CN107679651A - A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models - Google Patents

A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models Download PDF

Info

Publication number
CN107679651A
CN107679651A CN201710832600.8A CN201710832600A CN107679651A CN 107679651 A CN107679651 A CN 107679651A CN 201710832600 A CN201710832600 A CN 201710832600A CN 107679651 A CN107679651 A CN 107679651A
Authority
CN
China
Prior art keywords
power consumption
vector
moon
monthly
matrix
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
Application number
CN201710832600.8A
Other languages
Chinese (zh)
Inventor
蒋雪松
谭伦农
汪伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu University
Original Assignee
Jiangsu University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangsu University filed Critical Jiangsu University
Priority to CN201710832600.8A priority Critical patent/CN107679651A/en
Publication of CN107679651A publication Critical patent/CN107679651A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Tourism & Hospitality (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Water Supply & Treatment (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Operations Research (AREA)
  • Public Health (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Entrepreneurship & Innovation (AREA)

Abstract

The invention discloses a kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models, this method calculates every monthly index first, recycle per monthly index amendment moon power consumption data, logarithmic transformation then is made to the data of amendment and grey forecasting model is established with this, after prediction being fitted with the model, obtain the match value vector of moon power consumption, the predicted value vector and residual values vector of following moon power consumption, BP artificial nerve network models progress network training is inputted after residual values vector is done into calculating processing and is given a forecast, obtain residual prediction value vector, finally, value matrix must be finally predicted after predicted value vector after vectorial calculating processing with following moon power consumption is carried out into calculating processing.Compared to existing method, the present invention can significantly improve the precision of prediction of moon power consumption.

Description

A kind of moon power consumption based on grey forecasting model and BP artificial nerve network models is pre- Survey method
Technical field
The present invention relates to Load Prediction In Power Systems technical field, and grey forecasting model and BP are based on more particularly to one kind The moon electricity demand forecasting method of artificial nerve network model.
Background technology
The power system moon, electricity demand forecasting was the important component of Load Prediction In Power Systems, and it is appropriate to be mainly used in selection Machine set type, rational power supply architecture and determine fuel planning etc..If power system moon electricity demand forecasting result and future are real The border moon, power consumption compared to relatively low can cause net capacity insufficient, and power supply quality reduces；Otherwise it will cause to send out power transmission and transforming equipment utilization Rate deficiency, causes the waste of investment, reduces the economic benefit of power system.The characteristic that electric energy is unable to mass storage determines electric power The system moon electricity demand forecasting Operation of Electric Systems, scheduling in importance.
The power system moon electricity demand forecasting method mainly have regression analysis, time series analysis method, fuzzy prediction method, Grey method, artificial neural network method etc..The defects of these methods are individually present, as moon power consumption data row are rough When, the predicted value of grey method will appear from larger error.For another example the moon power consumption in obvious seasonal conversion when, the above method Rules of Seasonal Changes can not be all disclosed, therefore precision of prediction is not universal high.
Shanghai Dian Ji University proposed on July 2nd, 2013 entitled " power system load based on BP neural network is pre- The patent application of survey method ", Wenzhou University are entitled " based on improved grey model forecast model what is proposed on June 18th, 2014 Entitled " a kind of base that the patent application of electric load medium- and long-term forecasting method " and SanXia University proposed on May 4th, 2016 In the Mid-long term load forecasting method that improved grey model is theoretical " patent application all each propose a kind of Load Prediction In Power Systems side Method, but the above method is all only predicted using Individual forecast method, and method defect is very big and precision of prediction is not high.
The present invention is combined grey method and artificial neural network method by appropriate mode, and structure is based on gray prediction The moon electricity demand forecasting method of model and BP artificial nerve network models, number can not completely be disclosed by avoiding Individual forecast model The defects of according to changing rule, abundant mining data information, improve precision of prediction.
The content of the invention
For Shortcomings in the prior art, the invention provides one kind based on grey forecasting model and BP ANN The moon electricity demand forecasting method of network model, concrete methods of realizing are as follows：
A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models, its feature exist In comprising the following steps：
Step 1：More than N moon power consumption data are provided, its total moon number is n, wherein monthly power consumption is qij, wherein, I expression years, j expression months, i=1,2 ... N, 1≤j≤T, N >=4,1≤T≤12.
Step 2：Calculated according to monthly power consumption per monthly index sj：
Step 2-1：With monthly power consumption qijForming moon power consumption matrix A is： Wherein T represents that moon number and T≤12, N represent year, as monthly power consumption qijWhen can not fill up matrix A, mended with 0 element and fill up square Battle array A；
Step 2-2：Matrix A is expressed as with vectorial d：D=[d (1), d (2) ..., d (n)]=[q11,q12,...,q1T, q21,q22,...,q2T,...,qNj]；
Step 2-3：Calculate per monthly index sj：
Wherein, PjRepresent same period average, PzRepresent overall mean.
Step 3：Using the original moon power consumption data of every monthly index amendment, according to formulaObtain revised every Moon power consumption q 'ij；To revised monthly power consumption q 'ijMake logarithmic transformation, according to formula q "ij=ln (q 'ij), obtain again Revised monthly power consumption q "ij, with q "ijForming matrix A " is：When again Revised monthly power consumption q "ijMatrix A can not be filled up " when, mended with 0 element and fill up matrix A ", by matrix A " with vector x(0)Table It is shown as：x(0)=[x(0)(1),x(0)(2),....,x(0)(n)]=[q "11,q″12,...,q″1T,q″21,q″22,...,q ″2T,...,q″Nj]。
Step 4：According to revised monthly power consumption q againi″j, modeled with grey forecasting model, the gray prediction Model is GM (1,1) grey forecasting model, is fitted prediction to original moon power consumption data, obtains the fitting of monthly power consumption ValueThe predicted value of following monthly power consumptionAnd the monthly residual values e (k) of power consumption, wherein k=(1,2,3 ..., N), h=(1,2,3 ..., t), t represent total moon number of the following moon power consumption of prediction, and monthly the match value of power consumption forms moon electricity consumption The match value vector of amount The predicted value of following monthly power consumption forms the following moon The predicted value vector of power consumption Monthly the residual values of power consumption form residual values vector e,Minimum Residual difference wherein in residual values vector is mine (k).
Step 5：Residual values vector is handled：
Step 5-1：Reforming processing is carried out to residual values vector, obtaining the residual values vector e ' after reforming processing is： E '=e-mine (k)=[e ' (1), e ' (2), e ' (3) ... e ' (k) ..., e ' (n)]；
Step 5-2：Abnormal numerical value replacement is carried out to the abnormal numerical value in e ', method is as follows：
1. calculating the anomaly ratio λ (k) of each residual values in the residual values vector after reforming processing, it is：
2. to anomaly ratio according to being ranked up from big to small, and represented with vectorial β：
β=[β (1), β (2) ... β (n)]=sort [λ (1), λ (2) ... λ (k) .. λ (n)], wherein sort represent from Arrive float array function greatly；
3. calculate need to replace in the residual values vector after reforming processing abnormal residual values quantity m, m=ceil (c × N), wherein ceil represents downward bracket function, and c represents abnormal numerical value percentage；
4. if λ (k) >=β (m), makes e ' (k)=e ' (r),If R=1 or r=n,Wherein k=(1,2,3 ..., n)；
Step 5-3：Residual values vector e ", e "=[e ' (1), e ' (2) ..., e ' (r) ..., e ' after being handled (n)], wherein (r=2,3 ..., n-1).
Step 6：Residual values vector e " input BP artificial nerve network models after being handled carry out network training simultaneously Prediction residual valueObtain residual prediction value vector
Step 7：According to formulaCounter reforming processing is done to residual prediction value vector, must be repaiied Positive residual prediction value vector
Step 8：Will amendment residual prediction value be vectorial carries out calculating processing with the predicted value vector of following moon power consumption：
Step 8-1：Will amendment residual prediction value vectorWith the predicted value vector of following moon power consumptionIt is added, to being added Each numerical value in vector afterwards carries out exp function calculating, i.e.,WhereinAmong representing Variable is calculated, is usedForm vectorFor：
Step 8-2：By vectorWith matrixIt is expressed as：WhereinG represents total prediction year, works as vector In elementMatrix can not be filled upWhen, mended with 0 element and fill up matrix
Step 8-3：WillIt is multiplied by every monthly index sjObtain final predicted valueI.e.Final predicted valueForm Final prediction value matrixFor：WhereinFor matrixIn element, g represent prediction year Number, g=1,2 ... G, when final predicted valueMatrix can not be filled upWhen, mended with 0 element and fill up matrix
The beneficial effects of the present invention are：
1. monthly index method and logarithmic transformation method combine the grey forecasting model for being modified to moon electricity demand forecasting, fully anti- The seasonal conversion of moon power consumption is mirrored, data light slippery is substantially increased, so as to improve the precision of prediction of grey forecasting model.
2. vectorial the reforming of caused residual values is handled after grey forecasting model is predicted, greatly reduce residual values to Sign prediction error that amount is likely to occur when being predicted in BP artificial nerve network models and the prediction error increased.
3. caused residual values vector makees amendment outlier processing after grey forecasting model is predicted, data smoothing is improved Degree, so as to improve precision of prediction of the residual values vector in BP artificial nerve network models.
4. the formula that abnormal residual values replace processing makes abnormal numerical value percentage c flexibly to be adjusted according to different prediction cases It is whole, for example, the anomaly ratio of each residual values be all closer to, without big fluctuation when, abnormal numerical value percentage can be reduced；It is on the contrary Abnormal numerical value percentage can be increased.
5. as shown in moon power consumption matrix A, original moon power consumption data comprising all months in each year without monthly using Electricity, and nearest 1 year the moon power consumption data can be less than before each year moon power consumption data, this causes the inventive method Flexibility is very big during use.
6. making full use of BP artificial nerve network models to draw unstructuredness, the feature of non-precision rule, will do Cross reforming processing and abnormal numerical value is replaced the residual values vector input BP artificial nerve network models handled and given a forecast, must provide There are unstructuredness, the residual prediction value vector of non-precision rule, predict to obtain not with the vector corrected grey forecasting model The predicted value vector of next moon power consumption, improves precision of prediction.
7. compared with single use grey forecasting model or BP artificial nerve network models do moon electricity demand forecasting, the present invention Both are combined by appropriate mode, lacking for data variation rule can not completely be disclosed by avoiding Individual forecast model Fall into, abundant mining data information, improve precision of prediction.
Brief description of the drawings
Fig. 1 is that a kind of moon power consumption based on grey forecasting model and BP artificial nerve network models of the present invention is pre- Survey method flow diagram.
Fig. 2 is that a kind of moon power consumption based on grey forecasting model and BP artificial nerve network models of the present invention is pre- Survey method and each model predictive error broken line comparison diagram.
Embodiment
Below in conjunction with the accompanying drawings and specific embodiment the present invention is further illustrated, but protection scope of the present invention is simultaneously Not limited to this.
A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models, its feature exist In as shown in figure 1, comprising the following steps：
Step 1：More than N moon power consumption data are provided, its total moon number is n, wherein monthly power consumption is qij, wherein, I expression years, j expression months, i=1,2 ... N, 1≤j≤T, N >=4,1≤T≤12.
Step 2：Calculated according to monthly power consumption per monthly index sj：
Step 2-1：With monthly power consumption qijForming moon power consumption matrix A is： Wherein T represents that moon number and T≤12, N represent year, as monthly power consumption qijWhen can not fill up matrix A, mended with 0 element and fill up square Battle array A；
Step 2-2：Matrix A is expressed as with vectorial d：D=[d (1), d (2) ..., d (n)]=[q11,q12,...,q1T, q21,q22,...,q2T,...,qNj],；
Step 2-3：Calculate per monthly index sj：
Wherein, PjRepresent same period average, PzRepresent overall mean.
Step 3：Using the original moon power consumption data of every monthly index amendment, according to formulaObtain revised every Moon power consumption q 'ij；To revised monthly power consumption q 'ijMake logarithmic transformation, according to formula q "ij=ln (q 'ij), obtain again Revised monthly power consumption q "ij, with q "ijForming matrix A " is：When again Revised monthly power consumption q "ijMatrix A can not be filled up " when, mended with 0 element and fill up matrix A ", by matrix A " with vector x(0)Table It is shown as：x(0)=[x(0)(1),x(0)(2),....,x(0)(n)]=[q "11,q″12,...,q″1T,q″21,q″22,...,q ″2T,...,q″Nj]。
Step 4：According to revised monthly power consumption q " againij, modeled with grey forecasting model, the gray prediction Model is GM (1,1) grey forecasting model, is fitted prediction to original moon power consumption data, obtains the fitting of monthly power consumption ValueThe predicted value of following monthly power consumptionAnd the monthly residual values e (k) of power consumption, wherein k=(1,2,3 ..., N), h=(1,2,3 ..., t), t represent total moon number of the following moon power consumption of prediction, and monthly the match value of power consumption forms moon electricity consumption The match value vector of amount The predicted value of following monthly power consumption forms the following moon The predicted value vector of power consumption Monthly the residual values of power consumption form residual values vector e,Minimum Residual difference wherein in residual values vector is mine (k)；
The modeling procedure of grey forecasting model is as follows in the step 4：
Step 4-1：To x(0)(k) make that single order is cumulative to obtain x(1)(k), and vector x is formed(1)：x(1)=[x(1)(1),x(1) (2),....,x(1)(n)],Wherein k=(1,2 ..., n)；
Step 4-2：x(1)Do single order average generation and obtain x (k), and form vector x：X=[x (2), x (3) ... .x (n)], x (k)=- 1/2 [x(1)(k-1)+x(1)(k)], wherein, k=(2,3 ..., n)；
Step 4-3：To x(1)(k) differential equation of first order is established：Parameter a in formula, u can be by a most young waiters in a wineshop or an inn Multiplication is tried to achieve：Matrix B, data vector Y in formulaNFor：
Step 4-4：In primary conditionUnder, solve the differential equationObtain：
Wherein k=(0,1,2 ..., n-1)；
Step 4-5：According to formulaPredicted value is reduced, wherein k=(1, 2,…,n-1)；
Step 4-6：Reduce to obtain x by regressive(0)Model of fit be：
Wherein,
x(0)Forecast model be：
OrderObtain grey forecasting model.
Step 5：Residual values vector is handled：
Step 5-1：Reforming processing is carried out to residual values vector, obtaining the residual values vector e ' after reforming processing is： E '=e-mine (k)=[e ' (1), e ' (2), e ' (3) ... e ' (k) ..., e ' (n)]；
Step 5-2：Abnormal numerical value replacement is carried out to the abnormal numerical value in e ', method is as follows：
5. calculating the anomaly ratio λ (k) of each residual values in the residual values vector after reforming processing, it is：
6. to anomaly ratio according to being ranked up from big to small, and represented with vectorial β：
β=[β (1), β (2) ... β (n)]=sort [λ (1), λ (2) ... λ (k) .. λ (n)], wherein sort represent from Arrive float array function greatly；
7. calculate need to replace in the residual values vector after reforming processing abnormal residual values quantity m, m=ceil (c × N), wherein ceil represents downward bracket function, and c represents abnormal numerical value percentage；
8. if λ (k) >=β (m), makes e ' (k)=e ' (r),If R=1 or r=n,Wherein k=(1,2,3 ..., n)；
Step 5-3：Residual values vector e ", e "=[e ' (1), e ' (2) ..., e ' (r) ..., e ' after being handled (n)], wherein (r=2,3 ..., n-1).
Step 6：Residual values vector e " input BP artificial nerve network models after being handled carry out network training simultaneously Prediction residual valueObtain residual prediction value vector
Step 7：According to formulaCounter reforming processing is done to residual prediction value vector, must be repaiied Positive residual prediction value vector
Step 8：Will amendment residual prediction value be vectorial carries out calculating processing with the predicted value vector of following moon power consumption：
Step 8-1：Will amendment residual prediction value vectorWith the predicted value vector of following moon power consumptionIt is added, to being added Each numerical value in vector afterwards carries out exp function calculating, i.e.,WhereinAmong representing Variable is calculated, is usedForm vectorFor：
Step 8-2：By vectorWith matrixIt is expressed as：WhereinG represents total prediction year, works as vector In elementMatrix can not be filled upWhen, mended with 0 element and fill up matrix
Step 8-3：WillIt is multiplied by every monthly index sjObtain final predicted valueI.e.Final predicted valueForm Final prediction value matrixFor：WhereinFor matrixIn element, g represent prediction Year, g=1,2 ... G, when final predicted valueMatrix can not be filled upWhen, mended with 0 element and fill up matrix
Embodiment
Using 2008 to 2014 industrial moon power consumptions of certain electric company of city as initial data, establish according to the method described above Forecast model, and predict industrial moon power consumption in 2015.City's industry moon power consumption initial data such as table 1 below,
Certain the city's industry moon power consumption initial data of table 1
Unit：Ten thousand kilowatt hours
It is as follows to carry out network training and prediction process of the residual values vector of processing in BP artificial nerve network models： The residual values vector for carrying out processing using 2008 to 2013 inputs as BP artificial neural networks, and the conduct of 2014 is defeated Go out.Through analyzing repeatedly, network structure is defined as 3 layer networks：Input layer contains 6 neurons, and hidden layer contains 9 neurons, output Layer is 1 neuron.The transmission function of input layer to hidden layer uses tansig functions, the transmission function of hidden layer to output layer Using purelin functions, network training function uses traingdm functions.Weights and the initial value of threshold value use random number, training Target error is set to 0.001, and learning rate is set to 0.01, and momentum coefficient is set to 0.95.Premnmx functions are used during training first by institute There is sample to be normalized, treat that training is finished, make the predicted value obtained by sim functions at renormalization of postmnmx functions Manage to obtain residual prediction value vector.By nearly 18000 learning training, one group of suitable weights and threshold value are obtained.Afterwards with training Neural network forecast in January, 2015 to December residual prediction value vector.
For ease of comparing the prediction effect of different type GM (1,1) model, table 2 gives the pre- of Traditional GM (1,1) model Measured value B and error B ', the predicted value C and error C ' of GM (1,1) model based on logarithmic transformation, is based only upon every monthly index and logarithm The predicted value D and error D ' of GM (1,1) model of conversion, and the predicted value E and error E being predicted using the present invention '.Will Each error is depicted as line chart as shown in Figure 2.
2 each model predication value of table and errors table
It can be drawn with reference to table 2 and Fig. 2, the mean error of Traditional GM (1,1) model prediction is 13.07%.Based on logarithm The mean error of GM (1,1) model prediction of conversion is 13.51%.It is based only upon every monthly index and the gray prediction mould of logarithmic transformation The mean error of type prediction is 9.80%.Moon electricity demand forecasting side based on grey forecasting model Yu BP artificial nerve network models The mean error of method prediction is 5.17%.
As can be seen that Traditional GM (1,1) model and the utilization of the power consumption data seasonal variety of the moon in 1 year are not accounted for The prediction result of GM (1,1) model of logarithmic transformation is all unsatisfactory.Even the ash based on every monthly index and logarithmic transformation Color forecast model, although precision of prediction has large increase, because its model has residual error after fitting, greatly govern The raising of precision of prediction.Residual values vector reform of GM (1,1) grey forecasting model proposed by the present invention after fitting and BP artificial nerve network models are inputted after abnormal numerical value replacement processing and carry out network training and the worth residual prediction value of prediction residual Vector, finally use the prediction of the following moon power consumption of treated residual prediction value vector corrected GM (1,1) grey forecasting model Value vector, the results showed that prediction error is remarkably decreased.
The embodiment is preferred embodiment of the invention, but the present invention is not limited to above-mentioned embodiment, not Away from the present invention substantive content in the case of, those skilled in the art can make it is any it is conspicuously improved, replace Or modification belongs to protection scope of the present invention.

Claims (7)

  1. A kind of 1. moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models, it is characterised in that Comprise the following steps：
    Step 1：More than N moon power consumption data are provided, its total moon number is n, wherein monthly power consumption is qij, wherein, i tables Showing year, j represents month, i=1,2 ... N, 1≤j≤T, N >=4,1≤T≤12；
    Step 2：Calculated according to monthly power consumption per monthly index sj；
    Step 3：Using the original moon power consumption data of every monthly index amendment, revised monthly power consumption q ' is obtainedij, to amendment Monthly power consumption q ' afterwardsijMake logarithmic transformation, obtain revised monthly power consumption q " againij；
    Step 4：According to revised monthly power consumption q " againij, modeled with grey forecasting model, to original moon power consumption number According to prediction is fitted, the match value of monthly power consumption is obtainedThe predicted value of following monthly power consumptionAnd monthly use The residual values e (k) of electricity, wherein k=(1,2,3 ..., n), h=(1,2,3 ..., t), t represent to predict following moon power consumption The match value vector of total moon number, monthly the match value composition moon power consumption of power consumption The predicted value of following monthly power consumption forms the predicted value vector of following moon power consumption Monthly use The residual values of electricity form residual values vector e,Wherein in residual values vector Minimum Residual difference is mine (k)；
    Step 5：Residual values vector is handled；
    Step 6：Residual values vector e " input BP artificial nerve network models after being handled carry out network training and predicted Residual valuesObtain residual prediction value vector
    Step 7：Counter reforming processing is done to residual prediction value vector, residual prediction value vector must be corrected
    Step 8：Will amendment residual prediction value is vectorial carries out calculating processing with the predicted value vector of following moon power consumption, obtain most Prediction value matrix eventually.
  2. A kind of 2. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that per monthly index s in the step 2jComputational methods are as follows：
    Step 2-1：With monthly power consumption qijForming moon power consumption matrix A is：Its Middle T represents moon number and T≤12, N represent year, as monthly power consumption qijWhen can not fill up matrix A, mended with 0 element and fill up matrix A；
    Step 2-2：Matrix A is expressed as with vectorial d：
    D=[d (1), d (2) ..., d (n)]=[q11,q12,...,q1T,q21,q22,...,q2T,...,qNj]；
    Step 2-3：Calculate per monthly index sj：
    Wherein, PjRepresent same period average, PzRepresent overall mean.
  3. A kind of 3. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that to revised monthly power consumption q " again in the step 3ijThe implementation method of amendment is as follows：
    Step 3-1：According to formulaOriginal moon power consumption data are modified, obtain revised monthly power consumption q′ij；
    Step 3-2：According to formula q "ij=ln (q 'ij) to q 'ijLogarithmic transformation is done, obtains revised monthly power consumption again q″ij, with q "ijForming matrix A " is：When revised monthly power consumption again q″ijMatrix A can not be filled up " when, mended with 0 element and fill up matrix A "；
    Step 3-3：By matrix A " with vector x(0)It is expressed as：x(0)=[x(0)(1),x(0)(2),....,x(0)(n)]=[q "11, q″12,...,q″1T,q″21,q″22,...,q″2T,...,q″Nj]。
  4. A kind of 4. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that the method handled in the step 5 residual values vector is：
    Step 5-1：Reforming processing is carried out to residual values vector, obtaining the residual values vector e ' after reforming processing is：E '= E-mine (k)=[e ' (1), e ' (2), e ' (3) ... e ' (k) ..., e ' (n)]；
    Step 5-2：Abnormal numerical value replacement is carried out to the abnormal numerical value in e ',
    1. calculating the anomaly ratio λ (k) of each residual values in the residual values vector after reforming processing, it is：
    <mrow> <mi>&amp;lambda;</mi> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> <mo>=</mo> <mo>|</mo> <mfrac> <mrow> <msup> <mi>ne</mi> <mo>&amp;prime;</mo> </msup> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </mrow> <mrow> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msup> <mi>e</mi> <mo>&amp;prime;</mo> </msup> <mrow> <mo>(</mo> <mi>k</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mn>1</mn> <mo>|</mo> <mo>&amp;times;</mo> <mn>100</mn> <mi>%</mi> <mo>;</mo> </mrow>
    2. to anomaly ratio according to being ranked up from big to small, and represented with vectorial β：
    β=[β (1), β (2) ... β (n)]=sort [λ (1), λ (2) ... λ (k) .. λ (n)], wherein sort represent from greatly to Float array function；
    3. abnormal residual values the quantity m, m=ceil (c × n) for needing to replace in the residual values vector after reforming processing are calculated, Wherein ceil represents downward bracket function, and c represents abnormal numerical value percentage；
    4. if λ (k) >=β (m), makes e ' (k)=e ' (r),(r=2,3 ..., n-1), if r= 1 or r=n,Wherein k=(1,2,3 ..., n)；
    Step 5-3：Residual values vector e ", e "=[e ' (1), e ' (2) ..., e ' (r) ..., e ' (n)] after being handled, its In (r=2,3 ..., n-1).
  5. A kind of 5. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that the formula of counter reforming processing is in the step 7：
  6. A kind of 6. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that the computation processing method in the step 8 is：
    Step 8-1：Will amendment residual prediction value vectorWith the predicted value vector of following moon power consumptionIt is added, after addition Each numerical value in vector carries out exp function calculating, i.e.,WhereinRepresent intermediate computations Variable, useForm vectorFor：
    Step 8-2：By vectorWith matrixIt is expressed as：WhereinG represents total prediction year, works as vector In elementMatrix can not be filled upWhen, mended with 0 element and fill up matrix
    Step 8-3：WillIt is multiplied by every monthly index sjObtain final predicted valueI.e.Final predicted valueForm final Predict value matrixFor：WhereinFor matrixIn element, g represent prediction year, G=1,2 ... G, when final predicted valueMatrix can not be filled upWhen, mended with 0 element and fill up matrix
  7. A kind of 7. moon electricity demand forecasting based on grey forecasting model Yu BP artificial nerve network models according to claim 1 Method, it is characterised in that the grey forecasting model is GM (1,1) grey forecasting model.
CN201710832600.8A 2017-09-15 2017-09-15 A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models Pending CN107679651A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710832600.8A CN107679651A (en) 2017-09-15 2017-09-15 A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710832600.8A CN107679651A (en) 2017-09-15 2017-09-15 A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models

Publications (1)

Publication Number Publication Date
CN107679651A true CN107679651A (en) 2018-02-09

Family

ID=61136985

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710832600.8A Pending CN107679651A (en) 2017-09-15 2017-09-15 A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models

Country Status (1)

Country Link
CN (1) CN107679651A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109034905A (en) * 2018-08-03 2018-12-18 四川长虹电器股份有限公司 The method for promoting sales volume prediction result robustness
CN110610226A (en) * 2018-06-14 2019-12-24 北京德知航创科技有限责任公司 Generator fault prediction method and device
CN112102004A (en) * 2020-09-18 2020-12-18 合肥工业大学 Click rate prediction fusion method based on residual error learning

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110610226A (en) * 2018-06-14 2019-12-24 北京德知航创科技有限责任公司 Generator fault prediction method and device
CN109034905A (en) * 2018-08-03 2018-12-18 四川长虹电器股份有限公司 The method for promoting sales volume prediction result robustness
CN112102004A (en) * 2020-09-18 2020-12-18 合肥工业大学 Click rate prediction fusion method based on residual error learning
CN112102004B (en) * 2020-09-18 2022-08-30 合肥工业大学 Click rate prediction fusion method based on residual error learning

Similar Documents

Publication Publication Date Title
WO2021189739A1 (en) Production plan-based iron and steel plant oxygen load prediction method
Zhou et al. Optimizing carbon emission forecast for modelling China's 2030 provincial carbon emission quota allocation
CN109002942A (en) A kind of short-term load forecasting method based on stochastic neural net
CN107169633A (en) A kind of gas line network, gas storage peak regulating plan integrated evaluating method
CN106779277B (en) Classified evaluation method and device for network loss of power distribution network
CN112149879B (en) New energy medium-and-long-term electric quantity prediction method considering macroscopic volatility classification
Sun et al. Exploration of energy saving potential in China power industry based on Adaboost back propagation neural network
CN108876132B (en) Industrial enterprise energy efficiency service recommendation method and system based on cloud
CN107909253A (en) Intelligent distribution network scheduling controlling effect evaluation method based on interval based AHP
CN106529704A (en) Monthly maximum power load forecasting method and apparatus
CN112183849A (en) Short-term electric power quantity prediction method based on artificial neural network
CN104573857A (en) Power grid load rate prediction method based on intelligent algorithm optimization and combination
CN108596242A (en) Power grid meteorology load forecasting method based on wavelet neural network and support vector machines
CN109409770A (en) A kind of rural power grids level of intelligence evaluation method neural network based
CN107679651A (en) A kind of moon electricity demand forecasting method based on grey forecasting model Yu BP artificial nerve network models
CN112801687A (en) Overhead line engineering-based cost prediction model construction method
CN104834975A (en) Power network load factor prediction method based on intelligent algorithm optimization combination
CN110635518A (en) Source network load and storage optimization method based on photovoltaic high permeability
CN112054507A (en) Power distribution low-voltage distribution area theoretical line loss interval calculation method based on convolutional neural network
CN112085156B (en) Partial pressure line loss rate prediction and loss reduction potential evaluation method and system
CN112819225A (en) Carbon market price prediction method based on BP neural network and ARIMA model
CN114936742A (en) Water supply system scheduling agent decision method
CN117422165A (en) Urban water delivery system water quantity prediction method and system based on low carbon emission
Yılmaz Real measure of a transmission line data with load fore-cast model for the future
Liu et al. Collaborative optimization of dynamic grid dispatch with wind power

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20180209

RJ01 Rejection of invention patent application after publication