CN117035837A - Method for predicting electricity purchasing demand of power consumer and customizing retail contract - Google Patents
Method for predicting electricity purchasing demand of power consumer and customizing retail contract Download PDFInfo
- Publication number
- CN117035837A CN117035837A CN202311299832.3A CN202311299832A CN117035837A CN 117035837 A CN117035837 A CN 117035837A CN 202311299832 A CN202311299832 A CN 202311299832A CN 117035837 A CN117035837 A CN 117035837A
- Authority
- CN
- China
- Prior art keywords
- retail
- sample
- electricity
- samples
- feature
- 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.)
- Granted
Links
- 230000005611 electricity Effects 0.000 title claims abstract description 107
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000007246 mechanism Effects 0.000 claims abstract description 20
- 239000011159 matrix material Substances 0.000 claims description 77
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 10
- 238000012216 screening Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 230000009467 reduction Effects 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000010801 machine learning Methods 0.000 claims description 5
- 101100533306 Mus musculus Setx gene Proteins 0.000 claims description 3
- 101100273916 Schizosaccharomyces pombe (strain 972 / ATCC 24843) wip1 gene Proteins 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000001172 regenerating effect Effects 0.000 claims description 3
- 238000007405 data analysis Methods 0.000 abstract description 5
- 229910052799 carbon Inorganic materials 0.000 abstract description 3
- 238000010276 construction Methods 0.000 abstract description 2
- 238000005070 sampling Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/18—Legal services; Handling legal documents
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Tourism & Hospitality (AREA)
- Accounting & Taxation (AREA)
- General Business, Economics & Management (AREA)
- Finance (AREA)
- Artificial Intelligence (AREA)
- Development Economics (AREA)
- Evolutionary Computation (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Primary Health Care (AREA)
- Evolutionary Biology (AREA)
- Human Resources & Organizations (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Water Supply & Treatment (AREA)
- Technology Law (AREA)
- Public Health (AREA)
- Game Theory and Decision Science (AREA)
- Probability & Statistics with Applications (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for forecasting electricity purchasing demand of an electric power user and customizing retail contracts. According to the method, the user is described from three dimensions of the electric power user attribute, the electric power consumption behavior of the user and the user transaction preference, the retail user is finely classified in a multi-label mode, so that more-dimensional user images are obtained, guidance for making retail contracts is provided based on the fine classification, an efficient electric power retail platform data analysis function is realized, an electric power selling company and an electric power transaction mechanism are helped to expand the consumption channel of an electric power retail market, the transaction amount of retail packages is expanded, and a foundation is provided for continuous promotion of electric power retail market construction and realization of a double-carbon target.
Description
Technical Field
The invention relates to the technical field of green electric power market data analysis, in particular to a method for predicting electricity purchasing demand of electric power users and customizing retail contracts.
Background
In recent years, low-carbon operation has become a trend of future development. As the power market reforms and advances, it is important to push power consumers to participate in the power trade market and embody the key role of the market in resource allocation. However, as the power consumer scale continues to expand and the power demand progresses toward diversification, the market behaviors of the power consumer become more complex. Therefore, how to deeply analyze and understand the increasingly complex transaction behavior patterns of retail power users under new situation, predict the transaction demands of the users, improve the activity of retail package transactions, provide decision references for electric power companies and electric power transaction institutions, and become a problem to be solved in the demand side.
In the existing power retail market data analysis technology, comprehensive and deep analysis is not performed on the transaction demand prediction of retail power consumers, and multi-type and multi-dimensional data related to the power consumers in a power retail platform are difficult to effectively utilize and mine. Therefore, in the environment of the power retail market, the power consumer electricity purchasing demand prediction and retail contract customization method can overcome the problems to a certain extent, further achieve the high-efficiency and visual power consumer data analysis function, and improve the retail market transaction level, the new energy consumption level and the electricity selling company decision level.
In the prior art, a recommendation method, a system, equipment and a storage medium for electric power retail electric charge packages are provided, wherein the method comprises the steps of dividing users in a sample set into different basic electricity utilization types of the users according to electricity utilization attributes; according to the package selection, the selection times of the power retail electric charge packages are used as scoring basis, and a user preference scoring matrix of the basic electricity utilization type of the user is constructed; on the basis of the basic electricity utilization type of the user, carrying out secondary classification on the user according to the load data of the user, and establishing user subdivision categories; adjusting a user preference scoring matrix, sequencing the electric power retail electric charge packages, and selecting N high scoring packages; and determining the basic electricity utilization type of the user according to the electricity utilization attribute of the target user, calculating the distance between the target user and the clustering center of each subdivision category, selecting the subdivision category closest to the target user, and taking N high-scoring packages corresponding to the subdivision category as candidate package sets of the target user. However, in actual operation, the transaction platform is not allowed to provide unique package or recommendation actions of the electricity selling company for the user; meanwhile, the method classifies the users only based on the user load curve, and does not consider the user description angles of other dimensions such as regions, industries and the like, and the classification method lacks certain comprehensiveness (Liang Bo, wang Xin, jie Lei and the like; recommendation method, system, equipment and storage medium [ P ] of electric power retail electric charge packages, shandong province: CN116596640A, 2023-08-15).
Disclosure of Invention
In order to solve the technical problems, the invention provides the electricity purchasing demand prediction and retail contract customization method for the electric power consumers, which realizes the data analysis function of the efficient electric power retail platform, helps the electric power selling companies and the electric power transaction institutions to expand the consumption channels of the electric power retail market, enlarges the transaction amount of retail packages, and provides a foundation for the continuous promotion of the construction of the electric power retail market and the realization of the double-carbon targets.
The object of the invention is achieved by at least one of the following technical solutions.
