CN110428270A - The potential preference client recognition methods of the channel of logic-based regression algorithm - Google Patents
The potential preference client recognition methods of the channel of logic-based regression algorithm Download PDFInfo
- Publication number
- CN110428270A CN110428270A CN201910723897.3A CN201910723897A CN110428270A CN 110428270 A CN110428270 A CN 110428270A CN 201910723897 A CN201910723897 A CN 201910723897A CN 110428270 A CN110428270 A CN 110428270A
- Authority
- CN
- China
- Prior art keywords
- channel
- client
- preference
- potential
- data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- 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
- G06Q30/01—Customer relationship services
-
- 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
Abstract
The present invention relates to electric power data analysis fields, are the potential preference client recognition methods of channel of logic-based regression algorithm, comprising steps of data are acquired and pre-processed；Random sampling determines negative example sample, is modeled based on Information Entropy and determines positive example sample, handles positive and negative example imbalanced training sets problem, and stratified sampling generates training set and test set；Construction logic regression model recycles by grid search and adjusts ginseng, select to model and training obtains optimal models；Optimal models are verified on test set；By client characteristics data input optimal models to calculate prediction probability, the channel preference degree of potential customers is obtained, the potential preference client group of channel is determined by setting preference threshold value.The present invention utilizes customer information and magnanimity electric service data, excavates client characteristics, constructs model by big data analysis algorithm, identifies the potential preference client of channel, so that effectively support channel is precisely promoted, promotes channel shunting effect.
Description
Technical field
The present invention relates to electric power data analysis fields, and in particular to the potential preference client of the channel of logic-based regression algorithm
Recognition methods.
Background technique
The 95598 phone channels customer service channel traditional as Utilities Electric Co., the portfolio undertaken are significantly larger than online
Channel.In the power customer using phone channel, the overwhelming majority can choose manual telephone system service mode.And in online channel
Artificial online talk service belongs to manual service category, and the business model of the two with the manual telephone system service in phone channel
Enclose, service range, service feature etc. have certain homogeney, this phone channel is shunted to online channel become can
Energy.If customer demand can be fully understood and using the technological means of science, guidance client shunts from phone channel to online channel,
It then can largely alleviate artificial telephone traffic pressure, to guarantee customer service quality.Currently, Utilities Electric Co. just gradually promotes
Online channel, phone channel has tentatively been shown by online channel distributary phenomenon, but shunting effect is limited, thus needs a set of science
Effective method promotes channel shunting effect.
Summary of the invention
In view of the above-mentioned problems, the present invention provides the potential preference client recognition methods of channel of logic-based regression algorithm, it should
Method makes full use of customer information and magnanimity electric service data, excavates client characteristics, constructs mould by big data analysis algorithm
Type identifies the potential preference client of channel, so that effectively support channel is precisely promoted, promotes channel shunting effect.
The potential preference client recognition methods of the channel of logic-based regression algorithm according to the present invention, comprising the following steps:
S1, data acquisition, analysis identify related data with channel potential customers；
S2, data collected are pre-processed；
S3, random sampling determine negative example sample, are modeled based on Information Entropy and determine positive example sample, and is uneven to positive and negative example sample
Weighing apparatus problem is handled, and last stratified sampling generates training set and test set；
S4, construction logic regression model are recycled by grid search and are adjusted ginseng, carry out model selection using cross-validation method,
Model is trained to obtain optimal models；Optimal models are verified on test set, are examined by precision ratio, recall ratio optimal
Modelling effect；
S5, client characteristics data are inputted into optimal models to calculate prediction probability, obtains the channel preference of potential customers
Degree determines the potential preference client group of channel by setting preference threshold value.
In a preferred embodiment, step S3 constructs Information Entropy model based on the customers for using online channel, calculates
The channel preference degree of each storage client of online channel is measured the degree of stability of channel storage client according to preference height, is chosen
Preference online channel storage client in the top is as positive example sample.It is right also based on the customers for using phone channel
From client's random sampling of online channel was not used as negative number of cases evidence.
In a preferred embodiment, when step S3 handles positive and negative example imbalanced training sets problem, using being adopted on SMOTE
Quadrat method synthesizes new sample by way of interpolation for positive example sample, to realize that positive and negative example sample is balanced.
