frequency signals are obtained by performing signal separation on the historical original load sequence. The high-frequency and low-frequency signals are modally decomposed using the improved variational mode decomposition algorithm to obtain residual signals and several intrinsic mode components. The residual signals are further decomposed using wavelet basis functions. The three-level decomposition structure of high-pass and low-pass filtering algorithms, variational mode decomposition algorithms, and wavelet decomposition algorithms solves the problem of insufficient capture of the time series nonlinear characteristics of input features by traditional load forecasting methods, effectively solves the problems of modal aliasing and non-stationary residuals, can reduce feature extraction errors, and thus improve the accuracy of power load forecasting. In addition, the load forecasting model includes an LSTM-TCN hybrid model and an attention mechanism. The present application can also use the LSTM-TCN hybrid model, combined with the attention mechanism, to dynamically adjust the weights for different load types such as load peak high-frequency mutations and seasonal cycle low-frequency trends, thereby improving the stability of load forecasting. The method provided in this embodiment is a power load forecasting method based on multi-level decomposition and dynamic fusion.

[0040] In another exemplary embodiment of the present application, the above step 202 specifically includes: using a high-pass filter to separate the signals of the historical original load sequence to obtain a high-frequency signal; using a low-pass filter to separate the signals of the historical original load sequence to obtain a low-frequency signal.

[0041] Get historical raw load series , a dual-channel filtering algorithm (including high-pass filter and low-pass filter) is used to separate the high-frequency signal and the low-frequency signal. The calculation formula is shown in the following formula (1).

[0042] (1).

[0043] Where, is the cutoff frequency f c =0.1Hz Butterworth high-pass filter; is the cutoff frequency f c =0.05Hz low-pass filter; is a convolution operator used for signal filtering; It is the load signal after high-pass filtering, that is, high-frequency signal, used to represent short-term load fluctuations; It is the load signal after low-pass filtering, that is, the low-frequency signal, which is used to represent the long-term trend of the load.

[0044] In order to solve the problem of physical feature coupling, this application introduces load type based on traditional variational modal decomposition (VMD) The improved VMD algorithm is designed. After the load characteristic constraint term is introduced, the improved variational mode decomposition algorithm is expressed as shown in the following formula.

[0045] (2).

[0046] in, for Moment eigenmode components (decomposed sub-signals), =1,2,3,..., , is the preset number of decomposed modes, that is, the total number of eigenmode components; For the The center frequency of the eigenmode components; For time The partial derivative of is used to constrain the smoothness of the eigenmode components; Dirac Function, representing the impulse response; represents the imaginary unit, =-1; is the convolution operator; is the adaptive adjustment parameter; It is a load type related function, which dynamically adjusts the decomposition constraints according to the physical characteristics of the load. The load types are divided into industrial power charge, residential power charge and commercial power charge.

[0047] For industrial power loads, = , then the calculation formula of the load type related function corresponding to the industrial power charge is , industrial power charge is calculated by time. When the predicted time is between 8:00 and 18:00 every day, , when the predicted time is not between 8:00 and 18:00 every day, .

[0048] Regarding residential power load, = , then the calculation formula of the load type related function corresponding to the residential power charge is .

[0049] For commercial power loads, = , then the calculation formula of the load type related function corresponding to the commercial power load is .

[0050] In the calculation formula of the above load type related function: P base The basic load is the minimum power demand of this type of load that exists continuously and is relatively stable. is a normally distributed random variable (mean 0, standard deviation 0.2).

[0051] The high frequency signal and low-frequency signals As input, the improved VMD algorithm is decomposed into Eigenmode components { }, and separate The residual signal at time , expressed as shown below.

[0052] (3).

[0053] In the above step 204, the residual signal is further decomposed using the custom wavelet basis function , extract more refined time-frequency features , decompose the incompletely separated noise and non-stationary components to enhance noise robustness, perform multi-layer wavelet decomposition on the residual term, and then calculate the wavelet coefficients corresponding to the eigenmode components as follows: (4).

[0054] Where, For the The wavelet coefficients corresponding to the eigenmode components are expressed in scale and location The decomposition results on ; is the wavelet basis function, which is obtained by scaling and translation transformation of the mother wavelet. It is expressed as follows.

[0055] (5).

[0056] Among them, the custom wavelet basis function It is expressed as shown in the following formula.

[0057] (6).

[0058] in, is the attenuation coefficient, which controls the time domain width of the wavelet basis function (the larger the value, the narrower the time domain); is the nonlinear frequency modulation coefficient, which is used to adapt to the non-stationary characteristics of the load waveform; and Dynamic optimization is performed through gradient descent method, with the goal of minimizing the reconstruction error.

