A power load prediction method based on frequency domain feature clustering and weighted fusion

Abstract

The application discloses a power load prediction method based on frequency domain feature clustering and weighted fusion, decomposes power load signals by using Fourier transform, extracts frequency domain features of different frequency components, then carries out clustering processing on the frequency domain features, and individually establishes a personalized prediction model for each clustering category. When predicting, a plurality of models most similar to a target prediction sequence are selected, and the prediction results of the models are fused in a weighted average manner. Compared with existing power load prediction methods, the frequency feature extraction method can better capture important periodic features of power load; clustering can cluster user load data with similar frequency domain features, thereby better identifying different power consumption modes; and the prediction results of different models are fused by weighting, so that the error of a single model is effectively reduced, and the resistance to noise is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning and power load prediction, and particularly relates to a power load prediction method based on frequency feature clustering and weighted average fusion. BACKGROUND

[0002] In the process of marketization of the power system, accurate power load prediction is crucial to ensure the reliability and economy of the power system operation. Short-term power load is greatly influenced by weather, holidays, user activities and other factors, and often shows a certain randomness. Most of these influencing factors have periodicity, which makes the short-term power load predictable. Current prediction models rely on a large amount of historical data, extract features related to load, and use machine learning or deep learning algorithms for training. Reasonable data processing and algorithm selection can effectively capture complex load patterns and improve the accuracy and reliability of prediction.

[0003] Power load is influenced by multiple factors, resulting in significant volatility and non-linear characteristics. Machine learning can effectively identify and model these complex relationships, improving prediction accuracy. Through feature selection engineering, machine learning methods can extract key information from massive data, remove redundant features, and thus improve model performance. In practical applications, accurate feature extraction and rapid processing and analysis of real-time data are crucial for helping the power system to adjust load in a timely manner and ensure stable operation.

[0004] In existing power load prediction methods, prediction models mainly optimize data processing methods to fully utilize important features in the data, thereby improving the prediction accuracy of neural network models. In order to ensure the effectiveness of feature extraction, recent research mainly proposes three different methods: feature engineering method, hybrid model method, and self-attention mechanism method. The feature engineering method uses statistical analysis to extract important features related to power load, such as seasonality, weather factors and historical power consumption patterns. The hybrid model method combines multiple neural network architectures, such as combining convolutional neural networks (CNN) with long short-term memory networks (LSTM), using CNN to extract spatial features and LSTM to process temporal features, improving the prediction ability of the model. The self-attention mechanism method uses the self-attention mechanism to improve the model's attention to input features, enhancing the expression of important information, thereby improving prediction accuracy.

[0005] However, these methods often fail to fully utilize the frequency domain characteristics of power load, resulting in important frequency domain information not being effectively captured, affecting the prediction ability of the prediction model. In the face of complex load changes, due to the reliance on a single prediction model (such as a single type of neural network), the integration or hybrid model strategy is not adopted, resulting in insufficient flexibility of existing methods. The method clusters power load sequences according to frequency domain characteristics, obtains multiple independent personalized models, and synthesizes the load prediction results of multiple models through weighted averaging, fully utilizes the diversity between models, effectively reduces the error of a single prediction model, and makes the final power load prediction more stable. Compared with traditional methods, this method can better control the bias and variance of the prediction model. SUMMARY

[0006] In view of the problems existing in the existing method, the data processing technology of the existing method is changed, and the power load sequences are clustered into several categories according to the similarity of the power consumption behaviors of different users, and a power load prediction method based on frequency domain feature clustering and weighted fusion is proposed. The extraction of the frequency domain features of the power load retains important feature data in the power load, and the clustering based on the frequency domain features of the power load and the establishment of the personalized model ensure the reliability and accuracy of the load prediction.