Compared with prior art, the beneficial effects of the present invention are: the present invention make full use of Electricity customers essential information and
Channel use information construction feature system is divided by data prediction, specimen sample, and constitutive logic regression model passes through ginseng
Number tuning obtains stable Optimized model, predicts channel potential customers preference, identifies that channel is latent by setting preference threshold value
In preference client group, so that channel be supported precisely to promote, and then channel shunting effect is promoted.To data prediction, choose just
Negative number of cases handles imbalanced training sets problem according to composition sample data, using SMOTE top sampling method；It is patrolled based on sample data training
Regression model is collected, is recycled by grid search and adjusts ginseng, carries out model selection using cross-validation method, using F1 measurement as optimization
Target takes into account the precision and generalization ability of model；Prediction probability is calculated to channel potential customers, passes through precision ratio
(Precision) and recall ratio (Recall) assessment prediction effect, the validity of model is verified；It is general according to the prediction of potential customers
Rate generates channel preference degree, the potential preference client group of channel is identified by setting preference threshold value, to support channel accurate
It promotes, and then promotes channel shunting effect.
Detailed description of the invention
Fig. 1 is the flow chart of the potential preference client recognition methods of channel of logic-based regression algorithm.
Specific embodiment
The present invention is described in further detail with embodiment with reference to the accompanying drawings of the specification, but embodiment party of the invention
Formula is not limited to this.
As shown in Figure 1, the present invention is based on the potential preference client recognition methods of the channel of logistic regression algorithm, mainly include with
Lower step:
Step 1, data acquisition: analysis identifies related data with channel potential customers, to determine the data of building model
The time window of range, data sampling.
In present embodiment, the range of data acquisition is set as the number such as Electricity customers essential information and channel use information
According to；Using Electricity customers as research object, analysis identifies related data with channel potential customers, extracts the whole network and saves Electricity customers
Essential information, 1 year nearly (such as 2018 year) Electricity customers demand information and channel use information data.
Step 2, data prediction, comprising: data cleansing, feature machining and data conversion；
In present embodiment, the data such as Electricity customers essential information and channel use information are based on, processing client is substantially special
Sign, channel use feature and channel preference feature, by links such as Data Integration, cleaning, conversions, complete data prediction.In
Data cleansing link handles missing values and repetition values, handles jack per line client essential information colliding data.After cleaning
Electricity customers essential information and channel use information, processing client's essential characteristic, channel use feature and channel preference feature；It is right
Characteristic after processing is converted, and is numeric type feature by classifying type Feature Conversion.
In the present embodiment, when being converted to the characteristic after processing, numeric type feature is screened based on standard deviation, is based on
IV value sifting sort type feature；Based on WOE by classifying type Feature Conversion be numeric type feature.
Step 3, random sampling determine negative example sample, are modeled based on Information Entropy and determine positive example sample, up-sampled using SMOTE
The positive and negative example imbalanced training sets problem of method processing, last stratified sampling generate training set and test set.
In present embodiment, Select to use crosses the customers of online channel, which is online channel storage client, base
In the customers for using online channel, Information Entropy model is constructed, calculates the channel preference degree of each storage client of online channel, root
The degree of stability of channel storage client is measured according to preference height, the higher channel storage client characteristics of preference can more represent partially
Customer group's feature of the good channel, thus preference online channel storage client in the top is chosen as positive example data
(i.e. positive example sample).Based on the customers for using phone channel, make to from client's random sampling that online channel was not used
The number of cases that is negative is according to (i.e. negative example sample)；By positive example data and negative number of cases according to being included in sample, when actual samples, the quantity of positive example data
The quantity of usually less than negative number of cases evidence, this is because storage client's number of phone channel is much larger than the storage client of online channel
Number, and positive example data are strictly screened according to preference again, so the quantity that positive example data and negative number of cases evidence can occur is uneven
Weigh situation.Positive and negative number of cases is according to the unbalanced performance that will affect classification prediction model.
In order to guarantee that sample size is enough, while solving the problems, such as positive and negative number of cases according to unbalanced, present invention use up-sampling
Processing method, the quantity by increasing positive example data makes the equal number of positive and negative number of cases evidence.However, the facture of up-sampling
Repeated sampling simply cannot be carried out to positive example sample, otherwise will lead to the serious over-fitting of trained model.In present embodiment
Using SMOTE top sampling method, which is to be closed by way of interpolation for minority class (herein referring to positive example data) in short
The sample of Cheng Xin, to realize that positive and negative example sample is balanced.It is then based on balanced positive and negative example sample data, passes through stratified sampling
Method extracts 70% from positive and negative number of cases respectively in and generates training set, remaining 30% is used as test set.
