User load clustering method and system based on VMD and space-time network variational word coding

Abstract

The invention provides a user load clustering method and system based on VMD and space-time network variational word coding. Wherein modal decomposition and reconstruction are carried out on electric power signals by introducing variational modal decomposition, preliminary clustering is carried out by using a variational self-encoding model based on a space-time network and using a neural network, extraction of potential features of user electrical load data is realized, and then clustering is carried out by using a Mini Batch K-Means algorithm. Through verification, the variational self-encoding model based on the space-time network shows a good effect on the Davingberg index and the contour coefficient index through a dimension reduction and re-clustering algorithm, and a reference value can be provided for how to deal with the complexity and diversity problems of user power consumption behaviors in the industry.

Description

Technical Field

[0001] The invention relates to a user load clustering method and system based on VMD and spatiotemporal network variational word coding, belonging to the technical field of power system load analysis. Background Art

[0002] At present, the commonly used load characteristic analysis schemes in power systems are generally based on clustering algorithms. However, in power systems, users' electricity consumption behaviors are complex and changeable, especially with the widespread access to new energy and the diversification of user needs, traditional electricity consumption forecasting methods are facing increasing challenges. Summary of the invention

[0003] In view of the above problems, the present invention proposes a user load clustering method and system based on VMD and spatiotemporal network variational word coding to solve the problem that traditional methods cannot handle the complexity and diversity of user power consumption behaviors well.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] In a first aspect, an embodiment of the present invention provides a user load clustering method based on VMD and spatiotemporal network variational word coding, which includes:

[0006] Obtain the user's power load data and perform preprocessing;

[0007] The preprocessed power load data is decomposed using a variational mode decomposition algorithm to obtain multiple intrinsic mode components, and modal reconstruction is performed based on the intrinsic mode components;

[0008] The variational autoencoder model based on spatiotemporal network is used to extract potential features from the power load data after modal reconstruction, and preliminary clustering results are obtained.

[0009] Based on the preliminary clustering results, the Mini Batch K-Means algorithm is used to perform cluster analysis on the power load data to obtain accurate clustering results.

[0010] Based on the above method, optionally, obtaining the user's power load data and preprocessing it includes:

[0011] Selecting historical data of power load in a preset time period as basic data;

[0012] Correct or remove abnormal data in basic data;

[0013] Normalize the basic data after correcting or eliminating abnormal data.

[0014] Based on the above method, optionally, the preprocessed power load data is decomposed using a variational mode decomposition algorithm to obtain multiple eigenmode components, including:

[0015] By constructing a constrained variation function, the power load data is decomposed into several eigenmode components;

[0016] Through the Lagrange function, the constrained variational function for calculating the bandwidth is transformed into an unconstrained variational function, and then the optimal solution is obtained through iteration.

[0017] Based on the above method, optionally, the method for constructing the variational autoencoder model based on the spatiotemporal network includes:

[0018] Perform spatiotemporal network coding on the power load data after modal reconstruction;

[0019] The encoder of the variational autoencoder model is used to process the spatiotemporal network coding to obtain a reduced-dimensional feature vector;

[0020] The decoder of the variational autoencoder model is used to reconstruct the reduced-dimensional feature vector.

[0021] Clustering the reduced-dimensional feature vectors using a neural network model;

[0022] The clustering loss is combined with the KL divergence and reconstruction error to optimize the model and obtain a variational autoencoder model based on the spatiotemporal network.

[0023] Based on the above method, optionally, based on the preliminary clustering result, the Mini Batch K-Means algorithm is used to perform cluster analysis on the power load data to obtain an accurate clustering result, including:

[0024] Determine the optimal clustering K value within the set K value range;

[0025] Based on the determined optimal K value, the Mini Batch K-Means algorithm is run to process the power load data;

[0026] After the algorithm converges, the cluster centers are extracted from the final clustering results to obtain accurate clustering results.

[0027] Based on the above method, optionally, it also includes:

[0028] Based on the clustering results, the analysis results of the similarities and differences between the power load data of different users are generated.

