CN115526264A - User power consumption behavior classification analysis method based on self-encoder - Google Patents
User power consumption behavior classification analysis method based on self-encoder Download PDFInfo
- Publication number
- CN115526264A CN115526264A CN202211265987.0A CN202211265987A CN115526264A CN 115526264 A CN115526264 A CN 115526264A CN 202211265987 A CN202211265987 A CN 202211265987A CN 115526264 A CN115526264 A CN 115526264A
- Authority
- CN
- China
- Prior art keywords
- data
- user
- electricity utilization
- encoder
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000010224 classification analysis Methods 0.000 title claims abstract description 25
- 230000005611 electricity Effects 0.000 claims abstract description 101
- 230000006399 behavior Effects 0.000 claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 claims abstract description 37
- 238000011156 evaluation Methods 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000013145 classification model Methods 0.000 claims abstract description 7
- 238000004140 cleaning Methods 0.000 claims abstract description 6
- 230000002354 daily effect Effects 0.000 claims description 33
- 238000012360 testing method Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 3
- 230000003203 everyday effect Effects 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 abstract description 8
- 230000006870 function Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 11
- 238000003860 storage Methods 0.000 description 10
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000004590 computer program Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a user power consumption behavior classification analysis method based on a self-encoder, which comprises the following steps: s1, acquiring historical electricity consumption data of a user through an intelligent ammeter, and cleaning the data; s2, constructing a feature extraction model based on an incomplete self-encoder, and selecting a proper encoding ratio beta in an iterative mode; s3, constructing a characteristic optimization evaluation index, and obtaining an optimal user electricity utilization characteristic set by adopting a heuristic search method; s4, constructing a BP neural network-based user electricity utilization behavior classification model, taking the coding characteristics and the optimal electricity utilization characteristics as training characteristics of the BP neural network, taking the actual electricity utilization type of the user as a training label, and training the BP neural network; and S5, inputting the coding features and the optimal power utilization features of the users to be classified into the trained BP neural network to obtain the classification results of the power utilization behaviors of the users. The method can effectively improve the accuracy of classification and has higher calculation efficiency.
Description
Technical Field
The invention belongs to the technical field of user power consumption behavior analysis, relates to a user power consumption behavior classification analysis method, and particularly relates to a user power consumption behavior classification analysis method based on an autoencoder.
Background
In recent years, with the rapid development of smart power grids and the continuous improvement of informatization level, the construction of a user electricity utilization information acquisition system tends to be perfect. The intelligent electric meter is one of basic measuring devices of a power utilization information acquisition system, undertakes the tasks of acquisition, measurement and transmission of electric power data, and has a simple data analysis function. The popularization of the intelligent electric meter brings mass power consumption data for an electric power company, classification analysis is carried out on resident power consumption behaviors on the basis, more refined user power consumption characteristics can be mastered, and the demand response potential is fully excavated.
At present, scholars at home and abroad have conducted a lot of research on a classification analysis method of user electricity utilization behaviors. For example, the method for classifying the daily load curves of the users by using the singular value decomposition technology has the advantages of short running time and good robustness; the method has the advantages that the inter-class optimization and the intra-class optimization are utilized to enhance the separability of the load data, and certain self-healing optimization capacity is achieved; aiming at massive user load data with high dispersity, a distributed clustering algorithm is realized, and the effectiveness and robustness of load curve clustering are obviously improved; classifying and identifying load users based on a fuzzy clustering algorithm and curve similarity, and providing a basis for accurate service of a power supply company; the load clustering analysis method based on EMD considering the load longitudinal randomness comprehensively represents the electricity utilization behavior of the user from two angles of the transverse direction and the longitudinal direction; a load classification method based on information entropy segmentation aggregation approximation and spectral clustering is adopted, so that a good load clustering effect is obtained; the accuracy of short-term load prediction is improved by using the number of the electricity utilization modes of the users, and technical support is provided for the rapid development of the electricity marketing demand response business; and a characteristic optimization strategy is adopted, the limit learning machine is used for carrying out classification analysis on the electricity consumption behaviors of residents, network parameters are optimized, and the accuracy of detection is improved. The research has good effect on the aspect of classification and analysis of the power utilization behaviors of the users, but the applicability of the method in a specific application scene is not fully considered.