Entropy is the measurement of the unordered degree of system, can be used to measure the effective information and weight that given data is included.
Weight is determined by the calculating of entropy, and the weight of each index is exactly determined according to the difference degree of indices.It is built based on Information Entropy
Mould determines positive example data, and specific step is as follows:
(1) calculated specific gravity matrix.Specific formula is as follows:
In formula, xijFor i-th of user's jth item index value, if the dimension of indices or the order of magnitude are different, xijIt is i-th
A user's jth item criterionization treated value.yijFor the specific gravity of i-th of user under jth item index.
(2) information entropy is calculated.The information entropy e of jth item index is calculated according to the result of step (1)j, specific formula is such as
Under:
In formula, K=1/lnm.
(3) information utility value is calculated.D is worth according to the information utility that the result of step (2) calculates jth item indexj, specifically
Formula is as follows:
dj=1-ej
The comentropy e of i.e. i-th indexjWith the difference between 1.
(4) parameter weight.The weight w of jth item index is calculated according to the result of step (3)j, specific formula is as follows:
(5) comprehensive score is calculated.The comprehensive score of i-th of user is calculated according to step (1) and the result of (4), it is specific public
Formula is as follows:
According to the comprehensive score of calculating, the higher user of score value is chosen as positive example sample and is used for subsequent modeling.
Step 4, construction logic regression model recycle by grid search and adjust ginseng, carry out model choosing using cross-validation method
It selects, optimization aim training pattern is used as using F1 measurement, to take into account the precision and generalization ability of model；Training is obtained most
Excellent model is verified on test set, examines optimal models effect by precision ratio (Precision), recall ratio (Recall).
Logistic regression algorithm (logistic recurrence), can be divided into according to the value type of dependent variable: dualistic logistic regression and
Multivariate logistic regression.Dependent variable in dualistic logistic regression model can only take 0 and 1, and in multivariate logistic regression model because become
Amount can take multiple values, and the present invention mainly uses dualistic logistic regression, hereinafter referred to as logistic regression.
(1) logit is converted
Assuming that the probability that event occurs is p, to be predicted the probability that event occurs the value it is necessary to predict p.By
It is difficult simply in [0,1] section with linear model following in the value range of p, while when p is close to 0 or 1 endpoint value,
The situation of change of Probability p is difficult to be captured with commonsense method, therefore considers first to carry out conversion appropriate to p, then predict the value of p.
Usually consider logit transformation, i.e.,As the transfer function of p, which is a monotonic function of p, when
The value range of p transfer function log it p in [0,1] section value is (- ∞ ,+∞), this to work as p close to 0 or 1 end
When point value, functional value variation it is sensitive, thus overcome directly prediction p there are the problem of.
(2) Logic Regression Models
Linear relationship can be converted by the non-linear relation between dependent variable and independent variable using logit transformation: setting variable
Value x1,x2,K,xn, then equation of linear regression is established as Logic Regression Models:
Wherein, β0For constant term, βiFor the variation coefficient of i-th of variable in equation of linear regression.Do index in peer-to-peer both ends
Transformation, obtains:
If Y is two classified variables, Y=1, which is represented, to be occurred, and Y=0 representative does not occur, and both members transformation obtains predictive variable
Probability value when being 1, it may be assumed that
(3) regression coefficient and inspection
The parameter Estimation of above-mentioned Logic Regression Models can be estimated using Maximum Likelihood Estimation Method, and model coefficient is very big
Possibility predication be population sample parameter it is progressive it is unbiased be effectively estimated, approximation obeys classical normal distribution, therefore can be direct
Hypothesis testing is carried out to parameter, the method for inspection mainly used is Chi-square Test etc..
Grid-search algorithms are a kind of by traversing given parameter combination come the method for Optimized model.It is returned in construction logic
After returning model, is recycled by grid search and adjust ginseng, model tuning efficiency can be improved.
Cross-validation method is data set to be divided into n group, randomly selects n-1 group every time as training set, remaining one group
Collect as verifying, randomly selects n times, obtain n model, optimal model and parameter are evaluated by calculating target function.The party
Method can avoid model over-fitting using average thought to a certain extent.
