A method for power load forecasting based on optimizing the selection of typical daily load curves
By combining the improved PFCM and FLDA algorithms, the problem of inaccurate selection of typical daily load curves in power load prediction is solved, clustering accuracy and prediction efficiency are improved, and more efficient power load prediction is achieved.
Patent Information
- Application Number
- CN202111255211.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-27
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-10-27
AI Technical Summary
The existing power load prediction methods have problems of clustering inaccuracy and high computational costs when selecting typical daily load curves, which affect the accuracy and efficiency of prediction.
The improved probability fuzzy C-mean clustering algorithm (PFCM) and fuzzy linear discrimination method (FLDA) were used to determine the cluster center matrix through clustering analysis, and the sample point closest to the cluster center point was selected using the Pearson correlation coefficient. As a typical daily load curve, power load prediction is performed.
It improves the accuracy and clustering accuracy of typical daily load curve selection, reduces the calculation cost, and improves the accuracy and efficiency of power load prediction.
Smart Images

Figure CN114037005B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power load forecasting, and in particular to a power load forecasting method based on optimizing and selecting a typical daily load curve. Background Art
[0002] Electricity load forecasting is an important part of power system planning and the basis of economic operation of power system. It is extremely important for power system planning and operation. The key to load forecasting is to collect a large amount of historical data, establish a scientific and effective forecasting model, adopt effective algorithms, conduct a large number of experimental studies based on historical data, summarize experience, and continuously revise models and algorithms to truly reflect the law of load changes.
[0003] Before power load forecasting, it is often necessary to select a typical daily load curve in advance. The traditional approach is to select a typical daily load curve based on a certain characteristic indicator, or directly select the daily load curve with the largest load in the whole month, or select a fixed working day of the month as the daily load curve of the month. Among them, selecting a typical daily load curve based on a certain characteristic indicator will cause the overall description of the load curve to be biased; if the daily load curve with the largest load in the whole month or a fixed working day of the month is directly selected as the daily load curve of the month, there will be problems with applicability and lack of representativeness.
[0004] In addition, existing research also uses machine learning methods to select typical daily load curves, such as using a single clustering algorithm or an integrated clustering algorithm to obtain a cluster center matrix, and then using different methods to obtain the sample points closest to the cluster center as the typical daily load curve. However, the existing possible fuzzy C-means clustering algorithm (PCM) focuses on typicality and does not consider the compactness of sample points, which will result in cluster consistency; the probabilistic fuzzy C-means algorithm (PFCM) overcomes the shortcomings of PCM cluster consistency and FCM sensitivity to distorted data, but the objective function needs to calculate the parameters in FCM first, which increases the calculation cost. This will lead to a lack of accuracy in the selection of typical daily loads, thereby reducing the accuracy of subsequent power load forecasts. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a power load forecasting method based on optimizing the selection of typical daily load curves, so as to improve the accuracy of power load forecasting by optimizing and selecting typical daily load curves closer to the cluster center.
[0006] The object of the present invention can be achieved by the following technical solution: A method for predicting power load based on optimizing and selecting a typical daily load curve, comprising the following steps:
[0007] S1. Obtaining original load data;
[0008] S2. Preprocess the original load data to obtain characteristic indicators;
[0009] S3, using the improved PFCM algorithm, combined with the fuzzy linear discriminant method (FLDA), to perform cluster analysis on the pre-processed raw load data and determine the cluster center matrix;
[0010] S4. Determine the base date, and calculate the correlation between the base date of each month and each cluster center according to the Pearson correlation coefficient method to determine the category to which each month belongs;
[0011] S5. Taking the sample point with the largest membership degree in each category as the typical daily load curve, the power load is predicted based on the typical daily load curve to obtain the corresponding prediction result.
[0012] Furthermore, the step S2 specifically performs standardization processing and distortion data screening processing on the original load data in sequence.
[0013] Furthermore, the step S3 specifically includes the following steps:
[0014] S31, initial cluster center according to the number of clusters;
[0015] S32, respectively calculating the sample covariance, sample typical value, fuzzy membership matrix and cluster center matrix;
[0016] S33, further calculating and obtaining the fuzzy inter-class scatter matrix and the fuzzy total scatter matrix;
[0017] S34, determining eigenvalues and eigenvectors according to the fuzzy inter-class scatter matrix and the fuzzy total scatter matrix, and projecting the original sample into the feature space to obtain transformed sample data;
[0018] S35, calculating the fuzzy membership, typical value and cluster center matrix in the feature space;
[0019] S36, determine whether the iteration termination condition is met, if so, end the calculation and output the currently calculated cluster center matrix, otherwise return to step S33.