Considering from practical application, if the user power consumption behavior classification analysis function can be integrated in the intelligent electric meter, the data processing capacity of the intelligent electric meter can be fully utilized, the intelligence of the electric meter is increased, the data analysis efficiency of dispatching personnel can be effectively improved, and the labor cost is reduced. However, due to the limitation of the hardware condition of the current smart meter, the advanced function integration needs to consider the data capacity and the calculation efficiency. Therefore, on the premise of ensuring the accuracy of the algorithm, a user electricity consumption behavior classification analysis method for reducing the data volume and improving the calculation efficiency is a necessary condition for actually applying the function in the intelligent electric meter.
Through searching, the published patent documents which are the same as or similar to the invention are not found.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a user power consumption behavior classification analysis method based on an autoencoder, and has the advantages of effectively improving classification accuracy and being higher in calculation efficiency.
The invention solves the practical problem by adopting the following technical scheme:
a classification analysis method for user power consumption behaviors based on an auto-encoder comprises the following steps:
s1, acquiring historical electricity utilization data of a user through an intelligent ammeter, and cleaning the data;
s2, constructing a feature extraction model based on an incomplete self-encoder, selecting a proper encoding ratio beta in an iterative mode, and encoding the intelligent electric meter data obtained in the step S1 by adopting a nonlinear means to obtain encoding features so as to realize feature extraction of original data;
s3, constructing a characteristic optimization evaluation index, and obtaining an optimal user electricity utilization characteristic set by adopting a heuristic search method;
s4, constructing a BP neural network-based user electricity utilization behavior classification model, taking the coding characteristics and the optimal electricity utilization characteristics as training characteristics of the BP neural network, taking the actual electricity utilization type of the user as a training label, and training the BP neural network;
and S5, inputting the coding features and the optimal electricity utilization features of the users to be classified into the trained BP neural network to obtain the classification results of the electricity utilization behaviors of the users.
In addition, the dimensionality of the historical electricity consumption data of the smart meter user in the step S1 is 48 dimensions, that is, the data sampling frequency is 30 min/meter, and 48 data points are included each day.
In addition, the specific step of iteratively selecting the appropriate coding ratio β in step S2 includes:
(1) The coding ratio β in step S2 is defined as:
in the formula: n is the original ammeter data dimension, and M is the characteristic data dimension obtained by coding.
(2) The specific steps of selecting the appropriate coding ratio β in an iterative manner include:
(1) determining data acquisition frequency f of electric meter according to model of intelligent electric meter c Calculating the dimensionality N of the daily load data point, and setting the number M =1 of the under-complete self-encoder middle hidden layer units;
(2) encoding the original data of the electric meter by using an incomplete self-encoder to obtain characteristic data F of a middle hidden layer, and taking the characteristic data F as the input of a BP neural network;
(3) according to certain standard, user electricity consumption behavior labels are formulated, network training is carried out, and F of the test set is calculated 1 Indexes;
(4) calculating beta value according to M and N, recording F 1 An index, a coding ratio beta and a data occupation space W;
(5) let M = M +1, judge M = N? If not, returning to the step (2); if yes, the process ends.
In addition, the preferable characteristic evaluation index in step S3 is:
in the formula: j (x) is an evaluation value of a single electricity utilization characteristic x, I' (x; c) is normalized mutual information of the electricity utilization characteristic x and a user class c, J (Y) is an evaluation function of a preferred characteristic subset Y, J (Y) is an evaluation function of a characteristic Y in the preferred characteristic subset, and rho xy Correlation coefficients of the electricity utilization characteristics x and y;
the user electricity consumption feature set in the step S3 is { maximum daily load, minimum daily load, average daily load, peak-valley difference daily, daily load rate, peak-hour electricity consumption rate, valley electricity coefficient, flat electricity consumption percentage }.
The invention has the advantages and beneficial effects that:
1. the invention provides a user electricity consumption behavior classification analysis method based on a self-encoder. On the basis, the optimal coding ratio is optimized, and the typical electricity utilization characteristics of the user are combined, so that the accuracy of classification is improved. Experiments prove that the user electricity consumption behavior classification analysis method designed by the invention has the advantages of high accuracy, small classification error, less used data amount, higher calculation efficiency and higher practical engineering application value.
