Hybrid deep learning high fluctuation load prediction method and system based on secondary decomposition aggregation
Through a hybrid deep learning method based on quadratic decomposition and aggregation, combined with adaptive noise-complete ensemble empirical modal decomposition, sample entropy and K-mean reconstruction, variational modal decomposition and convolutional neural network and bidirectional long and short-term memory network, the problem of poor prediction accuracy in dynamic loads under high randomness and big data background is solved, and higher load prediction accuracy and robustness are achieved.
Patent Information
- Application Number
- CN202411330207.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2025-02-25
AI Technical Summary
When traditional load prediction methods face dynamic loads under the background of high randomness and big data, the prediction accuracy is poor.
Using a hybrid deep learning method based on quadratic decomposition and aggregation, a combination of CNN-BiLSTM hybrid deep learning prediction model is constructed through adaptive noise-complete ensemble empirical modal decomposition, sample entropy and K-mean reconstruction, variational modal decomposition and convolutional neural network and bidirectional long and short-term memory network.
It improves the accuracy of load prediction, can better capture complex features and laws in load data, and enhances the robustness and coping ability to high fluctuate loads.
Smart Images
Figure CN119518680A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of load forecasting, and in particular to a hybrid deep learning high-fluctuation load forecasting method and system based on secondary decomposition and aggregation. Background Art
[0002] In the energy sector, accurate load forecasting is crucial for energy management and optimization. With economic development and population growth, energy demand continues to increase, while energy resources are becoming increasingly scarce, making energy management and optimization an important issue in today's world. Traditional load forecasting methods are difficult to guarantee prediction accuracy with simple calculation models when faced with dynamic loads with high randomness and big data. In recent years, computing tools have been continuously upgraded, and the scale of training data has also increased significantly. Deep learning technology has been widely used in the field of load forecasting due to its powerful learning ability and automation advantages.
[0003] At present, there is an urgent need for a load forecasting method that can better ensure the prediction accuracy when facing dynamic loads with high randomness and big data background. Summary of the invention
[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.
[0005] In view of the above existing problems, the present invention is proposed.
[0006] Therefore, the present invention provides a hybrid deep learning high-volatility load forecasting method and system based on secondary decomposition and aggregation to solve the problem of poor prediction accuracy of traditional load forecasting methods when facing dynamic loads with high randomness and big data background.
[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0008] In a first aspect, the present invention provides a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation, comprising:
[0009] The original load sequence is obtained, and the original load sequence is decomposed once using the adaptive noise complete set empirical mode decomposition to obtain several inherent mode components;
[0010] Based on the intrinsic modal components, sample entropy and K-means are introduced to reconstruct the subsequences and filter out modal components with similar characteristics.
[0011] Based on the results of sample entropy reconstruction, variational mode decomposition is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence to obtain the decomposed subsequences;
[0012] Perform data preprocessing on the decomposed subsequences to obtain processed sequences;
[0013] Based on convolutional neural network and bidirectional long short-term memory network, a CNN-BiLSTM hybrid deep learning prediction model is constructed;
[0014] The processed sequences are input into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and the prediction results of each subsequence are superimposed to obtain the prediction value of the original high-volatility load sequence.
[0015] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0016] The adaptive noise complete set empirical mode decomposition is used to decompose the original load sequence once, including decomposing the original load sequence to obtain k subsequences, and the k subsequences include k-1 inherent mode components and 1 residual component.
[0017] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0018] The introduction of sample entropy and K-means to reconstruct the subsequence includes the following steps:
[0019] Calculate the sample entropy value for each modal component and residual component;
[0020] The calculated entropy values of k samples are used as input and divided into three categories using the K-means clustering algorithm;
[0021] According to the clustering results, the corresponding k subsequences are recombined into three new sequences, named Co-IMF0, Co-IMF1 and Co-IMF2.
[0022] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0023] The method of using variational mode decomposition to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence includes the following steps:
[0024] identifying components classified as high frequency sequences in the reconstruction of the sample entropy sequence;
[0025] By performing a preliminary empirical mode decomposition analysis on the high-frequency sequence, the number of target modes b of VMD decomposition is estimated using empirical mode decomposition;
[0026] According to the determined target modal number b, the VMD algorithm is applied to decompose the high-frequency sequence to obtain b subsequences.