A method for predicting electricity purchasing demand of power consumers and customizing retail contracts comprises the following steps:
s1, a retail power user characteristic matrix is initially constructed from 8 layers of retail transaction contract types, power utilization categories, industry classifications, user categories, power utilization levels, home power supply bureaus, power utilization conditions and transaction times;
s2, combining SMOTE (Synthetic Minority Over-sampling Technique) algorithm withkThe neighbor algorithm performs class unbalance processing on the transaction sample data of the retail power user;
s3, performing dimension reduction processing on the retail power consumer transaction sample data subjected to the class unbalance processing by adopting an automatic encoder;
s4, clustering the retail power consumer transaction sample data processed by the automatic encoder by adopting a K-means (K-means clustering) method, performing clustering effect judgment by adopting a contour coefficient method, and constructing a retail power consumer multidimensional label based on the clustering effect;
s5, optimizing the super parameters of LightGBM (Light Gradient Boosting Machine) by adopting a Bayesian optimization algorithm to obtain an optimized super parameter combination;
s6, performing retail power consumer feature optimization by a machine learning algorithm based on the LightGBM;
s7, constructing a neural network model based on an attention mechanism BiLSTM (Bidirectional Long Short-Term Memory);
s8, predicting retail power consumer transaction requirements and customizing retail contracts based on a LightGBM model and a neural network model based on an attention mechanism BiLSTM.
Further, in step S1, the method specifically includes the following steps:
s1.1, the text characteristics adopt label coding, each label is mapped to an integer value, and the increment is started from 0; the text class features include retail transaction contract type, electricity class, industry classification, user class, electricity class, and home power office;
s1.2, the electricity consumption condition comprises two characteristic variables of electricity consumption and electricity price, and the statistical characteristic calculation is carried out on the two characteristic variables of the electricity consumption and the electricity price, wherein the statistical characteristic calculation comprises the mean value, the variance and the median of the variables so as to expand the characteristics of data;
s1.3, converting each characteristic data into a value on a [0,1] interval by adopting a minimum-maximum normalization method, namely:
wherein,l max is the maximum value in the data and,l min as the minimum value in the data,is the converted data value;
s1.4, after the treatment, finally obtaining the firstqPersonal retail power consumer feature matrixx q ={l q,1 ,l q,2 ,...,l q,a0 }, whereina 0 For the initial feature quantity, at the same timeq={1,2,...,b}，bFor the total number of retail power customers,l q,p is the firstqFirst retail consumer of electricitypThe number of feature vectors is chosen to be the same,p={1,2,...,a 0 and get the total feature matrixA 0 ={x 1 ,x 2 ,...,x b }；
S1.5, in order to evaluate the correlation between each feature and other features, if a certain feature has no too great correlation with other features, the feature can be determined as a redundant feature, and the Pearson correlation coefficient method is adopted to evaluate the total feature matrixA 0 First of all retail power consumerspPerforming linear correlation between any two of the feature vectors, and performing feature preliminary screening, wherein the value interval of the Pearson correlation coefficient is [ -1,1]-1 represents a complete negative correlation, +1 represents a complete positive correlation; two characteristic variablesZAndYthe Pearson correlation coefficient calculation formula of (2) is:
、/>as a characteristic variable, a characteristic variable is used,p={1,2,...,a 0 }，rfor Pearson correlation coefficient, +.>And->Setting a threshold value of 0 as a feature screening condition for the variable mean value, judging features with Pearson correlation coefficients below 0 as redundant features, and removing the redundant features to generate an initial total user transaction feature matrixA={x 1 ,x 2 ,...,x b }，x={l 1 ,l 2 ,...,l a }，aIs the feature quantity after preliminary screening.
Further, in step S1.2, the mean value of the variables is specifically as follows:
wherein,for the average value of the electricity consumption or the electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n}，nis the total number of samples;
the variance of the variables is specifically as follows:
wherein,for the variance of the electricity consumption or electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n-a }; for a pair ofnThe electricity consumption or electricity price of each sample is ordered in descending order, and the number of bits is taken asl Med 。
Further, in step S2, the SMOTE algorithm is combined withkNeighbor algorithm is to initial total user transaction characteristic matrixA={x 1 ,x 2 ,...,x b Performing class unbalance processing, and assisting with sample class weights, wherein the specific method is as follows:
s2.1, acquiring a power consumer sample set from a transaction center, and finding out all minority samples from the power consumer sample setx i ，i=1~bA minority class sample is calculatedx i With other minority sample pointsx j Euclidean distance between:
other minority sample pointsx j And minority class samplesx i The Euclidean distance between the two is smaller than the set threshold value, and the samples are few samplesx i Is used to determine the neighbor of a (c),j=1~bobtaining minority class samplesx i A kind of electronic devicek 1 Neighbor samples and are marked asx i,near ，nearThe values were 1,2,3,k 1 ，k 1 is the number of neighbor samples and is based onk 1 The individual neighbor samples form a corresponding classification；
S2.2 slavex i A kind of electronic devicek 1 Randomly selecting one sample from among the neighbor samplesx i,r ，r=1~k 1 Regenerating a random number between 0 and 1On the basis, a new sample is synthesizedx new1 Wherein:
wherein, each time a neighbor sample is randomly selected, a new random number is corresponding to；
S2.3, repeating the step S2.2B 1 Next, obtainB 1 New samples, noted asx new,i ，i=1,2,...,B 1 Forming an expanded retail power consumer feature matrix，b'=b+B 1 The method comprises the steps of carrying out a first treatment on the surface of the Expanded retail power consumer feature matrix>In the first b elements are {x 1 ,x 2 ,...,x b Post (back)B 1 The individual elements are new samples and willB 1 New samplesx new,i Categorizing into respective corresponding samplesx i,r The category to which it belongs;
s2.4, input training setWherein->For different classification of the sample, the extended retail power consumer feature matrix>As training data;
s2.5, finding a sample from the sample datax i Nearest 3 points covering the 3 pointsx i The field is recorded asN 3 (x i ) Next, inN 3 (x i ) Wherein the decision is re-determined based on classification decision rulesxCategory of (2)：
Wherein, let theθ=1,...,w，ρ=1,...,wThe method comprises the steps of carrying out a first treatment on the surface of the Then do not meet the orderρ=1,...,wIs redundant extension sample;
s2.6, retail power consumer feature matrix after expansionRemoving redundant amplified samples to form a new retail power consumer characteristic matrix, and forming a new retail power consumer characteristic matrix +.>，b''=b+B 1 -B 2 ，B 2 Is the total number of redundant amplified samples.
Further, in step S3, the method specifically includes the following steps:
s3.1, new retail power consumer feature matrixInputting the samples in the automatic encoder network, and carrying out forward and backward propagation on the network until the set training times or convergence degree are reached;
s3.2, outputting the user characteristic matrix after dimension reduction，b' is the user characteristic matrix after dimension reduction +.>Is a sample count of the total number of samples in the sample.