Precision ratio (Precision) is to predict that how many is real positive example in the sample being positive, that is, it is inclined to be predicted as channel
How many is real channel preference client in hospitable family；Recall ratio (Recall) how many is predicted for the positive example in original sample
Correctly, i.e., the channel preference client in original sample how many be predicted correctly；F1 measurement is comprehensive evaluation index, can take into account and look into
Full rate and precision ratio are the harmonic averages of the two.The present invention selects F1 to measure the optimization aim as model, selects precision ratio
(Precision), recall ratio (Recall) two indices testing model effect.
The present embodiment is based on confusion matrix and precision ratio (Precision) and recall ratio (Recall) is calculated, and obscures square
Battle array is defined as follows:
Wherein,
TP (True Positive): true value is positive, and model is considered the quantity of positive；
FN (False Negative): true value is positive, and model is considered the quantity of negative；
FP (False Positive): true value is negative, and model is considered the quantity of positive；
TN (True Negative): true value is negative, and model is considered the quantity of negative.
The calculation formula of precision ratio (Precision) and recall ratio (Recall) is as follows:
F1 measurement is the harmonic average of precision ratio and recall ratio, and calculation formula is as follows:
It selects F1 to measure the optimization aim as model, the optimal models that training obtains is examined on test set, verify
As a result as follows:
Step 5, model application: after obtaining optimal models, client characteristics data are inputted into optimal models to calculate prediction
Probability determines the potential preference client group of channel by setting preference threshold value to obtain the channel preference degree of potential customers
Body.
This step predicts the channel preference of potential customers according to client characteristics based on constructed Logic Regression Models
Degree determines the potential preference client group of channel by setting preference threshold value, so that channel be supported precisely to promote, and then is promoted
Channel shunting effect.
The present invention is based on the potential preference client recognition methods of the channel of logistic regression algorithm, significantly enhance the saturating of model
Bright property, comprehensibility.Obviously, various changes and modifications can be made to the invention without departing from this hair by those skilled in the art
Bright spirit and scope.In this way, if these modifications and changes of the present invention belongs to the claims in the present invention and its equivalent technology
Within the scope of, then the present invention is also intended to include these modifications and variations.
Claims (9)
1. the potential preference client recognition methods of the channel of logic-based regression algorithm, which comprises the following steps:
S1, data acquisition, analysis identify related data with channel potential customers；
S2, data collected are pre-processed；
S3, random sampling determine negative example sample, are modeled based on Information Entropy and determine positive example sample, and asked positive and negative example imbalanced training sets
Topic is handled, and last stratified sampling generates training set and test set；
S4, construction logic regression model are recycled by grid search and are adjusted ginseng, model selection are carried out using cross-validation method, to mould
Type is trained to obtain optimal models；Optimal models are verified on test set, optimal models are examined by precision ratio, recall ratio
Effect；
S5, client characteristics data are inputted into optimal models to calculate prediction probability, obtains the channel preference degree of potential customers, leads to
Setting preference threshold value is crossed to determine the potential preference client group of channel.
2. the potential preference client recognition methods of channel according to claim 1, which is characterized in that in step S3, based on making
The customers of used online channel construct Information Entropy model, the channel preference degree of each storage client of online channel are calculated, according to inclined
Good degree height measures the degree of stability of channel storage client, chooses preference online channel storage client in the top as just
Example sample.
3. the potential preference client recognition methods of channel according to claim 2, which is characterized in that true based on Information Entropy modeling
Determining positive example sample, specific step is as follows:
(1) calculated specific gravity matrix:
In formula, xijFor i-th of user's jth item index value, if the dimension of indices or the order of magnitude are different, xijIt is used for i-th
Family jth item criterionization treated value；yijFor the specific gravity of i-th of user under jth item index；
(2) the information entropy e of jth item index is calculatedj:
In formula, K=1/lnm；
(3) information utility for calculating jth item index is worth dj:
dj=1-ej
(4) the weight w of jth item index is calculatedj:
(5) comprehensive score of i-th of user is calculated:
According to the comprehensive score of calculating, the higher user of score value is chosen as positive example sample and is used for subsequent modeling.
4. the potential preference client recognition methods of channel according to claim 1, which is characterized in that in step S3, based on making
The customers of used phone channel, to from client's random sampling of online channel was not used as negative number of cases evidence.
5. the potential preference client recognition methods of channel according to claim 1, which is characterized in that positive and negative example in step S3
When imbalanced training sets problem is handled, using SMOTE top sampling method, synthesized newly by way of interpolation for positive example sample
Sample, to realize that positive and negative example sample is balanced.
