Time-sharing electric quantity prediction method and device

Abstract

The invention is suitable for the technical field of electric quantity prediction, and provides a time-sharing electric quantity prediction method and device, and the method comprises the steps: obtaining initial time-sharing electric quantity data and initial meteorological data; performing symplectic geometric mode decomposition on the initial time-sharing electric quantity data, and performing de-noising processing on the decomposed initial time-sharing electric quantity data and initial meteorological data to respectively obtain de-noised time-sharing electric quantity data and a multi-dimensional meteorological sequence; performing multivariate variational mode decomposition on the denoised time-sharing electric quantity data and the multi-dimensional meteorological sequence to obtain same-frequency multivariate components; and inputting the same-frequency multivariate components into the space-time convolutional network, and obtaining a final prediction result based on a tuna swarm optimization algorithm. The time-sharing electric quantity decomposition prediction method and device can adapt to the spot market price uncertain environment, and the time-sharing electric quantity decomposition prediction precision is improved.

Description

Technical Field

[0001] The present application belongs to the technical field of power prediction, and in particular, relates to a time-sharing power prediction method and device. Background Art

[0002] Since electricity decomposition is an important task in the transaction process of each power company, and an effective electricity decomposition method can improve the power dispatching situation in the future, electricity decomposition plays a critical role in the operation of power companies, and many scholars have conducted research on this.

[0003] In related studies, a contract decomposition model for market managers to minimize the market power of power generators was constructed, and an integrated prediction method consisting of an empirical mode decomposition (EMD) algorithm and a deep learning method was proposed for load demand time series prediction. Other related studies consider that electric load data generally exhibits strong nonlinear and non-stationary characteristics, and data needs to be preprocessed to generate more regular subsequences. A prediction model based on improved variational mode decomposition and chaotic gray wolf algorithm optimization is proposed. Considering that a time series may contain multiple seasonal cycles of different lengths, a new combined model integrating multiple seasonal patterns, multiple neural networks, non-positive constraint theory and BFGS-FA is proposed. A prediction method combining curve decomposition with long short-term memory network (LSTM) considering the relevant factors of Bayesian optimization algorithm (BOA) is proposed, and a corresponding monthly electricity decomposition model is proposed for the multi-dimensional electricity curve decomposition problem involving multiple varieties, multiple provinces and multiple time scales in the market of giant hydropower stations. .

[0004] However, the above method only considers the electricity data itself, and does not consider other factors related to electricity, resulting in poor adaptability. In an environment of uncertain spot market prices, the prediction accuracy of time-based electricity decomposition is not high. Summary of the invention

[0005] The embodiments of the present application provide a time-sharing electricity forecasting method and device, which can adapt to the price uncertainty environment of the spot market and improve the accuracy of time-sharing electricity decomposition forecasting.

[0006] This application is implemented through the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a time-sharing power prediction method, comprising:

[0008] Obtain initial time-sharing electricity data and initial meteorological data;

[0009] The initial time-sharing electricity data is subjected to symplectic geometric mode decomposition, and the decomposed initial time-sharing electricity data and initial meteorological data are subjected to denoising, so as to obtain denoised time-sharing electricity data and multidimensional meteorological series respectively;

[0010] Perform multivariate variational mode decomposition on the denoised time-of-day electricity data and multidimensional meteorological series to obtain multivariate components with the same frequency.

[0011] The multivariate components with the same frequency are input into the spatiotemporal convolutional network, and the final prediction results are obtained based on the tuna school optimization algorithm.

[0012] In a possible implementation of the first aspect, symplectic geometric mode decomposition is performed on the initial time-sharing power data, and denoising is performed on the decomposed initial time-sharing power data to obtain denoised time-sharing power data, including:

[0013] Perform symplectic geometric mode decomposition on the initial time-sharing electricity data to obtain the electricity IMF components;

[0014] Determine the correlation between the IMF component of electricity and the initial time-sharing electricity data;

[0015] Based on the correlation, the electric quantity IMF component is divided into pure noise component, noisy electric quantity component and pure electric quantity component;

[0016] Perform improved wavelet threshold denoising on the noisy electrical quantity component to obtain the denoised noisy electrical quantity component;

[0017] The denoised noisy electricity component is synthesized with the pure electricity component to obtain the denoised time-sharing electricity data.

[0018] In a possible implementation of the first aspect, symplectic geometric mode decomposition is performed on the initial time-sharing electricity data to obtain electricity IMF components, including:

[0019] Convert the original signal of initial time-sharing electricity data into a phase space matrix;

[0020] Reconstruct the phase space matrix based on the delay time and the embedded dimension to obtain a reconstructed phase space matrix;

[0021] Based on the reconstructed phase space matrix, autocorrelation analysis is performed to obtain the covariance symmetric matrix;

[0022] Based on the covariance symmetric matrix, the initial single-component trajectory matrix is ​​obtained;

[0023] Based on the initial single-component trajectory matrix, the electric quantity IMF component is obtained.