[0059] The load forecasting model in step 205 includes an attention mechanism module and several LSTM-TCN hybrid predictors. The number of LSTM-TCN hybrid predictors is equal to the number of intrinsic mode components. The LSTM-TCN hybrid predictors include a Long Short-Term Memory (LSTM) model and a Temporal Convolutional Network (TCN) model. The attention mechanism module is used to adjust the dynamic weights corresponding to the LSTM-TCN hybrid predictors in real time.

[0060] An LSTM-TCN hybrid predictor corresponds to an eigenmode component and a wavelet coefficient corresponding to the eigenmode component. For any LSTM-TCN hybrid predictor, the LSTM model is used to predict the eigenmode component and the wavelet coefficient corresponding to the eigenmode component to obtain a first prediction result; the TCN model is used to predict the eigenmode component and the wavelet coefficient corresponding to the eigenmode component to obtain a second prediction result; the prediction result corresponding to the eigenmode component is determined based on the first prediction result, the second prediction result and the dynamic weight.

[0061] The eigenmode components { } and the wavelet coefficients of step 204 As input, each eigenmode component and its corresponding wavelet coefficient are input into a corresponding LSTM-TCN hybrid predictor respectively. An LSTM-TCN hybrid predictor is established for each decomposition component. The LSTM-TCN hybrid predictor is composed of an LSTM model and a TCN model. The prediction result corresponding to the eigenmode component is expressed as shown in the following formula.

[0062] (7).

[0063] Where: For the The predicted value of the decomposition component, that is, The prediction results corresponding to the eigenmode components; For the LSTM model The prediction result of the decomposition component, that is, The first prediction result corresponding to the eigenmode component; The TCN model is The prediction result of the decomposition component, that is, The second prediction result corresponding to the eigenmode component; The dynamic weight corresponding to the LSTM model has a value range of [0,1] and is used to fuse the prediction results of the LSTM model and the TCN model; is the dynamic weight corresponding to the TCN model.

[0064] Dynamic Weight Generated by the attention mechanism module, the attention mechanism module is based on the time-frequency characteristics of the intrinsic mode component (hidden state The weights of the LSTM and TCN models are adjusted in real time. For sudden spikes in load, the dynamic weight of the TCN model increases, while for gradual changes in periodic load, the weight of the LSTM model increases. The attention mechanism module assigns weights based on the importance of the intrinsic mode component features, as shown in the following formula.

[0065] (8).

[0066] Where, For the The hidden state corresponding to the intrinsic mode component is the hidden state of the LSTM model at each time step; is the attention query vector; is the trainable weight matrix; b is the bias vector; tanh() is the hyperbolic tangent activation function.

[0067] In another exemplary embodiment of the present application, the above step 206 specifically includes the following steps 301 to 304.

[0068] Step 301: For each of the eigenmode components, a gated recurrent unit is used to determine a dynamic fusion weight corresponding to the eigenmode component.

[0069] Step 302: Perform weighted summation based on the prediction result corresponding to the intrinsic mode component and the dynamic fusion weight to obtain the eigenmode component feature.

[0070] Step 303: Use the spatiotemporal convolution layer to perform feature extraction on the prediction results corresponding to all the intrinsic mode components to obtain spatiotemporal correlation features across components.

[0071] Step 304: Determine the final load forecast result based on the eigenmode component characteristics and the spatiotemporal correlation characteristics across components.

[0072] The final synthesis weights are dynamically adjusted through spatiotemporal convolutional layers and gated recurrent units (GRUs) to adapt to sudden load changes. A synthesis network with intersecting spatiotemporal features is designed. This network consists of a spatiotemporal convolutional layer and a gated recurrent unit. The final load forecast result is calculated as shown in the following formula.

[0073] (9).

[0074] Where, is the final composite load forecast value, that is Final load forecast result at the moment; For the The dynamic fusion weight corresponding to each intrinsic mode component, when the load changes suddenly, the weight of the high-frequency component will increase instantaneously; For the The prediction results corresponding to the eigenmode components; The weight coefficient of the convolution feature determined through training, that is, the weight coefficient corresponding to the spatiotemporal convolution layer; It is a one-dimensional convolution layer, that is, a spatiotemporal convolution layer, which is used to extract spatiotemporal correlation features across components.