[0007] The present application proposes a power load prediction method based on frequency domain feature clustering and weighted fusion of multiple model prediction results, wherein the frequency domain feature clustering refers to dividing the power load time sequence with a sliding window, then performing Fourier decomposition on each power load sequence segment, and then clustering the amplitude and phase in the frequency domain obtained as the frequency domain features of the power load sequence segment. The weighted fusion of multiple models in the method refers to clustering the frequency domain features, in order to more accurately apply the features learned from the historical training data to the new load sequence, training a power load prediction model for each class separately, and matching several most similar classes for the target power load prediction sequence. The prediction results of the prediction models corresponding to these classes are weighted and fused according to the similarity.

[0008] The technical solution of the present application is as follows:

[0009] Step one: frequency domain feature extraction on the power load sequence segment divided by the sliding window

[0010] S11 input: 1000 user power load data, select data time range, set data sampling interval to 30 minutes, and daily load sampling point to 48.

[0011] S12 sliding window divides the power load data: all user original power load sequences are divided by a sliding window with a length of 11 days of data, the window is slid by 1 day in length each time, and a plurality of power load sequence segments in time domain are obtained.

[0012] S13 extracts the frequency domain features of the power load sequence segments: performs fast Fourier transform on each segmented power load sequence segment in the time domain to obtain the amplitude and phase angle corresponding to the power load segment in the frequency domain, and uses the amplitude and phase angle as the frequency domain features of the sequence segment.

[0013] Step 2: Adaptively cluster the frequency domain features of the power load sequence segments

[0014] S21 establishes a power load data set: the frequency domain features of all power load sequence segments are marked one by one with their corresponding sequence segments in the original time domain, so as to facilitate the use of these frequency domain features in subsequent clustering.

[0015] S22 is based on the frequency domain characteristics of power load sequence fragments: the proximity propagation clustering (AP) algorithm is used to adaptively cluster the frequency domain characteristics of multiple power load sequence fragments to generate 12 clusters.

[0016] S23 establishes an intra-class power load data set: finds the original time domain sequence segments corresponding to the frequency domain features of all power load sequence segments in each class, thereby forming 12 classes consisting of the original time domain sequence segments.

[0017] Step 3: Establish a personalized power load model based on the adaptive clustering results

[0018] S31 establishes a power load data set within each class: after clustering, the time domain power load sequence segments within each class have similar features and patterns, and the data within each class is used as training data for that class.

[0019] S32 trains a personalized power load model: Uses power load sequence fragments within each class to train the corresponding long short-term memory (LSTM) model to optimize model performance and improve the prediction accuracy of power load data for that class.

[0020] Compared to previous technologies, this paper proposes a clustering method based on power load frequency domain features, which focuses on capturing the most important frequency domain features of power load sequences. It also constructs a clustering-based strategy for establishing a personalized power load forecasting model, aiming to capture similar features and patterns within clusters. This strategy not only enhances the power load model's resistance to noise but also significantly improves power load forecasting accuracy.

[0021] Frequency domain feature extraction module

[0022] Power load curves exhibit periodicity on different time scales, such as daily and weekly loads. They can be viewed as the superposition of multiple power load signals of varying frequencies and amplitudes. Fast Fourier transforms are used to extract frequency-domain features to identify the primary frequency components and periodic variations in the power load sequence. Time-domain power load signals are converted to frequency-domain signals, revealing hidden frequency characteristics. Frequency-domain feature extraction is used to distinguish power load signals from noise, thereby improving the model's prediction accuracy and reliability.

[0023] In order to decompose the signals of different frequencies in the power load data, the fast Fourier transform is used to decompose the power load data into frequency domain signals. The decomposition steps are as follows:

[0024] First, the discrete power load signal x[n] is obtained and its length N is determined. If N is not a power of 2, x[n] is zero-padded to make its length a power of 2 to improve the efficiency of fast Fourier transform decomposition.

[0025] Then, the discrete power load signal x[n] is decomposed according to the following formula to obtain the frequency domain signal X[k] of the power load. X[k] is a complex sequence, and each k corresponds to a frequency component.

[0026]

[0027] Where X[k] is the amplitude and phase angle information of the power load signal at the kth frequency component, i is the imaginary unit, k represents the frequency index (k = 0, 1, ..., N-1), and e is the natural base.