[0024] In a possible implementation manner of the first aspect, obtaining an initial single-component trajectory matrix based on a covariance symmetric matrix includes:

[0025] Based on the covariance symmetric matrix, construct the Hamiltonian matrix;

[0026] Based on the Hamiltonian matrix, construct a symplectic orthogonal matrix;

[0027] Based on the symplectic orthogonal matrix, calculate the transformation coefficient matrix;

[0028] Based on the transformation coefficient matrix, the initial single-component trajectory matrix is ​​obtained.

[0029] In a possible implementation manner of the first aspect, the frequency component of the i-th row and k-th column in the denoised noisy component It is expressed as:

[0030]

[0031] Among them, j is the scale, k is the number of layers, w is the frequency, and w j,k is the wavelet coefficient at scale j and layer k in the denoised noisy component; i represents the index number of the wavelet coefficient at a specific scale j and layer k; sgn() is the sign function; T i is the threshold of the wavelet coefficient with index number i.

[0032] In a possible implementation of the first aspect, multivariate variational mode decomposition is performed on the denoised time-sharing electricity data and the multidimensional meteorological series to obtain multivariate components with the same frequency, including:

[0033] Through multivariate variational mode decomposition, the denoised time-sharing electricity data and multidimensional meteorological series are analyzed in time and frequency synchronization to obtain the electricity component and multidimensional meteorological series component at each frequency.

[0034] The electrical quantity component at the same frequency is combined with the multidimensional meteorological sequence component to form the same-frequency multivariate component.

[0035] In a possible implementation manner of the first aspect, the objective function of the multivariate variational mode decomposition is expressed as:

[0036]

[0037] Among them, {u k,c} is the set of modal functions of the denoised time-sharing electricity data and multidimensional meteorological series; {ω k} is {u k,c} center frequency set; u k,c is the kth IMF of the cth channel; C is the number of channels, K is the number of IMFs; ω k for The center frequency of is the partial derivative operation with respect to time; The Hilbert transform operator is used to transform the vector signal u k,c The analytical representation of (t); x c (t) Input data of the cth channel.

[0038] In a possible implementation manner of the first aspect, the spatiotemporal convolutional network includes an input layer, a plurality of residual units, and an output layer;

[0039] Each residual unit includes a first one-dimensional hole causal convolution layer, a first normalization layer, a first activation function, a first regularization layer, a second one-dimensional hole causal convolution layer, a second normalization layer, a second activation function and a second regularization layer, which are connected in sequence.

[0040] In a possible implementation of the first aspect, the same-frequency multivariate components are input into a spatiotemporal convolutional network, and a final prediction result is obtained based on a tuna school optimization algorithm, including:

[0041] The multivariate components of the same frequency are input into the spatiotemporal convolutional network, and the hyperparameters of the spatiotemporal convolutional network are optimized synchronously using the tuna school optimization algorithm to obtain the optimal hyperparameters of the spatiotemporal convolutional network.

[0042] Input the same-frequency multivariate components into the spatiotemporal convolutional network with optimal hyperparameters to obtain the corresponding predicted values ​​of each same-frequency multivariate component;

[0043] The predicted values ​​corresponding to each multivariate component of the same frequency are added together to form the final prediction result.

[0044] In a second aspect, an embodiment of the present application provides a time-sharing power prediction device, which executes the time-sharing power prediction method of the first aspect, including:

[0045] A data acquisition module is used to acquire initial time-sharing electricity data and initial meteorological data;

[0046] The first decomposition module is used to perform symplectic geometric mode decomposition on the initial time-sharing electricity data, and perform denoising on the decomposed initial time-sharing electricity data and initial meteorological data to obtain denoised time-sharing electricity data and multidimensional meteorological series respectively;

[0047] The second decomposition module is used to perform multivariate variational mode decomposition on the denoised time-sharing electricity data and multidimensional meteorological series to obtain multivariate components with the same frequency;

[0048] The result output module is used to input the same-frequency multivariate components into the spatiotemporal convolutional network and obtain the final prediction results based on the tuna school optimization algorithm.

[0049] Compared with the related art, the embodiments of the present application have the following beneficial effects:

[0050] The time-sharing electricity forecasting method and device of the embodiment of the present application decomposes the initial electricity consumption data through symplectic geometric mode decomposition, and then denoises it. Multivariate variational mode decomposition is used for the denoised electricity consumption data and the multidimensional meteorological series, and time-frequency synchronization analysis is performed on the electricity consumption and the multidimensional meteorological series, which are decomposed into multivariate intrinsic mode functions to adapt to the price uncertainty environment of the spot market; finally, for each component data set with the same frequency, a spatiotemporal convolutional network with hyperparameter optimization based on the tuna school optimization algorithm is used for prediction, and the prediction results are evaluated after the components are merged, so as to improve the accuracy of intra-day time-sharing electricity decomposition.

[0051] The beneficial effects of the above-mentioned second aspect embodiment refer to the beneficial effects of the first aspect embodiment, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. 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 paying creative labor.