2. The user electricity consumption behavior classification analysis method based on the self-encoder, which is constructed in the steps 2-4, performs feature extraction on original electricity meter data based on the incomplete self-encoder, constructs a classification model by combining typical electricity consumption features of users, and simultaneously uses recessive features and dominant features to realize feature enhancement and feature dimension reduction, so that the difficulty in model training is reduced, and the classification accuracy can be effectively improved.
3. The user power consumption behavior classification analysis method based on the incomplete self-encoder, which is constructed in the step 2, can mine key features from mass data, effectively reduce data dimensionality and required storage space, save model training calculation time, improve algorithm accuracy by reducing redundancy, and has high practical engineering application value.
Drawings
FIG. 1 is a schematic diagram of the operation of a less complete auto-encoder (UAE) of the present invention;
FIG. 2 is a flowchart of a user electricity usage behavior classification analysis model of step 2-4 of the present invention;
FIG. 3 is a beta-F1 relationship diagram of the present invention.
Detailed Description
The embodiments of the invention will be described in further detail below with reference to the accompanying drawings:
a classification analysis method for user power consumption behaviors based on an auto-encoder comprises the following steps:
s1, acquiring historical electricity utilization data of a user through an intelligent ammeter, and cleaning the data;
the dimensionality of the historical electricity consumption data of the user of the intelligent ammeter in the step S1 is 48 dimensions, namely the data sampling frequency is 30 min/meter, and 48 data points are contained every day.
S2, constructing a feature extraction model based on an incomplete self-encoder, selecting a proper encoding ratio beta in an iterative mode, and encoding the intelligent electric meter data obtained in the step S1 by adopting a nonlinear means to obtain encoding features so as to realize feature extraction of original data;
the less-than-complete self-encoder in step S2 is constructed using the tenserflow, keras deep learning toolkit in python programming language.
The network structure of the under-complete self-encoder in the step S2 includes three layers, which are an input layer, a hidden layer and an output layer, respectively, the number of neurons in the input layer and the number of neurons in the output layer are both 128, and the number of neurons in the hidden layer is 48 × β.
The training method of the less-than-complete self-encoder in the step S2 comprises the following steps: and (3) taking daily load data of the user as input and output of the under-complete self-encoder at the same time, training the self-encoder, and realizing feature extraction of the original data by the middle hidden layer.
The specific step of selecting a suitable coding ratio β in an iterative manner in step S2 includes:
(1) The coding ratio β in step S2 is defined as:
in the formula: n is the original ammeter data dimension, and M is the characteristic data dimension obtained by coding.
The larger the beta value is, more data characteristics can be reserved, but the larger the required storage capacity is, the applicability on the intelligent electric meter is reduced; the beta value is too small, and important data characteristics are easily lost, so that the accuracy of the classification of the power utilization behaviors of the user is reduced. Only by selecting a reasonable beta value, the detection accuracy and the storage space occupied by the data can be considered.
(2) The specific steps of selecting the appropriate coding ratio β in an iterative manner include:
(1) determining data acquisition frequency f of electric meter according to model of intelligent electric meter c Calculating the dimensionality N of the daily load data point, and setting the number M =1 of the under-complete self-encoder middle hidden layer units;
(2) encoding the original data of the electric meter by using an incomplete self-encoder to obtain characteristic data F of a middle hidden layer, and taking the characteristic data F as the input of a BP neural network;
(3) the user electricity consumption behavior label is formulated according to a certain standard, network training is carried out, and F of the test set is calculated 1 Indexes;
(4) calculating beta value according to M and N, recording F 1 An index, a coding ratio beta and a data occupation space W;
(5) let M = M +1, judge M = N? If not, returning to the step (2); if yes, the process ends.
S3, constructing a feature optimal evaluation index, and obtaining an optimal user electricity utilization feature set by adopting a heuristic search method;
the preferable characteristic evaluation indexes in step S3 are:
in the formula: j (x) is an evaluation value of a single electricity utilization characteristic x, I' (x; c) is normalized mutual information of the electricity utilization characteristic x and a user class c, J (Y) is an evaluation function of a preferred characteristic subset Y, J (Y) is an evaluation function of a characteristic Y in the preferred characteristic subset, and rho xy Is the correlation coefficient of the power usage characteristics x and y. By constructing different feature sets and adopting a heuristic search method, the optimal user electricity utilization feature set can be searched according to the feature optimal selection evaluation indexes.