[0028] The frequency domain signal X[k] of the power load is expressed in the complex form of a+bi, where a and b represent the real part and the imaginary part respectively. Each complex number a+bi corresponds to a specific frequency component. The amplitude and phase angle of the frequency component can be obtained from this complex form.

[0029] X[k]=a+bi

[0030]

[0031]

[0032] Here, |X[k]| is the amplitude of the frequency component, and arg(X[k]) is the phase angle of the frequency component. Given X[k], the values ​​of a and b in the complex number are known, so the values ​​of |X[k]| and arg(X[k]) can be calculated.

[0033] Finally, each load sequence segment of length N generates N load frequency domain signals X[k], which in turn yield an N-length amplitude vector and a N-length phase angle vector. These two vectors are concatenated to form a 2N-length feature vector, which serves as the frequency domain feature vector for the load sequence segment. The amplitudes and phase angles of the different frequency components of the load sequence are used as the frequency domain features of the sequence. By analyzing the frequency components of the load, the cyclical variations and random fluctuations in the load data are extracted, improving the prediction accuracy for complex fluctuating loads.

[0034] Clustering module based on frequency domain features

[0035] Traditional clustering algorithms typically require a specific number of clusters. However, because power loads contain numerous different frequency components, the boundaries between these components are often difficult to define, making it difficult to predefine a reasonable number of categories. Therefore, adaptive clustering methods should be used to automatically determine the number of clusters. This paper proposes using the AP algorithm to classify frequency features. It adaptively determines cluster centers based on the similarity between data points, avoiding the uncertainty of manually setting the number of clusters.

[0036] The AP algorithm is used to classify frequency features. The core of the AP algorithm is based on the similarity matrix and determines the cluster center by iteratively updating the two values ​​of "responsibility" and "attribution". For a power load feature vector with a length of 2N, the clustering steps are as follows:

[0037] Construct similarity matrix: Calculate the similarity matrix of any two power load feature vectors x with a length of 2N i and x k The similarity between i and x k Represented by point i and point k respectively, negative Euclidean distance is used as the similarity measure s(i,k):

[0038] s(i,k)=-||x i -x k || 2

[0039] Update responsibility value: The responsibility value r(i,k) represents the support degree of data point i in selecting point k as the cluster center. The update formula is:

[0040]

[0041] Here, a(i,k') represents the degree of belonging of points other than k to point i, initially 0; s(i,k') represents the similarity of points other than k to point i. The max component is used to find the maximum suitability of all candidate cluster centers for point i, excluding k. This ensures that the responsibility value only reflects whether the current candidate k is more suitable as a cluster center.

[0042] Update the attribution degree: The attribution degree a(i,k) reflects the adaptability of data point k as the cluster center:

[0043]

[0044] Among them, r(k,k) is the responsibility value of data point k to take itself as the cluster center; the irresponsible value max(0,r(i',k)) means that only those responsibility values ​​that have positive support for point k are taken into consideration, so The irresponsible values ​​of all data points i' except point i are accumulated to measure the support of other points for k.

[0045] The AP algorithm is used to iterate until the responsibility value and the degree of attribution converge, and the cluster center is finally determined, completing the clustering. The AP algorithm automatically finds the best cluster center based on the similarity of the power load feature sequence, and explores the potential patterns and frequency characteristics in the load curve.

[0046] Network prediction stage

[0047] In the prediction stage, the eight most similar clusters are matched based on the Euclidean distance between the target power load forecast sequence and the 12 power load cluster centers. The prediction results of the power load forecast models corresponding to these clusters are weighted and fused according to the similarity. The fusion process of the prediction results is as follows:

[0048] First, calculate the target power load forecast sequence u and the i-th cluster center v i The Euclidean distance d(u,v i ):

[0049]

[0050] Then, record the 8 classes with the smallest Euclidean distance among the 12 classes, normalize these 8 Euclidean distances, and obtain the similarity vector s = [s0, s1, ..., s7] between the target power load forecast sequence and each class center. i The normalization process is as follows:

[0051]

[0052] Finally, the LSTM model corresponding to the eight classes is used to predict the target power load forecast sequence, and the prediction result vector f = [f0, f1, ..., f7] is obtained. Therefore, the final power load forecast result is:

[0053]

[0054] This paper proposes a method that uses weighted averaging to fuse the power load forecast results of the most similar forecast models, effectively reducing the error of individual models and making the final forecast more stable. Compared with traditional methods, this method can better control the deviation and variance of the forecast model.