[0029] In a second aspect, an embodiment of the present invention further provides a user load clustering system based on VMD and spatiotemporal network variational word coding, which includes:

[0030] A data acquisition module is used to acquire the user's power load data and perform preprocessing;

[0031] A decomposition and reconstruction module is used to decompose the preprocessed power load data using a variational mode decomposition algorithm to obtain multiple intrinsic mode components, and perform mode reconstruction based on the intrinsic mode components;

[0032] The first clustering module is used to extract potential features of the power load data after modal reconstruction using a variational autoencoder model based on a spatiotemporal network to obtain preliminary clustering results;

[0033] The second clustering module is used to perform cluster analysis on the power load data based on the preliminary clustering results by using the Mini Batch K-Means algorithm to obtain accurate clustering results.

[0034] The user load clustering method and system based on VMD and spatiotemporal network variational word coding provided by the present invention performs modal decomposition and reconstruction of power signals by introducing variational modal decomposition, and proposes to use a variational autoencoder model based on a spatiotemporal network, and use a neural network for preliminary clustering to extract potential features of user power load data, and then use the Mini Batch K-Means algorithm for clustering. It has been verified that the variational autoencoder model based on the spatiotemporal network has shown good results in both the Davis-St. Paul index and the silhouette coefficient indicators through the dimensionality reduction and re-clustering algorithm, which can provide reference value for how to deal with the complexity and diversity of user power consumption behavior in the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present invention, and together with the specification are used to explain the principles of the present invention. In addition, these drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by reference to specific embodiments.

[0036] Figure 1 A schematic flow chart of a user load clustering method based on VMD and spatiotemporal network variational word coding provided by one embodiment of the present invention;

[0037] Figure 2 A schematic diagram of the operation flow of variational mode decomposition and reconstruction provided by one embodiment of the present invention;

[0038] Figure 3 A schematic diagram of the structure of a user load clustering system based on VMD and spatiotemporal network variational word coding provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0040] In view of the complexity and diversity of user electricity consumption behaviors in the power system, especially in the context of widespread access to new energy and diversified user needs, the present invention proposes a user load clustering scheme based on VMD and spatiotemporal network variational word coding, aiming to improve the accuracy and reliability of electricity consumption prediction. The following non-limiting description of the specific implementation scheme is given through several examples or embodiments.

[0041] like Figure 1 As shown, the embodiment of the present invention provides a user load clustering method based on VMD and spatiotemporal network variational word coding, comprising the following steps:

[0042] Step S101: Acquire the user's power load data and perform preprocessing.

[0043] Specifically, the electricity load data, i.e., the data samples to be processed, can improve the accuracy and effectiveness of the samples through preprocessing.

[0044] Furthermore, in some embodiments, obtaining the user's power load data and performing preprocessing may specifically include:

[0045] 1. Select the historical data of power load in the preset time period as the basic data.

[0046] For example, you can select actual user electricity load data for key industries in a province. The data type is actual electricity consumption every 15 minutes. A total of 96 data points are collected every day, and the data frequency is daily data for the past three years.

[0047] 2. Correct or eliminate abnormal data in the basic data.

[0048] Specifically, some data may have more than 20 sampling points that are 0 or the sampling point values ​​do not change throughout the day. In such cases, direct deletion is performed.

[0049] If the value of a certain sampling point suddenly increases or decreases compared with the two sampling points before and after it, and the rate of change deviates from the sequence trend too much, the outlier will be replaced by the average value of the data before and after the outlier point.

[0050] By processing abnormal data, subsequent errors can be reduced.

[0051] 3. Normalize the basic data after correcting or eliminating abnormal data.

[0052] In order to eliminate the large differences in the order of magnitude of user power load data, improve the training rate of the neural network, and make it converge faster, the data set is normalized. The maximum and minimum value normalization (MinMax) is used, and the calculation formula is:

[0053]

[0054] In the formula, p ′ (j, t) is the data of the normalized sampling point; p(j, t) is the data to be normalized; p(j) min and p(j) max are the minimum and maximum values ​​of the jth sequence data respectively. The normalized data value range is [0, 1].

[0055] Step S102: Decompose the preprocessed power load data using a variational mode decomposition algorithm to obtain a plurality of intrinsic mode components, and perform mode reconstruction based on the intrinsic mode components.

[0056] Specifically, variational modal decomposition (VMD) shows good robustness in the process of processing nonlinear data noise. Therefore, this embodiment uses VMD to decompose and obtain components with different frequency characteristics.