Further, in step S4, the K-means method is adopted to cluster the retail power consumer transaction sample data processed by the automatic encoder, and the contour coefficient method is adopted to judge the clustering effect, and the retail power consumer multidimensional label is constructed based on the clustering effect, and the specific method is as follows:
s4.1, user characteristic matrix after dimension reductionRandom access in sample data in the sample datak 2 The individual samples are taken as cluster centers and are marked as +.>；
S4.2, adopting Euclidean distance as similarity measurement in the K-means method, wherein the loss function is the error square sum of each sample distance from the cluster center point, and the error square sum is specifically as follows:
wherein,e Ь representing a samplex i The cluster to which the cluster belongs is selected,Ьfor the number of clusters to be the number of clusters,representing clusterse Ь The corresponding center of the two-dimensional space is provided with a plurality of grooves,b' indicates the number of samples;
s4.3, orderFor the iterative step number, the following process is repeated until the algorithm calculation converges:
s4.3.1 for samplesx i Assigning it to the center nearest to it:
wherein,represent the firsttMultiple iterationsx i Cluster to which>Represent the firsttSub-iteration clusterCorresponding firstk 2 A center;
s4.3.2, for the center of each class, recalculate:
；
s4.4, the contour coefficient consists of two scores defined by the distance, and the samplex i The category is marked asThe nearest category is noted asC k Samples ofx i The corresponding contour coefficient is recorded asS(i)，d(x i ,C) For the samplex i The distance from the clustering center is calculated by the following method:
when there is only one point in the cluster, the contour coefficients are definedS(i) The profile coefficients of a set of data sets are the arithmetic mean of all sample profile coefficients in the data set, the range of values is [ -1,1]-1 represents false clustering, 1 represents that the data set is well divided, and the vicinity of 0 represents that various overlapping degrees are higher; determining cluster number from profile coefficients and actual transaction requirementsE；
S4.5, giving each data sample according to the clustering resultx i Adding a tag matrixx={l 1 ,l 2 ,...,l a ,U}, wherein i=[h i,1 ,h i,2 ,...,h i,E ]Is the firstiThe vector labels of the individual retail power consumers,hfor the tag identification number, if the user belongs to a certain categorynumClassification of tagsh num =1，num=1,2,...,EOtherwise 0, at this timep={1,2,...,a,...,a+E}；
S4.6, forming a final retail power consumer feature matrix of the final package labelWhereinx’={l 1 ,l 2 ,...,l a ,U}={l 1 ,l 2 ,...,l a ,h 1 ,h 2 ,...,h E }。
Further, in step S6, the machine learning algorithm based on the LightGBM performs feature optimization, and the specific method thereof is as follows:
s6.1, according to the final retail power consumer characteristic matrixConstructing a power purchase demand prediction model based on the LightGBM by the optimal super-parameter combination obtained in the step S5;
s6.2, classifying and predicting based on the constructed electricity purchasing demand prediction model based on the LightGBM, and outputting the characteristic importance of each characteristic by the electricity purchasing demand prediction model based on the LightGBM;
s6.3, nonlinear relation second derivative formed by feature quantity and feature accumulated importancef’’And (3) determining the number of effective features according to the number of the features corresponding to the position (0), screening out the effective features according to the feature importance ranking and the number of the effective features, deleting the ineffective features, and forming a new feature matrix.
Further, in step S6.1, the LightGBM model is adjusted by the combination of the super-optimal parameters obtained in step S5, and the model is input as the final retail power consumer feature matrixThe output is a user feature importance value, and the model is a power purchase demand prediction model based on the LightGBM.
Further, in step S7, the attention mechanism BiLSTM model includes an input layer, a BiLSTM hiding layer, an attention mechanism layer, a fully connected layer, and an output layer, where the fully connected layer performs local retail power consumer feature integration, and uses a new feature combination as an input.
Further, in step S8, the retail power consumer feature matrix is input into the LightGBM model, the feature importance data is output, the retail power consumer feature matrix is input into the attention mechanism BiLSTM prediction model, and the user multidimensional labels and occurrence frequencies are output, so that the retail contract is comprehensively optimized according to the user features and labels.
Compared with the prior art, the invention has the advantages that:
the invention describes the user from three dimensions of the power user attribute, the user electricity behavior and the user transaction preference, performs the fine classification of the retail user in a multi-label mode, thereby obtaining more-dimensional user portraits and providing guidance for making retail contracts based on the fine classification.