6. the potential preference client recognition methods of channel according to claim 1, which is characterized in that step S4 construction logic returns
Return model the following steps are included:
(1) logit transformation is carried out to the Probability p that event occurs, it willAs the transfer function of Probability p, then
The value of prediction probability p；
(2) value of variable is set as x1,x2,K,xn, equation of linear regression is established as Logic Regression Models:
Wherein, β0For constant term, βiFor the variation coefficient of i-th of variable in equation of linear regression；Do index change in peer-to-peer both ends
It changes, obtains:
If Y is two classified variables, Y=1, which is represented, to be occurred, and Y=0 representative does not occur, and both members transformation, obtaining predictive variable is 1
When probability value, it may be assumed that
(3) parameter of Maximum Likelihood Estimation Method estimation logic regression model is used, and the parameter of Logic Regression Models is blocked
It examines side.
7. the potential preference client recognition methods of channel according to claim 1, which is characterized in that in step S2, pretreatment
Including data cleansing, feature machining and data conversion.
8. the potential preference client recognition methods of channel according to claim 7, which is characterized in that in data cleansing link,
Missing values and repetition values are handled, jack per line client essential information colliding data is handled；It is basic based on the Electricity customers after cleaning
Information and channel use information, processing client's essential characteristic, channel use feature and channel preference feature；To the feature after processing
Data are converted, and are numeric type feature by classifying type Feature Conversion.
9. the potential preference client recognition methods of channel according to claim 8, which is characterized in that the characteristic after processing
When according to being converted, numeric type feature being screened based on standard deviation, is based on IV value sifting sort type feature；Based on WOE by classifying type
Feature Conversion is numeric type feature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910723897.3A CN110428270A (en) | 2019-08-07 | 2019-08-07 | The potential preference client recognition methods of the channel of logic-based regression algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910723897.3A CN110428270A (en) | 2019-08-07 | 2019-08-07 | The potential preference client recognition methods of the channel of logic-based regression algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110428270A true CN110428270A (en) | 2019-11-08 |
Family
ID=68412912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910723897.3A Pending CN110428270A (en) | 2019-08-07 | 2019-08-07 | The potential preference client recognition methods of the channel of logic-based regression algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428270A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111709532A (en) * | 2020-05-26 | 2020-09-25 | 重庆大学 | Model-independent local interpretation-based online shopping representative sample selection system |
| CN112116393A (en) * | 2020-09-23 | 2020-12-22 | 贝壳技术有限公司 | Method, device and equipment for realizing event user maintenance |
| CN112949301A (en) * | 2019-12-10 | 2021-06-11 | 中华电信股份有限公司 | System and method for public opinion analysis |
| CN113157763A (en) * | 2021-01-04 | 2021-07-23 | 北京汇达城数科技发展有限公司 | Accurate identification system and method for user with specified behavior information |
| CN113222632A (en) * | 2020-02-04 | 2021-08-06 | 北京京东振世信息技术有限公司 | Object mining method and device |
| WO2021253689A1 (en) * | 2020-06-15 | 2021-12-23 | 中山世达模型制造有限公司 | Multiple regression model-based method and system for predicting price of product processing |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096825A (en) * | 2016-06-03 | 2016-11-09 | 广东电网有限责任公司 | A kind of electrical power services channel analysis method and system |
| CN107103386A (en) * | 2017-04-13 | 2017-08-29 | 国网河南省电力公司电力科学研究院 | Power customer payment channel preference analyzing and predicting method based on RFM models |
| CN107220217A (en) * | 2017-05-31 | 2017-09-29 | 北京京东尚科信息技术有限公司 | Characteristic coefficient training method and device that logic-based is returned |
| CN108764663A (en) * | 2018-05-15 | 2018-11-06 | 广东电网有限责任公司信息中心 | A kind of power customer portrait generates the method and system of management |
| CN108830649A (en) * | 2018-06-05 | 2018-11-16 | 国网浙江省电力有限公司 | Change of title Electricity customers localization method for power marketing |
| CN108876076A (en) * | 2017-05-09 | 2018-11-23 | 中国移动通信集团广东有限公司 | The personal credit methods of marking and device of data based on instruction |
-
2019
