Deep learning-based short-term power load prediction method

Abstract

The invention provides a short-term power load prediction method based on deep learning. The method comprises the steps of collecting data, analyzing influence factors, performing preprocessing operation on the data, constructing a VMDCNNBiGRU model, training the VMDCNNBiGRU model based on the data after preprocessing operation, and performing short-term power load prediction based on the trained VMDCNNBiGRU model. According to the invention, the complex fluctuation of load data can be effectively dealt with, and the precision and stability of load prediction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a short-term power load forecasting method based on deep learning. Background Art

[0002] Power load forecasting is an important part of the power market energy management system. Short-term power load forecasting is based on the characteristics of randomness and uncertainty of short-term load. Taking into account weather conditions and social factors, it fully studies and utilizes existing historical data to establish a set of mathematical models that conform to the continuity and periodicity of historical data. In order to meet the requirements of high accuracy, it determines the power load in the next few hours, one day or several days.

[0003] However, with the rapid development of the power industry, short-term power load forecasting faces challenges such as large data fluctuations, complex influencing factors, and high prediction accuracy requirements. Therefore, it is necessary to design a short-term power load forecasting method based on deep learning. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a short-term power load forecasting method based on deep learning.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The present invention provides a short-term power load forecasting method based on deep learning, comprising:

[0007] Collect data and analyze influencing factors;

[0008] Perform preprocessing operations on the data;

[0009] Build the VMD_CNN_BiGRU model;

[0010] Train the VMD_CNN_BiGRU model based on the preprocessed data;

[0011] Short-term power load forecasting is performed based on the trained VMD_CNN_BiGRU model.

[0012] Preferably, data is collected and the influencing factors are analyzed, specifically:

[0013] Load data and meteorological factor data are collected, and the influencing factors are analyzed using the Pearson correlation coefficient. The weather characteristics, temperature characteristics, and date characteristics in the meteorological factor data are used as characteristic variables.

[0014] Preferably, the data is preprocessed, specifically:

[0015] Obtain load data and meteorological factor data, and perform feature screening, Pearson processing, missing value processing, outlier processing and normalization processing on them.

[0016] Preferably, a VMD_CNN_BiGRU model is constructed, specifically:

[0017] Build a BiGRU model and introduce CNN to form a CNN_BiGRU model;

[0018] Perform VMD decomposition on the collected load data to obtain multiple subsequences;

[0019] The obtained subsequence is used as part of the input data and together with the feature variables as the input of the CNN_BiGRU model;

[0020] For each subsequence and CNN_BiGRU model, build their own VMD_CNN_BiGRU model and get their respective predicted load values;

[0021] The predicted load values ​​of each subsequence are summed and inverse normalized to obtain the final load prediction value of the load data.

[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0023] The present invention provides a short-term power load forecasting method based on deep learning, the method comprising collecting data, analyzing influencing factors, performing preprocessing operations on the data, constructing a VMD_CNN_BiGRU model, training the VMD_CNN_BiGRU model based on the data after the preprocessing operations, and performing short-term power load forecasting based on the trained VMD_CNN_BiGRU model. The present invention can effectively cope with the complex volatility of load data and improve the accuracy and stability of load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. 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 paying creative labor.

[0025] Figure 1 A logic block diagram provided for an embodiment of the present invention;

[0026] Figure 2 This is the exploded view of VMD;

[0027] Figure 3This is a schematic diagram of the VMD_CNN_BiGRU model structure;

[0028] Figure 4 Fitting curves of predicted values ​​and true values ​​of different models;

[0029] Figure 5a Fitting curve graph for CNN_BiGRU model;

[0030] Figure 5b It is the VMD_BiGRU model fitting curve graph;

[0031] Figure 5c Fitting curve graph for EMD_CNN_BiGRU model;

[0032] Figure 5d Fitting curve graph for CEEMDAN_CNN_BiGRU model;

[0033] Figure 5e Fitting curve graph for VMD_CNN_BiGRU model;

[0034] Figure 6 A schematic diagram of a method flow chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0036] The purpose of this invention is to provide a short-term power load forecasting method based on deep learning, which uses the VMD_CNN_BiGRU model to predict short-term power load. The structure diagram of the model is as follows: Figure 3 As shown in Figure 3, the model combines the VMD decomposition method and the CNN_BiGRU model, aiming to effectively cope with the complex volatility of the original load data and improve the accuracy and stability of load forecasting.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Figure 1 A logic block diagram provided for an embodiment of the present invention, Figure 6 A schematic diagram of a method flow chart provided by an embodiment of the present invention, such as Figure 1 and Figure 6 As shown, the present invention provides a short-term power load forecasting method based on deep learning, comprising:

[0039] Step 100: Collect data and analyze influencing factors;

[0040] Step 200: pre-processing the data;

[0041] Step 300: construct a VMD_CNN_BiGRU model;

[0042] Step 400: training the VMD_CNN_BiGRU model based on the preprocessed data;

[0043] Step 500: Perform short-term power load forecasting based on the trained VMD_CNN_BiGRU model.