[0055] Improved forecast accuracy: Clustering power load frequency domain features effectively distinguishes different load patterns. By weighted fusion of similar patterns, the impact of individual outliers is eliminated, thereby improving forecast accuracy. This approach combines the advantages of different features to more accurately capture complex load fluctuations.

[0056] Enhance the generalization ability of the prediction model: After frequency domain clustering, the training data of the prediction model can better represent the overall power load characteristics within the class, thereby having better generalization ability under different power consumption environments, making the prediction model more stable when facing new data.

[0057] Improved Computational Efficiency: By clustering the power load data first, the prediction model reduces the amount of data it needs to process. This allows the weighted fusion phase to process only representative clustering results, rather than the entire power load dataset. This significantly improves computational efficiency, especially when applied to large-scale power load data. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a specific implementation flow chart of the present invention.

[0059] Figure 2 To determine the sliding window size for segmenting the power load sequence, the final performance is measured by the mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) corresponding to different window sizes.

[0060] Figure 3 To verify the effectiveness of using power load frequency features as the main clustering features, the frequency features of the power load fragment sequence are compared with the load values ​​of the power load fragment sequence in the time domain. The final performance is measured by MAE, MSE, and RMSE of the two methods.

[0061] Figure 4 To verify the effectiveness of the AP algorithm as a clustering method, the final performance is measured by MAE, MSE, and RMSE compared with using KMeans as a clustering method.

[0062] Figure 5 To verify the number of prediction models involved in weighted fusion for power load forecasting, the final performance is measured by the MAE, MSE, and RMSE corresponding to the number of prediction models involved in weighted fusion.

[0063] Figure 6 The load forecasting experiment results are shown on the Irish Energy Regulatory Commission dataset. The performance of different load forecasting methods is compared with each other and the final performance is measured by MAE, MSE, and RMSE on each method.

[0064] Figure 7 The figure shows the average accuracy trend of each power load forecasting method on the Irish Energy Regulatory Commission dataset. DETAILED DESCRIPTION

[0065] To accurately evaluate the present invention, we conducted experiments on the Irish Energy Regulatory Commission dataset. The Irish Energy Regulatory Commission dataset contains load profiles of over 1,000 residential consumers and small and medium-sized enterprises over a period of approximately one and a half years (July 1, 2009, to December 31, 2010). The data is sampled at a 30-minute interval and has 48 load sampling points per day.

[0066] When processing power load data, a sliding window of a certain size is required to intercept the original power load time series. In order to obtain the best prediction performance, the sliding window size should be determined through experiments. The sliding window size is in days, with 48 data points per day. Figure 2 As shown in the figure, different sliding window sizes yield different model prediction accuracies. Increasing the sliding window size from 6 to 11 days gradually increases prediction accuracy, reaching its highest accuracy with 11 days. This indicates that when the sliding window size is 11 days, the short series captured can fully reflect short-term load trends, resulting in accurate forecasts. Larger or smaller window sizes can lead to inaccurate clustering results, thus affecting subsequent forecast performance.

[0067] In order to verify the necessity of applying FFT to decompose time series, it is necessary to compare the differences in model prediction accuracy when clustering in the time domain and frequency domain. Figure 3 As shown in the figure, the power load frequency domain data after fast Fourier decomposition can better extract the periodic features in the sequence, thereby obtaining more accurate clustering results, which makes the prediction model perform better in the frequency domain than the traditional time domain analysis method.

[0068] In order to compare the impact of different clustering methods on model prediction accuracy, it is necessary to compare the differences in model prediction accuracy when using AP algorithm and KMeans clustering to process power load data. Figure 4 As shown in the figure, in the construction of the prediction model, the AP algorithm is used to cluster the power load data, which can effectively improve the performance of the model.

[0069] In order to verify the number of models k involved in weighted fusion in power load forecasting, it is necessary to select different k values ​​for comparative analysis and evaluate their impact on the forecast error.Figure 5 As shown in the figure, the model prediction performance is better when k=8. Under this value, the prediction results can better balance the diversity of the prediction model and the accuracy of the prediction.

[0070] On the Irish Energy Regulatory Commission dataset, e.g. Figure 6 As shown in , compared with the state-of-the-art global-local method, our method reduces MAE by 1.58%, MSE by 7.97%, and RMSE by 4.09%. Figure 6 As shown in Figure 3, the errors of our method are lower than those of other comparison methods, demonstrating excellent performance. This further demonstrates that our method can achieve higher accuracy in power load forecasting, verifying its effectiveness in power load forecasting.

Claims

1. A method for power load forecasting based on frequency domain feature clustering and weighted fusion, characterized in that: The implementation steps of this method are as follows: Step 1: Extract frequency domain features from the power load sequence segments segmented by the sliding window; S11 input: Power load data of 1000 users, select the data time range, set the data sampling interval to 30 minutes, and set the daily load sampling points to 48; S12 Sliding Window Segmentation of Power Load Data: Use a sliding window with a length of 11 days to segment the original power load sequence of all users. The window slides by 1 day each time to obtain several power load sequence segments in the time domain. S13 extracts frequency domain features of the power load sequence segments: performs fast Fourier transform on each segmented power load sequence segment in the time domain to obtain the amplitude and phase angle corresponding to the power load segment in the frequency domain, and uses the amplitude and phase angle as the frequency domain features of the sequence segment; Step 2: Adaptively cluster the frequency domain features of the power load sequence segments S21 establishes a power load data set: the frequency domain features of all power load sequence segments are marked one by one with their corresponding sequence segments in the original time domain, so as to facilitate the use of these frequency domain features in subsequent clustering; S22 is based on the frequency domain characteristics of power load sequence segments: the proximity propagation clustering (AP) algorithm is used to adaptively cluster the frequency domain characteristics of multiple power load sequence segments, thereby generating 12 clusters. S23 establishes a power load data set within each class: finds the original time domain sequence segments corresponding to the frequency domain features of all power load sequence segments within each class, thereby forming 12 classes consisting of the original time domain sequence segments; Step 3: Establish a personalized power load model based on the adaptive clustering results; S31 establishes a power load data set within each class: after clustering, the time domain power load sequence segments within each class have similar features and patterns, and the data within each class is used as training data for that class; S32 trains a personalized power load model: Uses power load sequence fragments within each class to train the corresponding long short-term memory (LSTM) model to optimize model performance and improve the prediction accuracy of power load data for that class.

2. The method for power load forecasting based on frequency domain feature clustering and weighted fusion according to claim 1, characterized in that: The implementation process of the frequency domain feature extraction module is as follows: The power load curve exhibits periodicity on different time scales, such as daily and weekly loads. The power load curve is considered to be the superposition of multiple power load signals of different frequencies and amplitudes. Fast Fourier transform is used to extract frequency domain features to identify the frequency components and periodic changes in the power load sequence. The time domain power load signal is converted into a frequency domain power load signal to reveal the hidden frequency characteristics in the signal. The power load signal is distinguished from noise through frequency domain feature extraction, thereby improving the accuracy and reliability of model prediction. In order to decompose the signals of different frequencies in the power load data, the fast Fourier transform is used to decompose the power load data into frequency domain signals. The decomposition steps are as follows: First, the discrete power load signal x[n] is obtained and its length N is determined. If N is not a power of 2, x[n] is zero-padded to make its length a power of 2 to improve the efficiency of fast Fourier transform decomposition. Then, the discrete power load signal x[n] is decomposed according to the following formula to obtain the frequency domain signal X[k] of the power load; X[k] is a complex sequence, and each k corresponds to a frequency component. Where X[k] is the amplitude and phase angle information of the power load signal at the kth frequency component, i is the imaginary unit, k represents the frequency index, k = 0, 1, ..., N-1, and e is the natural base; The frequency domain signal X[k] of the power load is represented by the complex form of a+bi, where a and b represent the real part and the imaginary part respectively. Each complex number a+bi corresponds to a specific frequency component. The amplitude and phase angle of the frequency component can be obtained from this complex form. X[k]=a+bi Where |X[k]| is the amplitude of the frequency component, and arg(X[k]) is the phase angle of the frequency component. When X[k] is obtained, the values ​​of a and b in the complex number are known, so the values ​​of |X[k]| and arg(X[k]) are calculated. Finally, each power load sequence segment of length N obtains N power load frequency domain signals X[k], and then obtains an amplitude vector of length N and a phase angle vector of length N. The two vectors are concatenated to obtain a feature vector of length 2N, which is used as the frequency domain feature vector of the power load sequence segment; the amplitude and phase angle of different frequency components of the power load sequence are used as the frequency domain characteristics of the sequence. By analyzing the frequency components of the load, the periodic changes and random fluctuations in the load data are extracted, thereby improving the prediction accuracy of complex fluctuating loads.

3. The power load forecasting method based on frequency domain feature clustering and weighted fusion according to claim 1 is characterized in that: The implementation process of the clustering module based on frequency domain features is as follows: An adaptive clustering method is used to automatically determine the number of clusters. The AP algorithm is used to classify frequency features. The core of the AP algorithm is based on the similarity matrix and determines the cluster center by iteratively updating the "responsibility" and "attribution" values. For a power load feature vector of length 2N, the clustering steps are as follows: Construct similarity matrix: Calculate the similarity matrix of any two power load feature vectors x with a length of 2N i and x k The similarity between i and x k Represented by point i and point k respectively, negative Euclidean distance is used as the similarity measure s(i,k): s(i,k)=-||x i -x k || 2 Update responsibility value: The responsibility value r(i,k) represents the support degree of data point i in selecting point k as the cluster center. The update formula is: Where a(i,k') represents the degree of belonging of points other than point k to point i, which is initially 0; s(i,k') represents the similarity of points other than point k to point i; the max part is used to find the maximum suitability of all candidate cluster centers of point i, excluding point k, to ensure that the responsibility value only reflects whether the current candidate point k is more suitable as a cluster center. Update the attribution degree: The attribution degree a(i,k) reflects the adaptability of data point k as the cluster center: Among them, r(k,k) is the responsibility value of data point k to take itself as the cluster center; the irresponsible value max(0,r(i',k)) means that only those responsibility values ​​that have positive support for point k are taken into consideration, so Accumulate the irresponsible values ​​of all data points i' except point i to measure the support of other points for k; Through iteration, until the responsibility value and attribution degree converge, the AP algorithm finally determines the cluster center and completes clustering; the AP algorithm is used to automatically find the optimal cluster center based on the similarity of the power load feature sequence and to explore the potential laws and frequency characteristics in the load curve.

4. The method for power load forecasting based on frequency domain feature clustering and weighted fusion according to claim 1, characterized in that: The implementation process of the network prediction stage is as follows: In the prediction stage, the eight most similar clusters are matched based on the Euclidean distance between the target power load forecast sequence and the 12 power load cluster centers. The prediction results of the power load forecast models corresponding to these clusters are weighted and fused according to the similarity. The fusion process of the prediction results is as follows: First, calculate the target power load forecast sequence u and the i-th cluster center v i The Euclidean distance d(u,v i ): Then, record the 8 classes with the smallest Euclidean distance among the 12 classes, normalize these 8 Euclidean distances, and obtain the similarity vector s = [s0, s1, ..., s7] between the target power load forecast sequence and each class center. i The normalization process is as follows: Finally, the LSTM model corresponding to the eight classes is used to predict the target power load forecast sequence, and the prediction result vector f = [f0, f1, ..., f7] is obtained. Therefore, the final power load forecast result is:

Images