A Regional New Energy Output and Load Prediction Method Based on Source-Load Co-mode

Abstract

This invention discloses a regional renewable energy output and load forecasting method based on source-load co-mode, comprising the following steps: Step 1) Collecting historical sequences of source and load and meteorological data of the same period; Step 2) Obtaining multimodal feature sequences of electricity load and renewable energy output sequences at different frequencies through variational mode decomposition of the original sequences; Step 3) Fusing the multimodal feature sequences, meteorological feature sequences, and time-coded feature sequences to obtain a multidimensional feature fusion sequence group; Step 4) Constructing a model containing a pyramid attention module based on a multi-branch tree structure; Step 5) Collecting the features given by the last node of all scales in the pyramid attention module, mapping them to all future time steps in the batch, and obtaining the final prediction through a fully connected layer across channel dimensions. Beneficial effects: Improves the co-mode effect under the influence of multiple factors such as supply and demand sides in complex coupling relationships.

Description

Technical Field

[0001] This invention relates to the fields of new energy output and load forecasting, and in particular to a regional new energy output and load forecasting method based on source-load co-mode. Background Technology

[0002] Existing source-load forecasting studies primarily model the renewable energy generation side and the load demand side separately. In terms of forecast type, the source side mainly forecasts wind turbine output and photovoltaic output; in terms of time scale, it can be divided into ultra-short-term, short-term, and medium-to-long-term forecasts; in terms of spatial scale, it can be divided into regional-level forecasts and station-level forecasts. Renewable energy output forecasts at different temporal and spatial scales differ in research methods and applications depending on accuracy requirements. The load side mainly forecasts electricity load demand, which can be further subdivided into regional-level total electricity consumption forecasts, regional-level annual peak load forecasts, regional-level long-cycle time-series load curve forecasts, and industry-specific electricity load forecasts. Traditionally, one approach involves acquiring meteorological data from numerical weather prediction (NWP) systems, using the Pearson correlation coefficient method or other correlation calculation methods to obtain meteorological characteristics strongly correlated with renewable energy output and regional load. These are then combined with historical renewable energy and load characteristics to independently build models for predicting renewable energy output and regional load. Another approach uses time series analysis, employing methods such as Empirical Mode Decomposition (EMD), Singular Spectral Analysis (SSA), STL decomposition, factor decomposition, and SEATS decomposition to decompose the original time series into long-term trends, seasonal trends, periodic fluctuations, and random noise. This allows for analysis of renewable energy output and load characteristics, followed by the use of probabilistic statistical methods, machine learning methods, and deep reinforcement learning methods to build predictive models. However, these analytical methods and approaches still present the following needs and challenges in achieving accurate regional power load and renewable energy output forecasting:

[0003] (1) The influencing factors are complex. The source load data itself is highly volatile and random, and is affected by many factors such as geographical location, seasonal changes, weather, social development, and policy planning, making it difficult to consider all aspects.

[0004] (2) The independent models are detached from each other in time and space. The wind power output prediction model, photovoltaic power output prediction model and load prediction model are constructed separately, but they fail to make full use of the spatiotemporal correlation information and fail to consider the joint effects of meteorological changes and different modes of supply and demand.

[0005] (3) The model has weak long-term learning ability. It lacks an adaptive feature learning method and manually combines features. Due to the differences between modalities, it cannot make good reference to different dimensions and more hidden information during long-term prediction, which affects the model's prediction ability. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for predicting regional renewable energy output and load based on source-load co-mode, specifically implemented by the following technical solution:

[0007] The method for predicting the output and load of new energy sources in the region includes the following steps:

[0008] Step 1) Collect historical sequence data and concurrent meteorological data from the source and load ends.