Drawings
FIG. 1 is a schematic diagram of steps of a method for predicting electricity purchasing demand of a power consumer and customizing a retail contract according to an embodiment of the present invention;
FIG. 2 is a diagram of the LightGBM variable importance corners in an embodiment of the invention;
fig. 3 is a schematic structural diagram of a attention mechanism BiLSTM model in an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
Examples:
in one embodiment, based on the Guangdong power trading center retail market model, a power consumer electricity purchasing demand prediction and retail contract customization method, as shown in FIG. 1, comprises the steps of:
s1, preliminarily constructing a retail power user characteristic matrix from 8 layers of retail transaction contract types, power consumption classes, industry classifications, user classes, power consumption levels, home power supply bureaus, power consumption conditions and transaction times, wherein the method specifically comprises the following steps of:
s1.1, the text characteristics adopt label coding, each label is mapped to an integer value, and the increment is started from 0; the text class features include retail transaction contract type, electricity class, industry classification, user class, electricity class, and home power office; in one embodiment, the retail transaction contract types include three characteristic variables of purchase power mode, sign-up mode, whether or not to green the electricity user;
TABLE 1
| Electricity purchasing mode | Subscription mode | Whether or not to green electricity users | Class of electricity | Industry classification | User category | Grade of electricity consumption | Home power supply office | Number of transactions |
| object | object | object | object | object | object | object | object | int |
The electricity purchasing mode comprises two necessary modes of fixed price and market linkage price, and three selectable modes of floating electricity charge, green electricity environment overflow price and coal-electricity linkage, wherein the three selectable modes are generated by combination, and the characteristic value epsilon [1,6]; the signing mode comprises two modes of a card hanging mode and an offer mode, and the characteristic value epsilon [1,2]; the green electricity user comprises a yes type and a no type, and the characteristic value epsilon [1,2]; the electricity consumption category comprises large industrial electricity consumption, common industry, business and other 4 categories, and the characteristic value is E [1,4]; the industry classification is counted according to national economy industry classification, and the characteristic value epsilon [1,129]; the user category comprises public variable users and private variable users, and the characteristic value epsilon [1,2]; the electricity consumption level comprises three types of alternating current 110kV, alternating current 20kV and alternating current 10kV, and the characteristic value epsilon [1,3]; the home power supply office comprises all power supply offices under Guangdong power grid, and characteristic values E [1,19];
s1.2, the electricity consumption condition comprises two characteristic variables of electricity consumption and electricity price, and the statistical characteristic calculation is carried out on the two characteristic variables of the electricity consumption and the electricity price, wherein the statistical characteristic calculation comprises the mean value, the variance and the median of the variables so as to expand the characteristics of data;
the mean values of the variables are specifically as follows:
wherein,for the average value of the electricity consumption or the electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n}，nis the total number of samples;
the variance of the variables is specifically as follows:
wherein,for the variance of the electricity consumption or electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n-a }; for a pair ofnThe electricity consumption or electricity price of each sample is ordered in descending order, and the number of bits is taken asl Med 。
S1.3, converting each characteristic data into a value on a [0,1] interval by adopting a minimum-maximum normalization method, namely:
wherein,l max is the maximum value in the data and,l min as the minimum value in the data,is the converted data value;
s1.4, after the treatment, finally obtaining the firstqPersonal retail power consumer feature matrixx q ={l q,1 ,l q,2 ,...,l q,a0 }, whereina 0 For the initial feature quantity, at the same timeq={1,2,...,b}，bFor the total number of retail power customers,l q,p is the firstqFirst retail consumer of electricitypThe number of feature vectors is chosen to be the same,p={1,2,...,a 0 and get the total feature matrixA 0 ={x 1 ,x 2 ,...,x b In one embodiment, the number of bits per frame,a 0 =14，b= 31890; total feature matrixA 0 As shown in table 2.
Table 2 transaction user total feature matrix table
S1.5, in order to evaluate the correlation between each feature and other features, if a certain feature has no too great correlation with other features, the feature can be determined as a redundant feature, and the Pearson correlation coefficient method is adopted to evaluate the total feature matrixA 0 First of all retail power consumerspAny two of the feature vectorsLinear correlation among the two is subjected to feature preliminary screening, and the value interval of the Pearson correlation coefficient is [ -1,1]-1 represents a complete negative correlation, +1 represents a complete positive correlation; two characteristic variablesZAndYthe Pearson correlation coefficient calculation formula of (2) is:
、/>as a characteristic variable, a characteristic variable is used,p={1,2,...,a 0 }，rfor Pearson correlation coefficient, +.>And->Setting a threshold value of 0 as a feature screening condition for the variable mean value, judging features with Pearson correlation coefficients below 0 as redundant features, and removing the redundant features to generate an initial total user transaction feature matrixA={x 1 ,x 2 ,...,x b }，x={l 1 ,l 2 ,...,l a }，aTo pre-screen the number of post-feature, in one embodiment,a=12，b=31890。
s2, combining SMOTE algorithmkNeighbor algorithm is to initial total user transaction characteristic matrixA={x 1 ,x 2 ,...,x b Performing class unbalance processing, and assisting with sample class weights, wherein the specific method is as follows:
s2.1, acquiring a power consumer sample set from a transaction center, and finding out all minority samples from the power consumer sample setx i ，i=1~bDifferent users have different retail contract types, and in one embodiment, a retail contract type with some maximum transaction amount is the largest sample, with the transaction amount being below the maximumThe retail contract types of 20% of the samples are minority samples, and minority samples are calculatedx i With other minority sample pointsx j Euclidean distance between:
other minority sample pointsx j And minority class samplesx i The Euclidean distance between the two is smaller than the set threshold value, and the samples are few samplesx i Is used to determine the neighbor of a (c),j=1~bobtaining minority class samplesx i A kind of electronic devicek 1 Neighbor samples and are marked asx i,near ，nearThe values were 1,2,3,k 1 ，k 1 is the number of neighbor samples and is based onk 1 The individual neighbor samples form a corresponding classificationThe method comprises the steps of carrying out a first treatment on the surface of the In one embodiment, fetchk 1 =3；
S2.2 slavex i A kind of electronic devicek 1 Randomly selecting one sample from among the neighbor samplesx i,r ，r=1~k 1 Regenerating a random number between 0 and 1On the basis, a new sample is synthesizedx new1 Wherein:
wherein, each time a neighbor sample is randomly selected, a new random number is corresponding to；
S2.3, repeating the step S2.2B 1 Next, obtainB 1 New samples, noted asx new,i ，i=1,2,...,B 1 Forming an expanded retail power consumer feature matrix，b'=b+B 1 The method comprises the steps of carrying out a first treatment on the surface of the Expanded retail power consumer feature matrix>In the first b elements are {x 1 ,x 2 ,...,x b Post (back)B 1 The individual elements are new samples and willB 1 New samplesx new,i Categorizing into respective corresponding samplesx i,r The category to which it belongs; in one embodiment of the present invention, in one embodiment,B 1 =12414，b'=b+B 1 =44304；