[0057] More specifically, the specific process of using the variational mode decomposition algorithm to decompose the preprocessed power load data to obtain multiple intrinsic mode components may include:

[0058] By constructing a constrained variational function, the power load data is decomposed into several eigenmode components; through the Lagrange function, the constrained variational function of the calculation bandwidth is transformed into an unconstrained variational function, and then the optimal solution is obtained through iteration.

[0059] Through VMD decomposition and modal reconstruction, components with different frequency characteristics can be extracted from complex signals, and the signal can be decomposed into multiple intrinsic modal components. This method shows good robustness when dealing with nonlinear data and noise, which helps to deeply understand the characteristics and structure of the signal. By introducing constrained variational functions and Lagrange functions, the constraints and solution process can be better optimized, and the effect of modal decomposition and reconstruction can be improved.

[0060] The principle of variational mode decomposition and reconstruction is as follows Figure 2 shown.

[0061] Step S103: using a variational autoencoder model based on a spatiotemporal network, extracting potential features from the power load data after modal reconstruction to obtain preliminary clustering results.

[0062] Specifically, the purpose of this step is to process the power load data after modal reconstruction through a variational autoencoder model based on a spatiotemporal network. This model combines a neural network for preliminary clustering, can efficiently capture the spatiotemporal correlation of power load data, and extract potential features through variational autoencoders.

[0063] In some embodiments, a method for constructing a variational autoencoder model based on a spatiotemporal network includes:

[0064] 1. Perform spatiotemporal network coding on the power load data after modal reconstruction;

[0065] Specifically, according to the spatiotemporal characteristics of user load data, a multidimensional spatiotemporal grid data matrix containing time series data and spatial information is constructed, and the spatiotemporal network data is represented in a form suitable for VAE (Variational Auto-Encoder) model input. Convolutional Neural Networks (CNN) are used to process spatial information, and Long Short-Term Memory (LSTM) is used to process time series information. The outputs of CNN and LSTM are combined to form a coded representation of the spatiotemporal network.

[0066] 2. Use the encoder of the variational autoencoder model to process the spatiotemporal network coding to obtain a reduced-dimensional feature vector.

[0067] VAE is a generative model that combines the concepts of deep learning and Bayesian inference. It can learn the potential representation of input data and generate new samples similar to the input data. The core idea of ​​VAE is to map the input data to the latent space through an encoder and generate output data from the latent space through a decoder. VAE introduces a probabilistic model, assuming that the input data is generated from latent variables, and uses variational inference methods to optimize model parameters.

[0068] Among them, the VAE encoder usually includes a multi-layer neural network, the last layer outputs the mean and standard deviation in the latent space, and uses the reparameterization technique to sample from the latent space distribution to obtain low-dimensional feature vectors, which contain the main information of the power load data.

[0069] 3. Use the decoder of the variational autoencoder model to reconstruct the reduced-dimensional feature vector.

[0070] Through the decoder part of VAE, the low-dimensional feature vector in the latent space is mapped back to the high-dimensional representation of the original data space. The output of the decoder should be as close to the original input data as possible to ensure the integrity of the information, and the reconstruction loss method is used to measure the difference between the decoder output and the original input. The present invention uses the mean square error MSE method to represent the reconstruction loss.

[0071] 4. Use the neural network model to cluster the reduced dimensionality feature vectors.

[0072] After extracting the reduced-dimensional feature vector from the decoder, the low-dimensional feature vector is standardized or normalized, and then a neural network is used for preliminary clustering, that is, each low-dimensional feature vector is classified using a pre-trained neural network model.

[0073] 5. The clustering loss is combined with the KL divergence and reconstruction error to optimize the model and obtain a variational autoencoder model based on the spatiotemporal network.

[0074] KL (Kullback-Leibler) is a divergence measure of the difference between the distribution of the encoder output and the prior distribution, encouraging the latent variables to follow the standard normal distribution. By jointly optimizing the clustering loss, KL divergence, and reconstruction error, a comprehensive model optimization goal can be formed to balance the clustering effect, model generalization ability, and data reconstruction accuracy.

[0075] In addition, in some embodiments, based on the clustering results, analysis results of similarities and differences between different users' power load data can also be generated to achieve extraction of potential features of user load data.

[0076] Step S104: Based on the preliminary clustering results, the Mini Batch K-Means algorithm is used to perform cluster analysis on the power load data to obtain accurate clustering results.