[0009] Step 2) Obtain the characteristic sequences of the multimode of electricity load and new energy output at different frequencies by using the original sequence of variational mode decomposition;

[0010] Step 3) Decompose the electricity load and renewable energy output sequences using different time scales to obtain feature sequences with different period lengths in the time dimension; calculate the correlation coefficient between the meteorological data of the corresponding region and the corresponding load and renewable energy output sequences using the Pearson correlation coefficient method, select meteorological features with larger correlation coefficients, and combine them into a meteorological feature sequence; perform one-hot encoding on the time dimension time labels to obtain a time-coded feature sequence; fuse the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence to obtain a multidimensional feature fusion sequence group;

[0011] Step 4) Construct a model containing a pyramid attention module based on a multi-branch tree structure, and optimize the model based on the set rules; the pyramid attention module constructs a multi-resolution representation of the original sequence at different time scales, and the C-ary tree with a pyramid structure is composed of nodes at different time scales;

[0012] Step 5) Collect the features given by the last node of all scales in the pyramid attention module, map them to all future time steps in the batch, obtain the final prediction through a fully connected layer across channel dimensions, and output all future prediction results linearly together.

[0013] A further design of the regional new energy output and load forecasting method is that the model further includes: a data embedding module, which embeds the acquired information including observation data, covariate data and location, and then adds up the parameters in the information to form a new feature representation;

[0014] The multi-time-resolution C-ary tree module uses a coarse scale to construct a multi-resolution C-ary tree, and completes the information transfer between different scales through nodes.

[0015] A further design of the regional renewable energy output and load forecasting method is that step 2) of variational mode decomposition specifically includes the following steps:

[0016] Step 2-1) Initialize the modal function set Center frequency set Lagrange operator λ 1 Initial iteration count b = 0;

[0017] Step 2-2) Set the iteration count b = b + 1;

[0018] Steps 2-3) When w≥0 is satisfied, update the spectrum of each mode.

[0019]

[0020] Where represents the penalty coefficient, which determines the bandwidth of the component mode function; w represents the center frequency, w k This represents the center frequency of the k-th mode; Represents the Lagrange operator of the signal to be decomposed; The frequency domain representation of the signal to be decomposed represents the update of the center frequency w. k :

[0021]

[0022] Steps 2-4) Update the Lagrange multiplier λ:

[0023]

[0024] Where ε is the noise tolerance parameter;

[0025] Step 2-5) Repeat steps 2-2) to 2-4) above until the termination condition is met:

[0026]

[0027] Where σ is the discrimination precision and is greater than 0.

[0028] A further design of the regional new energy output and load forecasting method is that the coarser-scale nodes in the multi-resolution C-tree summarize the corresponding finer-scale C-node information.

[0029] A further design of the regional renewable energy output and load prediction method is that the edges of the pyramid attention module can be divided into inter-scale connections and intra-scale connections, where the finest-scale nodes correspond to the smallest time point of the time series, and the coarser-scale nodes represent features with larger resolution.

[0030] A further design of the regional renewable energy output and load forecasting method involves optimizing the model based on predefined rules, specifically as follows:

[0031] Let s = 1, 2, ..., S represent the sequence of the finest scale at the bottom to the coarse scale at the top of the pyramid attention module. This represents the l-th node at scale s; each node can focus on a set of neighboring nodes at three scales. Adjacent A nodes of the same scale, including the node itself, are represented as follows: C child nodes are represented in a C-ary tree as follows: And its parent node in the C-ary tree is represented as Then we have:

[0032]

[0033] node Attention at y i The expression for calculation is:

[0034]

[0035] Where L represents the sequence length and A represents the number of adjacent nodes of the same scale; d represents the attention layer number of the l-th node at scale s; K This represents the length of the query vector, Q = XW Q q i ∈Q, K=XW K k i ∈K, V=XW V v l ∈V, X is the original input sample sequence, W Q W K W V It is a weight matrix, where Q, K, and V represent the query vector, key vector, and value vector in the attention mechanism, respectively. v l ∈V;

[0036] During training, attention information is expressed and optimized at different time scales of the input sequence to obtain key feature information and improve the model's ability to fit nonlinear sequences at any time step.

