User power consumption behavior dynamic sensing method and system based on self-organizing center K-means algorithm, and medium
By combining the SOM and K-means algorithms, and utilizing adaptive learning rate and multi-layer SOM network structure, the problem of initial cluster center influence is solved, achieving more accurate user electricity behavior classification and dynamic perception, which is suitable for load management of large industrial users.
Patent Information
- Application Number
- CN202411378572.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-07
AI Technical Summary
In existing technologies, the accuracy of the K-means algorithm is affected by the initial cluster centers, and the self-organizing map algorithm may lead to inaccurate clustering results, which cannot effectively cope with the dynamic changes in the electricity consumption behavior of power users.
The self-organizing map (SOM) algorithm is used for global and local clustering to obtain cluster centers, which are then used as the initial cluster centers for the K-means algorithm. Combined with an adaptive learning rate and a multi-layer SOM network structure, K-means clustering is performed to improve clustering accuracy and accelerate the convergence process.
It improves the accuracy and adaptability of clustering results, enabling more precise classification of user electricity consumption behavior categories, and is suitable for load forecasting and management of large-scale industrial users.
Smart Images
Figure CN119272077A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the application field of power storage technology, and in particular to a method, system and medium for dynamically sensing user power consumption behavior based on a self-organizing center K-means algorithm. Background Art
[0002] In recent years, the participation of power users in power consumption and auxiliary services on the grid side has become a hot topic at home and abroad. In the smart grid environment, the development and construction of smart terminals, power communication technology and advanced measurement technology will promote the realization of two-way interaction of smart power consumption, thereby laying the foundation for users to participate in automatic demand response and realize the connection with the grid. However, due to the characteristics of instant use of power resources, it is impossible to store a large amount of power resources, and it is very easy to waste resources due to the imbalance of supply and demand during peak and valley periods. For example, in the power consumption monitoring system of large-scale industrial users, power consumption behavior may change significantly due to external factors such as holidays and seasonal changes; for the monitoring of power consumption behavior of multiple users in distributed industrial parks, there may be uneven data distribution and sudden changes in power consumption behavior of individual users. Therefore, in the context of intelligent, interactive and digital power consumption, mining the power consumption behavior of power users and analyzing their adjustment potential, load power and other information are of great significance to grid load forecasting, demand-side management, power pricing, etc.
[0003] Cluster analysis is a multivariate statistical analysis method for quantitative analysis of multiple samples and is an important part of power demand side management. Clustering algorithms can adapt to changes in the electricity consumption behavior of multiple users, adjust clustering results in a timely manner and reflect the dynamic changes of users. At present, there are two main clustering methods for user loads: direct clustering and indirect clustering. The first method directly uses the original load data for clustering analysis. The main algorithms include fuzzy C-means, K-means, self-organizing neural networks, etc. The second method uses an intermediate algorithm to pre-process the daily load curve, including load dimension reduction, such as singular value decomposition and morphological feature measurement based on time series. Although the above studies have studied user clustering, most of them use the characteristics of a certain clustering algorithm, without considering the limitations of the K-means algorithm, such as the accuracy of the initial cluster center and the self-organizing maps (SOM) algorithm may lead to inaccurate clustering results. Summary of the invention
[0004] Purpose of the invention: The present invention aims to provide a method for dynamic perception of user electricity consumption behavior based on the self-organizing center K-means algorithm to improve the accuracy of clustering results; another purpose of the present invention is to provide a system and medium for dynamic perception of user electricity consumption behavior based on the self-organizing center K-means algorithm.
[0005] Technical solution: The method for dynamically sensing user electricity consumption behavior based on the self-organizing center K-means algorithm of the present invention comprises the following steps:
[0006] The SOM clustering algorithm is used to perform global clustering and local clustering on the pre-processed user historical load data to obtain the cluster center;
[0007] The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm for K-means clustering;
[0008] According to the results of K-means clustering, the user's electricity usage behavior categories are divided, and the dynamic perception model is trained to obtain the dynamic perception of the user's electricity usage behavior.