[0077] Specifically, the Mini Batch K-Means algorithm is an improved K-means clustering algorithm for processing large data sets. It is similar to the traditional K-means algorithm, but only uses the current mini-batch data instead of the entire data set when calculating each iteration step. This speeds up the algorithm and is suitable for processing large data sets that cannot be loaded into memory at once.

[0078] The basic steps of the Mini Batch K-Means algorithm are as follows:

[0079] 1. In the initialization stage, a certain number of samples are randomly selected from the data set as cluster centers.

[0080] 2. Randomly select a mini-batch of data samples from the dataset.

[0081] 3. For each sample data point, calculate the distance between it and the current cluster center point and assign it to the cluster with the nearest center point.

[0082] 4. Update the center point of the current cluster, use the new center point to calculate the value of the objective function, and check whether the convergence condition is met.

[0083] 5. If the algorithm does not converge, repeat steps 2 to 4 until the convergence condition is met.

[0084] In this embodiment, the Mini Batch K-Means algorithm is used to perform cluster analysis on user power load data. This algorithm significantly improves the calculation efficiency while maintaining a good clustering effect, and is suitable for processing large-scale power load data.

[0085] In some embodiments, based on the preliminary clustering results, the Mini Batch K-Means algorithm is used to perform cluster analysis on the power load data to obtain accurate clustering results, which may specifically include:

[0086] 1. Determine the optimal clustering K value within the set K value range.

[0087] The purpose of this step is to select a suitable clustering K value: first, determine a reasonable K value range based on the characteristics of the user's power load data and the clustering target. Then, within the selected K value range, run the Mini Batch K-Means algorithm with different K values, and select the optimal K value based on the evaluation criteria.

[0088] 2. Based on the determined optimal K value, run the Mini Batch K-Means algorithm to process the power load data.

[0089] In this step, the Mini Batch K-Means algorithm is run using the selected optimal K value. If the cluster center no longer changes significantly or reaches the preset number of iterations, the algorithm reaches the convergence condition.

[0090] 3. After the algorithm converges, extract the cluster center from the final clustering result to obtain an accurate clustering result.

[0091] The cluster center is the representative data point of each cluster, usually located in the center of the cluster. It can be extracted by visualizing the clustering results, calculating the distance between the cluster center and the points in the cluster, etc. In addition, the extracted cluster center can be verified to ensure that it can accurately reflect the clustering structure of the data.

[0092] Through the above scheme, clustering of user power load data can be achieved, which is convenient for subsequent analysis and processing. By introducing variational mode decomposition, the power signal is modally decomposed and reconstructed, and the variational autoencoder model based on the space-time network is proposed. The neural network is used for preliminary clustering to extract the potential features of the user power load data, and then the Mini Batch K-Means algorithm is used for clustering. It has been verified that the variational autoencoder model based on the space-time network has shown good results in both the Davis-St. Paul index and the silhouette coefficient indicators through the dimensionality reduction and clustering algorithm, which can provide reference value for how to deal with the complexity and diversity of user power consumption behavior in the industry.

[0093] The present invention aims to improve the accuracy and reliability of electricity consumption forecasting in view of the complexity and diversity of electricity consumption behaviors of users in the power system, especially in the context of widespread access to new energy and diversified user needs. To achieve this goal, the present invention innovatively introduces the variational mode decomposition technology to perform mode decomposition and reconstruction on the user's electricity load data, thereby deeply revealing the inherent structure and characteristics of the user's electricity load data. Furthermore, the present invention proposes a variational autoencoder model based on a spatiotemporal network, which combines a neural network for preliminary clustering, can efficiently capture the spatiotemporal correlation of power load data, and extract potential features through variational autoencoder. On this basis, the present invention further jointly optimizes the clustering loss, KL divergence, and reconstruction error to form a comprehensive model optimization goal, aiming to balance the clustering effect, model generalization ability, and data reconstruction accuracy. After the model is optimized, the present invention uses the Mini Batch K-Means algorithm to perform cluster analysis on the user's electricity load data. While maintaining a good clustering effect, the algorithm significantly improves the computational efficiency and is suitable for the processing of large-scale power load data.

[0094] The verification results show that the method of the present invention has excellent clustering effects on indicators such as Davies-Bouldin index (DBI) and silhouette coefficient (SC) through the dimensionality reduction and re-clustering algorithm, verifying its accuracy and stability. In summary, the present invention successfully solves the complexity and diversity of user electricity consumption behavior in the power system by introducing advanced signal processing, feature extraction and clustering analysis technology, provides strong technical support for the analysis and prediction of user electricity consumption behavior in the power industry, and helps promote the sustainable development of the power industry.

[0095] In addition, an embodiment of the present invention provides a user load clustering system based on VMD and spatiotemporal network variational word coding. Figure 3 As shown, the system includes:

[0096] The data acquisition module 31 is used to acquire the user's power load data and perform preprocessing;

[0097] A decomposition and reconstruction module 32 is used to decompose the preprocessed power load data using a variational mode decomposition algorithm to obtain multiple intrinsic mode components, and perform mode reconstruction based on the intrinsic mode components;

[0098] The first clustering module 33 is used to extract potential features from the power load data after modal reconstruction using a variational autoencoder model based on a spatiotemporal network to obtain a preliminary clustering result;

[0099] The second clustering module 34 is used to perform cluster analysis on the power load data based on the preliminary clustering result by using the Mini Batch K-Means algorithm to obtain an accurate clustering result.

[0100] Among them, the specific implementation methods of each module of the user load clustering system based on VMD and spatiotemporal network variational word coding can refer to the corresponding contents in the aforementioned method embodiments, which will not be repeated here.

[0101] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0102] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0103] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0104] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0105] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0106] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0107] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0108] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A user load clustering method based on VMD and spatiotemporal network variational word coding, characterized in that: include: Obtain the user's power load data and perform preprocessing; The preprocessed power load data is decomposed using a variational mode decomposition algorithm to obtain multiple intrinsic mode components, and modal reconstruction is performed based on the intrinsic mode components; The variational autoencoder model based on spatiotemporal network is used to extract potential features from the power load data after modal reconstruction, and preliminary clustering results are obtained. Based on the preliminary clustering results, the Mini Batch K-Means algorithm is used to perform cluster analysis on the power load data to obtain accurate clustering results.

2. The method according to claim 1, characterized in that The obtaining of the user's power load data and preprocessing includes: Selecting historical data of power load in a preset time period as basic data; Correct or remove abnormal data in basic data; Normalize the basic data after correcting or eliminating abnormal data.

3. The method according to claim 1, characterized in that The variational mode decomposition algorithm is used to decompose the preprocessed power load data to obtain multiple eigenmode components, including: By constructing a constrained variation function, the power load data is decomposed into several eigenmode components; Through the Lagrange function, the constrained variational function for calculating the bandwidth is transformed into an unconstrained variational function, and then the optimal solution is obtained through iteration.

4. The method according to claim 1, characterized in that The method for constructing the variational autoencoder model based on the spatiotemporal network includes: Perform spatiotemporal network coding on the power load data after modal reconstruction; The encoder of the variational autoencoder model is used to process the spatiotemporal network coding to obtain a reduced-dimensional feature vector; The decoder of the variational autoencoder model is used to reconstruct the reduced-dimensional feature vector. Clustering the reduced-dimensional feature vectors using a neural network model; The clustering loss is combined with the KL divergence and reconstruction error to optimize the model and obtain a variational autoencoder model based on the spatiotemporal network.

5. The method according to claim 1, characterized in that Based on the preliminary clustering results, the MiniBatch K-Means algorithm is used to perform cluster analysis on the power load data to obtain accurate clustering results, including: Determine the optimal clustering K value within the set K value range; Based on the determined optimal K value, the Mini Batch K-Means algorithm is run to process the power load data; After the algorithm converges, the cluster centers are extracted from the final clustering results to obtain accurate clustering results.

6. The method according to claim 1, characterized in that Also includes: Based on the clustering results, the analysis results of the similarities and differences between the power load data of different users are generated.

7. A user load clustering system based on VMD and spatiotemporal network variational word coding, characterized in that: include: A data acquisition module is used to acquire the user's power load data and perform preprocessing; A decomposition and reconstruction module is used to decompose the preprocessed power load data using a variational mode decomposition algorithm to obtain multiple intrinsic mode components, and perform mode reconstruction based on the intrinsic mode components; The first clustering module is used to extract potential features of the power load data after modal reconstruction using a variational autoencoder model based on a spatiotemporal network to obtain preliminary clustering results; The second clustering module is used to perform cluster analysis on the power load data based on the preliminary clustering results by using the Mini Batch K-Means algorithm to obtain accurate clustering results.

Images