[0053] Figure 1 This is an application scenario diagram of a time-sharing power prediction method and device provided in an embodiment of the present application;

[0054] Figure 2 It is a flowchart of a time-sharing power prediction method provided by an embodiment of the present application;

[0055] Figure 3 is a flowchart of a decomposition and denoising process provided by an embodiment of the present application;

[0056] Figure 4 It is a model flow chart of a time-sharing power prediction method provided by an embodiment of the present application;

[0057] Figure 5 is a first symplectic geometric mode decomposition component graph provided by an embodiment of the present application;

[0058] Figure 6 is a multivariate component frequency diagram of multivariate variational mode decomposition provided by an embodiment of the present application;

[0059] Figure 7 It is a secondary multivariate variational modal decomposition power consumption component diagram provided by an embodiment of the present application;

[0060] Figure 8 It is a schematic diagram of the structure of a spatiotemporal convolutional network provided in one embodiment of the present application;

[0061] Fig. 9It is a structural diagram of a time-sharing power prediction device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0062] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0063] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0064] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0065] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0066] References to "an embodiment", "one embodiment" or "some embodiments" described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0067] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0068] Reference Figure 1 1 is an application scenario diagram of the time-sharing power forecasting method and device of the present invention. The forecasting method and device run in the server 102. The server 102 obtains power data and related data from the power grid database 101.

[0069] With the continuous deepening of the power market reform, the power grid has temporarily changed from the traditional unified purchase and sales model to the role of agent power purchase. Under the influence of the agent power purchase policy and the time-sharing electricity price mechanism, users who purchase electricity through agents can feel the market price fluctuation signals in real time through the power grid enterprises, and reasonably adjust their electricity consumption behavior under the guidance of electricity prices. The types of power equipment on the customer side are complex and diverse. The impact of extreme weather, emergencies, etc. on power consumption is extremely complex. At the same time, profound changes in the power market and other environments have put forward higher requirements for power consumption. Due to the large number of types of power influencing factors and their strong coupling, it is difficult to accurately fit the development trends of various influencing factors when analyzing power data. Therefore, it is urgent to study the decomposition basis of time-sharing power indicators for different user groups with complex coupling factors, and realize the intelligent decomposition of intra-day time-sharing power based on daily and monthly power. The invention firstly decomposes the original electricity consumption sequence by using symplectic geometry mode decomposition (SGMD); then denoises based on correlation analysis and improved wavelet threshold method, and uses multivariate variational mode decomposition (MVMD) to perform time-frequency synchronization analysis on the denoised electricity consumption data and the multidimensional meteorological sequence, and decomposes them into frequency-aligned multivariate intrinsic mode functions; finally, for each component data set with the same frequency, a spatiotemporal convolutional network (TCN) after hyperparameter optimization based on a tuna swarm optimization algorithm (TSO) is used for prediction, and the prediction results are evaluated after the components are merged, so as to improve the accuracy of intra-day time-sharing electricity decomposition.

[0070] Figure 2 is a flow chart of a time-sharing power prediction method provided by an embodiment of the present application, referring to Figure 2 The time-sharing power forecasting method includes:

[0071] Step 201, obtaining initial time-sharing electricity data and initial meteorological data.

[0072] Exemplarily, the initial time-sharing electricity consumption data may include actual electricity consumption data of the power grid in the target area within a preset time period; the initial meteorological data may include temperature data, humidity data, precipitation data, wind speed data, etc.

[0073] Step 202 , performing symplectic geometric mode decomposition on the initial time-sharing electricity data, and performing denoising processing on the decomposed initial time-sharing electricity data and initial meteorological data, to obtain denoised time-sharing electricity data and multi-dimensional meteorological series respectively.

[0074] Each intrinsic mode function (IMF) of the initial time-sharing electricity data and each intrinsic mode function of the initial meteorological data are denoised and then synthesized to obtain the denoised time-sharing electricity data and multidimensional meteorological series.

[0075] Step 203, performing multivariate variational mode decomposition on the denoised time-sharing electricity data and the multidimensional meteorological series to obtain multivariate components with the same frequency.

[0076] Exemplarily, the denoised time-of-day electricity data and the multidimensional meteorological series are decomposed into a multivariate component set, which ensures that each prediction model is established at the same central frequency, which is conducive to improving the prediction accuracy.

[0077] Step 204, input the same frequency multivariate components into the spatiotemporal convolutional network, and obtain the final prediction result based on the tuna school optimization algorithm.

[0078] This embodiment decomposes the initial electricity consumption data through symplectic geometric mode decomposition, and then denoises it. Multivariate variational mode decomposition is used for the denoised electricity consumption data and the multidimensional meteorological series. Time-frequency synchronization analysis is performed on the electricity consumption and the multidimensional meteorological series, and they are decomposed into multivariate intrinsic mode functions to adapt to the price uncertainty environment of the spot market. Finally, for each component data set with the same frequency, a spatiotemporal convolutional network with hyperparameter optimization based on the tuna school optimization algorithm is used for prediction. The prediction results are evaluated after the components are merged, so as to improve the accuracy of intraday time-sharing electricity decomposition.