[0044] In step 100, data is collected and the influencing factors are analyzed, specifically:

[0045] Load data and meteorological factor data are collected, and the influencing factors are analyzed using the Pearson correlation coefficient. The weather characteristics, temperature characteristics, and date characteristics in the meteorological factor data are used as characteristic variables.

[0046] In step 200, the data is preprocessed, specifically:

[0047] Obtain load data and meteorological factor data, and perform feature screening, Pearson processing, missing value processing, outlier processing and normalization processing on them.

[0048] In step 300, a VMD_CNN_BiGRU model is constructed, specifically:

[0049] Build a BiGRU model and introduce CNN to form a CNN_BiGRU model;

[0050] The collected load data is subjected to VMD decomposition to obtain multiple subsequences, specifically:

[0051] Firstly, the VMD method is used to stabilize the original load data and decompose the original load data into multiple subsequences with different frequencies, so that each subsequence can better express the local characteristics of the original load data, thereby better capturing the complex change law of the load data.

[0052] The obtained subsequence is used as part of the input data and as the input of the CNN_BiGRU model together with the feature variables, specifically:

[0053] Next, each subsequence obtained after VMD decomposition is used as part of the input data and combined with the feature variables filtered by the Pearson correlation coefficient as the input of the CNN_BiGRU model. In this way, by integrating feature information of different scales and sources, the performance and generalization ability of the model can be further improved.

[0054] For each subsequence and CNN_BiGRU model, build their own VMD_CNN_BiGRU model to get their own predicted load values, specifically:

[0055] A VMD_CNN_BiGRU model is constructed for each merged subsequence to predict its load value. The VMD_CNN_BiGRU model combines convolutional neural network and bidirectional gated recurrent unit, which can effectively learn the characteristics and timing information of sequence data and improve the accuracy of load prediction.

[0056] The predicted load values ​​of each subsequence are summed and inverse normalized to obtain the final load prediction value of the load data, which is:

[0057] The prediction results of each merged subsequence are summed and inverse normalized to finally obtain the load prediction value of the original sequence. Through this process, the prediction results of each subsequence can be effectively integrated, making the final load prediction more accurate and reliable.

[0058] Combining the above steps, the VMD_CNN_BiGRU model can make full use of the characteristics of the original load data, better adapt to the complex changes of load data, improve the accuracy and stability of load forecasting, and thus provide more accurate prediction and decision-making basis for system operation.

[0059] It should be noted that in the VMD algorithm, the selection of the number of modal components k is crucial to the decomposition effect. If the k value is too large, modes with the same frequency characteristics may be incorrectly assigned to different components, causing modal aliasing; while if the k value is too small, the complete information of the load sequence may not be fully displayed, resulting in the loss of frequency details. Therefore, in order to obtain the best decomposition effect, the present invention uses the center frequency method to determine the optimal number of modal components k.

[0060] Table 1 Center frequency at different k values

[0061]

[0062] In Table 1, by observing the VMD decomposition results of different modal numbers (k values), it is found that when the modal numbers are 3 and 4, the center frequencies of some modal components are still not completely extracted, resulting in some information in the original load sequence being ignored. When the modal number increases to 6, it is found that the center frequencies of the third and fourth modes are over-decomposed, resulting in an increase in the similarity of the center frequencies between different components, and an obvious modal sharing phenomenon occurs. This shows that choosing an appropriate modal number is crucial for effectively extracting signal features to avoid overfitting and information loss. After comprehensive analysis, the decomposition modal number of the present invention is selected as 5. The specific decomposition results are shown in Figure 1. Figure 2 shown.

[0063] The present invention provides an embodiment: using the data of the 9th "China Electrical Engineering Society Cup" National Undergraduate Electrical Mathematical Modeling Competition in 2016 as a public data set. The public data set contains power load data and meteorological factor data from January 1, 2012 to January 10, 2015 in a certain area. The sampling frequency of power data in the data set is once every 15 minutes, 96 times a day, and the data volume is 106,176; meteorological data includes daily average temperature, daily maximum temperature, daily minimum temperature, daily relative humidity and daily rainfall, and the sampling frequency is once a day. The undecomposed model CNN_BiGRU and the VMD-decomposed model VMD_CNN_BiGRU were experimentally compared for load prediction. At the same time, in order to further explore the impact of VMD decomposition data on the prediction effect, the models after EMD and CEEMDAN decomposition were selected for comparison. In the experiment, the number of VMD decompositions k was selected as 5 based on the above VMD decomposition results. The prediction results are as follows Figure 4 Shown

[0064] from Figure 4Analysis shows that the models with decomposition algorithms, namely VMD_BiGRU, EMD_CNN_BiGRU, CEEMDAN_CNN_BiGRU, and VMD_CNN_BiGRU, are more outstanding in capturing load changes than the models without decomposition algorithms, namely CNN_BiGRU. The basis of this conclusion is that the models with decomposition algorithms can better handle the complexity and volatility of load data. Specifically, decomposition algorithms such as CEEMDAN, EMD, and VMD can decompose the original load data into subsequences of different frequencies, so that the model can more flexibly capture the frequency characteristics and change laws in the load data. In contrast, the model without decomposition may not be able to understand and utilize the frequency information in the load data well, resulting in insufficient capture of load changes. Therefore, by adding the decomposition algorithm, the model can make more full use of the frequency information of the load data, and improve the sensitivity and accuracy of load changes. This method of combining decomposition algorithms with deep learning models provides an effective way to optimize and improve the load forecasting model, which helps to improve the dispatching and operation efficiency of the power system.

[0065] analyze Figure 5a-5e From the separate fitting curve graphs in the figure, it can be found that in the three models of CNN_BiGRU, VMD_BiGRU, and VMD_CNN_BiGRU, whether adding CNN or VMD, the predicted values ​​at the peaks and troughs can be more consistent with the true values. The three models of EMD_CNN_BiGRU, CEEMDAN_CNN_BiGRU, and VMD_CNN_BiGRU combined with the decomposition algorithm can more accurately reflect the changing trend of the load in the load forecasting process, and the predicted curve is closer to the true value curve. Among these three models, the VMD_CNN_BiGRU model proposed in the present invention shows the best fitting effect. This shows that the VMD algorithm can better retain the important features of the load data during the data decomposition process, so that the model can more accurately predict future power loads. Therefore, the VMD_CNN_BiGRU model shows the best performance in the load forecasting task of the present invention.

[0066] Analyzing the evaluation index sizes in Table 2, EMD_CNN_BiGRU, CEEMDAN_CNN_BiGRU, and VMD_CNN_BiGRU are significantly improved compared to CNN_BiGRU in these four indicators, proving that the data decomposition algorithm has a significant effect on load data prediction, indicating that the subsequences after data decomposition may be relatively simple, making the model easier to learn and generalize to new data. And the CNN_BiGRU model with the VMD algorithm added has a better prediction effect than the model with the EMD and CEEMDAN algorithms added. This is because VMD is more effective in extracting signal features of load data, can more accurately capture key information in load data, and improves the prediction performance of the CNN_BiGRU model. The VMD_CNN_BiGRU model has a MAPE of 1.77%, an RMSE value of 18429.00NW, a MAE value of 12292.74NW, and an R2 of 0.94, which is the relatively optimal model among the models compared by the present invention. At the same time, through ablation experiments on the three models of CNN_BiGRU, VMD_BiGRU, and VMD_CNN_BiGRU, it is also proved that CNN and VMD have a significant impact on improving the prediction effect of the model.

[0067] Table 2 Prediction results of different models

[0068]

[0069] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0070] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A short-term power load forecasting method based on deep learning, characterized in that: include: Collect data and analyze influencing factors; Perform preprocessing operations on the data; Build the VMD_CNN_BiGRU model; Train the VMD_CNN_BiGRU model based on the preprocessed data; Short-term power load forecasting is performed based on the trained VMD_CNN_BiGRU model.

2. The method according to claim 1, characterized in that: Collect data and analyze the influencing factors, specifically: Load data and meteorological factor data are collected, and the influencing factors are analyzed using the Pearson correlation coefficient. The weather characteristics, temperature characteristics, and date characteristics in the meteorological factor data are used as characteristic variables.

3. The method according to claim 2, characterized in that Preprocess the data, specifically: Obtain load data and meteorological factor data, and perform feature screening, Pearson processing, missing value processing, outlier processing and normalization processing on them.

4. The method according to claim 3, characterized in that Construct the VMD_CNN_BiGRU model, specifically: Build a BiGRU model and introduce CNN to form a CNN_BiGRU model; Perform VMD decomposition on the collected load data to obtain multiple subsequences; The obtained subsequence is used as part of the input data and together with the feature variables as the input of the CNN_BiGRU model; For each subsequence and CNN_BiGRU model, build their own VMD_CNN_BiGRU model and get their respective predicted load values; The predicted load values ​​of each subsequence are summed and inverse normalized to obtain the final load prediction value of the load data.

Images