[0027] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0028] The VMD algorithm comprises the following steps:
[0029] Assume that the original signal f is decomposed into k components;
[0030] Assume the constraint that the sum of all modes is equal to the original signal;
[0031] Introducing Lagrangian function, the problem is transformed into an unconstrained variational problem;
[0032] The number of modal decompositions is determined by iteratively searching for the optimal solution of the variational model.
[0033] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0034] The decomposed subsequences are subjected to data preprocessing, including normalizing the decomposed subsequences. The normalization is expressed as:
[0035]
[0036] Among them, x′,x,x min ,x max They represent the normalized data, the real data, and the minimum and maximum values of the real data in the test set, respectively.
[0037] As a preferred solution of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation described in the present invention, wherein:
[0038] The CNN-BiLSTM hybrid deep learning prediction model, in which the state update and output of each BiLSTM layer are expressed as:
[0039] A i =f 1 (ω 1 X i +ω 2 A i-1 )
[0040] B i =f 2 (ω3 X i +ω 5 B i+1 )
[0041] Y i =f 3 (ω 4 A i +ω 6 B i )
[0042] Among them, X 1 ,X 2 ,X 3 ,...,X i ,...,X t Respectively represent the model at t 1 ,t 2 ,t 3 ,...,t t The input data corresponding to the moment, A 1 ,A 2 ,A 3 ,...,A i ,...,A t Represent the hidden layer state of the forward BiLSTM, B 1 ,B 2 ,B 3 ,...,B i ,...,B t Respectively represent the hidden layer state of the backward BiLSTM, Y 1 ,Y 2 ,Y 3 ,...,Y i ,...,Y t Respectively represent the model at t 1 ,t 2 ,t 3 ,...,t t The output data corresponding to the moment, ω 1 ,ω 2 ,ω 3 ,...,ω t Respectively represent the corresponding weight values of each layer in the model, f 1 ,f 2 ,f 3 They represent the activation functions between different hidden layers.
[0043] In a second aspect, the present invention provides a hybrid deep learning high-fluctuation load forecasting system based on secondary decomposition and aggregation, comprising:
[0044] The primary decomposition module is used to obtain the original load sequence, and the adaptive noise complete set empirical mode decomposition is used to decompose the original load sequence to obtain several intrinsic mode components;
[0045] The reconstruction module is used to reconstruct the subsequence based on the intrinsic modal components, introduce sample entropy and K-means, and screen out modal components with similar characteristics;
[0046] The secondary decomposition module is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence based on the result of sample entropy reconstruction by using variational mode decomposition to obtain the decomposed subsequence;
[0047] A preprocessing module is used to perform data preprocessing on the decomposed subsequences to obtain processed sequences;
[0048] Build a model module to build a CNN-BiLSTM hybrid deep learning prediction model based on convolutional neural network and bidirectional long short-term memory network;
[0049] The input module is used to input the processed sequence into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and superimpose the prediction results of each subsequence to obtain the prediction value of the original high-volatility load sequence.
[0050] In a third aspect, the present invention provides a computing device, comprising:
[0051] Memory, used to store programs;
[0052] A processor is used to execute the computer executable instructions, which, when executed by the processor, implement the steps of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation.
[0053] In a fourth aspect, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation are implemented.
[0054] Beneficial effects of the present invention: The present invention integrates a variety of deep learning algorithms, and through secondary decomposition and aggregation, it can better capture the complex characteristics and laws in the load data, help predict future load conditions, and achieve more accurate load forecasting; at the same time, it uses reinforcement learning and other methods to achieve real-time intelligent scheduling to improve energy utilization, achieve energy conservation and environmental protection, and has better robustness to various external influencing factors, and can effectively deal with the uncertainty and variability of highly fluctuating loads. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:
[0056] Figure 1 A basic flow chart of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation provided by an embodiment of the present invention;
[0057] Figure 2 A schematic diagram of a model of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation provided by an embodiment of the present invention;
[0058] Figure 3 A diagram of the primary decomposition results of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation provided by an embodiment of the present invention;
[0059] Figure 4 A subsequence sample entropy graph of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition aggregation provided by an embodiment of the present invention;
[0060] Figure 5 A sample entropy reconstruction result diagram of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation provided by an embodiment of the present invention;
[0061] Figure 6 A secondary decomposition result diagram of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition aggregation provided by an embodiment of the present invention;
[0062] Figure 7 A secondary decomposition spectrum diagram of a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition aggregation provided by an embodiment of the present invention;
[0063] Figure 8 A comparison chart of prediction results of a hybrid deep learning high-fluctuation load prediction method based on secondary decomposition and aggregation provided by an embodiment of the present invention;
[0064] Fig. 9 A comparison chart of prediction errors of a hybrid deep learning high-fluctuation load prediction method based on secondary decomposition and aggregation provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0065] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0066] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0067] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0068] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.
[0069] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0070] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0071] Example 1
[0072] Reference Figure 1-2, which is an embodiment of the present invention, provides a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation, comprising:
[0073] S1: Get the original load sequence, use adaptive noise complete set empirical mode decomposition to decompose the original load sequence once, and obtain several intrinsic mode components;
[0074] In the embodiment of the present application, the original load sequence is decomposed to obtain k subsequences, and the k subsequences include k-1 natural modal components and 1 residual component.
[0075] S2: Based on the intrinsic modal components, sample entropy and K-means are introduced to reconstruct the subsequences and filter out modal components with similar characteristics;
[0076] In the embodiment of the present application, sample entropy and K-means are introduced to reconstruct the subsequence, including the following steps:
[0077] Calculate the sample entropy value for each modal component and residual component;
[0078] The calculated entropy values of k samples are used as input and divided into three categories using the K-means clustering algorithm;
[0079] According to the clustering results, the corresponding k subsequences are recombined into three new sequences, named Co-IMF0, Co-IMF1 and Co-IMF2.
[0080] S3: Based on the results of sample entropy reconstruction, variational mode decomposition is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence to obtain the decomposed subsequence;
[0081] In the embodiment of the present application, variational mode decomposition is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence, including the following steps:
[0082] identifying components classified as high frequency sequences in the reconstruction of the sample entropy sequence;
[0083] By performing a preliminary empirical mode decomposition analysis on the high-frequency sequence, the number of target modes b of VMD decomposition is estimated using empirical mode decomposition;
[0084] According to the determined target modal number b, the VMD algorithm is applied to decompose the high-frequency sequence to obtain b subsequences.
[0085] In the embodiment of the present application, the VMD algorithm comprises the following steps:
[0086] Assume that the original signal f is decomposed into k components;
[0087] Assume the constraint that the sum of all modes is equal to the original signal;
[0088] Introducing Lagrangian function, the problem is transformed into an unconstrained variational problem;
[0089] The number of modal decompositions is determined by iteratively searching for the optimal solution of the variational model.
[0090] S4: performing data preprocessing on the decomposed subsequences to obtain processed sequences;
[0091] In the embodiment of the present application, the decomposed subsequences are subjected to data preprocessing, including normalizing the decomposed subsequences. The normalization processing is expressed as:
[0092]
[0093] Among them, x′,x,x min ,x max They represent the normalized data, the real data, and the minimum and maximum values of the real data in the test set, respectively.
[0094] S5: Based on convolutional neural network and bidirectional long short-term memory network, a CNN-BiLSTM hybrid deep learning prediction model is constructed;
[0095] In the embodiment of the present application, the CNN-BiLSTM hybrid deep learning prediction model, in which the state update and output of each layer of BiLSTM are expressed as:
[0096] A i =f 1 (ω 1 X i +ω 2 A i-1 )
[0097] B i =f 2 (ω 3 X i +ω 5 B i+1 )
[0098] Y i =f 3 (ω 4 A i +ω 6 B i )
[0099] Among them, X 1 ,X 2 ,X 3 ,...,X i ,...,X t Respectively represent the model at t 1 ,t2 ,t 3 ,...,t t The input data corresponding to the moment, A 1 ,A 2 ,A 3 ,...,A i ,...,A t Represent the hidden layer state of the forward BiLSTM, B 1 ,B 2 ,B 3 ,...,B i ,...,B i Represent the hidden layer state of the backward BiLSTM, Y 1 ,Y 2 ,Y 3 ,...,Y i ,...,Y t Respectively represent the model at t 1 ,t 2 ,t 3 ,...,t t The output data corresponding to the moment, ω 1 ,ω 2 ,ω 3 ,...,ω t Respectively represent the corresponding weight values of each layer in the model, f 1 ,f 2 ,f 3 They represent the activation functions between different hidden layers.
[0100] S6: The processed sequence is input into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and the prediction results of each subsequence are superimposed to obtain the prediction value of the original high-volatility load sequence.
[0101] It should be noted that the deep learning prediction model can be ARIMA, SVM with RBF Kernel, single LSTM, GRU; ARIMA is suitable for the prediction of linear trends and seasonal patterns, but its performance will be affected for nonlinear and non-stationary data, and it lacks the ability to capture complex patterns, especially when faced with a lot of noise or high-frequency fluctuations; SVM with RBF Kernel can handle nonlinear relationships and is suitable for small-scale data sets, but it has high computational complexity, especially slow training speed on large-scale data sets. For highly volatile data, SVM may require complex parameter adjustments to achieve better prediction results; a single LSTM is good at processing sequence data, especially time series prediction, but it only passes information forward and cannot utilize future contextual information. For complex data structures, a one-way LSTM may not be sufficient to capture all relevant patterns; GRU has high computational efficiency and a simplified structure, which reduces training time and resource requirements, but may not be as powerful as LSTM in handling long-term dependencies, so it may not be accurate enough when predicting volatile loads over a long time span. Our invention can effectively extract intrinsic modal components at different time scales from the original data by combining CEEMDAN and VMD, which helps to remove noise and reveal the essential characteristics of the data. Using CNN for feature extraction can capture local spatial patterns; BiLSTM can capture time dependencies from two directions, enhancing the expressiveness and generalization capabilities of the model. By combining a variety of signal processing technologies and deep learning methods, the model can better cope with highly volatile load sequences and improve the accuracy of predictions. Compared with a single model, this combined model can better handle non-stationary data and complex data patterns, thereby improving the stability and reliability of predictions.
[0102] This embodiment also provides a hybrid deep learning high-fluctuation load forecasting system based on secondary decomposition and aggregation, including:
[0103] The primary decomposition module is used to obtain the original load sequence, and the adaptive noise complete set empirical mode decomposition is used to decompose the original load sequence to obtain several intrinsic mode components;
[0104] The reconstruction module is used to reconstruct the subsequence based on the intrinsic modal components, introduce sample entropy and K-means, and screen out modal components with similar characteristics;
[0105] The secondary decomposition module is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence based on the result of sample entropy reconstruction by using variational mode decomposition to obtain the decomposed subsequence;
[0106] A preprocessing module is used to perform data preprocessing on the decomposed subsequences to obtain processed sequences;
[0107] Build a model module to build a CNN-BiLSTM hybrid deep learning prediction model based on convolutional neural network and bidirectional long short-term memory network;
[0108] The input module is used to input the processed sequence into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and superimpose the prediction results of each subsequence to obtain the prediction value of the original high-volatility load sequence.
[0109] Furthermore, it also includes:
[0110] Memory, used to store programs;
[0111] A processor is used to load the program to execute the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation.
[0112] This embodiment also provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the hybrid deep learning high-fluctuation load prediction method based on secondary decomposition and aggregation.
[0113] The storage medium proposed in this embodiment and the hybrid deep learning high-fluctuation load prediction method based on secondary decomposition and aggregation proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.
[0114] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.
[0115] Example 2
[0116] Reference Figure 3-Figure 9 and Table 1, are an embodiment of the present invention, which provides a hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation. In order to verify its beneficial effects, scientific demonstration is carried out through specific implementation methods and implementation effects.
[0117] In order to verify the performance of the prediction model proposed in the present invention, an actual data set is used for experimental test analysis. The data comes from the measured data of electric load from March 1, 2016 to May 31, 2016 provided by a certain area in southern my country, with a sampling frequency of 15 minutes and 96 sampling points per day. Each group of raw data contains four external environmental factors, namely, maximum temperature, minimum temperature, average temperature and rainfall, as well as load power data, totaling 8832 groups of data (7066 groups for training and 1766 groups for testing). Therefore, the present invention will conduct a study on power load prediction based on the above data.
[0118] Step 1.1 First mode decomposition
[0119] Set the sampling frequency to 4 and use the CEEMDAN algorithm to decompose the original time series into several IMF components, such as Figure 3 As shown (for obvious effect, only part is shown). The CEEMDAN algorithm decomposes the original time series into 14 IMF components, which respectively represent the characteristics of the original time series at different time scales. At the same time, from top to bottom, each IMF component can be roughly divided into three intervals: high frequency (IMF1, IMF2 and IMF3), medium frequency (IMF4, IMF5 and IMF6) and low frequency (IMF7, IMF8, IMF9, IMF10, IMF11, IMF12, IMF13 and IMF14). However, in order to accurately verify the above conjecture, the sample entropy-K-means clustering method is used to classify high, medium and low frequencies. The entropy values of the 14 subsequence samples are shown in the following figure. Figure 4 As shown. By changing the hyperparameter of sample entropy, it is found that the ranking of high-frequency components does not change, indicating that different hyperparameter combinations will not affect the determination of high, low, and medium frequency components. All sequences are integrated into three sequences by the sample entropy-K-means clustering method. The three subsequences IMF1, IMF2, and IMF3 with the highest sample entropy in the first decomposition sequence are integrated into the Co-IMF0 sequence, and the three subsequences IMF4, IMF5, and IMF6 with the second highest sample entropy are integrated into the Co-IMF1 sequence, and the remaining 8 subsequences are integrated into the Co-IMF2 sequence. The sample entropy reconstruction results are shown in Figure 5 shown.
[0120] Step 1.2 Secondary mode decomposition
[0121] The present invention performs VMD secondary modal decomposition on the high frequency component Co-IMF1 after K-means clustering. The decomposition number K=3 is specified, and the decomposition result is as follows: Figure 6 As shown in VMD-IMF1, VMD-IMF2 and VMD-IMF3, VMD-IMF4 and VMD-IMF5 are the above-mentioned Co-IMF2 and Co-IMF3. And the corresponding spectrum is as follows Figure 7 As shown in the figure, it can be seen that the load fluctuation frequency is significantly reduced after secondary decomposition.
[0122] Step 1.3 Comparison of prediction results
[0123] In order to verify the superiority of the proposed model CEEMDAN-VMD-CNN-BiLSTM (CVCB), it is compared with BiLSTM, CNN-BiLSTM (CB), and VMD-CNN-BiLSTM (VCB) models. Figure 8 As shown (only part is shown), the corresponding prediction error comparison is Fig. 9 The evaluation indicators are shown in Table 1.
[0124] Table 1 Prediction performance comparison table
[0125]
[0126] Combination Figure 8 , Fig. 9 It can be seen that the BiLSTM model has the worst effect in approximating the true value, followed by the CB model, the VCB model has a better fitting effect, and the CVCB model (the method in this paper) has the best fitting effect and the smallest prediction error.
[0127] From Table 1, we can see that the δ of the CVCB model MAE ,δ RMSE Compared with other models, the error can be reduced by up to 79.04% and 78.61%; the prediction accuracy of the four prediction models are 92.67%, 94.03%, 95.99% and 98.55% respectively. The prediction accuracy of the model proposed in this paper can be improved by up to 5.88%.
[0128] Based on the above experimental results, it can be concluded that the secondary decomposition technology and hybrid deep learning prediction method described in this paper have the best prediction performance in short-term power load forecasting.
[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation, characterized in that: include: The original load sequence is obtained, and the original load sequence is decomposed once using the adaptive noise complete set empirical mode decomposition to obtain several inherent mode components; Based on the intrinsic modal components, sample entropy and K-means are introduced to reconstruct the subsequences and filter out modal components with similar characteristics. Based on the results of sample entropy reconstruction, variational mode decomposition is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence to obtain the decomposed subsequences; Perform data preprocessing on the decomposed subsequences to obtain processed sequences; Based on convolutional neural network and bidirectional long short-term memory network, a CNN-BiLSTM hybrid deep learning prediction model is constructed; The processed sequences are input into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and the prediction results of each subsequence are superimposed to obtain the prediction value of the original high-volatility load sequence.
2. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 1 is characterized by: The adaptive noise complete set empirical mode decomposition is used to decompose the original load sequence once, including decomposing the original load sequence to obtain k subsequences, and the k subsequences include k-1 inherent mode components and 1 residual component.
3. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation according to claim 1 or 2, characterized in that: The introduction of sample entropy and K-means to reconstruct the subsequence includes the following steps: Calculate the sample entropy value for each modal component and residual component; The calculated entropy values of k samples are used as input and divided into three categories using the K-means clustering algorithm; According to the clustering results, the corresponding k subsequences are recombined into three new sequences, named Co-IMF0, Co-IMF1 and Co-IMF2.
4. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 3 is characterized by: The method of using variational mode decomposition to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence includes the following steps: identifying components classified as high frequency sequences in the reconstruction of the sample entropy sequence; By performing a preliminary empirical mode decomposition analysis on the high-frequency sequence, the number of target modes b of VMD decomposition is estimated using empirical mode decomposition; According to the determined target modal number b, the VMD algorithm is applied to decompose the high-frequency sequence to obtain b subsequences.
5. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 4 is characterized by: The VMD algorithm comprises the following steps: Assume that the original signal f is decomposed into k components; Assume the constraint that the sum of all modes is equal to the original signal; Introducing Lagrangian function, the problem is transformed into an unconstrained variational problem; The number of modal decompositions is determined by iteratively searching for the optimal solution of the variational model.
6. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 5 is characterized by: The decomposed subsequences are subjected to data preprocessing, including normalizing the decomposed subsequences. The normalization is expressed as: Among them, x′, x, x min , x max They represent the normalized data, the real data, and the minimum and maximum values of the real data in the test set, respectively.
7. The hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 6 is characterized by: The CNN-BiLSTM hybrid deep learning prediction model, in which the state update and output of each BiLSTM layer are expressed as: A i =f1(ω1X i +ω2A i-1 ) B i =f2(ω3X i +ω5B i+1 ) Y i =f3(ω4A i +ω6B i ) Among them, X1, X2, X3, ..., X i , ..., X t Respectively represent the model at t1, t2, t3, ..., t t The input data corresponding to the moment, A1, A2, A3, ..., A i , ..., A t Represent the hidden layer states of the forward BiLSTM, B1, B2, B3, ..., B i , ..., B t Represent the hidden layer states of the backward BiLSTM, Y1, Y2, Y3, ..., Y i , ..., Y t Respectively represent the model at t1, t2, t3, ..., t t The output data corresponding to the moment, ω1, ω2, ω3, ..., ω t They represent the corresponding weight values of each layer in the model, and f1, f2, and f3 represent the activation functions between different hidden layers.
8. A system based on the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as claimed in claim 1, characterized in that: The primary decomposition module is used to obtain the original load sequence, and the adaptive noise complete set empirical mode decomposition is used to decompose the original load sequence to obtain several intrinsic mode components; The reconstruction module is used to reconstruct the subsequence based on the intrinsic modal components, introduce sample entropy and K-means, and screen out modal components with similar characteristics; The secondary decomposition module is used to perform secondary decomposition on the strongly non-stationary sequence in the sample entropy reconstruction sequence based on the result of sample entropy reconstruction by using variational mode decomposition to obtain the decomposed subsequence; A preprocessing module is used to perform data preprocessing on the decomposed subsequences to obtain a processed sequence; Build a model module to build a CNN-BiLSTM hybrid deep learning prediction model based on convolutional neural network and bidirectional long short-term memory network; The input module is used to input the processed sequence into the CNN-BiLSTM hybrid deep learning prediction model for prediction respectively, and superimpose the prediction results of each subsequence to obtain the prediction value of the original high-volatility load sequence.
9. An electronic device, characterized in that: include: Memory, used to store programs; A processor, used to load the program to execute the steps of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the steps of the hybrid deep learning high-fluctuation load forecasting method based on secondary decomposition and aggregation as described in any one of claims 1 to 7 are implemented.