[0079] In one embodiment, see Figure 3 and Figure 4 Taking the initial time-sharing power data as an example, the decomposition and denoising process in step 202 is described in detail. The initial time-sharing power data is subjected to symplectic geometric mode decomposition, and the decomposed initial time-sharing power data is subjected to denoising to obtain denoised time-sharing power data, including:

[0080] Step 2021, perform symplectic geometric mode decomposition on the initial time-sharing electricity data to obtain electricity IMF components.

[0081] Step 2021 includes:

[0082] Step 1: Convert the original signal of the initial time-sharing electricity data into a phase space matrix.

[0083] Assume that n is the number of original signals of initial time-sharing electricity data, and the selected time series can be expressed as x = x1, x2, ..., x n Based on Takens embedding theorem, the original signal is transformed into the phase space matrix X.

[0084]

[0085] Step 2: Reconstruct the phase space matrix based on the delay time and the embedding dimension to obtain the reconstructed phase space matrix.

[0086] The values ​​of the delay time τ and the embedding dimension d are the key to phase space reconstruction. The expression of the reconstructed phase space I(τ) with respect to the delay time τ is:

[0087]

[0088] Among them, p(x i ,x i+τ ) is x i With x i+τ The joint frequency, p(x i ) is x i The probability in the reconstruction matrix.

[0089] The expression of the reconstructed phase space E(d) with embedding dimension d is:

[0090]

[0091] Among them, X i (d) is the i-th reconstruction vector in the embedding dimension d, X n(i,d) (d) is the distance X i (d) The nearest vector.

[0092] Step 3: Based on the reconstructed phase space matrix, autocorrelation analysis is performed to obtain the covariance symmetric matrix.

[0093] On the basis of reconstructing the phase space matrix, autocorrelation analysis is performed on X to obtain the covariance symmetric matrix A:

[0094] A=X T X (4)

[0095] Step 4: Based on the covariance symmetric matrix, obtain the initial single-component trajectory matrix.

[0096] The fourth step includes:

[0097] Based on the covariance symmetric matrix, the Hamiltonian matrix M is constructed.

[0098]

[0099] Then N=M 2Also the Hamiltonian matrix.

[0100] Based on the Hamiltonian matrix N, construct the symplectic orthogonal matrix Q T NQ.

[0101]

[0102] Among them, R is the submatrix after matrix transformation; B is an upper triangular matrix, and the eigenvalues ​​are λ1,λ2,λ3,…,λ d ; B T is the transposed matrix of B.

[0103]

[0104] Among them, σ i The distribution of is the symplectic geometric spectrum of A, i = 1, 2,…, d.

[0105] Based on the symplectic orthogonal matrix, calculate the transformation coefficient matrix S i .

[0106]

[0107] Based on the transformation coefficient matrix S i , and obtain the initial single-component trajectory matrix.

[0108] For S i Transform to obtain a single component Z i :

[0109] Z i =Q i S i (9)

[0110] Then the initial single-component trajectory matrix Z is:

[0111] Z=Z1+Z2+…+Z d (10)

[0112] Among them, Z∈R m×d ;d=min(m,d);n=m+(d-1)τ.

[0113] Step 5: Based on the initial single-component trajectory matrix, obtain the electric quantity IMF component.

[0114] make:

[0115]

[0116] The matrix Z is an m×d matrix, so it is necessary to reorder the pure signal matrix Z, and the reconstructed time series is as follows:

[0117]

[0118] Among them, m * = max(m, d). After this diagonal averaging, the matrix Z is converted into the sequence Y = [y1, y2, …, y n ]. Thus, the original signal of the initial time-sharing electricity data can be decomposed into components with different trends and frequency bands, such as Figure 5 shown.

[0119] Step 2022, determining the correlation between the electric quantity IMF component and the initial time-sharing electric quantity data.

[0120] Exemplarily, the correlation is determined by calculating the Pearson coefficient between each electricity IMF component and the initial time-sharing electricity data set.

[0121] Step 2023: Based on the correlation, the electric quantity IMF component is divided into a pure noise component, a noisy electric quantity component and a pure electric quantity component.

[0122] Step 2024, performing improved wavelet threshold denoising on the noisy electrical quantity component to obtain a denoised noisy electrical quantity component.

[0123] For example, the improved wavelet threshold method is used to obtain the wavelet coefficient with index number i in the layer k of the noise component after denoising. It is expressed as:

[0124]

[0125] Among them, j is the scale, k is the number of layers, w is the frequency, and w j,k is the wavelet coefficient at scale j and layer k in the denoised noisy component; i represents the index number of the wavelet coefficient at a specific scale j and layer k; sgn() is the sign function; T i is the threshold of the wavelet coefficient with index number i.

[0126] The above improved threshold filtering method can not only retain detail information, but also achieve good filtering effect according to the adaptive change of wavelet decomposition scale.

[0127] Step 2025, synthesize the denoised noisy electricity component and the pure electricity component to obtain denoised time-sharing electricity data.

[0128] Exemplarily, while obtaining the denoised time-sharing electricity data, the initial meteorological data is also denoised to obtain a multi-dimensional meteorological sequence.