[0075] It should be noted that the weight coefficient of the convolution feature determined by training is , will change according to the input historical original load sequence. For example, when the historical original load sequence from the 1st time step to the 100th time step is used to predict the load at the 101st time step, the weight coefficient corresponding to the spatiotemporal convolution layer is When predicting the load at the 102nd time step, the load sequence from the 2nd time step to the 101st time step can be used as the historical original load sequence. At this time, the weight coefficient corresponding to the spatiotemporal convolution layer will change with the change of the historical original load sequence, which is recorded as , The weight coefficients corresponding to the spatiotemporal convolution layer are dynamically generated through the attention mechanism module and are adaptively adjusted according to the temporal characteristics of the input historical original load sequence. The calculation formula is shown in formula (8). Replace with That's it.

[0076] Final prediction results The error with the actual load is back-propagated to each step and updated through the gated recurrent unit. The update formula is shown below.

[0077] (10).

[0078] Where, The current time step The corresponding dynamic fusion weights of the eigenmode components; is the Sigmoid function, which maps the output to the interval [0, 1]; is the trainable weight matrix; is the hidden state of the previous time step, that is The hidden state of the moment; is the context feature of the current time step, that is The contextual features of the moment come from the output of the temporal convolutional layer; For offset items.

[0079] In order to address the problem that traditional load forecasting methods cannot adequately capture the temporal nonlinear characteristics of input features, this application proposes a power load forecasting method based on multi-level decomposition and dynamic fusion by using high-pass and low-pass filtering to separate the original signal, extracting the main mode and residual features step by step based on the improved variational mode decomposition algorithm and wavelet decomposition, allocating long-short-term memory networks and time convolution networks according to the component frequency characteristics, combining a dynamic weight adaptive selection model, and using gated recurrent units and spatiotemporal convolution to fuse the prediction results of each component.

[0080] The present application also provides an application scenario, which applies the above-mentioned power load forecasting method. Specifically: the power load forecasting method provided in this embodiment can be applied in the residential power load forecasting scenario. The residential power load forecasting scenario includes a historical charge generation link and a power load forecasting link; the historical original load sequence enters the power load forecasting link from the historical charge generation link to obtain corresponding content features. The power load forecasting method provided in this embodiment belongs to the power load forecasting link. Specifically, in the power load forecasting link process for the historical original load sequence, the historical original load sequence can be subjected to signal separation to obtain high-frequency signals and low-frequency signals. The high-frequency signals and low-frequency signals are subjected to modal decomposition using an improved variational mode decomposition algorithm to obtain residual signals and several intrinsic mode components. The residual signal is decomposed using a wavelet basis function to obtain the wavelet coefficients corresponding to each intrinsic mode component. For each intrinsic mode component, the eigenmode component and the wavelet coefficients corresponding to the eigenmode component are input into the trained load forecasting model to obtain the prediction results corresponding to each eigenmode component. The final load forecast result is determined based on the prediction results corresponding to all eigenmode components.

[0081] Based on the same inventive concept, the present application also provides an electric load forecasting device for implementing the above-mentioned electric load forecasting method. The implementation solution provided by this device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations of one or more electric load forecasting device embodiments provided below can be referred to the limitations of the electric load forecasting method above and will not be repeated here.

[0082] In an exemplary embodiment, Figure 4 As shown, a power load prediction device is provided, which includes the following modules.

[0083] The historical original load sequence acquisition module T1 is used to obtain the historical original load sequence.

[0084] The signal separation module T2 is used to perform signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal.

[0085] The modal decomposition module T3 is used to use an improved variational modal decomposition algorithm to perform modal decomposition on the high-frequency signal and the low-frequency signal to obtain a residual signal and several intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence.

[0086] The residual signal decomposition module T4 is used to decompose the residual signal using a wavelet basis function to obtain a wavelet coefficient corresponding to each of the eigenmode components.

[0087] The load prediction module T5 is used to input the eigenmode components and the wavelet coefficients corresponding to the eigenmode components into a trained load prediction model to obtain prediction results corresponding to each of the eigenmode components.

[0088] The final load prediction result determination module T6 is used to determine the final load prediction result according to the prediction results corresponding to all the eigenmode components.

[0089] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store power load forecast data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a power load forecasting method is implemented.

[0090] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0091] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0092] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0093] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0094] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0095] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0096] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0097] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for predicting power load, characterized in that: The power load forecasting method comprises: Get historical original load series; Performing signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal; Using an improved variational modal decomposition algorithm, modal decomposition is performed on the high-frequency signal and the low-frequency signal to obtain a residual signal and a plurality of intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence; Decomposing the residual signal using a wavelet basis function to obtain a wavelet coefficient corresponding to each of the eigenmode components; Inputting the eigenmode components and the wavelet coefficients corresponding to the eigenmode components into a trained load prediction model to obtain prediction results corresponding to each of the eigenmode components; The final load prediction result is determined according to the prediction results corresponding to all the eigenmodal components.