The user electricity utilization feature set in the step S3 is { daily maximum load, daily minimum load, daily average load, daily peak-valley difference rate, daily load rate, peak-hour electricity consumption rate, valley electricity coefficient, flat section electricity utilization percentage }.
S4, constructing a BP neural network-based user electricity utilization behavior classification model, taking the coding characteristics and the optimal electricity utilization characteristics as training characteristics of the BP neural network, taking the actual electricity utilization type of the user as a training label, and training the BP neural network;
and S5, inputting the coding features and the optimal electricity utilization features of the users to be classified into the trained BP neural network to obtain the classification results of the electricity utilization behaviors of the users.
The working principle of the invention is as follows:
the invention provides a user power consumption behavior classification analysis method based on an autoencoder. Firstly, acquiring historical electricity utilization data of a user through an intelligent ammeter, and cleaning the data; then constructing a feature extraction model based on an incomplete self-encoder, selecting a proper encoding ratio by adopting an iteration mode, and encoding the data of the intelligent electric meter by adopting a nonlinear means to obtain encoding features so as to realize feature extraction of the original data; then constructing a characteristic optimization evaluation index, and obtaining an optimal user electricity utilization characteristic set by adopting a heuristic search method; and finally, constructing a user electricity utilization behavior classification model based on the BP neural network, taking the coding characteristics and the optimal electricity utilization characteristics as training characteristics of the BP neural network, taking the actual electricity utilization type of the user as a training label, training the BP neural network, and inputting the coding characteristics and the optimal electricity utilization characteristics of the user to be classified into the trained BP neural network to obtain a user electricity utilization behavior classification result.
Example 1
In this embodiment, an irish smart meter data set is used as an experimental object, and the method for classifying and analyzing the user electricity consumption behavior based on the self-encoder of the present invention is used for classifying and analyzing the user electricity consumption behavior, and includes the following steps:
s1, acquiring historical electricity utilization data of a user through an intelligent ammeter, and cleaning the data;
s2, constructing a feature extraction model based on an incomplete self-encoder, selecting a proper encoding ratio beta in an iterative mode, and encoding the intelligent electric meter data obtained in the step S1 by adopting a nonlinear means to obtain encoding features so as to realize feature extraction of original data;
an Auto-Encoder (AE) is a three-layer neural network unsupervised learning algorithm composed of an Encoder and a decoder, and an input signal X is encoded to obtain a new signal Y through nonlinear mapping of an intermediate hidden layer and then decoded back to X. Therefore, the input form and the output form of the self-encoder are almost the same. If the dimension of the hidden layer is larger than that of the input layer, an overcomplete self-encoder is obtained. If the dimension of the constrained implicit layer is smaller than the input dimension, an incomplete Auto-encoder (UAE) can be obtained, and learning the incomplete representation will force the Auto-encoder to capture the most significant features in the training data. The UAE operating principle is shown in fig. 1. The less-than-complete self-coding can well overcome the defect that a common self-coder is easy to over-fit, and the middle hidden layer can perform efficient feature extraction, but the problem exists in that the reconstruction process is difficult due to the fact that the number of units of the middle hidden layer is extremely small. When only the first half part of the network is used, the advantages of the incomplete self-encoder can be exerted, and the dimension reduction of the data of the intelligent electric meter is realized. Compared with linear transformation between each layer of traditional Principal Component Analysis (PCA), the nonlinear transformation adopted by the incomplete self-encoder can learn more important and more comprehensive data characteristics, high-dimensional original data collected by the intelligent electric meter is reduced to low-dimensional data capable of retaining the important characteristics, the redundancy of electric power data is effectively reduced, the storage space required by the data is reduced, the operation efficiency of the algorithm is improved, and the application scene on the intelligent electric meter is adapted.
When encoding using an under-complete autocoder, the coding ratio β is defined as
In the formula: n is the original ammeter data dimension, and M is the characteristic data dimension obtained by coding. The larger the beta value is, more data characteristics can be reserved, but the larger the required storage capacity is, the applicability on the intelligent electric meter is reduced; the beta value is too small, and important data characteristics are easily lost, so that the accuracy of the classification of the power utilization behaviors of the user is reduced. Only by selecting a reasonable beta value, the detection accuracy and the storage space occupied by the data can be considered.