s2.4, input training setWherein->For different classification of the sample, the extended retail power consumer feature matrix>As training data;
s2.5, finding a sample from the sample datax i Nearest 3 points covering the 3 pointsx i The field is recorded asN 3 (x i ) Next, inN 3 (x i ) Wherein the decision is re-determined based on classification decision rulesxCategory of (2)：
Wherein, let theθ=1,...,w，ρ=1,...,wThe method comprises the steps of carrying out a first treatment on the surface of the Then do not meet the orderρ=1,...,wIs redundant extension sample;
s2.6, retail power consumer feature matrix after expansionRemoving redundant amplified samples to form a new retail power consumer characteristic matrix, and forming a new retail power consumer characteristic matrix +.>，b''=b+B 1 -B 2 ，B 2 To be a redundant total number of amplified samples, in one embodiment,b''=b+B 1 -B 2 =44152。
s3, performing dimension reduction processing on the retail power consumer transaction sample data subjected to class unbalance processing by adopting an automatic encoder, wherein the method specifically comprises the following steps of:
s3.1, new retail power consumer feature matrixInputting the samples in the automatic encoder network, and carrying out forward and backward propagation on the network until the set training times or convergence degree are reached;
s3.2, outputting the user characteristic matrix after dimension reduction，b' is the user characteristic matrix after dimension reduction +.>In one embodiment,b'''=41572。
s4, clustering the retail power consumer transaction sample data processed by the automatic encoder by adopting a K-means method, judging the clustering effect by adopting a contour coefficient method, and constructing a retail power consumer multidimensional label based on the clustering effect, wherein the specific method is as follows:
s4.1, user characteristic matrix after dimension reductionRandom access in sample data in the sample datak 2 The individual samples are taken as cluster centers and are marked as +.>；
S4.2, adopting Euclidean distance as similarity measurement in the K-means method, wherein the loss function is the error square sum of each sample distance from the cluster center point, and the error square sum is specifically as follows:
wherein,e Ь representing a samplex i The cluster to which the cluster belongs is selected,Ьfor the number of clusters to be the number of clusters,representing clusterse Ь The corresponding center of the two-dimensional space is provided with a plurality of grooves,b' indicates the number of samples;
s4.3, orderFor the iterative step number, the following process is repeated until the algorithm calculation converges:
s4.3.1 for samplesx i Assigning it to the center nearest to it:
wherein,represent the firsttMultiple iterationsx i Cluster to which>Represent the firsttThe second iteration cluster corresponds tok 2 A center;
s4.3.2, for the center of each class, recalculate:
；
s4.4, the contour coefficient consists of two scores defined by the distance, and the samplex i The category is marked asThe nearest category is noted asC k Samples ofx i The corresponding contour coefficient is recorded asS(i)，d(x i ,C) For the samplex i The distance from the clustering center is calculated by the following method:
when there is only one point in the cluster, the contour coefficients are definedS(i) The profile coefficients of a set of data sets are the arithmetic mean of all sample profile coefficients in the data set, the range of values is [ -1,1]-1 represents false clustering, 1 represents that the data set is well divided, and the vicinity of 0 represents that various overlapping degrees are higher; determining cluster number from profile coefficients and actual transaction requirementsE，Determining cluster number in combination with actual transaction requirementsE=9；
S4.5, giving each data sample according to the clustering resultx i Adding a tag matrixx={l 1 ,l 2 ,...,l a ,U}, wherein i=[h i,1 ,h i,2 ,...,h i,E ]Is the firstiThe vector labels of the individual retail power consumers,hfor the tag identification number, if the user belongs to a certain categorynumClassification of tagsh num =1，num=1,2,...,EOtherwise 0, at this timep={1,2,...,a,...,a+E}；
S4.6, forming a final retail power consumer feature matrix finally comprising labelsWhereinx’={l 1 ,l 2 ,...,l a ,U}={l 1 ,l 2 ,...,l a ,h 1 ,h 2 ,...,h E }。
S5, optimizing the superparameter of LightGBM (Light Gradient Boosting Machine) by adopting a Bayesian optimization algorithm to obtain an optimized superparameter combination, wherein the method specifically comprises the following steps of:
s5.1, setting initial parameters of a model as follows: the number of leaf nodes is (3,128), the maximum depth range of the tree is (3, 25), the sample sub-sampling proportion range is (0.5, 1), the characteristic sub-sampling proportion range is (0.5, 1), and the L1 regularization penalty term coefficient range is (0, 1);
s5.2, outputting to-be-adjusted super parameter combination in retail power consumer feature matrix LightGBMvAn overshoot parameter table 3 is obtained.
Table 3 super parameter tuning table
| Parameter name | Parameter optimization range | Bayesian optimization values |
| Number of leaf nodes | (3,128) | 103 |
| Maximum depth of tree | (3,25) | 13 |
| Sample sub-sampling ratio | (0.5,1) | 0.84 |
| Feature sub-sample ratio | (0.5,1) | 0.71 |
| L1 regularization penalty term coefficient | (0,1) | 0.83 |
S6, performing retail power consumer feature optimization by a machine learning algorithm based on the LightGBM, wherein the specific method is as follows:
s6.1, according to the final retail power consumer characteristic matrixConstructing a power purchase demand prediction model based on the LightGBM by the optimal super-parameter combination obtained in the step S5;
in one embodiment, the superoptimal parameter combination obtained in step S5 adjusts the LightGBM model, the input of which is the final retail power consumer feature matrixThe output is a user feature importance value, and the model is a power purchase demand prediction model based on the LightGBM.
S6.2, classifying and predicting based on the constructed electricity purchasing demand prediction model based on the LightGBM, and outputting the characteristic importance of each characteristic by the electricity purchasing demand prediction model based on the LightGBM;
s6.3, nonlinear relation second derivative formed by feature quantity and feature accumulated importancef’’The number of effective features is determined by the number of corresponding features at=0, and the effective features are selected by the feature importance ranking and the number of effective features, the ineffective features are deleted,a new feature matrix is formed. As shown in FIG. 2, redundant features are deleted to form a new feature matrix, 9 features of electric quantity average value, electricity purchasing mode, industry classification, home power supply office, electricity price variance, signing mode, user class, transaction times and electricity consumption class are selected as effective features according to descending order of importance of various variables to form a fine feature matrix，x’={l 1 ,l 2 ,...,l a’ ,U}, whereina' =9, as shown in table 4.
TABLE 4 transaction user fine feature matrix
S7, constructing a neural network model based on an attention mechanism BiLSTM (Bidirectional Long Short-Term Memory);
as shown in fig. 3, the attention mechanism BiLSTM model includes an input layer, a BiLSTM hidden layer, an attention mechanism layer, a fully connected layer and an output layer, the fully connected layer performs local retail power consumer feature integration, takes a new feature combination as input, and completes model training, and a weight coefficient calculation formula of the attention mechanism can be expressed as:
wherein,represent the firsttOutput vector from hidden layer of neural network at momenth t The determined attention probability distribution value;uand->As the weight coefficient of the light-emitting diode,u=[0.2,0.8,0.5]，/>=[0.4,0.2,0.7]；/>for bias factor +.>=[0.8,0.3,0.4]；s t Is in the mechanism of AttentiontAnd outputting the time.
In one embodiment, the attention mechanism BiLSTM model is constructed as follows:
firstly, constructing an input dense layer, wherein the number of the layer units is 64 units, and the layer output is used as the input of a Bi-LSTM layer; the Bi-LSTM layer is of a 2-layer structure, 128 units are arranged in each layer, and the Bi-LSTM layer is output as the input of the attention layer; the attention layer is of a 1-layer structure, the number of the units of the layer is 32, and the output is used as the input of the full-connection layer; the full-connection layer is of a 1-layer structure, the number of the units of the layer is 32, and the output is used as the input of the output layer; the output layer has a 1-layer structure with the number of units of the layerOutputFor the classification prediction model, outputOutputThe number of units corresponds to the number of categories that one wants to output.
S8, predicting retail power consumer transaction requirements and customizing retail contracts based on a LightGBM model and a neural network model based on an attention mechanism BiLSTM;
in one embodiment, the transaction requirements are analyzed by inputting the retail power consumer feature matrix into a LightGBM model, outputting feature importance data, inputting the retail power consumer feature matrix into a attention mechanism BiLSTM predictive model, outputting the consumer multidimensional labels and frequency of occurrence, as shown in table 5.
TABLE 5 user transaction prediction list
And if the number of the industrial users is large, the electricity consumption of the users is large, and the electricity consumption stability requirement is high, the two parties are subjected to default and compensation setting to certain selectable options when the retail contract is set, so that the electric power user requirement of a large amount of high-stability electricity consumption is ensured.
The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above examples, and any other modifications, substitutions, combinations, and simplifications without departing from the spirit and principles of the present invention should be made in the equivalent manner, and are included in the scope of the present invention.
Claims (10)
1. The method for predicting the electricity purchasing demand of the power consumer and customizing the retail contract is characterized by comprising the following steps of:
s1, a retail power user characteristic matrix is initially constructed from 8 layers of retail transaction contract types, power utilization categories, industry classifications, user categories, power utilization levels, home power supply bureaus, power utilization conditions and transaction times;
s2, combining SMOTE algorithmkThe neighbor algorithm performs class unbalance processing on the transaction sample data of the retail power user;
s3, performing dimension reduction processing on the retail power consumer transaction sample data subjected to the class unbalance processing by adopting an automatic encoder;
s4, clustering the retail power consumer transaction sample data processed by the automatic encoder by adopting a K-means method, performing clustering effect judgment by adopting a contour coefficient method, and constructing a retail power consumer multidimensional label based on the clustering effect;
s5, optimizing the super parameters of the LightGBM by adopting a Bayesian optimization algorithm to obtain an optimized super parameter combination;
s6, performing retail power consumer feature optimization by a machine learning algorithm based on the LightGBM;
s7, constructing a neural network model based on an attention mechanism BiLSTM;
s8, predicting retail power consumer transaction requirements and customizing retail contracts based on a LightGBM model and a neural network model based on an attention mechanism BiLSTM.
2. The method for predicting electricity purchasing demand and customizing retail contracts for electricity consumers according to claim 1, wherein the step S1 comprises the steps of:
s1.1, the text characteristics adopt label coding, each label is mapped to an integer value, and the increment is started from 0; the text class features include retail transaction contract type, electricity class, industry classification, user class, electricity class, and home power office;
s1.2, the electricity consumption condition comprises two characteristic variables of electricity consumption and electricity price, and the statistical characteristic calculation is carried out on the two characteristic variables of the electricity consumption and the electricity price, wherein the statistical characteristic calculation comprises the mean value, the variance and the median of the variables so as to expand the characteristics of data;
s1.3, converting each characteristic data into a value on a [0,1] interval by adopting a minimum-maximum normalization method, namely:
wherein,l max is the maximum value in the data and,l min as the minimum value in the data,is the converted data value;
s1.4, after the treatment, finally obtaining the firstqPersonal retail power consumer feature matrixx q ={l q,1 ,l q,2 ,...,l q,a0 }, whereina 0 For the initial feature quantity, at the same timeq={1,2,...,b}，bFor the total number of retail power customers,l q,p is the firstqFirst retail consumer of electricitypThe number of feature vectors is chosen to be the same,p={1,2,...,a 0 and get the total feature matrixA 0 ={x 1 ,x 2 ,...,x b }；
S1.5, adopt PearMethod for evaluating total feature matrix by son correlation coefficientA 0 First of all retail power consumerspPerforming linear correlation between any two of the feature vectors, and performing feature preliminary screening, wherein the value interval of the Pearson correlation coefficient is [ -1,1]-1 represents a complete negative correlation, +1 represents a complete positive correlation; two characteristic variablesZAndYthe Pearson correlation coefficient calculation formula of (2) is:
、/>as a characteristic variable, a characteristic variable is used,p={1,2,...,a 0 }，rfor Pearson correlation coefficient, +.>And->Setting a threshold value of 0 as a feature screening condition for the variable mean value, judging features with Pearson correlation coefficients below 0 as redundant features, and removing the redundant features to generate an initial total user transaction feature matrixA={x 1 ,x 2 ,...,x b }，x={l 1 ,l 2 ,...,l a }，aIs the feature quantity after preliminary screening.