[0009] Furthermore, the SOM clustering algorithm is used to perform global clustering and local clustering to obtain the cluster centers, as follows:
[0010] Based on the self-organizing map (SOM) clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced. Global clustering is first performed to distinguish input data with significant differences. Then, in the global clustering, objects with approximate values on the clustering standard are adaptively adjusted according to the feature points formed after aggregation by using the local characteristics of the data itself, and clustered in local areas.
[0011] Furthermore, K-means clustering is as follows:
[0012] The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
[0013] Furthermore, the user electricity usage behavior categories include peak-period electricity users, off-peak-period electricity users, continuous and stable electricity users, periodically fluctuating electricity users, and users with abnormal electricity usage behavior.
[0014] The user power consumption behavior dynamic perception system based on the self-organizing center K-means algorithm of the present invention comprises:
[0015] Data preprocessing module, used to preprocess user historical load data;
[0016] The clustering algorithm module is used to perform global clustering and local clustering using the SOM clustering algorithm, and to perform K-means clustering using the cluster center obtained by the SOM clustering algorithm as the initial cluster center of the K-means algorithm;
[0017] User classification module, used to classify user electricity consumption behavior categories based on the results of K-means clustering;
[0018] The dynamic perception model module is used to train the dynamic perception model and obtain dynamic perception of the user's electricity consumption behavior.
[0019] Furthermore, the clustering algorithm module, the SOM clustering algorithm performs global clustering and local clustering to obtain cluster centers, as follows:
[0020] The SOM clustering algorithm is used for global clustering and local clustering to obtain the cluster centers, as follows:
[0021] Based on the self-organizing map (SOM) clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced. Global clustering is first performed to distinguish input data with significant differences. Then, in the global clustering, objects with approximate values on the clustering standard are adaptively adjusted according to the feature points formed after aggregation by using the local characteristics of the data itself, and clustered in local areas.
[0022] Furthermore, the clustering algorithm module, K-means clustering is specifically as follows:
[0023] The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
[0024] Furthermore, the user classification module divides the user's electricity consumption behavior into categories according to the results of K-means clustering, including peak-period electricity users, off-peak-period electricity users, continuous and stable electricity users, periodically fluctuating electricity users and users with abnormal electricity consumption behavior.
[0025] The computer device of the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the steps of the above method are implemented when the processor executes the computer program.
[0026] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program implements the steps of the above method when executed by a processor.
[0027] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. The self-organizing center K-means algorithm adopted in the present invention uses the clustering center obtained by the SOM clustering algorithm as the initial clustering center of the K-means algorithm, cleverly combining the significant role of the SOM network in improving the clustering accuracy of the self-organizing preliminary clustering center with the advantages of the K-means algorithm in enhancing the clustering effect and accelerating the convergence process. 2. The present invention also makes up for the shortcomings of the SOM network's long convergence time and the poor clustering effect caused by improper selection of the initial clustering center of the K-means algorithm. It has very good application properties, accurately completes the clustering task, and has high user universality, and is suitable for any type of large industrial users. BRIEF DESCRIPTION OF THE DRAWINGS
[0028] Figure 1 This is a flow chart of clustering in the present invention. DETAILED DESCRIPTION
[0029] The present invention will be further described below in conjunction with the accompanying drawings.
[0030] The method for dynamically sensing user electricity consumption behavior based on the self-organizing center K-means algorithm of the present invention comprises the following steps:
[0031] The SOM clustering algorithm is used to perform global clustering and local clustering on the pre-processed user historical load data to obtain the cluster center;
[0032] (1) Obtain user historical load data and perform preprocessing.
[0033] (2) Initialization of SOM network weights to complete initial clustering. Based on the self-organizing map SOM clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced. Global clustering is first performed to distinguish input data with significant differences. Then, in the global clustering, objects with similar values on the clustering standard are adaptively adjusted based on the feature points formed after aggregation, using the local characteristics of the data itself, and clustered in local areas.
[0034] The weight vector W connecting the input node to the jth output node j (j=1,2,…,p) assigns a random number and sets the initial number of loops.