2. The power load forecasting method according to claim 1, characterized in that: Determining the final load prediction result based on the prediction results corresponding to all the eigenmodal components specifically includes: For each of the intrinsic mode components, using a gated recurrent unit, determining a dynamic fusion weight corresponding to the intrinsic mode component; Performing weighted summation based on the prediction result corresponding to the intrinsic mode component and the dynamic fusion weight to obtain the eigenmode component feature; Using a spatiotemporal convolutional layer to extract features from the prediction results corresponding to all the intrinsic mode components, and obtain spatiotemporal correlation features across components; The final load forecast result is determined based on the characteristics of the eigenmode components and the spatiotemporal correlation characteristics across components.

3. The power load forecasting method according to claim 2, characterized in that: The calculation formula of the final load forecast result is: ； Where, for Final load forecast result at the moment; For the The dynamic fusion weights corresponding to the eigenmode components, =1,2,3,..., , is the total number of eigenmode components; For the The prediction results corresponding to the eigenmode components; is the weight coefficient corresponding to the spatiotemporal convolution layer; It is the spatiotemporal convolution layer.

4. The power load forecasting method according to claim 1, characterized in that: Signal separation is performed on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal, specifically including: Using a high-pass filter, performing signal separation on the historical original load sequence to obtain a high-frequency signal; A low-pass filter is used to perform signal separation on the historical original load sequence to obtain a low-frequency signal.

5. The power load forecasting method according to claim 1, characterized in that: The improved variational mode decomposition algorithm is expressed as: ； in, for Moment eigenmode components, =1,2,3,..., , is the total number of eigenmode components; For the The center frequency of the eigenmode components; For time The partial derivative of Dirac function; represents an imaginary unit; is the convolution operator; is the adaptive adjustment parameter; is a function related to load type.

6. The power load forecasting method according to claim 1, characterized in that: The load forecasting model includes an attention mechanism module and a plurality of LSTM-TCN hybrid predictors; the number of LSTM-TCN hybrid predictors is the number of intrinsic mode components; the LSTM-TCN hybrid predictor includes an LSTM model and a TCN model; The attention mechanism module is used to adjust the dynamic weights corresponding to the LSTM-TCN hybrid predictor in real time; An LSTM-TCN hybrid predictor corresponds to an eigenmode component and a wavelet coefficient corresponding to the eigenmode component. For any LSTM-TCN hybrid predictor, the LSTM model is used to predict the eigenmode component and the wavelet coefficient corresponding to the eigenmode component to obtain a first prediction result; the TCN model is used to predict the eigenmode component and the wavelet coefficient corresponding to the eigenmode component to obtain a second prediction result; the prediction result corresponding to the eigenmode component is determined based on the first prediction result, the second prediction result and the dynamic weight.

7. The power load forecasting method according to claim 1, characterized in that: The calculation formula of the wavelet coefficient corresponding to the eigenmode component is as follows: ； Where, For the The wavelet coefficients corresponding to the eigenmode components; for The residual signal at time t; is the wavelet basis function.

8. A power load forecasting device for implementing the power load forecasting method according to any one of claims 1 to 7, characterized in that: The power load prediction device includes: A historical original load sequence acquisition module is used to obtain the historical original load sequence; A signal separation module is used to perform signal separation on the historical original load sequence to obtain a high-frequency signal and a low-frequency signal; a modal decomposition module for performing modal decomposition on the high-frequency signal and the low-frequency signal using an improved variational modal decomposition algorithm to obtain a residual signal and a plurality of intrinsic modal components; the improved variational modal decomposition algorithm introduces a load type correlation function; the load type correlation function is determined by the load type of the historical original load sequence; A residual signal decomposition module, configured to decompose the residual signal using a wavelet basis function to obtain a wavelet coefficient corresponding to each of the intrinsic mode components; A load prediction module, configured to input the intrinsic mode components and the wavelet coefficients corresponding to the eigenmode components into a trained load prediction model to obtain prediction results corresponding to each eigenmode component; The final load forecast result determination module is used to determine the final load forecast result according to the prediction results corresponding to all the eigenmode components.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the power load forecasting method according to any one of claims 1 to 7.

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

Images