- 2019-08-07 CN CN201910723897.3A patent/CN110428270A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096825A (en) * | 2016-06-03 | 2016-11-09 | 广东电网有限责任公司 | A kind of electrical power services channel analysis method and system |
| CN107103386A (en) * | 2017-04-13 | 2017-08-29 | 国网河南省电力公司电力科学研究院 | Power customer payment channel preference analyzing and predicting method based on RFM models |
| CN108876076A (en) * | 2017-05-09 | 2018-11-23 | 中国移动通信集团广东有限公司 | The personal credit methods of marking and device of data based on instruction |
| CN107220217A (en) * | 2017-05-31 | 2017-09-29 | 北京京东尚科信息技术有限公司 | Characteristic coefficient training method and device that logic-based is returned |
| CN108764663A (en) * | 2018-05-15 | 2018-11-06 | 广东电网有限责任公司信息中心 | A kind of power customer portrait generates the method and system of management |
| CN108830649A (en) * | 2018-06-05 | 2018-11-16 | 国网浙江省电力有限公司 | Change of title Electricity customers localization method for power marketing |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112949301A (en) * | 2019-12-10 | 2021-06-11 | 中华电信股份有限公司 | System and method for public opinion analysis |
| CN113222632A (en) * | 2020-02-04 | 2021-08-06 | 北京京东振世信息技术有限公司 | Object mining method and device |
| CN111709532A (en) * | 2020-05-26 | 2020-09-25 | 重庆大学 | Model-independent local interpretation-based online shopping representative sample selection system |
| CN111709532B (en) * | 2020-05-26 | 2023-09-22 | 重庆大学 | Online shopping representative sample selection system based on model-independent local interpretation |
| WO2021253689A1 (en) * | 2020-06-15 | 2021-12-23 | 中山世达模型制造有限公司 | Multiple regression model-based method and system for predicting price of product processing |
| CN112116393A (en) * | 2020-09-23 | 2020-12-22 | 贝壳技术有限公司 | Method, device and equipment for realizing event user maintenance |
| CN113157763A (en) * | 2021-01-04 | 2021-07-23 | 北京汇达城数科技发展有限公司 | Accurate identification system and method for user with specified behavior information |
| CN113157763B (en) * | 2021-01-04 | 2023-10-13 | 北京汇达城数科技发展有限公司 | Accurate identification system and method for user with specified behavior information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110428270A (en) | The potential preference client recognition methods of the channel of logic-based regression algorithm | |
| Shao et al. | Nickel price forecast based on the LSTM neural network optimized by the improved PSO algorithm | |
| CN109118067A (en) | A kind of Renewable Energy Development potential evaluation method | |
| CN110111024A (en) | Scientific and technological achievement market valuation method based on AHP model of fuzzy synthetic evaluation | |
| CN111369070A (en) | Envelope clustering-based multimode fusion photovoltaic power prediction method | |
| CN110765268B (en) | Client appeal-based accurate distribution network investment strategy method | |
| CN108364187A (en) | A kind of power failure sensitive users based on power failure sensitivity characteristic determine method and system | |
| CN108345908A (en) | Sorting technique, sorting device and the storage medium of electric network data | |
| CN106845777A (en) | Case workload evaluation method, judge's performance evaluation techniques, modeling method, computing device and computer-readable recording medium | |
| CN111260198A (en) | Method and system for judging degree of rationality of line loss in transformer area synchronization and terminal equipment | |
| CN106251241A (en) | A kind of feature based selects the LR Bagging algorithm improved | |
| CN101739614A (en) | Hierarchy-combined prediction method for communication service | |
| CN111062539A (en) | Total electric quantity prediction method based on secondary electric quantity characteristic clustering analysis | |
| Pan et al. | A study on the prediction of book borrowing based on ARIMA-SVR model | |
| CN112101673B (en) | Power grid development trend prediction method and system based on hidden Markov model | |
| CN112950048A (en) | National higher education system health evaluation based on fuzzy comprehensive evaluation | |
| CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
| CN116662860A (en) | User portrait and classification method based on energy big data | |
| CN111310974A (en) | Short-term water demand prediction method based on GA-ELM | |
| CN111160719A (en) | Two-network fused multi-index evaluation method | |
| Karimi et al. | Analyzing the results of buildings energy audit by using grey incidence analysis | |
| CN115169819A (en) | Power supply business hall efficiency data processing method | |
| CN113361776A (en) | Power load probability prediction method based on user power consumption behavior clustering | |
| Jiang et al. | Discharge estimation based on machine learning | |
| CN113780888A (en) | Comprehensive evaluation method for power quality |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |
|
| RJ01 | Rejection of invention patent application after publication |