[0035] First, calculate the clustering objects Z according to the following formula: m (m=1,2,...,n) weight vector W j The winner weight vector W in g With Z m Distance:
[0036]
[0037] The Euclidean distance between any two d-dimensional vectors E and F can be expressed as
[0038]
[0039] Next, define N u (t) A neighborhood centered on the winner (i.e., the best matching unit, BMU). When unit u is determined to be the winner of the current input sample, the connection weight vectors of each unit in the neighborhood around it will begin to converge to the characteristics of the input sample, allowing the network to gradually adapt to and reflect the distribution characteristics of the input data. As the number of training times continues to increase, the algorithm will continue to perform the above process for each new input sample, that is, to find the winner and update the weights of the units in its neighborhood. When the training reaches a certain number of iterations or the algorithm reaches convergence, that is, the change in the weight vector is very small and almost no longer changes, the connection weights of the network will tend to stabilize. At this time, the network can better reflect the internal structure of the input data. Based on this stable network state, the initial clustering of the input samples is achieved by observing the responses of each node on the output layer. The learning equation is:
[0040]
[0041] Where: v jk Represents the connection weight between the jth input node and the kth output node, where k∈N k (t); represents the excitation input of the jth node of sample K; η(t) represents the tth learning rate which decreases as the number of training times increases.
[0042] The data is input into the self-organizing map (SOM) network for initial clustering. During the SOM training process, the present invention introduces an adaptive learning rate, aggregates objects with approximate values on the selected clustering standard near certain feature points, and performs adaptive adjustments based on the local characteristics of the data. A multi-layer SOM network structure is introduced, and coarse-grained global clustering is first performed, and then fine-grained secondary SOM training is performed in the local area to obtain preliminary and accurate clustering results.
[0043] (3) Apply the self-organizing map (SOM) algorithm to obtain the cluster center G = (G1, G2, ..., G k ), as the initial cluster center point of the K-means clustering algorithm, to improve the clustering effect and accelerate convergence.
[0044] The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
[0045] Calculate the sum of the squared distances J(C) of all samples to the cluster center of their category, and divide each sample into the cluster center closest to it. Samples with the same cluster center belong to the same category.
[0046]
[0047] Where: l(c p ,z q ) is the distance from the sample to the cluster center of its category; z q is other sample data in the class; c p is the sample cluster center; u pq is a binary variable, u pq = 0 means that the qth sample does not belong to the pth class, u pq =1 means it belongs to this category;
[0048] Then the class center is updated. According to the division result calculated in the previous step, the least squares method and the Lagrange principle, the center c of the K classes is updated. p , until the convergence condition is met.
[0049]
[0050] (4) According to the clustering results of K-means clustering, the electricity consumption behaviors of industrial users are divided into different categories, and the dynamic perception model is trained to obtain the dynamic perception of user electricity consumption behaviors. The user electricity consumption behavior categories include peak electricity users, off-peak electricity users, continuous and stable electricity users, periodic fluctuation electricity users and abnormal electricity users.
[0051] If new users’ daily load data is added, the trained model will be used to parse and process the data to achieve dynamic perception of the user’s electricity consumption behavior.
[0052] The user power consumption behavior dynamic perception system based on the self-organizing center K-means algorithm of the present invention comprises:
[0053] Data preprocessing module, used to preprocess user historical load data;
[0054] The clustering algorithm module is used to perform global clustering and local clustering using the SOM clustering algorithm, and to perform K-means clustering using the cluster center obtained by the SOM clustering algorithm as the initial cluster center of the K-means algorithm;
[0055] User classification module, used to classify user electricity consumption behavior categories based on the results of K-means clustering;
[0056] The dynamic perception model module is used to train the dynamic perception model and obtain dynamic perception of the user's electricity consumption behavior.
[0057] Furthermore, the clustering algorithm module, the SOM clustering algorithm performs global clustering and local clustering to obtain cluster centers, as follows:
[0058] The SOM clustering algorithm is used for global clustering and local clustering to obtain the cluster centers, as follows:
[0059] Based on the self-organizing map (SOM) clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced. Global clustering is first performed to distinguish input data with significant differences. Then, in the global clustering, objects with approximate values on the clustering standard are adaptively adjusted according to the feature points formed after aggregation by using the local characteristics of the data itself, and clustered in local areas.
[0060] Furthermore, the clustering algorithm module, K-means clustering is specifically as follows:
[0061] The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
[0062] Furthermore, the user classification module divides the user's electricity consumption behavior into categories according to the results of K-means clustering, including peak-period electricity users, off-peak-period electricity users, continuous and stable electricity users, periodically fluctuating electricity users and users with abnormal electricity consumption behavior.
[0063] The computer device of the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the steps of the above method are implemented when the processor executes the computer program.
[0064] The computer-readable storage medium of the present invention stores a computer program thereon, and the computer program implements the steps of the above method when executed by a processor.
Claims
1. A method for dynamically sensing user electricity consumption behavior based on a self-organizing center K-means algorithm, characterized in that: The following steps are involved: The SOM clustering algorithm is used to perform global clustering and local clustering on the pre-processed user historical load data to obtain the cluster center; The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm for K-means clustering; According to the results of K-means clustering, the user's electricity usage behavior categories are divided, and the dynamic perception model is trained to obtain the dynamic perception of the user's electricity usage behavior.
2. According to claim 1, the user power consumption behavior dynamic perception method based on the self-organizing center K-means algorithm is characterized by: The SOM clustering algorithm is used for global clustering and local clustering to obtain the cluster centers, as follows: Based on the self-organizing map (SOM) clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced to perform global clustering first. In the global clustering, objects with approximate values in the clustering standard are adaptively adjusted according to the feature points formed after aggregation by using the local characteristics of the data itself, and clustered in the local area.
3. According to claim 2, the user power consumption behavior dynamic perception method based on the self-organizing center K-means algorithm is characterized by: K-means clustering is as follows: The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
4. According to claim 3, the user power consumption behavior dynamic perception method based on the self-organizing center K-means algorithm is characterized in that: User electricity usage behavior categories include peak-period electricity users, off-peak-period electricity users, continuous and stable electricity users, periodically fluctuating electricity users, and users with abnormal electricity usage behavior.
5. A user electricity consumption behavior dynamic perception system based on self-organizing center K-means algorithm, characterized in that: include Data preprocessing module, used to preprocess user historical load data; The clustering algorithm module is used to perform global clustering and local clustering using the SOM clustering algorithm, and to perform K-means clustering using the cluster center obtained by the SOM clustering algorithm as the initial cluster center of the K-means algorithm; User classification module, used to classify user electricity consumption behavior categories based on the results of K-means clustering; The dynamic perception model module is used to train the dynamic perception model and obtain dynamic perception of the user's electricity consumption behavior.
6. According to claim 5, the user power consumption behavior dynamic perception system based on the self-organizing center K-means algorithm is characterized in that: The clustering algorithm module, SOM clustering algorithm performs global clustering and local clustering to obtain cluster centers, as follows: The SOM clustering algorithm is used for global clustering and local clustering to obtain the cluster centers, as follows: Based on the self-organizing map (SOM) clustering network, an adaptive learning rate and a multi-layer SOM network structure are introduced to perform global clustering first. In the global clustering, objects with approximate values in the clustering standard are adaptively adjusted according to the feature points formed after aggregation by using the local characteristics of the data itself, and clustered in the local area.
7. The user electricity consumption behavior dynamic perception system based on the self-organizing center K-means algorithm according to claim 6 is characterized in that: The clustering algorithm module, K-means clustering is specifically as follows: The cluster center obtained by the SOM clustering algorithm is used as the initial cluster center of the K-means algorithm. The squared distance between all user historical load data and the cluster center of their category is calculated, and the initial cluster center is updated. When the updated cluster center meets the convergence condition of the K-means algorithm, the user historical load data corresponding to each updated cluster center is used as the user historical load data of each category.
8. The user electricity consumption behavior dynamic perception system based on the self-organizing center K-means algorithm according to claim 7 is characterized in that: The user classification module divides the user's electricity consumption behavior into categories according to the result of K-means clustering, including peak-period electricity users, off-peak-period electricity users, continuous and stable electricity users, periodically fluctuating electricity users and users with abnormal electricity consumption behavior.
9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.