[0037] A further design of the regional new energy output and load forecasting method is that, in step 3), the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence are fused using a random selection method or the maximum entropy principle to obtain a multidimensional feature fusion sequence group.

[0038] The present invention also provides an electronic device, including a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the aforementioned method for predicting regional renewable energy output and load based on source-load co-mode.

[0039] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned method for predicting regional renewable energy output and load based on source-load co-mode.

[0040] The advantages of this invention are as follows:

[0041] The regional renewable energy output and load forecasting method of the present invention combines renewable energy output, electricity load data, meteorological data and other elements to form a model, which enables the model to make full use of the spatiotemporal correlation information of the sequence and improve the common mode effect of complex coupling relationship under the influence of multiple factors such as supply and demand.

[0042] This method improves the adaptive feature learning capability by using variational mode decomposition to decompose random fluctuation sequences into multiple more regular modes in the frequency domain.

[0043] This method uses a pyramidal C-ary tree structure for feature representation, which improves the model's learning efficiency and reduces time complexity in both time and space dimensions.

[0044] This method, through end-to-end model training, can learn the sequence input and output at any time step, improving the model's long-term learning ability. It also considers the data dependencies between the source and payload ends, better references different dimensions and more hidden information, and strengthens the model's predictive ability. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the regional renewable energy output and load prediction method based on source-load co-mode of the present invention. Detailed Implementation

[0046] The technical solution of the present invention will be described in detail below.

[0047] like Figure 1 The regional renewable energy output and load forecasting method based on source-load co-mode in this embodiment includes the following steps:

[0048] Step 1) Collect historical sequence data of the source load and meteorological data of the same period.

[0049] Step 2) By using the original sequence of variational mode decomposition, the characteristic sequences of the electricity load and new energy output sequence at different frequencies are obtained.

[0050] Step 3) Decompose the electricity load and renewable energy output sequences using different time scales to obtain feature sequences with different period lengths in the time dimension; calculate the correlation coefficient between the meteorological data of the corresponding region and the corresponding load sequence and renewable energy output sequence using the Pearson correlation coefficient method, select meteorological features with larger correlation coefficients, and combine them into a meteorological feature sequence; perform one-hot encoding on the time dimension time label to obtain a time-coded feature sequence; fuse the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence to obtain a multidimensional feature fusion sequence group.

[0051] Step 4) Construct a model containing a pyramid attention module based on a multi-branch tree structure, and optimize the model based on the set rules; the pyramid attention module constructs a multi-resolution representation of the original sequence at different time scales, and the C-ary tree with a pyramid structure is composed of nodes at different time scales.

[0052] Step 5) Collect the features given by the last node of all scales in the pyramid attention module, map them to all future time steps in the batch, obtain the final prediction through a fully connected layer across channel dimensions, and output all future prediction results linearly together.

[0053] The model in this embodiment also includes a data embedding module and a multi-temporal resolution C-ary tree module. The data embedding module embeds the acquired information, including observation data, covariate data, and location, and then adds the parameters from the information together to form a new feature representation. The multi-temporal resolution C-ary tree module (CSCM module) constructs a multi-resolution C-ary tree using a coarse scale, and completes information transfer between different scales through nodes. This model introduces a pyramid attention module (PAM module), enabling the inter-scale tree structure to summarize features at different resolutions, and the intra-scale adjacent connections to model the temporal dependencies across different ranges.

[0054] Step 2) involves Variational Mode Decomposition (VMD), a signal decomposition method that estimates individual signal components by solving a frequency domain variational optimization problem. This decomposes the original sequence into a specified number of mode components in the frequency domain. Since the collected new energy output and electricity load sequences contain fluctuating signals at different frequencies, decomposition and combination make it easier for the model to capture the joint changes in the sequence's macroscopic trends and microscopic abrupt changes. VMD assumes that all components are narrowband signals concentrated near their respective center frequencies. Therefore, VMD establishes a constrained optimization problem based on the narrowband conditions of the components to estimate the center frequencies of the signal components and reconstruct the corresponding components. The decomposition process of VMD is essentially the solution process of a variational problem. The algorithm mainly includes the construction and solution of the variational problem.

[0055] First, intrinsic mode components (IMFs) are defined. Then, a variational problem is constructed, requiring that the sum of the bandwidths at the center frequencies of each mode component be minimized, and the sum of all mode components equal to the original signal. For each mode function, the one-sided spectrum is obtained through Hilbert transform, and then multiplied by the estimated center frequency to shift the one-sided spectrum of each mode to the estimated center frequency. The bandwidth of each mode signal is then calculated using the square norm of the gradient. Finally, by introducing the Lagrange operator and a quadratic penalty factor, the inequality constraints are transformed into equality constraints for solution. Specifically, the steps include:

[0056] Step 2-1) Initialize the modal function set Center frequency set Lagrange operator λ 1 Initial iteration count b = 0;

[0057] Step 2-2) Set the iteration count b = b + 1;

[0058] Steps 2-3) When w≥0 is satisfied, update the spectrum of each mode.

[0059]

[0060] Update center frequency w k :

[0061]

[0062] Steps 2-4) Update the Lagrange multiplier λ:

[0063]

[0064] Step 2-5) Repeat steps 2-2) to 2-4) above until the termination condition is met:

[0065]

[0066] Where σ is the discrimination precision and is greater than 0.

[0067] A further design of the regional new energy output and load forecasting method is that the coarser-scale nodes in the multi-resolution C-tree summarize the corresponding finer-scale C-node information.

[0068] A further design of the regional renewable energy output and load prediction method is that the edges of the pyramid attention module can be divided into inter-scale connections and intra-scale connections, where the finest-scale nodes correspond to the smallest time point of the time series, and the coarser-scale nodes represent features with larger resolution.

[0069] A further design of the regional renewable energy output and load forecasting method involves optimizing the model based on predefined rules, specifically as follows:

[0070] Let s = 1, 2, ..., S represent the finest scale at the bottom and the coarsest scale at the top of the pyramid attention module, respectively. This represents the l-th node at scale s; each node can focus on a set of neighboring nodes at three scales. Adjacent A nodes of the same scale, including the node itself, are represented as follows: C child nodes are represented in a C-ary tree as follows: And its parent node in the C-ary tree is represented as Then we have:

[0071]

[0072] point Attention at y i The expression for calculation is:

[0073]

[0074] N represents the number of attention layers.

[0075] Virtual Mode Decomposition (VMD) is a non-recursive, adaptive signal decomposition method that can decompose complex, non-stationary, and nonlinear signals, effectively suppressing mode aliasing. In this embodiment, the IMFs decomposed by the VMD method all have independent center frequencies and exhibit sparsity in the frequency domain.

[0076] In step 3), the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence are fused using a random selection method or the maximum entropy principle to obtain a multidimensional feature fusion sequence group.

[0077] This embodiment also provides an electronic device, including a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the regional renewable energy output and load prediction method based on source-load co-mode.

[0078] This embodiment also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the regional renewable energy output and load prediction method based on source-load co-mode.

[0079] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting regional renewable energy output and load based on source-load co-mode, characterized in that... Includes the following steps: Step 1) Collect historical sequence data and concurrent meteorological data from the source and load ends. Step 2) Obtain the characteristic sequences of the multimode of electricity load and new energy output at different frequencies by using the original sequence of variational mode decomposition; Step 3) Decompose the electricity load and renewable energy output sequences using different time scales to obtain feature sequences with different period lengths in the time dimension; calculate the correlation coefficient between the meteorological data of the corresponding region and the corresponding load and renewable energy output sequences using the Pearson correlation coefficient method, select meteorological features with larger correlation coefficients, and combine them into a meteorological feature sequence; perform one-hot encoding on the time labels in the time dimension to obtain a time-coded feature sequence; fuse the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence to obtain a multidimensional feature fusion sequence group; Step 4) Construct a model containing a pyramid attention module based on a multi-branch tree structure, and optimize the model based on the set rules; the pyramid attention module constructs a multi-resolution representation of the original sequence at different time scales, and the C-ary tree with a pyramid structure is composed of nodes at different time scales; Step 5) Collect the features given by the last node of all scales in the pyramid attention module, map them to all future time steps in the batch, obtain the final prediction through a fully connected layer across channel dimensions, and output all future prediction results linearly together. In step 3), the multimodal feature sequence, meteorological feature sequence, and time-coded feature sequence are fused using a random selection method or the maximum entropy principle to obtain a multidimensional feature fusion sequence group.

2. The regional renewable energy output and load forecasting method according to claim 1, characterized in that... The model also includes: The data embedding module constructs a main input variable containing observation data, covariate data, and location information through mode decomposition and multi-feature selection based on importance. The main input variable is embedded, and a tree-based attention mechanism is constructed to combine the various parameters in the information to form a new high-dimensional feature representation. The multi-time-resolution C-ary tree module uses a coarse scale to construct a multi-resolution C-ary tree, and completes the information transfer between different scales through nodes.

3. The regional renewable energy output and load forecasting method according to claim 1, characterized in that... Step 2) of variational mode decomposition specifically includes the following steps: Step 2-1) Initialize the modal function set Center frequency set Lagrange operator λ 1 Initial iteration count b = 0; Step 2-2) Set the iteration count b = b + 1; Steps 2-3) When w≥0 is satisfied, update the spectrum of each mode. Where α represents the penalty coefficient, which determines the bandwidth of the component mode function; w represents the center frequency, w k This represents the center frequency of the k-th mode; Represents the Lagrange operator of the signal to be decomposed; The frequency domain representation of the signal to be decomposed; Update center frequency w k : Steps 2-4) Update the Lagrange multiplier λ: Where ε is the noise tolerance parameter; Step 2-5) Repeat steps 2-2) to 2-4) above until the termination condition is met: Where σ is the discrimination precision and is greater than 0.

4. The regional renewable energy output and load forecasting method according to claim 2, characterized in that... In the multi-resolution C-ary tree, the coarser-scale nodes summarize the corresponding finer-scale node information.

5. The regional renewable energy output and load forecasting method according to claim 2, characterized in that... The edges of the pyramid attention module can be divided into inter-scale connections and intra-scale connections. The finest-scale nodes correspond to the smallest time point in the time series, while the coarser-scale nodes represent features with larger resolution.

6. The regional renewable energy output and load forecasting method according to claim 2, characterized in that... The optimization of the model based on the set rules is specifically as follows: Let s = 1, 2, ..., S represent the sequence of the finest scale at the bottom to the coarse scale at the top of the pyramid attention module. This represents the l-th node at scale s; each node can focus on a set of neighboring nodes at three scales. Adjacent A nodes of the same scale, including the node itself, are represented as follows: C child nodes are represented in a C-ary tree as follows: Its parent node in the C-ary tree is represented as P. l (s) Then we have: node Attention at y i The expression for calculation is: Where L represents the sequence length, and A represents the number of adjacent nodes of the same scale; d K This represents the length of the query vector, Q = XW Q ,q i ∈Q, K=XW K ,k i ∈K, V=XW V ,v l ∈V, X is the original input sample sequence, W Q W K W V It is a weight matrix, where Q, K, and V represent the query vector, key vector, and value vector in the attention mechanism, respectively. i ∈Q、 v l ∈V; During training, attention information is expressed and optimized at different time scales of the input sequence to obtain key feature information and improve the model's ability to fit nonlinear sequences at any time step.

7. An electronic device, characterized in that, It includes a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the regional renewable energy output and load prediction method based on source-load co-mode as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that... The storage medium stores a computer program, which is executed by a processor to implement the regional renewable energy output and load prediction method based on source-load co-mode as described in any one of claims 1-6.

Images