[0020] Furthermore, the sample covariance calculation formula in step S32 is:
[0021]
[0022]
[0023] Where n is the number of samples, x j is the sample vector, σ 2 is the covariance matrix, is the sample vector average; the typical value of the sample is calculated as:
[0024]
[0025] Among them, c is the number of clusters, m is the fuzzy weighted number, t ij For sample x j For the typical value of the i-th class, p is the first parameter set, and γ is the current number of iterations;
[0026] The calculation formula of the fuzzy membership matrix of the sample is:
[0027]
[0028] Among them, d ij For sample x j To cluster center V i The Euclidean distance, u ij is the membership degree of the jth sample point to the i-th category;
[0029] The calculation formula of the cluster center matrix of the sample is:
[0030]
[0031] Among them, a and b are the second and third parameters set respectively, and v ij is the cluster center matrix of the jth sample point belonging to the i-th class.
[0032] Furthermore, the fuzzy inter-class scatter matrix in step S33 is specifically:
[0033]
[0034] The fuzzy total scatter matrix is specifically:
[0035]
[0036] Furthermore, the fuzzy total scatter matrix needs to be a non-singular matrix. If the calculated fuzzy total scatter matrix is a singular matrix, regularization processing is required.
[0037] Furthermore, the step S34 specifically includes the following steps:
[0038] S341. Calculate eigenvalues and eigenvectors:
[0039] S fT -1 S fB ω=λω
[0040] Among them, λ is the eigenvalue and ω is the eigenvector;
[0041] S342, project the sample into the feature space:
[0042] y j =ω T x j
[0043] Among them, y j is the transformed sample vector.
[0044] Furthermore, the specific process of step S35 is as follows:
[0045] Convert the cluster center matrix to feature space:
[0046] v i '(γ) =ω T v i (γ)
[0047] Among them, v i is the cluster center vector;
[0048] Compute the fuzzy membership function in the feature space:
[0049]
[0050] d ij '(γ) =|y i -v i '(γ) |
[0051] Compute typical values in feature space:
[0052]
[0053] Compute the cluster center matrix in feature space:
[0054]
[0055] Furthermore, the iteration termination condition includes a first termination condition and a second termination condition. When the first termination condition or the second termination condition is met, the iteration termination condition is met. The first termination condition is specifically that the current number of iterations is greater than a set maximum number of iterations.
[0056] The second termination condition is specifically that the absolute value of the difference between the current cluster center matrix and the previous cluster center matrix is less than or equal to a preset threshold.
[0057] Furthermore, the reference date is specifically a monthly average load day.
[0058] Compared with the prior art, the present invention aims at the problem of inaccurate clustering when selecting the typical daily load curve in the current power load forecast. By improving the PFCM algorithm and combining it with the FLDA algorithm, a new integrated clustering algorithm is realized. The new integrated clustering algorithm is used for clustering of load curves, which can fully consider the compactness of samples in the data set and improve the clustering accuracy; then combined with FLDA, iterative calculation is performed on the optimal feature space to obtain the best cluster center matrix, and the clustering effect is further optimized; finally, the sample point closest to the cluster center point is selected as the typical daily load curve through the Pearson correlation coefficient. In this way, the accuracy of the selection of the typical daily load curve is guaranteed, and the calculation cost is reduced, thereby effectively improving the accuracy and efficiency of power load forecasting. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] Figure 1 It is a schematic diagram of the method flow of the present invention;
[0060] Figure 2 This is a schematic diagram of the integrated clustering algorithm process proposed in the present invention. DETAILED DESCRIPTION
[0061] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0062] Example
[0063] like Figure 1 As shown, a power load forecasting method based on optimizing and selecting a typical daily load curve comprises the following steps:
[0064] S1. Obtaining original load data;
[0065] S2. Preprocess the original load data to obtain characteristic indicators, specifically, perform standardization processing and distortion data screening processing on the original load data in sequence;
[0066] S3, using the improved PFCM algorithm, combined with the fuzzy linear discriminant method (FLDA), to perform cluster analysis on the pre-processed raw load data and determine the cluster center matrix;
[0067] S4. Determine the base date. According to the Pearson correlation coefficient method, calculate the correlation between the base date of each month and each cluster center, and determine the category to which each month belongs. Specifically, the base date is the average load day of the month.
[0068] S5. Taking the sample point with the largest membership degree in each category as the typical daily load curve, the power load is predicted based on the typical daily load curve to obtain the corresponding prediction result.
[0069] This embodiment applies the above method, combines the improved probability C-means clustering algorithm (PFCM) with the fuzzy linear discriminant method (FLDA) to propose a new integrated clustering method. The algorithm first improves the original PFCM to obtain an improved PFCM; then combines the improved PFCM with FLDA, and applies the integrated clustering algorithm to the clustering of load curves.