3. The method for predicting electricity purchasing demand and customizing retail contracts for electric power consumers according to claim 1, wherein in step S1.2, the average value of the variables is specifically as follows:
wherein,for the average value of the electricity consumption or the electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n}，nis the total number of samples;
the variance of the variables is specifically as follows:
wherein,for the variance of the electricity consumption or electricity price of the sample,l υ is the firstυThe electricity consumption or electricity price of each sample,υ={1,2,…,n-a }; for a pair ofnThe electricity consumption or electricity price of each sample is ordered in descending order, and the number of bits is taken asl Med 。
4. The method for forecasting electricity purchasing demand and customizing a retail contract of an electric power consumer according to claim 2, wherein in step S2, the SMOTE algorithm is combined with the retail contractkNeighbor algorithm is to initial total user transaction characteristic matrixAClass unbalance processing is carried out, and sample class weights are assisted, and the specific method is as follows:
s2.1, acquiring a power consumer sample set from a transaction center, and finding out all minority samples from the power consumer sample setx i ，i=1~bA minority class sample is calculatedx i With other minority sample pointsx j Euclidean distance between:
other minority sample pointsx j And minority class samplesx i The Euclidean distance between the two is smaller than the set threshold value, and the samples are few samplesx i Is used to determine the neighbor of a (c),j=1~bobtaining minority class samplesx i A kind of electronic devicek 1 Neighbor samples and are marked asx i,near ，nearThe values were 1,2,3,k 1 ，k 1 is the number of neighbor samples and is based onk 1 The individual neighbor samples form a corresponding classification；
S2.2 slavex i A kind of electronic devicek 1 Randomly selecting one sample from among the neighbor samplesx i,r ，r=1~ k 1 Regenerating a random number between 0 and 1On the basis, a new sample is synthesizedx new1 Wherein:
wherein, each time a neighbor sample is randomly selected, a new random number is corresponding to；
S2.3, repeating the step S2.2B 1 Next, obtainB 1 New samples, noted asx new,i ，i=1,2,...,B 1 Forming an expanded retail power consumer feature matrix，b'=b+B 1 The method comprises the steps of carrying out a first treatment on the surface of the Expanded retail power consumer feature matrix>In the first b elements are {x 1 ,x 2 ,...,x b Post (back)B 1 The individual elements are new samples and willB 1 New samplesx new,i Categorizing into respective corresponding samplesx i,r The category to which it belongs;
s2.4, input training setWherein->For different classification of the sample, the extended retail power consumer feature matrix>As training data;
s2.5, finding a sample from the sample datax i Nearest 3 points covering the 3 pointsx i The field is recorded asN 3 (x i ) Next, inN 3 (x i ) Wherein the decision is re-determined based on classification decision rulesxCategory of (2)：
Wherein, let theθ=1,...,w，ρ=1,...,wThe method comprises the steps of carrying out a first treatment on the surface of the Then do not meet the orderρ=1,...,wIs redundant extension sample;
s2.6, retail power consumer feature matrix after expansionRemoving redundant amplified samples to form a new retail power consumer characteristic matrix, and forming a new retail power consumer characteristic matrix +.>，b''=b+B 1 -B 2 ，B 2 Is the total number of redundant amplified samples.
5. The method for predicting electricity purchasing demand and customizing retail contracts for electricity consumers as claimed in claim 4, wherein the step S3 comprises the steps of:
s3.1, new retail power consumer feature matrixInputting the samples in the automatic encoder network, and carrying out forward and backward propagation on the network until the set training times or convergence degree are reached;
s3.2, outputting the user characteristic matrix after dimension reduction，b' is the user characteristic matrix after dimension reduction +.>Is a sample count of the total number of samples in the sample.
6. The method for predicting electricity purchasing demand and customizing retail contracts of power consumers according to claim 5, wherein in step S4, the K-means method is adopted to cluster the transaction sample data of the retail power consumers processed by the automatic encoder, the contour coefficient method is adopted to judge the clustering effect, and the multi-dimensional labels of the retail power consumers are constructed based on the clustering effect, and the specific method is as follows:
s4.1, user characteristic matrix after dimension reductionRandom access in sample data in the sample datak 2 The individual samples are taken as cluster centers and are marked as +.>；
S4.2, adopting Euclidean distance as similarity measurement in the K-means method, wherein the loss function is the error square sum of each sample distance from the cluster center point, and the error square sum is specifically as follows:
wherein,e Ь representing a samplex i The cluster to which the cluster belongs is selected,Ьfor the number of clusters to be the number of clusters,representing clusterse Ь The corresponding center of the two-dimensional space is provided with a plurality of grooves,b' indicates the number of samples;
s4.3, orderFor the iterative step number, the following process is repeated until the algorithm calculation converges:
s4.3.1 for samplesx i Assigning it to the center nearest to it:
wherein,represent the firsttMultiple iterationsx i Cluster to which>Represent the firsttThe second iteration cluster corresponds tok 2 A center;
s4.3.2, for the center of each class, recalculate:
；
s4.4, the profile coefficient consists of two scores defined by the distance,sample ofx i The category is marked asThe nearest category is noted asC k Samples ofx i The corresponding contour coefficient is recorded asS(i)，d(x i ,C) For the samplex i The distance from the clustering center is calculated by the following method:
when there is only one point in the cluster, the contour coefficients are definedS(i) The profile coefficients of a set of data sets are the arithmetic mean of all sample profile coefficients in the data set, the range of values is [ -1,1]-1 represents false clustering, 1 represents that the data set is well divided, and the vicinity of 0 represents that various overlapping degrees are higher; determining cluster number from profile coefficients and actual transaction requirementsE；
S4.5, giving each data sample according to the clustering resultx i Adding a tag matrixx={l 1 ,l 2 ,...,l a ,U}, wherein i=[h i,1 ,h i,2 ,...,h i,E ]Is the firstiThe vector labels of the individual retail power consumers,hfor the tag identification number, if the user belongs to a certain categorynumClassification of tagsh num =1，num=1,2,...,EOtherwise 0, at this timep={1,2,...,a,...,a+E}；
S4.6, forming a final retail power consumer feature matrix finally comprising labelsWhereinx’={l 1 ,l 2 ,...,l a ,U}={l 1 ,l 2 ,...,l a ,h 1 ,h 2 ,...,h E }。
7. The method for predicting electricity purchasing demand and customizing retail contracts for electric power consumers as claimed in claim 6, wherein in step S6, the machine learning algorithm based on LightGBM is used for feature optimization, and the specific method is as follows:
s6.1, according to the final retail power consumer characteristic matrixConstructing a power purchase demand prediction model based on the LightGBM by the optimal super-parameter combination obtained in the step S5;
s6.2, classifying and predicting based on the constructed electricity purchasing demand prediction model based on the LightGBM, and outputting the characteristic importance of each characteristic by the electricity purchasing demand prediction model based on the LightGBM;
s6.3, nonlinear relation second derivative formed by feature quantity and feature accumulated importancef’’And (3) determining the number of effective features according to the number of the features corresponding to the position (0), screening out the effective features according to the feature importance ranking and the number of the effective features, deleting the ineffective features, and forming a new feature matrix.
