CN107944617A - A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns - Google Patents
A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns Download PDFInfo
- Publication number
- CN107944617A CN107944617A CN201711154727.5A CN201711154727A CN107944617A CN 107944617 A CN107944617 A CN 107944617A CN 201711154727 A CN201711154727 A CN 201711154727A CN 107944617 A CN107944617 A CN 107944617A
- Authority
- CN
- China
- Prior art keywords
- stealing
- function
- gradient
- weights
- logic
- 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
- 238000005457 optimization Methods 0.000 title claims abstract description 16
- 238000000034 method Methods 0.000 title claims abstract description 15
- 238000010276 construction Methods 0.000 claims abstract description 10
- 238000011478 gradient descent method Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 41
- 230000007423 decrease Effects 0.000 claims description 4
- 230000001419 dependent effect Effects 0.000 claims description 4
- 230000005611 electricity Effects 0.000 abstract description 4
- 238000007476 Maximum Likelihood Methods 0.000 abstract description 3
- 238000007477 logistic regression Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012795 verification Methods 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of doubtful stealing theme influence factor weight optimization method that logic-based returns, using maximum likelihood method construction logic regression forecasting function, and the logarithmic function that can be led by establishing arbitrary order is used as loss function, and globally optimal solution can be obtained using using gradient descent method optimizing factors weights, it is final to obtain high accurancy and precision volume anticipation function, so as to draw the doubtful stealing theme influence factor weight for meeting history electricity stealing sample, it is low to solve conventional stealing probabilistic forecasting precision, too rely on the problem of artificial, and improve the precision of prediction.
Description
Technical field
The present invention relates to a kind of weight of doubtful stealing theme influence factor, particularly a kind of logic-based returns doubtful
Stealing theme influence factor weight optimization method.
Background technology
With the fast development of China's economy, people are continuously increased for the demand of electric power, and power grid enterprises develop rapidly, steal
Electric problem is more serious than ever.Also by having turned to corporation from single individual, the concealment of stealing is also stealing main body
It is higher and higher, if supervision of power consumption personnel specialty quality is not profound, insight is inadequate, it is difficult to find the row of this stealing electric power
For.State Grid Corporation of China proposed metering on-line monitoring and intelligent diagnostics model, its stealing model energy included in 2015
Enough power utility check personnel are helped to analyze stealing problem, but since the decision rule of the model is higher to the dependency degree of subjective experience,
Be difficult to accurately catch the diversified stealing of user behavioural characteristic, there is rate of false alarm it is excessive the problems such as.
The content of the invention
It is an object of the invention to overcome the deficiencies in the prior art part, by dividing into row conventional stealing sample data
Analysis, adjustment is optimized to metering on-line monitoring with stealing model in intelligent diagnostics model, each using logistic regression algorithm optimization
The weights of correlative factor, construct the computation model that can predict stealing user, so as to precisely predict the general of user's stealing
Rate, reduces concerned power economic loss.
A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns, (1) uses Sigmoid letters
Number construction logic regression forecasting function, the anticipation function are：Wherein, θiRepresent corresponding factor xiWeights, i ∈ (0,
1,...k)；
(2) loss function returned using logarithmic function construction logic, the loss function are：Wherein yiFor dependent variable, i.e., whether be stealing user this
As a result, above required hθ(x) probability that y takes 1 is represented, wherein y takes 1 expression stealing user；
(3) using gradient descent method renewal weights：Weights θ is initialized first0,θ1,...,θk, threshold value η and learning rate α；
Comprise the following steps that：1) gradient of the loss function of current location is determined first, for θi, its pressure gradient expression formula
For：
2) gradient of loss function is multiplied by with learning rate, the distance i.e. step-length for obtaining current location decline is：Wherein, learning rate can in the range of 0.1~0.5 multiple exploitation, according to convergence rate and approaching optimal solution
Situation selection Optimal learning efficiency value；
3) determine that the distance that each θ values gradient declines is both less than η, if less than η, then algorithm termination, otherwise enters step
Suddenly (4), wherein, the 1% of step-length first may be selected in gradient falling-threshold value；
4) θ values are updated according to the following formula：Then step (1) is gone back to, it is continuous by above step
Iteration, can draw the weights of optimization, form final anticipation function, so as to precisely predict the probability of user's stealing.
Traditional consideration stealing influence factor weight is too dependent on artificial judgement, be subject to subjective factor influenced compared with
Greatly.This method carries out weight optimization by the way of logistic regression to the influence factor of stealing theme, on the one hand can be objective anti-
The correlation degree between data factors and stealing result is reflected, on the other hand also can be by stealing verification as a result, further right
Model optimization, by constantly verifying and revision models weight, the self study and self-optimizing of implementation model.
Logistic regression algorithm is applied on power grid enterprises identification stealing user by the present invention, can be relatively accurately fitted pre-
Function is surveyed, predicts stealing user.The present invention is based on Sigmond functions, utilizes the spy of its change rate maximum when z values are 0.5
Property, based on 1:1 stealing and non-stealing sample architecture anticipation function.Maximum-likelihood estimation is based on, is built and damaged using logarithmic function
Lose function so that there is the local optimum in the loss function solution procedure of convex function characteristic to be approximately equal to globally optimal solution, keep away
Exempt from because stealing number of users is big cause weight optimization to be absorbed in local optimum the problem of.Weights are updated using gradient descent method, are made
Obtain weights and solve and be most worth along the most fast direction finding of function change, since the factor involved by stealing user is numerous, using ladder
Degree descent method can obtain optimal solution faster.
In summary, present invention advantage following compared with prior art：
The present invention uses maximum likelihood method construction logic regression forecasting function, and the logarithm letter that can be led by establishing arbitrary order
Number is used as loss function, and can obtain globally optimal solution using using gradient descent method optimizing factors weights, and final acquisition is high
Precision volume anticipation function, so as to draw the doubtful stealing theme influence factor weight for meeting history electricity stealing sample, solves
Conventional stealing probabilistic forecasting precision is low, it is undue rely on the problem of artificial, and improve the precision of prediction.
Brief description of the drawings
Fig. 1 is the flow chart for the doubtful stealing theme influence factor weight optimization method that the logic-based of the present invention returns.
Embodiment
The present invention is described in more detail with reference to embodiment.
Embodiment 1
A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns, (1) uses Sigmoid letters
Number construction logic regression forecasting function, the anticipation function are：
Wherein, θiRepresent corresponding factor xiWeights, i ∈ (0,1 ... k)；
(2) loss function returned using logarithmic function construction logic, the loss function are：Wherein yiFor dependent variable, i.e., whether be stealing user this as a result,
Above required hθ(x) probability that y takes 1 is represented, wherein y takes 1 expression stealing user；
(3) using gradient descent method renewal weights：Weights θ is initialized first0,θ1,...,θk, threshold value η and learning rate α；
Comprise the following steps that：1) gradient of the loss function of current location is determined first, for θi, its pressure gradient expression formula
For：
2) gradient of loss function is multiplied by with learning rate, the distance i.e. step-length for obtaining current location decline is：Wherein, learning rate can in the range of 0.1~0.5 multiple exploitation, according to convergence rate and approaching optimal solution
Situation selection Optimal learning efficiency value；
3) determine that the distance that each θ values gradient declines is both less than η, if less than η, then algorithm termination, otherwise enters step
Suddenly (4), wherein, the 1% of step-length first may be selected in gradient falling-threshold value；
4) θ values are updated according to the following formula：Then step (1) is gone back to, by above step not
Disconnected iteration, can draw the weights of optimization, form final anticipation function, so as to precisely predict the probability of user's stealing.
The flow chart for the doubtful stealing theme influence factor weight optimization method that logic-based as shown in Figure 1 returns, it is first
First construction logic regression forecasting function, then construction logic recurrence loss function, using gradient descent method, calculates and stealing phase
The gradient of the factor weights of pass, tries to achieve the step-length of this decline, if step-length is less than threshold value, algorithm terminates, no after calculating every time
Then, weights are just updated, gradient and step-length is recalculated, is thus iterated, finally solve anticipation function, obtain each factor power
Weight values.
Illustrate the technique effect of the present invention with the example of concrete application of the present invention below：Experimental data nets certain electric power public affairs with state
Take charge of the 2000 stealing users chosen and its related electricity consumption data, 2000 non-stealing users and its a period of time during stealing
Related electricity consumption data, collectively constitute sample.Wherein there are 22 correlative factors, a pair of of correlation combiner factor.
All correlative factors and correlation combiner factor are substituted into logistic regression computational methods, declining step according to gradient carries out
Calculate.It is 0.5, η=0.1 to initialize all θ values, and learning rate takes α=0.1 and α=0.5 respectively, the θ updated per this iteration
Value, updates anticipation function and is predicted sample substitution.
After final iteration 152 times, solving result no longer changes, and sample substitution anticipation function is calculated, what is obtained is flat
Equal accuracy rate is 94.1%, and average rate of precision is 96.6%, and average recall rate is 95.8%.
The not described part of the present embodiment is same as the prior art.
Claims (1)
1. a kind of doubtful stealing theme influence factor weight optimization method that logic-based returns, it is characterised in that step is as follows：
(1) Sigmoid function construction logic regression forecasting functions are used, the anticipation function is：Wherein, θiRepresent corresponding factor xiWeights, i ∈ (0,
1,...k)；
(2) loss function returned using logarithmic function construction logic, the loss function are：Wherein yiFor dependent variable, i.e., whether be stealing user this
As a result, above required hθ(x) probability that y takes 1 is represented, wherein y takes 1 expression stealing user；
(3) using gradient descent method renewal weights：Weights θ is initialized first0,θ1,...,θk, threshold value η and learning rate α；
Comprise the following steps that：1) gradient of the loss function of current location is determined first, for θi, its pressure gradient expression formula is：
2) gradient of loss function is multiplied by with learning rate, the distance i.e. step-length for obtaining current location decline is：Its
In, learning rate multiple exploitation, foundation convergence rate can select most in the range of 0.1~0.5 with the situation for approaching optimal solution
Good learning rate value；
3) determine that the distance that each θ values gradient declines is both less than η, if less than η, then algorithm termination, otherwise enters step
(4), wherein, the 1% of step-length first may be selected in gradient falling-threshold value；
4) θ values are updated according to the following formula：Then step (1) is gone back to, by the continuous iteration of above step,
It can draw the weights of optimization, form final anticipation function, so as to precisely predict the probability of user's stealing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711154727.5A CN107944617A (en) | 2017-11-20 | 2017-11-20 | A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711154727.5A CN107944617A (en) | 2017-11-20 | 2017-11-20 | A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107944617A true CN107944617A (en) | 2018-04-20 |
Family
ID=61931957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711154727.5A Pending CN107944617A (en) | 2017-11-20 | 2017-11-20 | A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107944617A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765340A (en) * | 2018-05-29 | 2018-11-06 | Oppo（重庆）智能科技有限公司 | Fuzzy image processing method, apparatus and terminal device |
| 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 |
| CN112070559A (en) * | 2020-09-17 | 2020-12-11 | 贝壳技术有限公司 | State acquisition method and device, electronic equipment and storage medium |
| CN113205219A (en) * | 2021-05-12 | 2021-08-03 | 大连大学 | Agricultural water quality prediction method based on gradient descent optimization logistic regression algorithm |
| CN114548048A (en) * | 2022-02-23 | 2022-05-27 | 南京审计大学 | Bank false alarm detection method based on text theme index |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104751374A (en) * | 2015-03-27 | 2015-07-01 | 深圳供电局有限公司 | Metering automation system wireless communication failure prediction method |
| CN106845107A (en) * | 2017-01-19 | 2017-06-13 | 中国南方电网有限责任公司 | User's stealing probability forecasting method, device and equipment based on Trust Region Algorithm |
| CN107145966A (en) * | 2017-04-12 | 2017-09-08 | 山大地纬软件股份有限公司 | Logic-based returns the analysis and early warning method of opposing electricity-stealing of probability analysis Optimized model |
| CN112101471A (en) * | 2020-09-21 | 2020-12-18 | 国网辽宁省电力有限公司电力科学研究院 | Electricity stealing probability early warning analysis method |
-
2017
- 2017-11-20 CN CN201711154727.5A patent/CN107944617A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104751374A (en) * | 2015-03-27 | 2015-07-01 | 深圳供电局有限公司 | Metering automation system wireless communication failure prediction method |
| CN106845107A (en) * | 2017-01-19 | 2017-06-13 | 中国南方电网有限责任公司 | User's stealing probability forecasting method, device and equipment based on Trust Region Algorithm |
| CN107145966A (en) * | 2017-04-12 | 2017-09-08 | 山大地纬软件股份有限公司 | Logic-based returns the analysis and early warning method of opposing electricity-stealing of probability analysis Optimized model |
| CN112101471A (en) * | 2020-09-21 | 2020-12-18 | 国网辽宁省电力有限公司电力科学研究院 | Electricity stealing probability early warning analysis method |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765340A (en) * | 2018-05-29 | 2018-11-06 | Oppo（重庆）智能科技有限公司 | Fuzzy image processing method, apparatus and terminal device |
| 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 |
| CN112070559A (en) * | 2020-09-17 | 2020-12-11 | 贝壳技术有限公司 | State acquisition method and device, electronic equipment and storage medium |
| CN113205219A (en) * | 2021-05-12 | 2021-08-03 | 大连大学 | Agricultural water quality prediction method based on gradient descent optimization logistic regression algorithm |
| CN114548048A (en) * | 2022-02-23 | 2022-05-27 | 南京审计大学 | Bank false alarm detection method based on text theme index |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107944617A (en) | A kind of doubtful stealing theme influence factor weight optimization method that logic-based returns | |
| CN106453293B (en) | A kind of network security situation prediction method based on improved BPNN | |
| CN106485089B (en) | The interval parameter acquisition methods of harmonic wave user's typical condition | |
| CN109547431A (en) | A kind of network security situation evaluating method based on CS and improved BP | |
| CN111598224B (en) | Air conditioner cold load prediction method for optimizing BP neural network based on simulated annealing algorithm | |
| CN110837915B (en) | Low-voltage load point prediction and probability prediction method for power system based on hybrid integrated deep learning | |
| CN105631532A (en) | Power system load prediction method using fuzzy decision-based neural network model | |
| CN113780684A (en) | Intelligent building user energy consumption behavior prediction method based on LSTM neural network | |
| CN115310727B (en) | Building cooling, heating and power load prediction method and system based on transfer learning | |
| CN104539601A (en) | Reliability analysis method and system for dynamic network attack process | |
| CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
| CN113762591B (en) | Short-term electric quantity prediction method and system based on GRU and multi-core SVM countermeasure learning | |
| CN115130830A (en) | Non-intrusive load decomposition method based on cascade width learning and sparrow algorithm | |
| Lingqing et al. | Detection method for power theft based on SOM neural network and K-means clustering algorithm | |
| CN113361776A (en) | Power load probability prediction method based on user power consumption behavior clustering | |
| CN117875752A (en) | Power system flexible operation domain assessment method based on self-organizing map decision tree | |
| CN106599541B (en) | A kind of structure and parameter on-line identification method of dynamic power load model | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| Zhang | Short‐Term Power Load Forecasting Based on SAPSO‐CNN‐LSTM Model considering Autocorrelated Errors | |
| Ma et al. | Traffic flow forecasting research based on Bayesian normalized Elman neural network | |
| CN114091782B (en) | Medium-long term power load prediction method | |
| CN113884936B (en) | ISSA coupling DELM-based lithium ion battery health state prediction method | |
| CN113255223B (en) | Short-term prediction method and system for air conditioner load | |
| CN115619028A (en) | Clustering algorithm fusion-based power load accurate prediction method | |
| CN112327046B (en) | Non-invasive load monitoring method based on fuzzy clustering and support vector regression |
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 |