[0129] In this embodiment, the complex data set is decomposed into a series of independent modal components IMF through SGMD, thereby achieving data dimensionality reduction and compression. The denoised time-sharing electricity data and multidimensional meteorological series will be closer to the actual situation, reducing the impact of noise on the data analysis results. The denoised time-sharing electricity data and meteorological data are fused and analyzed to explore the complex relationship between power load and meteorological factors, providing a scientific basis for the optimized operation of the power system.

[0130] In one embodiment, see Figure 4 , the specific implementation method of performing multivariate variational mode decomposition on the denoised time-sharing electricity data and multidimensional meteorological series is described in detail, step 203 includes:

[0131] The denoised time-sharing electricity data and multidimensional meteorological series are analyzed in time and frequency synchronization through multivariate variational mode decomposition to obtain the electricity components and multidimensional meteorological series components at each frequency.

[0132] Exemplarily, a multi-constrained variational optimization problem is constructed, the goal is to minimize the sum of the bandwidths of the multivariate intrinsic mode function set, then the objective function of the multivariate variational mode decomposition is expressed as:

[0133]

[0134] Among them, {u k,c} is the set of modal functions of the denoised time-sharing electricity data and multidimensional meteorological series; {ω k} is {u k,c} center frequency set; u k,c is the kth IMF of the cth channel; C is the number of channels, K is the number of IMFs; ω k for The center frequency of is the partial derivative operation with respect to time t; The Hilbert transform operator is used to transform the vector signal u k,c The analytical representation of (t); x c (t) Input data of the cth channel, where C is the number of channels.

[0135] Construct an augmented Lagrangian function to remove the constraints of the multivariate variational optimization problem. The augmented Lagrangian function contains two penalty terms, of which the quadratic term To ensure the reconstruction accuracy, the Lagrange multiplier term To ensure that the constraints are strictly met, the formula is as follows:

[0136]

[0137] The variational optimization problem in equation (15) is solved by the alternating direction method of multipliers (ADMM). The advantage of ADMM is that it can transform the entire complex optimization problem into multiple sub-optimization problems, thereby reducing the difficulty of solving. The update equation of is shown in equation (16), the update equation of the center frequency is shown in equation (17), and the Lagrange multiplier is The update equation of is shown in formula (18).

[0138]

[0139]

[0140]

[0141] in, are the Fourier transforms of x(t), λ(t), and u(t) respectively.

[0142] Finally, after multiple iterations, the MVMD decomposition result of the multivariate original signal is output, such as Figure 6 Multivariate variational mode decomposition Multivariate component frequency diagram and Figure 7 The power consumption components of the quadratic multivariate mode decomposition are shown in the figure.

[0143] The electrical quantity component at the same frequency is combined with the multidimensional meteorological sequence component to form the same-frequency multivariate component.

[0144] In this embodiment, by minimizing the sum of the bandwidths of the modal functions, different frequency components in the signal are more accurately separated, which helps to more accurately identify the key frequency components in the power data and meteorological series, thereby improving the accuracy of time-frequency analysis. The power components and multidimensional meteorological series components at each frequency obtained by MVMD decomposition can reflect the interaction between power changes and meteorological factors, providing a new perspective for in-depth understanding of the operating laws of the power system. Using the ADMM algorithm to solve the MVMD optimization problem can decompose the complex optimization problem into multiple sub-problems for parallel solution, thereby significantly improving the computational efficiency.

[0145] In one embodiment, see Figure 4 and Figure 8 , in step 204, the spatiotemporal convolutional network includes an input layer, a plurality of residual units and an output layer.

[0146] Each residual unit includes the first one-dimensional hole causal convolution layer, the first normalization layer, the first activation function, the first regularization layer, the second one-dimensional hole causal convolution layer, the second normalization layer, the second activation function and the second regularization layer. The time domain convolutional network is a special convolutional network with good applicability in the field of time series prediction. TCN does not have to process time prediction problems in chronological order, but can be processed as a whole. At the same time, the size of the receptive field can be flexibly changed in the TCN network, and the relationship between longer time series can be obtained.

[0147] Step 204 specifically includes:

[0148] The multivariate components of the same frequency are input into the spatiotemporal convolutional network, and the hyperparameters of the spatiotemporal convolutional network are simultaneously optimized using the tuna swarm optimization algorithm to obtain the optimal hyperparameters of the spatiotemporal convolutional network.

[0149] Exemplarily, the multivariate component composed of the electric quantity component and the multidimensional meteorological sequence component is input into TCN, and the average value of MAE, MRE, RMSE and TIC is used as the objective function. The three hyperparameters (number of filters, filter size and Dropout rate) in TCN are simultaneously optimized by TSO to obtain the optimal hyperparameters of the TCN model.

[0150] The same-frequency multivariate components are input into the spatiotemporal convolutional network with optimal hyperparameters to obtain the corresponding predicted values ​​of each same-frequency multivariate component.

[0151] The predicted values ​​corresponding to each multivariate component of the same frequency are added together to form the final prediction result.

[0152] In one embodiment, actual electricity consumption data of a regional power grid is used to perform simulation analysis to further describe the specific implementation methods of the present invention.

[0153] The initial electricity consumption data covers a period of 359 days from January 1, 2022 to December 31, 2022. The initial electricity consumption data is sampled at one-hour intervals as historical electricity consumption data, with a total of 24 sets of electricity consumption data per day.

[0154] First, a symplectic geometric mode decomposition is performed on the initial time-sharing power data. The results are as follows: Figure 5 As shown:

[0155] Calculate the Pearson correlation coefficient (R) between each IMF and the original sequence of initial time-sharing electricity data i), and the results are shown in Table 1. The thresholds are set to 0.1 and 0.2 according to the correlation coefficient results. The IMFs between 0 and 0.1 (IMF 5, IMF 6, IMF 7, and IMF 8) are removed as noise components, and the IMFs between 0.1 and 0.2 (IMF 3 and IMF 4) are treated as noisy components and reconstructed by wavelet threshold processing and synthesized with the noise-free components (IMF 1 and IMF 2) to achieve the denoising effect.

[0156] Table 1 Correlation coefficients between the original sequence and the component sequences

[0157]

[0158] The denoised initial time-sharing electricity data and multivariate meteorological series (temperature x1, humidity x2, precipitation x3, wind speed x4) are subjected to multivariate variational mode decomposition to achieve time-frequency synchronization analysis, and the denoised initial time-sharing electricity data and multidimensional meteorological series are decomposed into a multivariate component set. This ensures that each prediction model is established at the same center frequency, which is conducive to improving the prediction accuracy. The frequency of each multivariate component set is as follows: Figure 6 As shown in the figure, the decomposition results of electricity consumption are as follows Figure 7 As shown:

[0159] In order to improve the performance of the overall model and avoid the increased workload caused by manually setting hyperparameters, TSO is used to optimize the hyperparameters inside the TCN module. The hyperparameter values ​​are shown in Table 2.

[0160] Table 2 TSO-TCN hyperparameter optimization results

[0161]

[0162] The multivariate component set is input into the TCN model after TSO hyperparameter optimization for model training, the final component prediction value is determined, and the sum is added to form the final prediction data.

[0163] The predicted time-sharing electricity data is calibrated and the calibration index considered is R 2 , the results are shown in Table 3.

[0164] Table 3 Calibration values ​​of different decomposition prediction methods

[0165]

[0166] According to Table 3, the R 2 It is 0.9767, which proves that the designed model is effective in time-sharing power forecasting.

[0167] In summary, the time-sharing electricity forecasting method provided in the embodiment of the present application predicts the time-sharing electricity based on secondary decomposition and TSO-TCN. First, the initial time-sharing electricity curve is decomposed, and denoising is performed based on correlation analysis and improved wavelet threshold method. Then, the denoised electricity data and multidimensional meteorological series are decomposed for multivariate variational mode decomposition. The spatiotemporal convolutional network after hyperparameter optimization based on the tuna school optimization algorithm is used to predict each set of components with the same frequency separately. After the components are merged, the prediction results are evaluated to improve the accuracy of intra-day time-sharing electricity decomposition.

[0168] Among them, based on the correlation coefficient of each type of IMF component, each IMF is clustered into three categories: pure noise component, noisy power component, and pure power component. Different processing methods are used to denoise the three types of components. Then, through the secondary decomposition combining symplectic geometric mode decomposition and multivariate variational mode decomposition, a set of multivariate components with the same frequency is obtained. Each multivariate component set is brought into the prediction model for training, ensuring that each prediction model is established at the same center frequency, which is conducive to improving the prediction accuracy.

[0169] See also Fig. 9 The embodiment of the present application provides a time-sharing power prediction device, which executes the time-sharing power prediction method as in the above embodiment, including a data acquisition module 301, a first decomposition module 302, a second decomposition module 303 and a result output module 304.

[0170] The data acquisition module 301 is used to acquire initial time-sharing power data and initial meteorological data;

[0171] The first decomposition module 302 is used to perform symplectic geometric mode decomposition on the initial time-sharing power data, and perform denoising on the decomposed initial time-sharing power data and initial meteorological data to obtain denoised time-sharing power data and multidimensional meteorological series respectively;

[0172] The second decomposition module 303 is used to perform multivariate variational mode decomposition on the denoised time-sharing electricity data and the multidimensional meteorological series to obtain multivariate components with the same frequency;

[0173] The result output module 304 is used to input the same frequency multivariate components into the spatiotemporal convolutional network and obtain the final prediction result based on the tuna school optimization algorithm.

[0174] Exemplarily, in the first decomposition module 302, the initial time-sharing power data is subjected to symplectic geometric mode decomposition, and the decomposed initial time-sharing power data is subjected to denoising to obtain denoised time-sharing power data, including:

[0175] Perform symplectic geometric mode decomposition on the initial time-sharing electricity data to obtain the electricity IMF components;

[0176] Determine the correlation between the IMF component of electricity and the initial time-sharing electricity data;

[0177] Based on the correlation, the electric quantity IMF component is divided into pure noise component, noisy electric quantity component and pure electric quantity component;

[0178] Perform improved wavelet threshold denoising on the noisy electrical quantity component to obtain the denoised noisy electrical quantity component;

[0179] The denoised noisy electricity component is synthesized with the pure electricity component to obtain the denoised time-sharing electricity data.

[0180] Exemplarily, the initial time-sharing electricity data is subjected to symplectic geometric mode decomposition to obtain electricity IMF components, including:

[0181] Convert the original signal of initial time-sharing electricity data into a phase space matrix;

[0182] Reconstruct the phase space matrix based on the delay time and the embedded dimension to obtain a reconstructed phase space matrix;

[0183] Based on the reconstructed phase space matrix, autocorrelation analysis is performed to obtain the covariance symmetric matrix;

[0184] Based on the covariance symmetric matrix, the initial single-component trajectory matrix is ​​obtained;

[0185] Based on the initial single-component trajectory matrix, the electric quantity IMF component is obtained.

[0186] Exemplarily, based on the covariance symmetric matrix, an initial single-component trajectory matrix is ​​obtained, including:

[0187] Based on the covariance symmetric matrix, construct the Hamiltonian matrix;

[0188] Based on the Hamiltonian matrix, construct a symplectic orthogonal matrix;

[0189] Based on the symplectic orthogonal matrix, calculate the transformation coefficient matrix;

[0190] Based on the transformation coefficient matrix, the initial single-component trajectory matrix is ​​obtained.

[0191] Exemplarily, the frequency component of the i-th row and k-th column in the denoised noisy component It is expressed as:

[0192]

[0193] Among them, j is the scale, k is the number of layers, w is the frequency, and w j,k is the wavelet coefficient at scale j and layer k in the denoised noisy component; i represents the index number of the wavelet coefficient at a specific scale j and layer k; sgn() is the sign function; T iis the threshold of the wavelet coefficient with index number i.

[0194] Exemplarily, the second decomposition module 303 is specifically used for:

[0195] Through multivariate variational mode decomposition, the denoised time-sharing electricity data and multidimensional meteorological series are analyzed in time and frequency synchronization to obtain the electricity component and multidimensional meteorological series component at each frequency.

[0196] The electrical quantity component at the same frequency is combined with the multidimensional meteorological sequence component to form the same-frequency multivariate component.

[0197] Exemplarily, the objective function of multivariate variational mode decomposition is expressed as:

[0198]

[0199] Among them, {u k,c} is the set of modal functions of the denoised time-sharing electricity data and multidimensional meteorological series; {ω k} is {u k,c} center frequency set; u k,c is the kth IMF of the cth channel; C is the number of channels, K is the number of IMFs; ω k for The center frequency of is the partial derivative operation with respect to time; The Hilbert transform operator is used to transform the vector signal u k,c The analytical representation of (t); x c (t) Input data of the cth channel.

[0200] Exemplarily, in the result output module 304, the spatiotemporal convolutional network includes an input layer, a plurality of residual units, and an output layer;

[0201] Each residual unit includes a first one-dimensional hole causal convolution layer, a first normalization layer, a first activation function, a first regularization layer, a second one-dimensional hole causal convolution layer, a second normalization layer, a second activation function and a second regularization layer, which are connected in sequence.

[0202] Exemplarily, the same-frequency multivariate components are input into the spatiotemporal convolutional network, and the final prediction result is obtained based on the tuna school optimization algorithm, including:

[0203] The multivariate components of the same frequency are input into the spatiotemporal convolutional network, and the hyperparameters of the spatiotemporal convolutional network are optimized synchronously using the tuna school optimization algorithm to obtain the optimal hyperparameters of the spatiotemporal convolutional network.

[0204] Input the same-frequency multivariate components into the spatiotemporal convolutional network with optimal hyperparameters to obtain the corresponding predicted values ​​of each same-frequency multivariate component;

[0205] The predicted values ​​corresponding to each multivariate component of the same frequency are added together to form the final prediction result.

[0206] It should be noted that although several units / modules or sub-units / modules of the time-sharing power prediction device are mentioned in the above detailed description, this division is only exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules for embodiment.

[0207] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0208] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the time-sharing power prediction method provided in the above embodiment of the present application is implemented.

[0209] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0210] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A time-sharing electricity prediction method, characterized in that: include: Obtain initial time-sharing electricity data and initial meteorological data; Performing symplectic geometric mode decomposition on the initial time-sharing electricity data, and performing denoising on the decomposed initial time-sharing electricity data and the initial meteorological data, to obtain denoised time-sharing electricity data and a multidimensional meteorological sequence, respectively; Performing multivariate variational mode decomposition on the denoised time-sharing electricity data and multidimensional meteorological series to obtain multivariate components with the same frequency; The same-frequency multivariate components are input into the spatiotemporal convolutional network, and the final prediction result is obtained based on the tuna school optimization algorithm.

2. The time-sharing power prediction method according to claim 1, characterized in that: The initial time-sharing power data is subjected to symplectic geometric mode decomposition, and the decomposed initial time-sharing power data is subjected to denoising to obtain denoised time-sharing power data, including: Perform symplectic geometric mode decomposition on the initial time-sharing electricity data to obtain the electricity IMF components; Determine the correlation between the electric quantity IMF component and the initial time-sharing electric quantity data; Based on the correlation, the electric quantity IMF component is divided into a pure noise component, a noisy electric quantity component and a pure electric quantity component; Performing improved wavelet threshold denoising on the noisy electrical quantity component to obtain a denoised noisy electrical quantity component; The denoised noisy electric quantity component is synthesized with the pure electric quantity component to obtain denoised time-sharing electric quantity data.

3. The time-sharing electricity prediction method according to claim 2, characterized in that: The symplectic geometric mode decomposition is performed on the initial time-sharing electricity data to obtain the electricity IMF components, including: Convert the original signal of initial time-sharing electricity data into a phase space matrix; Reconstruct the phase space matrix based on the delay time and the embedding dimension to obtain a reconstructed phase space matrix; Based on the reconstructed phase space matrix, autocorrelation analysis is performed to obtain a covariance symmetric matrix; Based on the covariance symmetric matrix, an initial single-component trajectory matrix is ​​obtained; Based on the initial single-component trajectory matrix, the electric quantity IMF component is obtained.

4. The time-sharing electricity prediction method according to claim 3, characterized in that: The initial single-component trajectory matrix is ​​obtained based on the covariance symmetric matrix, including: Based on the covariance symmetric matrix, construct a Hamiltonian matrix; Based on the Hamiltonian matrix, construct a symplectic orthogonal matrix; Based on the symplectic orthogonal matrix, calculating a transformation coefficient matrix; Based on the transformation coefficient matrix, the initial single-component trajectory matrix is ​​obtained.

5. The time-sharing electricity prediction method according to claim 2, characterized in that: The frequency component of the i-th row and k-th column in the noise-containing component after denoising It is expressed as: Among them, j is the scale, k is the number of layers, w is the frequency, and w j,k is the wavelet coefficient at scale j and layer k in the denoised noisy component; i represents the index number of the wavelet coefficient at a specific scale j and layer k; sgn() is the sign function; T i is the threshold of the wavelet coefficient with index number i.

6. The time-sharing electricity prediction method according to claim 1, characterized in that: The multivariate variational mode decomposition is performed on the denoised time-sharing electricity data and the multidimensional meteorological series to obtain the same-frequency multivariate components, including: Performing time-frequency synchronization analysis on the denoised time-sharing electricity data and the multidimensional meteorological series through multivariate variational mode decomposition to obtain electricity components and multidimensional meteorological series components at each frequency; The electrical quantity component at the same frequency is combined with the multidimensional meteorological sequence component to form the same-frequency multivariate component.

7. The time-sharing electricity prediction method according to claim 6, characterized in that: The objective function of the multivariate variational mode decomposition is expressed as: Among them, {u k,c } is the set of modal functions of the denoised time-sharing electricity data and the multidimensional meteorological sequence; {ω k } is {u k,c } center frequency set; u k,c is the kth IMF of the cth channel; C is the number of channels, K is the number of IMFs; ω k for The center frequency of is the partial derivative operation with respect to time; The Hilbert transform operator is used to transform the vector signal u k,c The analytical representation of (t); x c (t) Input data of the cth channel.

8. The time-sharing electricity prediction method according to claim 1, characterized in that: The spatiotemporal convolutional network includes an input layer, a plurality of residual units and an output layer; Each of the residual units includes a first one-dimensional hole causal convolution layer, a first normalization layer, a first activation function, a first regularization layer, a second one-dimensional hole causal convolution layer, a second normalization layer, a second activation function and a second regularization layer connected in sequence.

9. The time-sharing electricity prediction method according to claim 1, characterized in that: The step of inputting the same-frequency multivariate components into a spatiotemporal convolutional network and obtaining a final prediction result based on a tuna school optimization algorithm includes: Inputting the same-frequency multivariate components into a spatiotemporal convolutional network, and using a tuna school optimization algorithm to synchronously optimize the hyperparameters of the spatiotemporal convolutional network to obtain the optimal hyperparameters of the spatiotemporal convolutional network; Inputting the same-frequency multivariate components into a spatiotemporal convolutional network with the optimal hyperparameters to obtain prediction values ​​corresponding to each same-frequency multivariate component; The predicted values ​​corresponding to each multivariate component of the same frequency are added together to form the final prediction result.

10. A time-sharing electricity prediction device, characterized in that: Applying the time-sharing power forecasting method according to any one of claims 1 to 9 comprises: A data acquisition module is used to acquire initial time-sharing electricity data and initial meteorological data; The first decomposition module is used to perform symplectic geometric mode decomposition on the initial time-sharing electricity data, and perform denoising on the decomposed initial time-sharing electricity data and the initial meteorological data to obtain denoised time-sharing electricity data and multidimensional meteorological series respectively; The second decomposition module is used to perform multivariate variational mode decomposition on the denoised time-sharing electricity data and the multidimensional meteorological series to obtain multivariate components with the same frequency; The result output module is used to input the same-frequency multivariate components into the spatiotemporal convolutional network and obtain the final prediction result based on the tuna school optimization algorithm.

Images