8. The method for predicting electricity purchasing demand and customizing retail contracts of electric power consumers as claimed in claim 7, wherein in step S6.1, the super-optimal parameter combination obtained in step S5 adjusts the LightGBM model, and the model is input as the final retail electric power consumer feature matrixThe output is a user feature importance value, and the model is a power purchase demand prediction model based on the LightGBM.
9. The method for predicting electricity purchasing demand and customizing retail contracts of a power consumer according to claim 1, wherein in step S7, the attention mechanism BiLSTM model comprises an input layer, a BiLSTM hiding layer, an attention mechanism layer, a full connection layer and an output layer, wherein the full connection layer performs local retail power consumer feature integration, and new feature combinations are used as inputs.
10. The method for predicting electricity purchasing demand and customizing retail contracts according to any one of claims 1 to 9, wherein in step S8, the retail electricity consumer feature matrix is input into a LightGBM model, the feature importance data is output, the retail electricity consumer feature matrix is input into a attention mechanism BiLSTM prediction model, and the multidimensional labels and occurrence frequencies of the consumers are output, so that the retail contracts are comprehensively optimized according to the user features and the labels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311299832.3A CN117035837B (en) | 2023-10-09 | 2023-10-09 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311299832.3A CN117035837B (en) | 2023-10-09 | 2023-10-09 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117035837A true CN117035837A (en) | 2023-11-10 |
| CN117035837B CN117035837B (en) | 2024-01-19 |
Family
ID=88641678
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311299832.3A Active CN117035837B (en) | 2023-10-09 | 2023-10-09 | Method for predicting electricity purchasing demand of power consumer and customizing retail contract |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117035837B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111428766A (en) * | 2020-03-17 | 2020-07-17 | 深圳供电局有限公司 | Power consumption mode classification method for high-dimensional mass measurement data |
| CN112396301A (en) * | 2020-11-05 | 2021-02-23 | 国网天津市电力公司 | Power consumer demand response characteristic control method based on energy big data driving |
| CN113962364A (en) * | 2021-10-22 | 2022-01-21 | 四川大学 | Multi-factor power load prediction method based on deep learning |
| CN114722810A (en) * | 2022-03-21 | 2022-07-08 | 浙江工业大学 | Real estate customer portrait method and system based on information extraction and multi-attribute decision |
| CN115526264A (en) * | 2022-10-13 | 2022-12-27 | 国网天津市电力公司 | User power consumption behavior classification analysis method based on self-encoder |
| US20230032739A1 (en) * | 2021-07-29 | 2023-02-02 | Dell Products L.P. | Propensity modeling process for customer targeting |
-
2023
- 2023-10-09 CN CN202311299832.3A patent/CN117035837B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111428766A (en) * | 2020-03-17 | 2020-07-17 | 深圳供电局有限公司 | Power consumption mode classification method for high-dimensional mass measurement data |
| CN112396301A (en) * | 2020-11-05 | 2021-02-23 | 国网天津市电力公司 | Power consumer demand response characteristic control method based on energy big data driving |
| US20230032739A1 (en) * | 2021-07-29 | 2023-02-02 | Dell Products L.P. | Propensity modeling process for customer targeting |
| CN113962364A (en) * | 2021-10-22 | 2022-01-21 | 四川大学 | Multi-factor power load prediction method based on deep learning |
| CN114722810A (en) * | 2022-03-21 | 2022-07-08 | 浙江工业大学 | Real estate customer portrait method and system based on information extraction and multi-attribute decision |
| CN115526264A (en) * | 2022-10-13 | 2022-12-27 | 国网天津市电力公司 | User power consumption behavior classification analysis method based on self-encoder |
Non-Patent Citations (1)
| Title |
|---|
| 汪波 等: "基于深层标签和Ｋ⁃Ｍｅａｎｓ＋＋算法的电力用户画像研究", 鞍山师范学院学报, vol. 24, no. 6, pages 43 - 48 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117035837B (en) | 2024-01-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063945B (en) | Value evaluation system-based 360-degree customer portrait construction method for electricity selling company | |
| CN108897791B (en) | Image retrieval method based on depth convolution characteristics and semantic similarity measurement | |
| CN110990567A (en) | Electric power audit text classification method for enhancing domain features | |
| CN112966114A (en) | Document classification method and device based on symmetric graph convolutional neural network | |
| CN110689162B (en) | Bus load prediction method, device and system based on user side classification | |
| CN108345908A (en) | Sorting technique, sorting device and the storage medium of electric network data | |
| CN102088709A (en) | Method for predicting telephone traffic based on clustering and autoregressive integrated moving average (ARIMA) model | |
| CN113554241B (en) | User layering method and prediction method based on user electricity complaint behaviors | |
| Zhu et al. | Loan default prediction based on convolutional neural network and LightGBM | |
| CN113591947A (en) | Power data clustering method and device based on power consumption behaviors and storage medium | |
| CN117035837B (en) | Method for predicting electricity purchasing demand of power consumer and customizing retail contract | |
| Daneshmandi et al. | A hybrid data mining model to improve customer response modeling in direct marketing | |
| Ou et al. | On data mining for direct marketing | |
| CN116303386A (en) | Intelligent interpolation method and system for missing data based on relational graph | |
| CN115689201A (en) | Multi-criterion intelligent decision optimization method and system for enterprise resource supply and demand allocation | |
| Zheng | Application of silence customer segmentation in securities industry based on fuzzy cluster algorithm | |
| CN114238852A (en) | Operation data analysis method and device, storage medium and electronic equipment | |
| Xi et al. | Improved AHP model and neural network for consumer finance credit risk assessment | |
| Harikrishna et al. | Credit scoring using support vector machine: a comparative analysis | |
| CN114092123A (en) | Satisfaction intelligent analysis system | |
| Davoodabadi et al. | Building C ustomers’ Credit Scoring Models with Combination of Feature Selection and Decision Tree Algorithms | |
| Huang et al. | A clustering-based method for business hall efficiency analysis | |
| Wang et al. | SOMEDGRA: A case retrieval method for machine tool product configuration design | |
| CN112836926B (en) | Enterprise operation condition evaluation method based on electric power big data | |
| Yang et al. | Research on Distributed Photovoltaic Power Station Builders Segmentation Based on Data Mining |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |