Power consumption mode consistency analysis method, system and equipment based on load time-frequency domain feature fusion and medium

Abstract

The invention discloses a power utilization mode consistency analysis method, system and device based on load time-frequency domain feature fusion and a medium, and the method comprises the steps: obtaining power utilization load mode data of a client, extracting time domain features through dimension reduction, carrying out the Fourier transformation of the power utilization load mode data, calculating the amplitude of current harmonic waves, and obtaining the consistency of the current harmonic waves; selecting a preset number of significant harmonics to construct a frequency domain load feature vector; inputting the load characteristic template library and the transient current waveform template library into a clustering algorithm for time domain dominant clustering to obtain an initial clustering result, performing quality analysis, and triggering frequency domain clustering correction if the initial clustering result does not meet the standard; and the initial clustering result is combined with the frequency domain load feature vector to be input into an improved following leader algorithm to be corrected, so that a final consistency result is obtained. According to the method, time domain and frequency domain collaborative optimization is combined with the dynamic cluster number to adapt to a complex scene to ensure quality analysis reliability, accurate customer classification is achieved, and the method is suitable for personalized services of the electricity market.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a method, system, device and medium for electricity consumption pattern consistency analysis based on load time-frequency domain feature fusion. Background Technology

[0002] With the gradual opening and intelligent development of the electricity market, the demand for refined analysis of customer electricity load patterns in the power system is increasing. Currently, by analyzing customer electricity consumption behavior, power suppliers can optimize power resource allocation, design differentiated electricity pricing strategies, and provide personalized services. This requires in-depth research into the diversity and regularity of customer electricity load to achieve more efficient power management and market operation. At present, the analysis of customer electricity consumption patterns mainly relies on the statistical processing and feature extraction of load data, combined with external factors such as weather conditions, geographical location, and industry type, as well as internal factors such as daily electricity consumption curves and peak-valley characteristics for classification. Based on these characteristics, customers are usually divided into macro categories, such as residential, commercial, and industrial users, and further subdivided according to time dimensions, such as weekdays vs. weekends, and summer vs. winter, to identify different electricity consumption patterns.

[0003] In existing technologies, customer classification is typically based on the analysis of time-domain features, such as the peak and trough values ​​of daily load curves and total electricity consumption, grouping customers using statistical methods or traditional clustering algorithms like K-means. Additionally, frequency-domain analysis is sometimes introduced, extracting harmonic components of load patterns through Fourier transform to capture periodic characteristics. These methods can, to some extent, distinguish the electricity consumption characteristics of different customer groups, such as the high load stability of industrial users versus the regular operating hours of commercial users. However, existing technologies still suffer from limitations when dealing with complex and heterogeneous electricity consumption patterns. The diversity of individual user load patterns is not fully considered, and there is a lack of effective integration mechanisms for time-domain and frequency-domain features. This results in the frequency-domain features not playing a full role in classification, and the high computational complexity also affects the accuracy of feature extraction. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for electricity consumption pattern consistency analysis based on the fusion of load time-frequency domain features. This solves the problems that, when dealing with complex and heterogeneous electricity consumption patterns, the diversity of individual user load patterns is not fully considered, the fusion of time-domain and frequency-domain features lacks an effective integration mechanism, the role of frequency-domain features in classification cannot be fully utilized, and the computational complexity is also high.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for analyzing the consistency of electricity consumption patterns based on the fusion of time-frequency domain characteristics of load, including: acquiring customer electricity load pattern data;

[0008] The power load pattern data is subjected to dimensionality reduction processing, and the feature vector of daily load is extracted to construct a load feature template library, and waveform segments are extracted to construct a transient current waveform template library;

[0009] The maximum meaningful harmonic order is determined based on the sampling parameters of the power load pattern data. Fourier transform is performed on the power load pattern data to calculate the amplitude of the current harmonics. Based on the maximum meaningful harmonic order and amplitude, a preset number of significant harmonics are selected to construct a frequency domain load feature vector.

[0010] The load feature template library and transient current waveform template library are input into the clustering algorithm for time-domain dominant clustering to obtain the initial clustering results;

[0011] If the initial clustering results do not meet the preset standards, frequency domain clustering correction is triggered. The frequency domain clustering correction combines the initial clustering results with the frequency domain load feature vector input to improve the follow-leader algorithm, so as to obtain the final consistency result of the customer's electricity load pattern.

[0012] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the method includes: extracting the feature vector of daily load to construct a load feature template library, comprising:

[0013] Construct the covariance matrix of electricity load pattern data and obtain the matrix eigenvalues;

[0014] The eigenvalues ​​are sorted to obtain the eigenvector matrix;

[0015] Based on the feature vector matrix, the fundamental energy of daily load, the fluctuation characteristics of daily load, and the peak load characteristics are extracted, and a load feature template library is constructed.

[0016] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the method includes: extracting waveform segments to construct a transient current waveform template library, comprising:

[0017] The peak detection algorithm is used to locate characteristic segments in the load current.

[0018] The feature segments are aligned with load current waveforms of different lengths through dynamic time warping, and a transient current waveform template library is generated through hierarchical clustering.

[0019] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the method includes: determining the maximum meaningful harmonic order based on the sampling parameters of the electricity load pattern data; performing a Fourier transform on the electricity load pattern data to calculate the amplitude of the current harmonics; and constructing a frequency domain load feature vector based on the maximum meaningful harmonic order and amplitude by selecting a preset number of significant harmonics, including:

[0020] Determine the sampling parameters for each customer in the electricity load pattern data, including the sampling time interval, base frequency, event interval, and sampling frequency;

[0021] Based on Nyquist's theorem, the maximum meaningful harmonic order is extracted from the data by combining the sampling parameters to determine the boundary of the harmonic analysis range.

[0022] Based on the harmonic analysis range limit, relevant data are selected from the power load pattern data and converted into frequency domain signals through short-time discrete Fourier transform, decomposing them into harmonic components of different frequencies.

[0023] Each harmonic corresponds to an amplitude; calculate the amplitude of all harmonics within the harmonic analysis range limit.

[0024] For each customer, select the nth harmonic with the largest amplitude from the corresponding harmonics to form a feature vector;

[0025] For all customers, a frequency domain load feature vector is generated based on each customer's feature vector.

[0026] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the load feature template library and transient current waveform template library are input into a clustering algorithm for time-domain dominant clustering to obtain initial clustering results, including:

[0027] The fundamental energy of daily load, the fluctuation characteristics of daily load, the peak load characteristics, and the transient current waveform template library are input into the improved density peak clustering algorithm;

[0028] The load characteristics of different customers and the distance between the transient current waveform template libraries are calculated respectively, and the final time-domain clustering structure is determined by a voting mechanism to obtain the initial clustering results.

[0029] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the method includes: performing quality analysis on the initial clustering results; if the results do not meet preset standards, triggering frequency domain clustering correction, including:

[0030] The quality of the initial clustering results was analyzed using the mean index, cluster dispersion index, and Davies-Bouldin index.

[0031] If any one of the indices does not meet the first preset threshold requirement, frequency domain clustering correction is triggered.

[0032] As a preferred embodiment of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion described in this invention, the initial clustering results are combined with the frequency domain load feature vector input to an improved follow-leader algorithm for correction, in order to obtain the final consistency result of the customer's electricity consumption load pattern, including:

[0033] Based on the initial clustering results, the centroid is extracted to obtain time-domain features, which are then combined with the frequency-domain load feature vector to obtain a comprehensive feature vector;

[0034] The initial cluster assignment and centroid for each customer are obtained based on the initial clustering results;

[0035] Set a modified Euclidean distance threshold, and perform iterative optimization with the initial cluster assignment and centroid as the initial state;

[0036] Calculate the corrected Euclidean distance between the comprehensive feature vector of each customer and the current cluster centroid, allocate customers based on the second preset threshold, automatically determine the number of clusters and update the centroids;

[0037] Iteratively reassign clients to the cluster with the smallest distance, dynamically updating the centroid until the cluster assignment is stable;

[0038] Output stable cluster assignments and centroids, generating a final consistent result describing the power consumption patterns of each cluster.

[0039] The beneficial effects of this preferred technical solution are as follows: the combination of quality analysis and frequency domain correction forms a closed-loop optimization process, introducing frequency domain features for correction only when necessary, thus balancing computational efficiency and classification accuracy. This ensures the consistency and reliability of the final classification results in diverse scenarios.

[0040] Secondly, the present invention provides a power consumption pattern consistency analysis system based on load time-frequency domain feature fusion, comprising:

[0041] The acquisition module is used to acquire customer electricity load pattern data;

[0042] The first feature extraction module is used to perform dimensionality reduction processing on the electricity load pattern data, extract the feature vector of daily load to construct a load feature template library, and extract waveform segments to construct a transient current waveform template library.

[0043] The second feature extraction module is used to determine the maximum meaningful harmonic order based on the sampling parameters of the power load pattern data, perform Fourier transform on the power load pattern data, calculate the amplitude of the current harmonics, and select a preset number of significant harmonics based on the maximum meaningful harmonic order and amplitude to construct a frequency domain load feature vector.

[0044] The clustering module is used to input the load feature template library and the transient current waveform template library into the clustering algorithm to perform time-domain dominant clustering and obtain the initial clustering results.

[0045] The analysis and correction module is used to perform quality analysis on the initial clustering results. If the results do not meet the preset standards, frequency domain clustering correction is triggered. The frequency domain clustering correction combines the initial clustering results with the frequency domain load feature vector input and an improved follow-leader algorithm for cluster correction. After correction, the final consistency result of the customer's electricity load pattern is obtained.

[0046] Thirdly, the present invention provides a computer device, comprising:

[0047] Memory and processor;

[0048] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion.

[0049] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the power consumption pattern consistency analysis method based on load time-frequency domain feature fusion.

[0050] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention reduces computational complexity by extracting time-domain features through dimensionality reduction, and enhances the comprehensiveness of analysis by combining it with frequency-domain analysis. Dynamic harmonic selection can enhance the specificity of classification discrimination. The collaborative optimization of time and frequency domains, combined with dynamic cluster number, adapts to complex scenarios and ensures the reliability of quality analysis. Ultimately, it achieves efficient and accurate customer classification, suitable for optimized dispatching and personalized services in the power market, overcoming the limitations of traditional methods in addressing individual diversity, and significantly improving classification accuracy and flexibility. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the overall process of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion according to an embodiment of the present invention. Detailed Implementation

[0053] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0054] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for electricity consumption pattern consistency analysis based on load time-frequency domain feature fusion is provided, including:

[0055] S100: Acquire customer electricity load pattern data;

[0056] S200: Perform dimensionality reduction processing on the electricity load pattern data, extract the feature vector of daily load to construct a load feature template library, and extract waveform segments to construct a transient current waveform template library;

[0057] S300: Determine the maximum meaningful harmonic order based on the sampling parameters of the power load pattern data, perform Fourier transform on the power load pattern data, calculate the amplitude of the current harmonics, and construct a frequency domain load feature vector based on the maximum meaningful harmonic order and amplitude by selecting a preset number of significant harmonics.

[0058] S400: Input the load feature template library and transient current waveform template library into the clustering algorithm to perform time-domain dominant clustering and obtain the initial clustering results;

[0059] S500: Perform quality analysis on the initial clustering results. If they do not meet the preset standards, trigger frequency domain clustering correction. The frequency domain clustering correction combines the initial clustering results with the improved follow-leader algorithm based on the frequency domain load feature vector input to perform cluster correction. After correction, obtain the final consistency result of the customer's electricity load pattern.

[0060] It should be noted that the diversity of individual user load patterns is mainly reflected in residential users or some small commercial users. Their electricity consumption patterns are affected by factors such as living habits or business hours, family structure or customer traffic, and preferences for home appliances, exhibiting a high degree of randomness and diversity. Currently, most classifications focus on group characteristics, making it difficult to accurately capture the personalized load patterns of individual customers, which leads to insufficient granularity in the classification results. The high-order harmonics of users differ significantly from their low-order harmonics, but existing clustering methods cannot fully utilize this frequency domain information to optimize classification.

[0061] To address the aforementioned main issues, steps S100-S500 are designed to integrate time-domain and frequency-domain features and dynamically cluster to determine the number of clusters. This adapts to the diversity of individual users, optimizes classification accuracy, overcomes the limitations of existing technologies in analyzing customer electricity load patterns, and provides support for the refined management of the electricity market.

[0062] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a method for electricity consumption pattern consistency analysis based on load time-frequency domain feature fusion is provided.

[0063] In this embodiment of the application, customer electricity load pattern data is obtained in step S100;

[0064] In one alternative implementation, the acquisition method can be to use smart meters deployed on the client side to collect high-frequency electricity consumption data. The smart meters can capture detailed electricity load patterns, including daily load curves and short-term fluctuations, which is suitable for a wide range of individual users and can be used for subsequent time-domain feature extraction in S200 and frequency-domain analysis in S300.

[0065] In another alternative implementation, the data can be obtained by using monitoring equipment in the power distribution system to collect customer electricity load data. The monitoring equipment is usually installed at key nodes of the power distribution network, such as transformers or distribution cabinets, and records the total load data of a specific area or customer. It can be used for load data collection in centralized residential communities.

[0066] In this embodiment of the application, step S200 involves dimensionality reduction processing of the electricity load pattern data, extracting the feature vector of the daily load to construct a load feature template library, including the following steps A1-A3:

[0067] A1: Construct the covariance matrix of electricity load pattern data and obtain the matrix eigenvalues;

[0068] A2: Sort the eigenvalues ​​to obtain the eigenvector matrix;

[0069] A3: Extract the daily load fundamental energy, daily load fluctuation characteristics and peak load characteristics based on the feature vector matrix, and construct a load feature template library.

[0070] In another alternative implementation, dimensionality reduction can also be performed using LDA;

[0071] For example, 24-hour time-series data of M customers are obtained from S100, where the data for each customer is a high-dimensional vector representing the daily load pattern;

[0072] After preprocessing, initial category labels are set based on customer type or customer preferences;

[0073] Organize the daily load data of M customers into a matrix X with dimensions M×D, where D is the original dimension;

[0074] Calculate the intra-class scatter matrix, which is to calculate the variance of the feature vectors of customers within each class, to reflect the closeness of customers in the same class;

[0075] Calculate the inter-class scatter matrix, which is to calculate the variance between the centroids (i.e., the average eigenvectors) of different classes, to reflect the separation between classes;

[0076] By optimizing the ratio of intra-class scatter matrix to inter-class scatter matrix, the projection matrix W is solved to maximize the inter-class distance and minimize the intra-class distance after projection.

[0077] The daily load vector of each customer is projected onto a low-dimensional space through a projection matrix W to generate a low-dimensional feature vector. The low-dimensional vector encodes category distinguishing information and may include the fundamental energy of the daily load, the daily load fluctuation characteristics, and the peak load characteristics.

[0078] It should be noted that dimensionality reduction can ensure the efficiency and robustness of feature extraction, avoid noise interference in high-dimensional data, and enable subsequent clustering algorithms to focus on the most representative electricity consumption pattern features, improve the computational efficiency of classification, and maintain the quality of features.

[0079] In this embodiment of the application, step S200, which extracts waveform segments to construct a transient current waveform template library, includes the following steps B1-B2:

[0080] B1: Locate characteristic segments in the load current using a peak detection algorithm;

[0081] B2: The feature segments are aligned with load current waveforms of different lengths through dynamic time warping, and a transient current waveform template library is generated through hierarchical clustering.

[0082] It should be noted that aligning load current waveforms of different lengths is mainly used to eliminate the effects of time offset.

[0083] In this embodiment of the application, step S300 determines the maximum meaningful harmonic order based on the sampling parameters of the electricity load pattern data, performs a Fourier transform on the electricity load pattern data, calculates the amplitude of the current harmonics, and selects a preset number of significant harmonics based on the maximum meaningful harmonic order and amplitude to construct a frequency domain load feature vector, including the following steps C1-C6:

[0084] C1: Determine the sampling parameters for each customer in the electricity load pattern data, including the sampling time interval T, the base frequency f = 1 / T, the event interval T1, and its sampling frequency f1 = 1 / T1;

[0085] C2: Based on Nyquist's theorem and combined with sampling parameters, extract the maximum meaningful harmonic order h(2T1)max from the data. max To determine the limits of the harmonic analysis range;

[0086] C3: Based on the harmonic analysis range limit, relevant data are selected from the power load pattern data and converted into frequency domain signals through short-time discrete Fourier transform, decomposing them into harmonic components of different frequencies;

[0087] C4: Each harmonic corresponds to an amplitude; calculate the amplitude of all harmonics within the harmonic analysis range limit.

[0088] C5: For each customer, select the nth harmonic with the largest amplitude from the corresponding harmonics to form a feature vector;

[0089] Specifically, the choice of n depends on the specific application and can be based on experience or data analysis, mainly to ensure that the main features are captured without introducing too much noise.

[0090] C6: For all customers, generate frequency domain load feature vectors based on each customer's feature vector.

[0091] It should be noted that the short-time Fourier transform was chosen because it can decompose into harmonic components of different frequencies. The fundamental frequency f may be the daily cycle frequency of electricity consumption, while higher harmonics such as 2f and 3f may reflect nonlinear characteristics introduced by certain electrical equipment or behaviors. The amplitude can quantify the importance of each harmonic in the load pattern. For example, some customers' load patterns may have larger amplitudes in lower harmonics and smaller amplitudes in higher harmonics, indicating that their electricity consumption behavior is mainly low-frequency periodic. The setting of the maximum meaningful harmonic order is primarily to avoid erroneous features caused by sampling limitations affecting clustering and to enhance the discriminative power of customer classification.

[0092] It should also be noted that the frequency domain feature vectors provide additional information for subsequent clustering corrections, helping to distinguish customer electricity consumption patterns that are difficult to differentiate in the time domain features.

[0093] In this embodiment of the application, step S400 involves inputting the load feature template library and the transient current waveform template library into a clustering algorithm for time-domain dominant clustering to obtain initial clustering results, including the following steps E1-E2:

[0094] E1: Input the fundamental energy of daily load, the fluctuation characteristics of daily load, the peak load characteristics, and the transient current waveform template library into the improved density peak clustering algorithm;

[0095] E2: Calculate the load characteristics of different customers and the distance between the transient current waveform template libraries, and determine the final time-domain clustering structure using a voting mechanism to obtain the initial clustering results.

[0096] For example, the fundamental energy of daily load [x1,x2,...,x] M Daily load fluctuation characteristics [y1, y2, ..., y M Peak load characteristics [z1, z2, ..., z M ] and input of transient current waveform template library;

[0097] Taking the fundamental energy of daily load as an example:

[0098]

[0099] Then for each element x in feature x i Calculate the local density ρ i According to formula (2):

[0100] ρ i =∑ j∈I,j≠i χ(d ij -a) (2)

[0101] In the formula, a represents the neighborhood radius, and χ(·) satisfies equation (3):

[0102]

[0103] The density sequence ρ = [ρ1, ρ2, ..., ρ] corresponding to the load event samples is obtained. N Sort ρ in descending order to obtain the index sequence q = [q1, q2, ..., q]. N Furthermore, the distribution sequence is calculated according to equation (4).

[0104]

[0105] Calculate the density threshold ρ according to equation (5) respectively. th The distribution threshold δ is calculated according to equation (6). th :

[0106] ρ th =30% * N (5)

[0107] δ th =10%*max(x) (6)

[0108] For each element in x, if x i Simultaneously satisfy ρ i >ρ th as well as Then x i Selecting [c1, c2, ..., c3] as cluster centers, let all calculated cluster centers be C = [c1, c2, ..., c3]. L ], where L is the number of cluster centers, and the other samples in x are clustered according to the cluster centers C.

[0109] Other temporal features are also clustered according to this pattern, and the final temporal clustering structure is determined by a voting mechanism.

[0110] In this embodiment of the application, step S500 performs a quality analysis on the initial clustering results. If the results do not meet the preset standards, frequency domain clustering correction is triggered, including steps F1-F2:

[0111] F1: Quality analysis of the initial clustering results is performed using the mean index, cluster dispersion index, and Davies-Bouldin index;

[0112] It should be noted that the Mean Index (MIA) depends on the distance between the representative load pattern of each class and the load pattern belonging to the corresponding cluster; the Cluster Dispersion Index (CDI) depends on the distance between load patterns within the same cluster and the distance between the representative load patterns of the class; and the Davies-Bouldin index represents the average similarity measure of each cluster to its most similar cluster across the system, expressed in Euclidean form.

[0113] F2: If any one of the indices does not meet the first preset threshold requirement, frequency domain clustering correction is triggered.

[0114] It should be noted that, compared with traditional analysis methods that rely solely on the time domain or frequency domain, this approach, by integrating both features, can significantly improve the comprehensiveness and accuracy of classification, thereby avoiding information loss caused by single-dimensional analysis and providing a richer data foundation for subsequent clustering.

[0115] In this embodiment of the application, the frequency domain clustering correction in step S500 combines the initial clustering results with the improved follow-leader algorithm based on the frequency domain load feature vector input to perform clustering correction, so as to obtain the final consistency result of the customer's electricity load pattern, including the following steps G1-G6:

[0116] G1: Extract time-domain features based on the centroid of the initial clustering results, and combine them with the frequency-domain load feature vector to obtain a comprehensive feature vector;

[0117] G2: Obtain the initial cluster assignment and centroid for each customer based on the initial clustering results;

[0118] G3: Set a modified Euclidean distance threshold, and perform iterative optimization with the initial cluster assignment and centroid as the initial state;

[0119] G4: Calculate the corrected Euclidean distance between the comprehensive feature vector of each customer and the current cluster centroid, allocate customers based on the second preset threshold, automatically determine the number of clusters and update the centroid;

[0120] G5: Iteratively reassign clients to the cluster with the smallest distance, dynamically updating the centroid until the cluster assignment is stable;

[0121] G6: Outputs stable cluster assignments and centroids, generating a final consistent result describing the power consumption patterns of each cluster.

[0122] In one alternative implementation, if the corrected Euclidean distance is less than a second preset threshold, the customer is assigned to the nearest cluster; if the corrected Euclidean distance is greater than the second preset threshold, a new cluster is created with the customer's feature vector as the centroid.

[0123] In another alternative implementation, the improved follow-leader algorithm can be replaced by the K-means algorithm, which combines the time-domain features of S200 and the frequency-domain features of S300 into a comprehensive feature vector as input, uses the initial centroid of S400 as the initial centroid of K-means, presets the number of clusters based on the number of clusters of S400, iteratively assigns clients to the nearest cluster, updates the centroids, until convergence, and generates the final classification.

[0124] It should be noted that our preferred improved follow-leader algorithm is an iterative clustering method based on a distance threshold. It automatically adjusts customer assignments to different clusters until a stable classification is formed. Unlike traditional K-means, it does not pre-determine the number of clusters but dynamically determines the number of clusters through a distance threshold and optimizes the classification by updating centroids. The unsupervised nature of K-means may result in lower accuracy than the improved follow-leader algorithm, especially when there are many initial clustering errors or complex customer types.

[0125] In summary, this invention achieves accurate classification and evaluation of user electricity consumption patterns by jointly analyzing time-domain and frequency-domain features; it optimizes computational efficiency through feature dimensionality reduction and fusion, providing a feasible solution for large-scale user data analysis; the extracted harmonic features provide a new technical path for appliance-level electricity consumption diagnosis and non-intrusive load monitoring; and the dynamic two-stage clustering mechanism can adaptively adjust the classification granularity, supporting both refined energy efficiency management for residential users and providing data support for the formulation of power grid demand response strategies, which is of great significance for promoting the development of personalized services in smart grids.

[0126] Example 3 illustrates a schematic scheme for a power consumption pattern consistency analysis method based on load time-frequency domain feature fusion. It should be noted that the technical solution of this system for power consumption pattern consistency analysis based on load time-frequency domain feature fusion is based on the same concept as the technical solution of the aforementioned power consumption pattern consistency analysis method based on load time-frequency domain feature fusion. Details not described in detail in this embodiment can be found in the description of the aforementioned technical solution of the power consumption pattern consistency analysis method based on load time-frequency domain feature fusion.

[0127] This embodiment also provides an electricity consumption pattern consistency analysis system based on load time-frequency domain feature fusion, including:

[0128] The acquisition module is used to acquire customer electricity load pattern data;

[0129] The first feature extraction module is used to perform dimensionality reduction processing on the electricity load pattern data, extract the feature vector of daily load to construct a load feature template library, and extract waveform segments to construct a transient current waveform template library.

[0130] The second feature extraction module is used to determine the maximum meaningful harmonic order based on the sampling parameters of the power load pattern data, perform Fourier transform on the power load pattern data, calculate the amplitude of the current harmonics, and select a preset number of significant harmonics based on the maximum meaningful harmonic order and amplitude to construct a frequency domain load feature vector.

[0131] The clustering module is used to input the load feature template library and the transient current waveform template library into the clustering algorithm to perform time-domain dominant clustering and obtain the initial clustering results.

[0132] The analysis and correction module is used to perform quality analysis on the initial clustering results. If the results do not meet the preset standards, frequency domain clustering correction is triggered. The frequency domain clustering correction combines the initial clustering results with the frequency domain load feature vector input and an improved follow-leader algorithm for cluster correction. After correction, the final consistency result of the customer's electricity load pattern is obtained.

[0133] This embodiment also provides a computer device applicable to the case of electricity consumption pattern consistency analysis based on load time-frequency domain feature fusion, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as proposed in the above embodiment.

[0134] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as proposed in the above embodiments.

[0135] The storage medium proposed in this embodiment and the method for implementing power consumption pattern consistency analysis based on load time-frequency domain feature fusion proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0136] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing the consistency of electricity consumption patterns based on the fusion of load time-frequency domain characteristics, characterized in that, include: Obtain customer electricity load pattern data; The power load pattern data is subjected to dimensionality reduction processing, and the feature vector of daily load is extracted to construct a load feature template library, and waveform segments are extracted to construct a transient current waveform template library; The maximum meaningful harmonic order is determined based on the sampling parameters of the power load pattern data. Fourier transform is performed on the power load pattern data to calculate the amplitude of the current harmonics. Based on the maximum meaningful harmonic order and amplitude, a preset number of significant harmonics are selected to construct a frequency domain load feature vector. The load feature template library and transient current waveform template library are input into the clustering algorithm for time-domain dominant clustering to obtain the initial clustering results; If the initial clustering results do not meet the preset standards, frequency domain clustering correction is triggered. The frequency domain clustering correction combines the initial clustering results with the frequency domain load feature vector input to improve the follow-leader algorithm, thereby obtaining the final consistent result of the customer's electricity load pattern.

2. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 1, characterized in that, Extracting feature vectors from daily loads to construct a load feature template library, including: Construct the covariance matrix of electricity load pattern data and obtain the matrix eigenvalues; The eigenvalues ​​are sorted to obtain the eigenvector matrix; Based on the feature vector matrix, the fundamental energy of daily load, the fluctuation characteristics of daily load, and the peak load characteristics are extracted, and a load feature template library is constructed.

3. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 2, characterized in that, Extract waveform segments to construct a transient current waveform template library, including: The peak detection algorithm is used to locate characteristic segments in the load current. The feature segments are aligned with load current waveforms of different lengths through dynamic time warping, and a transient current waveform template library is generated through hierarchical clustering.

4. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 3, characterized in that, The maximum meaningful harmonic order is determined based on the sampling parameters of the electricity load pattern data. A Fourier transform is performed on the electricity load pattern data to calculate the amplitude of the current harmonics. Based on the maximum meaningful harmonic order and amplitude, a preset number of significant harmonics are selected to construct a frequency domain load feature vector, including: Determine the sampling parameters for each customer in the electricity load pattern data, including the sampling time interval, base frequency, event interval, and sampling frequency; Based on Nyquist's theorem, the maximum meaningful harmonic order is extracted from the data by combining the sampling parameters to determine the boundary of the harmonic analysis range. Based on the harmonic analysis range limit, relevant data are selected from the power load pattern data and converted into frequency domain signals through short-time discrete Fourier transform, decomposing them into harmonic components of different frequencies. Each harmonic corresponds to an amplitude; calculate the amplitude of all harmonics within the harmonic analysis range limit. For each customer, select the nth harmonic with the largest amplitude from the corresponding harmonics to form a feature vector; For all customers, a frequency domain load feature vector is generated based on each customer's feature vector.

5. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 4, characterized in that, The load characteristic template library and the transient current waveform template library are input into the clustering algorithm for time-domain dominant clustering to obtain the initial clustering results, including: The fundamental energy of daily load, the fluctuation characteristics of daily load, the peak load characteristics, and the transient current waveform template library are input into the improved density peak clustering algorithm; The load characteristics of different customers and the distance between the transient current waveform template libraries are calculated respectively, and the final time-domain clustering structure is determined by a voting mechanism to obtain the initial clustering results.

6. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 5, characterized in that, A quality analysis is performed on the initial clustering results. If they do not meet the preset standards, frequency domain clustering correction is triggered, including: The quality of the initial clustering results was analyzed using the mean index, cluster dispersion index, and Davies-Bouldin index. If any one of the indices does not meet the first preset threshold requirement, frequency domain clustering correction is triggered.

7. The electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in claim 6, characterized in that, The initial clustering results are combined with the frequency domain load feature vector input to refine the improved follow-leader algorithm, thereby obtaining the final consistent results of customer electricity load patterns, including: Based on the initial clustering results, the centroid is extracted to obtain time-domain features, which are then combined with the frequency-domain load feature vector to obtain a comprehensive feature vector; The initial cluster assignment and centroid for each customer are obtained based on the initial clustering results; Set a modified Euclidean distance threshold, and perform iterative optimization with the initial cluster assignment and centroid as the initial state; Calculate the corrected Euclidean distance between the comprehensive feature vector of each customer and the current cluster centroid, allocate customers based on the second preset threshold, automatically determine the number of clusters and update the centroids; Iteratively reassign clients to the cluster with the smallest distance, dynamically updating the centroid until the cluster assignment is stable; Output stable cluster assignments and centroids, generating a final consistent result describing the power consumption patterns of each cluster.

8. A power consumption pattern consistency analysis system based on load time-frequency domain feature fusion, wherein the method described in any one of claims 1-7 is characterized in that, include: The acquisition module is used to acquire customer electricity load pattern data; The first feature extraction module is used to perform dimensionality reduction processing on the electricity load pattern data, extract the feature vector of daily load to construct a load feature template library, and extract waveform segments to construct a transient current waveform template library. The second feature extraction module is used to determine the maximum meaningful harmonic order based on the sampling parameters of the power load pattern data, perform Fourier transform on the power load pattern data, calculate the amplitude of the current harmonics, and select a preset number of significant harmonics based on the maximum meaningful harmonic order and amplitude to construct a frequency domain load feature vector. The clustering module is used to input the load feature template library and the transient current waveform template library into the clustering algorithm to perform time-domain dominant clustering and obtain the initial clustering results. The analysis and correction module is used to perform quality analysis on the initial clustering results. If the results do not meet the preset standards, frequency domain clustering correction is triggered. The frequency domain clustering correction combines the initial clustering results with the improved follow-leader algorithm based on the frequency domain load feature vector input to perform cluster correction. After correction, the final consistency result of the customer's electricity load pattern is obtained.

9. A computer device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the power consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a processor, implement the steps of the electricity consumption pattern consistency analysis method based on load time-frequency domain feature fusion as described in any one of claims 1 to 7.

Images