[0070] The main process includes:
[0071] 1. Improved PFCM algorithm
[0072] For the PFCM method, it is necessary to first calculate the problem of increased computational cost due to parameters in the FCM algorithm, and first improve the PCM algorithm to:
[0073]
[0074]
[0075] In the formula: c is the number of clusters; n is the number of samples; m is the fuzzy weighted number; d ij For sample x j To cluster center V i The Euclidean distance of x j is the sample vector; u ij is the membership degree of the jth sample point to the ith class; t ij For sample x j Typical value for the i-th class; σ 2 is the covariance matrix; is the sample vector mean.
[0076] Improved PFCM objective function:
[0077]
[0078] Typical values for the sample:
[0079]
[0080] Class center matrix:
[0081]
[0082] Fuzzy membership matrix:
[0083]
[0084] Where: a, b, p are the first, second and third parameters set; γ is the number of iterations.
[0085] 2. FLDA-Improved PFCM Algorithm
[0086] In the processed data space, FLDA is used to find the optimal transformation vector to project the original matrix onto the optimal space, which can make the distance between classes larger and the samples within the class more compact.
[0087] Fuzzy between-class scatter matrix:
[0088]
[0089] Fuzzy total scatter matrix:
[0090]
[0091] Fuzzy total scatter matrix S ft In subsequent calculations, it must be a non-singular matrix. If it is a singular matrix, it must be regularized first.
[0092] Calculation of eigenvalues and eigenvectors:
[0093] S fT -1 S fB ω=λω (9)
[0094] In the formula, λ is the eigenvalue and ω is the eigenvector;
[0095] Project the samples into feature space:
[0096] y j =ω T x j (10)
[0097] The effective combination of the two algorithms can improve the clustering effect. The fuzzy linear discriminant algorithm (FLDA) iteratively calculates the optimal transformation vector in the optimal feature space, and combines it with the fuzzy clustering algorithm to improve its clustering accuracy. Combining FLDA with the improved PFCM algorithm, the following algorithm is obtained.
[0098] The cluster center matrix is transformed into the feature space:
[0099] v i '(γ) =ω T v i (γ) (11)
[0100] Calculate the fuzzy membership function value in the feature space:
[0101] d ij '(γ) =|y i -v i '(γ) | (12)
[0102]
[0103] Compute typical values in feature space:
[0104]
[0105] Compute the cluster center matrix in feature space:
[0106]
[0107] The above FLDA-improved PFCM algorithm is applied to the clustering of load curves, such as Figure 2 As shown, the specific steps are as follows:
[0108] Step 1: Initialize the improved PFCM algorithm and calculate the sample covariance σ using formula (2) 2 , and u is obtained by formula (4-6) ij , t ij , v ij The value of
[0109] Step 2: Calculate the fuzzy inter-class scatter matrix S using formulas (7, 8) fB and the fuzzy total scatter matrix S fT ;
[0110] Step 3: Calculate the eigenvalue λ and eigenvector ω using formula (9);
[0111] Step 4: Use formula (10) to convert the original sample x j Projecting to the feature space to get y j ;
[0112] Step 5: Get u through formula (11-15) ij ’(γ+1) , t ij ’(γ+1) , v ij ’(γ+1) ;
[0113] Step 6: γ=γ+1, and u ij ’(γ+1) , t ij ’(γ+1) , v ij ’(γ+1) The values of ij ’(γ) , t ij ’(γ) , v ij ’(γ) Then, repeat Step 2. If γ≤γ max or |v ij ‘(γ+1) -vij ‘(γ) |≤ζ, the iteration terminates.
[0114] 3. Selection of typical daily load curve
[0115] The base day (monthly average load day) is calculated, and the maximum value of the Pearson correlation coefficient between the base day vector of each month and the center vector of each cluster is used as the classification basis. The sample point with the largest membership degree in each class is selected as the typical daily load curve of this class, that is, the typical daily load curve of this class containing months.
[0116] In summary, the present invention first reduces the dimension of the original data using feature indicators and introduces σ 2 The concept of covariance is used to optimize the parameters in the existing PCM and then combine it with FCM to obtain an improved PFCM method, which takes into account the compactness of the samples in the data set and improves the clustering accuracy. Then, FLDA is used to iteratively calculate in the optimal feature space to obtain the optimal cluster center matrix to further optimize the clustering effect. The sample point closest to the cluster center point is selected as the typical daily load curve through the Pearson correlation coefficient. On the one hand, the improved PFCM method makes up for the shortcomings of PCM consistency and the lack of sensitivity of FCM to distorted data. On the other hand, FLDA is used to further alternately optimize in the fuzzy space to improve the clustering accuracy. Finally, the classification result is determined by the Pearson correlation coefficient, and the sample point with the largest membership in each category is taken as the typical day; through parameter optimization, the calculation cost is reduced and the work efficiency is improved.
Claims
1. A method for predicting power load based on optimizing and selecting typical daily load curves, characterized in that: The following steps are involved: S1. Obtaining original load data; S2. Preprocess the original load data to obtain characteristic indicators; S3, using the improved PFCM algorithm, combined with the fuzzy linear discriminant method, to perform cluster analysis on the pre-processed raw load data and determine the cluster center matrix; S4. Determine the base date. According to the Pearson correlation coefficient method, calculate the correlation between the base date of each month and each cluster center, and determine the category to which each month belongs. Specifically, the base date is the average load day of the month. S5. Taking the sample point with the largest membership degree in each category as the typical daily load curve, and performing power load forecasting based on the typical daily load curve to obtain corresponding forecasting results; Step S3 specifically includes the following steps: S31, initial cluster center according to the number of clusters; S32, respectively calculating the sample covariance, sample typical value, fuzzy membership matrix and cluster center matrix; S33, further calculating and obtaining the fuzzy inter-class scatter matrix and the fuzzy total scatter matrix; S34, determining eigenvalues and eigenvectors according to the fuzzy inter-class scatter matrix and the fuzzy total scatter matrix, and projecting the original sample into the feature space to obtain transformed sample data; S35, calculating the fuzzy membership, typical value and cluster center matrix in the feature space; S36, determine whether the iteration termination condition is met, if so, end the calculation and output the currently calculated cluster center matrix, otherwise return to step S33.
2. The method for predicting power load based on optimizing and selecting typical daily load curve according to claim 1, characterized in that: The step S2 specifically performs standardization processing and distortion data screening processing on the original load data in sequence.
3. The method for predicting power load based on optimizing and selecting typical daily load curve according to claim 1, characterized in that: The formula for calculating the covariance of the samples in step S32 is: Where n is the number of samples, x j is the sample vector, σ 2 is the covariance matrix, is the sample vector average; The typical value of the sample is calculated as: Among them, c is the number of clusters, m is the fuzzy weighted number, t ij For sample x j For the typical value of the i-th class, p is the first parameter set, and γ is the current number of iterations; The calculation formula of the fuzzy membership matrix of the sample is: Among them, d ij For sample x j To cluster center V i The Euclidean distance, u ij is the membership degree of the jth sample point to the i-th category; The calculation formula of the cluster center matrix of the sample is: Among them, a and b are the second and third parameters set respectively, and v ij is the cluster center matrix of the jth sample point belonging to the i-th class.
4. The method for predicting power load based on optimizing and selecting typical daily load curve according to claim 3 is characterized in that: The fuzzy inter-class scatter matrix in step S33 is specifically: The fuzzy total scatter matrix is specifically:
5. The method for predicting power load based on optimizing and selecting typical daily load curves according to claim 4, characterized in that: The fuzzy total scatter matrix needs to be a non-singular matrix. If the calculated fuzzy total scatter matrix is a singular matrix, regularization processing is required.
6. The method for predicting power load based on optimizing and selecting typical daily load curve according to claim 4, characterized in that: The step S34 specifically includes the following steps: S341. Calculate eigenvalues and eigenvectors: S fT -1 S fB oh = I Among them, λ is the eigenvalue and ω is the eigenvector; S342, project the sample into the feature space: y j =ω T x j Among them, y j is the transformed sample vector.
7. The method for predicting power load based on optimizing and selecting typical daily load curves according to claim 6, characterized in that: The specific process of step S35 is as follows: Convert the cluster center matrix to feature space: v i ' (γ) =ω T v i (γ) Among them, v i is the cluster center vector; Compute the fuzzy membership function in the feature space: d ij ' (γ) =|y i -v i ' (γ) Compute typical values in feature space: Compute the cluster center matrix in feature space:
8. The method for predicting power load based on optimizing and selecting typical daily load curves according to claim 1, characterized in that: The iteration termination condition includes a first termination condition and a second termination condition. When the first termination condition or the second termination condition is met, the iteration termination condition is met. The first termination condition is specifically that the current number of iterations is greater than the set maximum number of iterations. The second termination condition is specifically that the absolute value of the difference between the current cluster center matrix and the previous cluster center matrix is less than or equal to a preset threshold.
Citation Information
Patent Citations
Tea near-infrared spectrum classification method for fuzzy identification information extraction
CN110378374A
Distributed energy station typical daily load obtaining method based on k-means clustering
CN111583059A
Kernel fuzzy c-means fast clustering algorithm with integrated spatial constraints
WO2021007744A1