F 1 The index is a common evaluation index in a multi-classification problem, gives consideration to classification accuracy and recall rate, and is defined as follows:
in the formula: pr is the accuracy rate, which means the ratio of true positive samples in the samples predicted to be positive; re is recall, which means the proportion of samples that are predicted to be positive among samples that are truly positive. Selection of F in the invention 1 The index is used as an evaluation index of the classification effect of the power consumption behaviors of the user and is detectedThe optimal coding ratio of the less-than-complete self-encoder is researched, and the design scheme flow is as follows:
1) Determining data acquisition frequency f of electric meter according to model of intelligent electric meter c Calculating the dimensionality N of daily load data points, and setting the number M =1 of the under-complete self-encoder middle hidden layer units;
2) Encoding the original data of the electric meter by using an incomplete self-encoder to obtain characteristic data F of a middle hidden layer, and taking the characteristic data F as the input of a BP neural network;
3) The user electricity consumption behavior label is formulated according to a certain standard, network training is carried out, and F of the test set is calculated 1 Indexes;
4) Calculating beta value according to M and N, recording F 1 An index, a coding ratio beta and a data occupation space W;
5) Let M = M +1, judge M = N? If not, returning to the step 2); if yes, the process ends.
S3, constructing a characteristic optimization evaluation index, and obtaining an optimal user electricity utilization characteristic set by adopting a heuristic search method;
the daily load curve of the user contains some key electricity utilization characteristics, and the electricity utilization characteristics of the user, such as daily maximum load, daily peak-valley difference, daily load rate and the like, can be effectively reflected. The feature extraction is carried out on the electric meter data through the incomplete self-encoder, so that the storage capacity is effectively reduced, the calculation efficiency is improved, and the accuracy of partial detection is sacrificed. Therefore, the model is optimized by combining the typical electricity utilization characteristics of the user, and the model and the coded data are used as the input of the BP neural network together, so that the classification accuracy is improved.
The Feature Selection Strategy (FSS) is an important method for studying closeness between the electricity consumption characteristic index and the electricity consumption behavior of the user. The method carries out quantitative analysis on the electricity consumption behavior characteristics based on mutual information and correlation coefficients, comprehensively considers the effectiveness and complementarity of the electricity consumption information characteristics on the analysis performance, and constructs characteristic optimization evaluation indexes as
In the formula: j (x) is an evaluation value of a single electricity utilization characteristic x, I' (x; c) is normalized mutual information of the electricity utilization characteristic x and a user class c, J (Y) is an evaluation function of a preferred characteristic subset Y, J (Y) is an evaluation function of a characteristic Y in the preferred characteristic subset, and rho xy Is the correlation coefficient of the power utilization characteristics x and y. By constructing different feature sets and adopting a heuristic search method, the optimal user electricity utilization feature set can be searched according to the feature optimal selection evaluation indexes.
The method selects a user electricity utilization feature set as { daily maximum load, daily minimum load, daily average load, daily peak-valley difference rate, daily load rate, peak-time power consumption rate, valley power coefficient and flat section electricity utilization percentage }, adopts a feature optimization strategy to screen out three typical features from the feature set, and uses the three typical features together with coded data as the input of a BP neural network for network training.
S4, constructing a BP neural network-based user electricity utilization behavior classification model, taking the coding features and the optimal electricity utilization features as training features of the BP neural network, taking the actual electricity utilization type of the user as a training label, and training the BP neural network, wherein the algorithm flow is shown in figure 2.
And S5, inputting the coding features and the optimal electricity utilization features of the users to be classified into the trained BP neural network to obtain the classification results of the electricity utilization behaviors of the users.
By way of example, in the present embodiment, an ireland smart meter data set is used as an experimental object, and the data set covers 536 groups of daily load curves of 4000 residential users, each daily load curve is 48 points, namely, the collection frequency of the meter is 30 minutes/time. The following are detailed experimental results:
the Ireland resident user loads are divided into 10 types and used for BP neural network training, and the classification details are shown in a table 1.
TABLE 1 Ireland resident load types
Optimization of the coding ratio beta, plotting beta-F 1 The relationship diagram is shown in fig. 3. As can be seen from FIG. 3, F for the training set and the test set 1 The index increases as the coding ratio β increases. The inflection point of the curve is located at approximately β =0.33, and before the inflection point, the curve is steeper, F 1 The value increases faster; after reaching the inflection point, the curve becomes very gentle and the F1 index slowly increases. Therefore, the present invention selects 0.33 as the optimal coding ratio, and the use of the less-than-complete auto-encoder to encode the original meter data results in F 1 The loss of value will be compensated by model optimization.
And applying a characteristic optimization strategy under the optimal coding ratio to obtain typical user electricity utilization characteristics of { average daily load, valley power coefficient and percentage of flat electricity utilization }, using the typical user electricity utilization characteristics and coded data as input of the BP neural network for network training, and performing classification analysis on the user electricity utilization behaviors.
In order to verify the accuracy of the method, this embodiment is compared with a combined method of a BP neural network, a Random Forest (RF), a Support Vector Machine (SVM), and a BP neural network and an under-complete self-encoder, where experimental results are shown in table 2 and effect comparison is shown in table 3.
TABLE 2 comparison of the results
TABLE 3 comparison of the results of the methods
Note: the results in Table 3 are the percentage of the effect of the method of the invention that was improved (or reduced).
As can be seen from tables 2 and 3, the indexes obtained by using the original electric meter data in combination with the BP neural network method to perform the classification analysis of the electricity consumption behavior of the user are poor in performance, low in prediction accuracy and efficiency, and large in data occupation space. When the electric meter data is encoded by the incomplete self-encoder, the training time, the testing time and the occupied space of the data of the program are obviously reduced, but partial prediction accuracy is lost, and the lost accuracy on the training set and the testing set is 0.99% and 0.89% respectively. The Random Forest (RF) method improves the accuracy of classification to 92.01% and 91.85 on the training and test sets, respectively, but requires longer training and testing times. Although the training time and the testing time of the Support Vector Machine (SVM) method are short, the accuracy is not obviously improved. The method provided by the invention combines the less-complete self-encoder with the typical characteristics of the user, the accuracy rates on the training set and the test set respectively reach 94.34% and 94.26, which are respectively improved by 4.21% and 4.72% compared with the traditional BP neural network, and the method has the best performance in the five methods. Meanwhile, the data occupied space is reduced by 64.74%, and the training time and the prediction time are also reduced remarkably, namely 17.64s and 0.16s respectively. The result shows that the method provided by the invention can effectively improve the accuracy and the calculation efficiency of the classification of the power utilization behaviors of the user while saving the data storage space.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Claims (4)
1. A classification analysis method for user power consumption behaviors based on an autoencoder is characterized in that: the method comprises the following steps:
s1, acquiring historical electricity consumption data of a user through an intelligent ammeter, and cleaning the data;
s2, constructing a feature extraction model based on an incomplete self-encoder, selecting a proper encoding ratio beta in an iterative mode, and encoding the intelligent electric meter data obtained in the step S1 by adopting a nonlinear means to obtain encoding features so as to realize feature extraction of original data;
s3, constructing a feature optimal evaluation index, and obtaining an optimal user electricity utilization feature set by adopting a heuristic search method;
s4, constructing a BP neural network-based user electricity utilization behavior classification model, taking the coding characteristics and the optimal electricity utilization characteristics as training characteristics of the BP neural network, taking the actual electricity utilization type of the user as a training label, and training the BP neural network;
and S5, inputting the coding features and the optimal electricity utilization features of the users to be classified into the trained BP neural network to obtain the classification results of the electricity utilization behaviors of the users.
2. The self-encoder-based classification analysis method for the power utilization behavior of the user according to claim 1, characterized in that: the dimensionality of the historical electricity consumption data of the user of the intelligent ammeter in the step S1 is 48 dimensions, namely the data sampling frequency is 30 min/meter, and 48 data points are contained every day.
3. The self-encoder-based classification analysis method for the power utilization behavior of the user according to claim 1, characterized in that: the specific step of selecting a suitable coding ratio β in an iterative manner in step S2 includes:
(1) The coding ratio β in step S2 is defined as:
in the formula: n is the original electric meter data dimension, and M is the feature data dimension obtained by coding.
(2) The specific steps of selecting the appropriate coding ratio β in an iterative manner include:
(1) determining data acquisition frequency f of electric meter according to model of intelligent electric meter c Calculating the dimensionality N of the daily load data point, and setting the number M =1 of the under-complete self-encoder middle hidden layer units;
(2) encoding the original data of the electric meter by using an incomplete self-encoder to obtain characteristic data F of a middle hidden layer, and taking the characteristic data F as the input of a BP neural network;
(3) according to a certain proportionThe standard of (2) sets up a user electricity consumption behavior label, performs network training, and calculates F of a test set 1 Indexes;
(4) calculating beta value according to M and N, recording F 1 An index, a coding ratio beta and a data occupation space W;
(5) let M = M +1, judge M = N? If not, returning to the step (2); if yes, the process ends.
4. The self-encoder-based classification analysis method for the power utilization behavior of the user according to claim 1, characterized in that: the characteristic optimization evaluation indexes in step S3 are:
in the formula: j (x) is an evaluation value of a single electricity utilization characteristic x, I' (x; c) is normalized mutual information of the electricity utilization characteristic x and a user class c, J (Y) is an evaluation function of a preferred characteristic subset Y, J (Y) is an evaluation function of a characteristic Y in the preferred characteristic subset, and rho xy Correlation coefficients of the electricity utilization characteristics x and y;
the user electricity utilization feature set in the step S3 is { daily maximum load, daily minimum load, daily average load, daily peak-valley difference rate, daily load rate, peak-hour electricity consumption rate, valley electricity coefficient, flat section electricity utilization percentage }.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211265987.0A CN115526264A (en) | 2022-10-13 | 2022-10-13 | User power consumption behavior classification analysis method based on self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211265987.0A CN115526264A (en) | 2022-10-13 | 2022-10-13 | User power consumption behavior classification analysis method based on self-encoder |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115526264A true CN115526264A (en) | 2022-12-27 |
Family
ID=84701537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211265987.0A Pending CN115526264A (en) | 2022-10-13 | 2022-10-13 | User power consumption behavior classification analysis method based on self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115526264A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117035837A (en) * | 2023-10-09 | 2023-11-10 | 广东电力交易中心有限责任公司 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
-
2022
- 2022-10-13 CN CN202211265987.0A patent/CN115526264A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117035837A (en) * | 2023-10-09 | 2023-11-10 | 广东电力交易中心有限责任公司 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
| CN117035837B (en) * | 2023-10-09 | 2024-01-19 | 广东电力交易中心有限责任公司 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Sparse and redundant representation-based smart meter data compression and pattern extraction | |
| CN110781332A (en) | Electric power resident user daily load curve clustering method based on composite clustering algorithm | |
| CN111008726B (en) | Class picture conversion method in power load prediction | |
| CN110689162B (en) | Bus load prediction method, device and system based on user side classification | |
| CN111368904A (en) | Electrical equipment identification method based on electric power fingerprint | |
| CN111563827A (en) | Load decomposition method based on electrical appliance physical characteristics and residential electricity consumption behaviors | |
| CN115526264A (en) | User power consumption behavior classification analysis method based on self-encoder | |
| CN115271161A (en) | Short-term prediction method for multi-energy load | |
| CN114548586A (en) | Short-term power load prediction method and system based on hybrid model | |
| CN111612319A (en) | Load curve depth embedding clustering method based on one-dimensional convolution self-encoder | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN114202012A (en) | High-dimensional load clustering method based on recursive graph and convolution self-encoder | |
| Miraftabzadeh et al. | Knowledge Extraction from PV Power Generation with Deep Learning Autoencoder and Clustering-Based Algorithms | |
| CN113033089B (en) | Method and device for identifying abnormal electricity consumption user | |
| Han et al. | Non‐intrusive load monitoring by voltage–current trajectory enabled asymmetric deep supervised hashing | |
| CN114358367A (en) | Non-invasive load decomposition method and device | |
| Wang et al. | Evsense: A robust and scalable approach to non-intrusive ev charging detection | |
| CN116777066B (en) | Photovoltaic power probability prediction method and device based on foundation cloud image guided depth decomposition | |
| CN116338322A (en) | Power grid line impedance prediction method and system | |
| CN116561569A (en) | Industrial power load identification method based on EO feature selection and AdaBoost algorithm | |
| CN116304295A (en) | User energy consumption portrait analysis method based on multivariate data driving | |
| Liu et al. | A clustering-based feature enhancement method for short-term natural gas consumption forecasting | |
| Yan et al. | A clustering method for power time series curves based on improved self-organizing mapping algorithm | |
| He et al. | Residential power user segmentation based on k-means clustering method in the context of big data | |
| CN117833243B (en) | Method, system, equipment and medium for predicting short-term demand of electric power |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |