CN104809522B - A kind of comprehensive energy prediction technique - Google Patents
A kind of comprehensive energy prediction technique Download PDFInfo
- Publication number
- CN104809522B CN104809522B CN201510228035.5A CN201510228035A CN104809522B CN 104809522 B CN104809522 B CN 104809522B CN 201510228035 A CN201510228035 A CN 201510228035A CN 104809522 B CN104809522 B CN 104809522B
- Authority
- CN
- China
- Prior art keywords
- sample
- energy
- layer
- consuming
- time
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000013507 mapping Methods 0.000 claims abstract description 28
- 238000005265 energy consumption Methods 0.000 claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 238000011161 development Methods 0.000 claims abstract description 12
- 239000013598 vector Substances 0.000 claims description 77
- 230000035945 sensitivity Effects 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 24
- 238000012545 processing Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000005611 electricity Effects 0.000 claims description 12
- 230000008859 change Effects 0.000 claims description 11
- 241001269238 Data Species 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 238000013139 quantization Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 8
- 239000006185 dispersion Substances 0.000 claims description 8
- 239000003245 coal Substances 0.000 claims description 6
- 238000000205 computational method Methods 0.000 claims description 5
- 238000004519 manufacturing process Methods 0.000 claims description 5
- 238000013404 process transfer Methods 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 6
- 239000000843 powder Substances 0.000 abstract description 2
- 230000000875 corresponding effect Effects 0.000 description 11
- 230000014509 gene expression Effects 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 239000010779 crude oil Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000013604 expression vector Substances 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000003915 air pollution Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000000295 fuel oil Substances 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to energy-consuming electric powder prediction more particularly to a kind of comprehensive energy prediction technique, specifically a kind of energy-consuming prediction techniques.Including:Region factors historical data and energy-consuming demand measured value are by year chosen as data sample;In conjunction with data in upper step, relationship between different year, different types of data is excavated, searching region factors sample is for the sample weights of energy-consuming demand measured value, influence degree of the region factors for energy-consuming demand for determining different year;Using the prediction algorithm based on Linear Mapping, the annual consumption demand total value for the somewhere to current year is predicted.It can ensure that the relationship for objectively responding each factor and prediction result, improve efficiency of algorithm, keep energy consumption predictions process more fast and effective, conducive to the energy forecast accuracy improved under the new normality of economic development, energy environment Condition of Strong Constraint, and then community energy balance is calculated, the final energy development and Protection Policies for determining reasonable.
Description
Technical field
The present invention relates to energy-consuming electric powder prediction more particularly to a kind of comprehensive energy prediction technique, specifically one
Kind energy-consuming prediction technique.
Background technology
Currently, China is in the new normality developing stage of economic development, the major economic indicators based on GDP present relatively steady
Fixed low growth trend is oscillated in quotations, naturally with the propulsion of prevention and control of air pollution action plan by international crude oil price
The primary energy price such as gas, coal continued jitters, high energy-consuming industry multifactor impacts, the energy development such as limit output face new
Development environment.Therefore, the prediction of total energy consumption becomes one of the important link of energy development planning.
The prior art mostly uses the prediction technique based on learning training to predict the energy demand total amount in somewhere, however,
The true problem of currently used method generally existing forecasting inaccuracy.In this kind of prediction technique, core link is learner
Training mostly uses the method construct learner of weak typing to ensure the calculating speed of prediction, and Weak Classifier is for sample spy
The classification of sign causes wrong study there may be the judgement of mistake.Since energy-consuming demand is by social development situation, industry
Structural adjustment, environmental impact factor, regional economic development degree, administrative planning etc. factors influence, and factor feature is general
All over conspicuousness and regularity is lacked, cause current many prediction technique mistakes study situation than more serious, it can not be accurately
Suitable for energy-consuming requirement forecasting problem.
Up to now, for medium-term and long-term energy-consuming forecasting problem mainly use intelligence learning algorithm, gray system theory,
The methods of expert system, support vector machines are not directed to and consider that industry restructuring, carbon emission constraint, new energy access power grid
Prediction technique, only by a large amount of historical datas carry out indifferenceization input calculate, not to designated area energy-consuming predict
Influence factor carries out analysis selection, and resultant error is big, with a low credibility.Simultaneously as energy-consuming prediction calculates required basic number
Big according to amount, Electric Power Network Planning department acquisition prediction result workload is heavy, therefore designs for a large amount of new energy access electricity of consideration
After net, the work of the energy-consuming prediction technique under the multifactor restriction of energy environment economy seems of crucial importance.
Invention content
Place in view of the deficiency of the prior art, the present invention provide a kind of comprehensive energy prediction technique.Purpose exists
In by establishing a set of rationally accurate prediction theory system and forecasting system, energy-consuming prediction accuracy is solved the problems, such as.
The present invention is achieved by the following technical solutions:
A kind of comprehensive energy prediction technique,:Include the following steps:
Step 1:Region factors historical data and energy-consuming demand measured value are by year chosen as data sample;
Step 2:It is found in conjunction with the data in step 1 by excavating relationship between different year, different types of data
Region factors sample for energy-consuming demand measured value sample weights, for determining the region factors of different year for energy
The influence degree of source consumption demand;
Step 3:Using the prediction algorithm based on Linear Mapping, the annual consumption demand total value for the somewhere to current year
It is predicted.
The step 1 by year chooses region factors historical data and energy-consuming demand measured value as data
Sample;
First, determination will predict the time of energy-consuming demand;
Then, after the time is determined, before being chosen at the time, at least 5 years in each time the region because
The historical data of element and energy-consuming demand measured value corresponding with the year, this area;
The region factors historical data includes following 10 item data:Total number of people, secondary industry total value, national product
Total value, inhabitant's consumption level, average electricity price, the practical operation amount of electric power primary energy, electricity consumption proportion, production of energy elasticity system
Number, the energy process transfer efficiency, consumption of coal total amount.
In the step 2, steps are as follows for the calculating of weight:
(1) hierarchical structure is built, the hierarchical structure is by up of three layers, specially:
Destination layer:For before this area's energy-consuming demand time to be predicted, each time corresponding energy at least 5 years
Consumption demand measured value;Different year is set objectives layer respectively;
Rule layer:Based on the time of destination layer, before taking the time, at least 3 years times, sample weights year is formed
Part;Each destination layer determines the respective sample weights time, respectively constitutes respective rule layer;
Solution layer:According to rule layer determine time, calculate under the time each region factors historical data identical
Changing sensitivity under the variation of energy-consuming demand measured value;Each solution layer determines that factor respective variation in different regions is sensitive
Degree, respectively constitutes solution layer;
(2) standardization processing is carried out to the data in different levels, specially:
Destination layer:Destination layer representative will solve the problems, such as that data are not required to standardization processing;
Rule layer:Standardization processing is carried out using scale quantization criterion;
Solution layer:Standardize respectively according to different year determined by rule layer firstly, for the items of solution layer
Processing, i.e. normalized;
Then, it is determined that the sensitivity that all samples change energy-consuming;
The sensitivity of the energy-consuming variation reflects that energy-consuming requirements change and adjacent year between the adjacent time
Relationship between part between sample changed;
(3) by development of judgment matrix, weight and criterion of each sample of solution layer for the corresponding rule layer time are sought
Weight of the layer each time for destination layer energy-consuming demand;
(4) to solution layer for the synthetic weights severe W of destination layeriIt is calculated, that is, determines every sample in destination layer time
Originally the influence weights for energy-consuming requirements, formula are:
Wi=Σjpjvij
In formula, pjThe n that for the correlation between rule layer time and destination layer energy-consuming requirements, and has j=1 ...,
Middle n >=3;vijCorrelation between solution layer and rule layer, and have i=1 ... m, wherein m=10, i indicate i-th of factor, m
Indicate m-th of factor.
The step 3:Using the prediction algorithm based on Linear Mapping, the annual energy for the somewhere power grid to current year
Source consumption demand total value predicted, specially:
Step 3.1:The region factors sample for treating the prediction regional energy-consuming demand time and several times before the year
Region factors sample, energy-consuming demand sample is normalized;And to the region factors sample after normalized
According to different year i, the influence weights W determined with step 2iMake multiplication processing;
Step 3.2:Original sample collection is constructed, as the input based on Linear Mapping prediction algorithm;
It is input sample by 10 region factors samples in step (3.1)And full community energy consumption demand amount is made
For output quantityEstablish classified sample set S;It is as follows:
In formula, N indicates that the quantity of input sample, the representative sample time in sample set S, L indicate initial sample number 10, R
The desirable set of real numbers of representative sample;
The classified sample set obtained in step 3.2 is converted into two class classified sample sets, the two classes sample set is combined,
Obtain the original sample collection D of the prediction algorithm based on Linear Mapping;It is as follows:
D={ (xi,yi)|xi∈RL,yi∈ { -1 ,+1 }, i=1,2 ..., N }
In formula, xi10 samples in step 3.1 are represented to be constituted with value of the energy-consuming demand measured value after translation
Vector;I indicates that sample serial number, N indicate the quantity of sample;yiFor the desired value of the original sample collection of construction, actual value be 1 or-
1;The desirable set of real numbers of R representative samples;I indicates i-th of factor.
Step 3.3:Using the original sample collection D constructed in step 3.2, initial sample weights vector d is calculated1,i, according to yi
Sample is considered as two classes by value for 1 or -1, and for each sample in every one kind, weight average point is carried out according to total sample number amount
Match, and ensures that totally the sum of all sample weights are 1;
Establish sample average mt,k, k=1 or -1, t represent iterations;Representative sample mean value corresponds to two kinds and is classified as two
It is a;Each sample average it is expected by weighted mean method to calculate sample by meeting a kind of sample interior, and divided by the category in
Total weight obtain, since sample has already passed through normalized, any element is not to be exceeded 1 in sample average;
Step 3.4:Establish the Linear Mapping factor;
The sample average that using sample be 1 and sample is -1 constructs sample standard deviation value difference, and in L+1 dimension spaces after the conversion
Construction between-class scatter and within-class scatter determine the Linear Mapping factor using two dispersions;Sample
Inter _ class relationship is with within-cluster variance calculation formula:
In formula, St,kFor the inter _ class relationship of sample, St,WFor the within-cluster variance of sample;dt,iFor the sample of current iteration layer t
This weight vectors, value rely on the right value update in each iteration;E representation dimensions and sample xiIdentical vector, internal member
Element is 1, mt,kFor two class sample average in current iteration layer;I indicates sample serial number;xiAs abovementioned steps meaning, all it is
The region factors factor, T indicate transposed matrix.
The Linear Mapping factorComputational methods are the quotient of inter _ class relationship and within-cluster variance;The Linear Mapping factor can be with
Original sample is mapped to new sample space, sample meets inter _ class relationship within this space and within-cluster variance ratio is maximum
Change;
Step 3.5:Establish Weak Classifier;
By the Weak Classifier in best projection map construction current iteration layer:Sample in mapping space is more than threshold values
When, be divided into one kind, otherwise be used as it is another kind of, the Weak Classifier formula of foundation is:
ht={ ht,l| l=1...L+1 }
In formula, htTo be constructed Weak Classifier, xi,lIndicate xiIn l elements,ForIn l elements,
L=1 ... L+1 representative sample types, i indicate sample serial number, θtIt is threshold values vector, is determined by the classification error rate in previous iteration layer
It is fixed, (θt)lIt is its internal l element;Others indicates other situations.
The threshold value vector calculation formula be:
In formula, mtFor mean vector, dimension and sample xiIdentical, inner element is the flat of the sample average in current iteration
Mean, i.e. mt,1With mt,-1Mean value,To meet y in previous iterationiThe classification error rate of=1 sample,For previous iteration
In meet yiThe classification error rate of=- 1 sample, wherein classification error rate is 0 when iterative calculation t=1 for the first time;
Step 3.6:Determine that classification error rate, formula are:
In formula, ht(xi) indicate sample classification results, ht(xi)lFor its inside l element, i indicates sample serial number, ht
(xi)l≠ 1 indicates to meet yi=1 i-th of sample xiIn, l sample elements xl,iBy mistake point in current iteration layer t
Class, ht(xi)l≠ -1 meaning is similar, and I () indicates discriminant function, if content in bracket is set up, I=1, otherwise I=0, most
Afterwards to the weight d of all misjudged samplest,l,iRespectively with yi=1 or yi=-1 carries out classification weighting, obtains in current iteration layer
Two class classification error rates;K is a stochastic variable, represents 1 or -1.
Step 3.7:Ballot parameter alpha is determined using the classification error rate in step (3.6)t, formula is:
In formula, εtFor the classification error rate of all samples in current iteration layer;Ballot parameter alphatObtained by the t times iteration
To a kind of measurement of grader classification accuracy rate, when it acts as indicating finally to weight grader in all layers, each layer grader
Weight, the t times Iterative classification error rate is smaller, indicate the t time Iterative classification device for totality it is more effective;
Step 3.8:Weight vectors are updated by classification error rate and ballot parameter, the weight vectors formula after updating is such as
Under:
dt+1,i=dt,i*exp(-αtyiht(xi))
In formula, dt,iFor original weight vectors, yiIndicate that score value 1 or -1, value correspond to xiMiddle element;Exp functions it is interior
Portion's element is vector, and exp is using e as the truth of a matter, using each element in vector in bracket as index, carries out exponential function calculating;It is public
No. * expression vector point multiplication in formula, the i.e. internal respective items of vector are multiplied, and result of calculation is the identical vector of dimension, that is, is updated
D afterwardst+1,i;
Weight in weight vectors should meet adduction and be equal to 1, therefore formula is normalized to weight vectors and is:
In formula, Zt is equal to the adduction of weight vectors interior element;
Step 3.1-step 3.8 is repeated, until completing the training of all samples;
Step 3.9:Weak Classifier in all iteration layers finally determined to step 3.5 is joined with the ballot that step 3.7 determines
Number be used as respective weights, using weighting integrate all graders, obtain regression equation, the regression equation describe sample data with
Implicit function relationship between target energy-consuming requirements, input are 10 master datas in target time, and output is area
The predicted value of energy-consuming demand data.
The calculating for the sensitivity that all samples change energy-consuming is determined in (2) of the step 2 in solution layer
Method is:
10 sampled data values of n before the time are denoted as ci,j, i=1 ... m, m=10, j=1 ... n, n>=3;The energy
Consumption demand measured value is denoted as aj, j=1 ... n;The quantization in rule layer time is denoted as b before the timej, j=1 ... n;The rule layer time
Correlation between destination layer energy-consuming requirements is pj, solution layer changing sensitivity is denoted as qij, i=1 ... m, solution layer
Correlation between rule layer is vij, for the time in each rule layer, calculated in respective party pattern layer by following formula
Every quantized value qij, i.e., sensitivity that all samples change energy-consuming:
In formula, qijMeaning is:Variation spirit of i-th of the sample in j-th of time for j-th of time energy-consuming requirements
Sensitivity is qij;aj-1、Ci,j-1Preceding 1 year data are indicated in above formula.
Weight of each sample of solution layer for the corresponding rule layer time is sought described in (3) in the step 2, and
Rule layer each time for destination layer energy-consuming demand weight, specially:
There is 1 for rule layer judgment matrix A, the time quantity that dimension n includes by solution layer determines that wherein element is:
In above formula:b、b1、bnAll it is solution layer judgment matrix in above formula, is found out by following matrix;
There are n for the judgment matrix B of solution layer, dimension m is determined by the sample size participated in step 1, for j-th
Matrix wherein element is:
In above formula:q、q1j、qmjSensitivity is indicated in above formula, is found out by formula above;
The maximum eigenvalue and feature vector for calculating each matrix, by feature vector come acquire the weight of disparity items to
Amount, weight vectors are denoted as respectively:
P=(p1,p2,...pj,...pn), j=1..n
V=(v1j,v2j..., vij,...vmj), i=1..m
In formula, weight of the different year for energy-consuming demand in p expressiveness layers;Vj was indicated in rule layer jth year
Weight of the different samples of part to energy-consuming demand.
Beneficial effects of the present invention:
The present invention is a kind of comprehensive energy prediction technique of Linear Mapping, more in economy of energy Environmental Industry structural adjustment etc.
Under weight environment, in the energy-consuming requirement forecasting calculating of a large amount of clean energy resource access power grids, the industrial structure, carbon is added in analysis selection
Exhaust emission constraint, energy prices etc. are multifactor and quantify to it, such as average electricity price, the practical operation amount of electric power primary energy, state
Border crude oil price, domestic gas price, China Diesel price, internal fuel oil price lattice, coal price etc., to influencing energy-consuming
Each factor of variation calculated from the excellent Changeable weight that becomes, it is ensured that objectively responds the relationship of each factor and prediction result, establishes
High-dimensional feature space, and therefrom extraction and Combinatorial Optimization go out the low-dimensional feature of most discriminating power, train up data, it is maximum
Change avoids generating overlearning, improves efficiency of algorithm, keeps energy consumption predictions process more fast and effective, is conducive to improve economic hair
The energy forecast accuracy under new normality, energy environment Condition of Strong Constraint is opened up, and then calculates community energy balance, final determine is closed
Manage feasible energy development and Protection Policies.
Specific implementation mode
Embodiment specifically includes following steps:
Step 1, region factors historical data and energy-consuming demand measured value are by year chosen as data sample.
First, present embodiment determination will predict that the time of energy-consuming demand is 2013, and user can be as needed,
Voluntarily select the year to be predicted.
Then, before choosing 2013 in 6 years in each time the historical data of the regional factor and with the year, the ground
The corresponding energy-consuming demand measured value in area, the region factors historical data include following 10 item data:Total number of people, the
Two industry total values, gross national product, inhabitant's consumption level, average electricity price, the practical operation amount of electric power primary energy, electricity consumption
Proportion, energy production elasticity, the energy process transfer efficiency, primary energy total quantity consumed.It is specific as shown in Table 1 and Table 2.
As shown in Table 1, the factor historical data for influencing this area's energy-consuming in 2005 is as follows:Total number of people 201.56
Ten thousand, 86.51 hundred million yuan of secondary industry total value, 94.71 hundred million yuan of gross national product, 4.6 ten thousand yuan/year of inhabitant's consumption level, average electricity
0.351 yuan of valence/degree, electricity consumption proportion 3.1%, energy production elasticity 0.97, the energy process transfer efficiency 74.7, coal
9,880,000 tons of standard coals of total quantity consumed
As shown in Table 2, energy-consuming demand measured value in 2005 is 12,880,000 tons of standard coals.
For other times, if 2006~2013 annual this area factor historical datas can be found shown in 1, again not
It is repeating.And energy-consuming demand measured value corresponding with each time can then be found according to table 2, not repeated.
Step 2:Relationship between excavation different year, different types of data, searching ground are passed through to the sample data of step 1
Area's factor sample carries out weight calculation, the ground for determining different year for the sample weights of energy-consuming demand measured value
Influence degree of area's factor sample for energy-consuming demand measured value.
The sample weights calculate, and are as follows:
(1) hierarchical structure is built, for the hierarchical structure by up of three layers, structure rule is as follows:
Destination layer:For before this area's energy-consuming demand time to be predicted, each time corresponding energy at least 5 years
Consumption demand measured value;Different year is set objectives layer respectively.
In present embodiment, using 2013 as this area's energy-consuming demand time to be predicted, then hierarchical structure is established
When, there are 5 different destination layers, respectively 2008 energy-consuming measured values are as first object layer, energy-consuming in 2009
Measured value as the second destination layer ..., until, energy-consuming measured value is as the 5th destination layer within 2012.
Rule layer:Based on the time of destination layer, 3 years before time times are taken, form the sample weights time.It is right
For present embodiment, for first object layer, 3 years before taking 2008 are as corresponding with first object layer first
Rule layer, i.e., 2005,2006 and 2007 are the first rule layer.
Each destination layer determines the respective sample weights time, respectively constitutes respective rule layer:
Second rule layer is:2006,2007 and 2008;
5th rule layer is:2009,2010 and 2011.
Solution layer:According to rule layer determine time, calculate under the time each region factors historical data identical
Changing sensitivity under the variation of energy-consuming demand measured value;Each solution layer determines that factor respective variation in different regions is sensitive
Degree, respectively constitutes solution layer.
By taking present embodiment as an example, for the first rule layer:Determine 10 region factors historical datas in 2005 for energy
The sensitivity of source consumption demand measured value, specifically includes:Total number of people changing sensitivity in 2005, secondary industry total value in 2005
Changing sensitivity, 2005 ... ..., total energy consumption changing sensitivity in 2005.
……
5th rule layer:Determine 10 region factors historical datas in 2009 for the sensitive of energy-consuming demand measured value
Degree, specifically includes:Total number of people changing sensitivity in 2009, secondary industry total value changing sensitivity in 2009,2009
Year ... ..., total energy consumption changing sensitivity in 2009.
Establish the perfectly correlated property structure in target time, i.e. arbitrary element in guarantee hierarchical structure in every adjacent two layers all
With correlation.
(2) standardization processing is carried out to the data in different levels, specially:
Destination layer:Destination layer representative will solve the problems, such as that data are not required to standardization processing.
Rule layer:Standardization processing is carried out using scale quantization criterion.Specially:
The different year energy-consuming demand of rule layer does not quantify meaning, needs empirically to carry out rule layer weight
New quantization.Quantitative criteria can be used as according to the difference of energy-consuming demand between the time.Different year is passed through into 1-9 scale amounts
Change criterion to quantify, specific criterion is:1-9 scales quantify criterion to quantify.1 indicates that significance level is most weak, and 9 indicate most strong.
Solution layer:Standardization processing is carried out respectively according to different year determined by rule layer for the items of solution layer,
Process is consistent with normalized.
Then, 10 sampled data values of n before the time are denoted as ci,j, i=1 ... m, m=10, j=1 ... n, n>=
3;Energy-consuming demand measured value is denoted as aj, j=1 ... n;The quantization in rule layer time is denoted as b before the timej, j=1 ... n;Criterion
Correlation between layer time and destination layer energy-consuming requirements is pj, solution layer changing sensitivity is denoted as qij, i=1 ... m,
Correlation between solution layer and rule layer is vij, for the time in each rule layer, respective party is calculated by following formula
Every quantized value q in pattern layerij, i.e., sensitivity that all samples change energy-consuming:
In formula, qijMeaning is:Changing sensitivity of i-th of the sample in j-th of time for j-th of time energy-consuming value
For qij;aj-1、Ci,j-1Preceding 1 year data are indicated in above formula;Changing sensitivity reflects, energy-consuming value between the adjacent time
Relationship between variation and adjacent time between sample changed;
(3) by development of judgment matrix, weight and criterion of each sample of solution layer for the corresponding rule layer time are sought
Layer each time for destination layer energy-consuming demand weight, specially:
There is 1 for rule layer judgment matrix A, the time quantity that dimension n includes by solution layer determines that wherein element is:
In above formula:b、b1、bnAll it is solution layer judgment matrix in above formula, is found out by following matrix;
There are n for the judgment matrix B of solution layer, dimension m is determined by the sample size participated in step 1, for jth
A matrix wherein element is:
In above formula:q、q1j、qmjSensitivity is indicated in above formula, is found out by formula above;
The maximum eigenvalue and feature vector for calculating each matrix, by feature vector come acquire the weight of disparity items to
Amount.Weight vectors are denoted as respectively:
P=(p1,p2,...pj,...pn), j=1..n
V=(v1j,v2j,...,vij,...vmj), i=1..m
Weight of the different year for energy-consuming demand in p expressiveness layers.Vj indicate the rule layer jth time not
With sample to the weight of energy-consuming demand.
(4) solution layer calculates the synthetic weights severe of destination layer, that is, determines every sample in destination layer time
For the weighing factor of energy-consuming requirements, formula is:
Wi=Σjpjvij
In formula, pjThe n that for the correlation between rule layer time and destination layer energy-consuming requirements, and has j=1 ...,
Middle n >=3;vijCorrelation between solution layer and rule layer, and have i=1 ... m, wherein m=10, i indicate i-th of factor, m
Indicate m-th of factor.
If the training goal in step 3 is prediction community energy consumption demand value in 2013, then sample should include 2008-
Sample data and 2008-2012 energy-consuming demands in 2013 surveys Value Data.For any one in 2008-2012
The sample weights in year all should be to consider with this time, be pushed forward 3 years tissue input datas to establish level computational algorithm.Such as it is right
Sample weights calculating in 2008 needs 2005-2007 sample datas and calculates calculation to establish sample weights level in 2008
Method.The input quantity weights in 2008-2012 each times are determined successively.
Step 3:Using the prediction algorithm based on Linear Mapping, the annual energy-consuming for the somewhere power grid to current year
Demand total value predicted, specially:
(3.1) the region factors sample for treating the prediction regional energy-consuming demand time and several times before this year
Region factors sample, energy-consuming demand sample are normalized.
The region factors sample in energy-consuming demand time to be predicted is normalized, 10 features in step 1
Data are sample characteristics, and the time is shown in Table 1 with sample class, specially:Total number of people, secondary industry total value, national product are total
Value, inhabitant's consumption level, average electricity price, the practical operation amount of electric power primary energy, electricity consumption proportion, production of energy elasticity system
Number, the energy process transfer efficiency, consumption of coal total amount.
For the ease of carry out sample quantization coefficient normalized, in quantized samples choose maximum absolute value value with absolutely
To being worth minimum value.Such as 2008-2012 total number of people samples, maximum value is 381.69 ten thousand people, minimum value 294.10
Ten thousand people.
Input sample is normalized, normalization formula is:
X=(x0-xmin)/(xmax-xmin)。
In formula, x is the input value after normalized, is 10 feature samples for influencing total energy consumption in this example
This.x0Sample is characterized without the initial data before normalized, xmaxFor input feature vector sample maximum;xminFor input
Feature samples minimum value.
To the region factors sample after normalized according to different year i, the influence weights W determined with step 2iMake phase
Multiply processing.
(3.2) original sample collection is constructed, as the input based on Linear Mapping prediction algorithm.
With the quantization parameter input value of 10 characteristics and total energy consumption in table 1, input feature vector number is arranged
According to output total energy consumption data, construct original sample collection.It is input by 10 region factors samples in step (3.1)
SampleAnd full community energy consumption figure is as output quantityEstablish classified sample set S;Formula is:
In formula, xi is i-th of input vector, represents 1 year sample, containing for different type sample is represented inside vector
Flexible strategy evidence;Yi is i-th of output valve, represents 1 year energy-consuming demand measured value;N is that original sample concentrates total sample number,
Take be trained in this example within first 5 years.Ordinal number i represents the time of input sample.In formula, representative sample time, L in sample set S
Indicate initial sample number 10, the desirable set of real numbers of R representative samples;
Given training parameter:Maximum training iterations T, translational movement δ (are used as in present embodiment using Weak Classifier
The effect of weak learner, translational movement is by the way that by one section of fixed range of desired value upper and lower translation, original sample set is converted to two
A classified sample set realizes weak study to make learning method become two classification convenient for learner, reduces mistake study probability),
Translational movement is an experience value, and 0.3 is taken as in present embodiment.
The classified sample set obtained in step (3.2) is converted into two class classified sample sets, by the two classes sample set sample
This collection combines, and obtains the original sample collection D for instructing the prediction algorithm based on Linear Mapping, process is as follows:
Sample set S will be returned and be converted into classified sample set S1, method is as follows.
By by desired value yiIt translates fixed range δ respectively up and down, converts S to two class classified sample sets.
The input vector of classified sample set by 6 influence total energy consumption factor normalization coefficient input vector and
The recurrence sample object value composition translated.New desired value is demarcated as 1 and -1 respectively, and corresponding two desired values are 1 and -1
Classification samples, transformed two classes sample set is respectively:
In formula, xi is identical with yi meanings and set S, L representative sample number of types, is 10 kinds in this example, and N is the time,
It is 5 years in this example.
Sample set is combined to obtain the input layer of training algorithm, it is as follows.
D={ (xi,yi)|xi∈RL+1,yi∈ { -1 ,+1 }, i=1,2 ..., N }
In formula, xiFor new samples vector, the xi sample vectors in being gathered by original D1 and D2 and the yi groups after translation
At.Y in formulaiFor the desired value of the original sample collection of construction, practical presentation class value is+1 or -1;What R representative samples can use
Set of real numbers;I indicates i-th of factor.
If not specified otherwise, following sample standard deviation indicates the sample in sample set D.
(3.3) the original sample collection D constructed in step (3.2) is utilized, sample weights vector is calculated.Calculate initial sample
Weight vectors d1,i, according to yiSample is considered as two classes by value for 1 or -1, for each sample in every one kind, according to total sample number
Amount carries out weight average distribution, and ensures that totally the sum of all sample weights are 1.
Primary iteration number t=1, initialization weight vectors d are set1,i, the different serial numbers of i expressions.If yi=1 sample number
Amount is N+, yi=-1 sample size is N-, d1,iThe initial value setting of inner element is as follows.
d1,iFor row vector, wherein l indicates different sample types.
By the weight vectors d of each samplet,iIt is expected by calculated with weighted average method sample, and is distributed by total weight
To sample average mt,k.Since input layer sample is all by normalized, therefore sample average is not to be exceeded 1.
mt,k={ mt,l,k| l=1...L+1 }, k=1, -1
In formula, mt,kFor dimension and sample xiIt is all the vector of L+1;K=1 or -1, t represent current iteration hierachy number;It represents
Sample average corresponds to two kinds and is classified as two;Each sample average is calculated by meeting a kind of sample interior by weighted mean method
Sample it is expected, and divided by the category in total weight obtain.
dt,l,iFor weight vectors dt,iIn l elements;xl,iFor sample xiIn l elements;Adduction condition yi=k
Expression will meet corresponding yiValue is that all sample elements of k are summed up according to ordinal number i, hereafter similar.
(3.4) the Linear Mapping factor is established.The sample average that using sample be 1 and sample is -1 constructs sample standard deviation value difference,
And between-class scatter and within-class scatter are constructed in L+1 dimension spaces after the conversion, it is discrete using described two
Degree, determines the Linear Mapping factor.
Wherein:Between-class scatter St,k, formula is as follows:
Inter _ class relationship characterizes the dispersion degree between different classes of sample, in formula, dt,iI.e. repeatedly for sample in current iteration
For the sample weights vector of layer t, value relies on the right value update in each iteration;xiIndicate sample, e representation dimensions and sample xi
Identical vector, inner element are 1, mt,kFor the sample average in current iteration;I indicates sample serial number;
Construct within-class scatter St,W, formula is as follows:
Within-cluster variance characterizes the dispersion degree between the same category sample.Each element meaning and inter _ class relationship in formula
Formula is identical.
According to linear decision rule, make classification projection mapping x the most accuratei→xi=wtxiDispersion should be mapped as
The equal value difference of sample.Obtain optimum linear mapping-factorBest projection mapsComputational methods are in inter _ class relationship and class
Original sample can be mapped to new sample space by the quotient of dispersion, and sample meets inter _ class relationship and class within this space
Interior dispersion ratio maximizes.Computational methods are:
In formula,It is mapped for best projection to be asked, St,WFor within-cluster variance.
(3.5) Weak Classifier is established.When sample in mapping space is more than threshold values, it is divided into one kind, is otherwise used as another
Class, by the Weak Classifier in best projection map construction current iteration layer, formula is:
ht={ ht,l| l=1...L+1 }
In formula, htTo be constructed Weak Classifier.xi,lIndicate xiIn l elements,ForIn l elements,
L=1 ... L+1 representative sample types.I indicates sample serial number.θtIt is threshold values vector, is determined by the classification error rate in previous iteration layer
It is fixed, (θt)lIt is its internal l element;Others indicates other situations.
Weak Classifier is a kind of efficient learner, and the form of expression can be adjusted according to the actual needs.One
As for the classification value of Weak Classifier be constant, different points may be implemented in the Rule of judgment for adjusting classification according to actual needs
Class standard.
A kind of threshold values of present embodiment selection passes through the classification error rate of preceding an iteration as criteria for classification, threshold values, adjusts
Sample average in whole space of linear mapping obtains.The effect of threshold values be for the grader in current iteration layer provide classification according to
According to.In present embodiment, the calculation formula of threshold value is:
In formula, mtFor mean vector, dimension and sample xiIdentical, inner element is the flat of the sample average in current iteration
Mean, i.e. mt,1With mt,-1Mean value.To meet y in previous iterationiThe classification error rate of=1 sample,For in previous iteration
Meet yiThe classification error rate of=- 1 sample, classification error rate is 0 when paying attention to iterating to calculate t=1 for the first time.
The Weak Classifier of construction is vector, pays attention to judging that the value of sentence is different from every time, so the sample being classified is every
It is secondary to have differences.In general a rational Weak Classifier, due to the presence of linear projection mapping-factor, with iteration time
Several increases, classification judge that sentence can reduce Weak Classifier and classify for the mistake of sample automatically, reduce classification error rate.Into
And improve the forecasting accuracy of whole prediction algorithm.
ht={ ht,l| l=1...L+1 }
(3.6) by Weak Classifier identification and classification value weight distribution vector contradictory with desired value, and meet this by all
Class weight distribution sums it up, and obtains classification error rate in current iteration layer, formula is:
In formula, ht(xi) indicate sample classification results, ht(xi)lFor its inside l element, i indicates sample serial number.ht
(xi)l≠ 1 indicates to meet yi=1 i-th of sample xiIn, l sample elements xl,iBy mistake point in current iteration layer t
Class, ht(xi)l≠ -1 meaning is similar.I () indicates discriminant function, if content in bracket is set up, I=1, otherwise I=0, most
Afterwards to the weight d of all misjudged samplest,l,iRespectively with yi=1 or yi=-1 carries out classification weighting, obtains in current iteration layer
Two class classification error rates.
Often there is mistake classification in Weak Classifier, by calculating classification error rate, can adjust the classifying rules of next time, make
Assorting process is close to the direction for reducing mistake classification, this process relies on classification threshold values θtIt realizes.
(3.7) ballot parameter alpha is calculated by classification error ratet, formula is:
In formula, εtFor the classification error rate of all samples in current iteration layer.Ballot parameter alphatObtained by the t times iteration
To a kind of measurement of grader classification accuracy rate, when it acts as indicating finally to weight grader in all layers, each layer grader
Weight.In general, the t times Iterative classification error rate is smaller, indicates the t times Iterative classification device for overall more effective.
(3.8) weight vectors are updated by classification error rate and ballot parameter, the weight vectors formula after updating is as follows:
In formula, dt,iFor original weight vectors, αtFor parameter of voting, yiIndicate that score value 1 or -1, value correspond to xiMiddle member
Element.The inner element of exp functions is vector, and exp, using each element in vector in bracket as index, is referred to using e as the truth of a matter
Number function calculates.No. * expression vector point multiplication in formula, the i.e. internal respective items of vector are multiplied, and result of calculation is that dimension is identical
Vector, i.e., updated dt+1,i。
Weight in weight vectors should meet adduction and be equal to 1, therefore formula is normalized to weight vectors and is:
In formula, Zt is equal to the adduction of weight vectors interior element.
Step (3.1)~step (3.8) is repeated, until completing the training of all samples.
(3.9) grader in all iteration is finally integrated, regression equation is obtained by weighting.I.e. finally to step 3.5
Weak Classifier in determining all iteration layers integrates institute using the ballot parameter that step 3.7 determines as respective weights using weighting
There is grader, obtains regression equation.The regression equation describes the implicit function between sample data and target energy-consuming requirements
Relationship, input are 10 master datas in target time, and output is the predicted value of community energy consumption demand data;
Regression equation is as follows:
In formula, y0Community energy consumption demand sample between the several years comprising the energy-consuming demand time to be predicted to
Amount, x0The region factors sample matrix formed for the several years sample of 10 basic input datas containing weight coefficient.αtIt is
The ballot parameter of t iteration, ht,lIndicate the classification value about l samples in the t times iteration.C is the constant of compressive classification.
H (x) represents a compressive classification value on set D, training result C after transformation and illustrates more ideal H (x0,
y0) should be closer to C values.Although the value of regression function be by grader h operations obtain as a result, however meeting the y of above formula0
With x0It is inevitable to meet H (x simultaneously0,y0Equality constraint of the)=C regression equations on set D, if for example, wishing to predict 2013
It can be considered as parameter by energy-consuming requirements, constitute the new (x of sample set containing undetermined parameter0,y0) and participate in regression equation
It establishes, and final classification value is made to be equal to C.
Actually any one is to weight the implicit function relationship that integrated regression equation can be between inner element clear.
Using undetermined coefficient, i.e., community energy consumption demand y to be predicted0With new region factors sample set x0Regression equation is clear,
It obtains on set S, i.e. energy consumption data y0About feature vector x0Integrated Models:
y0=F (x0)
In formula, y0For amount to be predicted, wherein including undetermined parameter y0, i.e. energy-consuming requirements in 2013;x0It is 2013
And its sample set (given data group) of preceding 10 region factors of several years composition.
The each basic input quantity for substituting into the target time, after obtaining weight coefficient by step 2, brings into Integrated Models,
Obtain the predicted value of energy-consuming demand data, the energy-consuming total demand of as 2013 this areas.
Step 4:According to the regional factor of actual area, the regression equation determined using step 3 calculates the energy in the region
Consumption demand total value, using the energy-consuming demand total value, to adjust every profession and trade energy consumption
It should be noted that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although with reference to preferable
Embodiment describes the invention in detail, it will be understood by those of ordinary skill in the art that, it can be to the technology of the present invention
Scheme is modified or replaced equivalently, and without departing from the principle and essential scope of technical solution of the present invention, should all be covered
In scope of the presently claimed invention.
Table 1 is the somewheres 2005-2013 energy-consuming and its correlative factor table.
Table 2 is the somewheres 2005-2013 energy-consuming table.
Claims (3)
1. a kind of comprehensive energy prediction technique, it is characterised in that:Include the following steps:
Step 1:Region factors historical data and energy-consuming demand measured value are by year chosen as data sample;
Step 2:In conjunction with the data in step 1 area is found by excavating relationship between different year, different types of data
Factor sample for energy-consuming demand measured value sample weights, for determining that the region factors of different year disappear for the energy
Take the influence degree of demand;
Step 3:Using the prediction algorithm based on Linear Mapping, the annual consumption demand total value for the somewhere to current year carries out
Prediction;
The step 1 by year chooses region factors historical data and energy-consuming demand measured value as data sample;
First, determination will predict the time of energy-consuming demand;
Then, after the time is determined, before being chosen at the time, this area's factor in each time at least 5 years
Historical data and energy-consuming demand measured value corresponding with the year, this area;
The region factors historical data includes following 10 item data:Total number of people, secondary industry total value, national product are total
Value, inhabitant's consumption level, average electricity price, the practical operation amount of electric power primary energy, electricity consumption proportion, production of energy elasticity system
Number, the energy process transfer efficiency, consumption of coal total amount;
In the step 2, steps are as follows for the calculating of weight:
(1) hierarchical structure is built, the hierarchical structure is by up of three layers, specially:
Destination layer:For before this area's energy-consuming demand time to be predicted, each time corresponding energy-consuming at least 5 years
Demand measured value;Different year is set objectives layer respectively;
Rule layer:Based on the time of destination layer, before taking the time, at least 3 years times, the sample weights time is formed;
Each destination layer determines the respective sample weights time, respectively constitutes respective rule layer;
Solution layer:According to rule layer determine time, calculate under the time each region factors historical data in the identical energy
Changing sensitivity under the variation of consumption demand measured value;Each solution layer determines the respective changing sensitivity of different regions factor, point
Solution layer is not constituted;
(2) standardization processing is carried out to the data in different levels, specially:
Destination layer:Destination layer representative will solve the problems, such as that data are not required to standardization processing;
Rule layer:Standardization processing is carried out using scale quantization criterion;
Solution layer:Standardization processing is carried out respectively according to different year determined by rule layer firstly, for the items of solution layer,
That is normalized;
Then, it is determined that the sensitivity that all samples change energy-consuming;
The sensitivity of the energy-consuming variation reflects that energy-consuming requirements changed between the adjacent time between the adjacent time
Relationship between sample changed;
(3) by development of judgment matrix, it is each for the weight and rule layer in corresponding rule layer time to seek each sample of solution layer
Weight of the time for destination layer energy-consuming demand;
(4) to solution layer for the synthetic weights severe W of destination layeriCalculated, that is, determine the destination layer time every sample for
The influence weights of energy-consuming requirements, formula are:
Wi=∑jpjvij
In formula, pjFor the correlation between rule layer time and destination layer energy-consuming requirements, and have j=1 ... n, wherein n >=
3;vijCorrelation between solution layer and rule layer, and have i=1 ... m, wherein m=10, i indicate that i-th of factor, m indicate
M-th of factor;
The step 3:Using the prediction algorithm based on Linear Mapping, the annual energy for the somewhere power grid to current year disappears
Expense demand total value predicted, specially:
Step 3.1:Treat the region factors sample in prediction regional energy-consuming demand time and before this year several times ground
Area's factor sample, energy-consuming demand sample are normalized;And to the region factors sample after normalized according to
Different year i, the influence weights W determined with step 2iMake multiplication processing;
Step 3.2:Original sample collection is constructed, as the input based on Linear Mapping prediction algorithm;
It is input sample by 10 region factors samples in step (3.1)And full community energy consumption demand amount is as defeated
OutputEstablish classified sample set S;It is as follows:
In formula, N indicates that the quantity of input sample, the representative sample time in sample set S, L indicate that initial sample number 10, R represent
The desirable set of real numbers of sample;
The classified sample set obtained in step 3.2 is converted into two class classified sample sets, the two classes sample set is combined, is obtained
The original sample collection D of prediction algorithm based on Linear Mapping;It is as follows:
D={ (xi,yi)|xi∈RL,yi∈ { -1 ,+1 }, i=1,2 ..., N }
In formula, xiRepresent 10 samples and value constituted vector of the energy-consuming demand measured value after translation in step 3.1;i
Indicate that sample serial number, 2N indicate the quantity of sample;yiFor the desired value of the original sample collection of construction, actual value is 1 or -1;R generations
The desirable set of real numbers of table sample sheet;I indicates i-th of factor;
Step 3.3:Using the original sample collection D constructed in step 3.2, initial sample weights vector d is calculated1,i, according to yiValue is 1
Or sample is considered as two classes by -1, for it is every it is a kind of in each sample, carry out weight average distribution according to total sample number amount, and protect
The sum of overall all sample weights of card are 1;
Establish sample average mt,k, k=1 or -1, t represent iterations;Representative sample mean value corresponds to two kinds and is classified as two;Often
A sample average it is expected by weighted mean method to calculate sample by meeting a kind of sample interior, and divided by the category in totality
Weight obtains, and since sample has already passed through normalized, any element is not to be exceeded 1 in sample average;
Step 3.4:Establish the Linear Mapping factor;
The sample average that using sample be 1 and sample is -1 constructs sample standard deviation value difference, and is constructed in L+1 dimension spaces after the conversion
Between-class scatter and within-class scatter determine the Linear Mapping factor using two dispersions;Between sample class
Dispersion is with within-cluster variance calculation formula:
In formula, St,kFor the inter _ class relationship of sample, St,WFor the within-cluster variance of sample;dt,iIt is weighed for the sample of current iteration layer t
Weight vector, value rely on the right value update in each iteration;E representation dimensions and sample xiIdentical vector, inner element are equal
It is 1, mt,kFor two class sample average in current iteration layer;I indicates sample serial number;xiAll it is area as abovementioned steps meaning
Factors, T indicate transposed matrix;
The Linear Mapping factorComputational methods are the quotient of inter _ class relationship and within-cluster variance;The Linear Mapping factor can will be original
Sample is mapped to new sample space, and sample meets inter _ class relationship within this space and within-cluster variance ratio maximizes;
Step 3.5:Establish Weak Classifier;
By the Weak Classifier in best projection map construction current iteration layer:When sample in mapping space is more than threshold value, point
For one kind, otherwise it is used as another kind of, the Weak Classifier formula of foundation is:
ht={ ht,l| l=1...L+1 }
In formula, htTo be constructed Weak Classifier, xi,lIndicate xiIn l elements,ForIn l elements, l=
1 ... L+1 representative sample types, i indicate sample serial number, θtIt is threshold vector, is determined by the classification error rate in previous iteration layer,
(θt)lIt is its internal l element;Others indicates other situations;
The calculation formula of threshold vector is:
In formula, mtFor mean vector, dimension and sample xiIdentical, inner element is the average of the sample average in current iteration,
That is mt,1With mt,-1Mean value,To meet y in previous iterationiThe classification error rate of=1 sample,To meet y in previous iterationi
The classification error rate of=- 1 sample, wherein classification error rate is 0 when iterative calculation t=1 for the first time;
Step 3.6:Determine that classification error rate, formula are:
In formula, ht(xi) indicate sample classification results, ht(xi)lFor its inside l element, i indicates sample serial number, ht(xi)l
≠ 1 indicates to meet yi=1 i-th of sample xiIn, l sample elements xl,iClassified by mistake in current iteration layer t, ht
(xi)l≠ -1 meaning is similar, and I (g) indicates discriminant function, if content in bracket is set up, I=1, otherwise I=0, finally to institute
There is the weight d of misjudged samplet,l,iRespectively with yi=1 or yi=-1 carries out classification weighting, obtains two classes point in current iteration layer
Class error rate;
Step 3.7:Ballot parameter alpha is determined using the classification error rate in step (3.6)t, formula is:
In formula, εtFor the classification error rate of all samples in current iteration layer;Ballot parameter alphatAs obtained by the t times iteration points
A kind of measurement of class device classification accuracy rate, when it acts as indicating finally to weight grader in all layers, the power of each layer grader
Weight, the t times Iterative classification error rate is smaller, indicates the t times Iterative classification device for overall more effective;
Step 3.8:Weight vectors are updated by classification error rate and ballot parameter, the weight vectors formula after updating is as follows:
dt+1,i=dt,i*exp(-αtyiht(xi))
In formula, dt,iFor original weight vectors, yiIndicate that score value 1 or -1, value correspond to xiMiddle element;The inside member of exp functions
Element is vector, and exp is using e as the truth of a matter, using each element in vector in bracket as index, carries out exponential function calculating;* in formula
Number indicate that vector point multiplication, the i.e. internal respective items of vector are multiplied, result of calculation is the identical vector of dimension, i.e., updated
dt+1,i;
Weight in weight vectors should meet adduction and be equal to 1, therefore formula is normalized to weight vectors and is:
In formula, Zt is equal to the adduction of weight vectors interior element;
Step 3.1-step 3.8 is repeated, until completing the training of all samples;
Step 3.9:Weak Classifier in all iteration layers finally determined to step 3.5 is made with the ballot parameter that step 3.7 determines
For respective weights, all graders are integrated using weighting, obtain regression equation, which describes sample data and target
Implicit function relationship between energy-consuming requirements, input are 10 master datas in target time, and output is community energy
The predicted value of consumption demand data.
2. a kind of comprehensive energy prediction technique according to claim 1, it is characterised in that:
The computational methods for the sensitivity that all samples change energy-consuming are determined in (2) of the step 2 in solution layer
For:
10 sampled data values of n before the time are denoted as ci,j, i=1 ... m, m=10, j=1 ... n, n>=3;Energy-consuming
Demand measured value is denoted as aj, j=1 ... n;The quantization in rule layer time is denoted as b before the timej, j=1 ... n;Rule layer time and mesh
It is p to mark the correlation between layer energy-consuming requirementsj, solution layer changing sensitivity is denoted as qij, i=1 ... m, solution layer with it is accurate
Then the correlation between layer is vij, for the time in each rule layer, calculated by following formula each in respective party pattern layer
Item quantized value qij, i.e., sensitivity that all samples change energy-consuming:
In formula, qijMeaning is:Changing sensitivity of i-th of the sample in j-th of time for j-th of time energy-consuming requirements
For qij;aj-1、Ci,j-1Preceding 1 year data are indicated in above formula.
3. a kind of comprehensive energy prediction technique according to claim 1, it is characterised in that:(3) institute in the step 2
Each sample of solution layer of seeking stated disappears for the destination layer energy for the weight in corresponding rule layer time and rule layer each time
Take the weight of demand, specially:
There is 1 for rule layer judgment matrix A, the time quantity that dimension n includes by solution layer determines that wherein element is:
In above formula:b、b1、bnAll it is solution layer judgment matrix in above formula, is found out by following matrix;
There are n for the judgment matrix B of solution layer, dimension m is determined by the sample size participated in step 1, for j-th of matrix
Wherein element is:
In above formula:q、q1j、qmjSensitivity is indicated in above formula, is found out by formula above;
The maximum eigenvalue and feature vector for calculating each matrix, the weight vectors of disparity items are acquired by feature vector,
Weight vectors are denoted as respectively:
P=(p1,p2,...pj,...pn), j=1..n
V=(v1j,v2j,...,vij,...vmj), i=1..m
In formula, weight of the different year for energy-consuming demand in p expressiveness layers;Vj was indicated in the rule layer jth time
Weight of the different samples to energy-consuming demand.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510228035.5A CN104809522B (en) | 2015-05-07 | 2015-05-07 | A kind of comprehensive energy prediction technique |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510228035.5A CN104809522B (en) | 2015-05-07 | 2015-05-07 | A kind of comprehensive energy prediction technique |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104809522A CN104809522A (en) | 2015-07-29 |
CN104809522B true CN104809522B (en) | 2018-10-02 |
Family
ID=53694331
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510228035.5A Active CN104809522B (en) | 2015-05-07 | 2015-05-07 | A kind of comprehensive energy prediction technique |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104809522B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748938A (en) * | 2017-11-06 | 2018-03-02 | 国网福建省电力有限公司 | A kind of electric power demand forecasting method based on Vector Autoression Models |
CN108182549B (en) * | 2018-01-30 | 2021-10-01 | 国网河南省电力公司经济技术研究院 | Regional energy efficiency measuring and calculating method considering environmental cost |
CN109145345A (en) * | 2018-03-27 | 2019-01-04 | 燕山大学 | A kind of tailor welded square box shape piece parameter optimization in forming method and system |
CN109086312B (en) * | 2018-06-26 | 2020-09-25 | 湘潭大学 | Data representation method and system for regional comprehensive energy system multi-energy flow network |
CN109829575B (en) * | 2019-01-17 | 2021-05-28 | 新奥数能科技有限公司 | Energy data prediction processing method and device, readable medium and electronic equipment |
CN112966959B (en) * | 2021-03-23 | 2024-04-09 | 海南电网有限责任公司 | Comprehensive evaluation method for grid structure of power distribution network considering load release |
US20230196378A1 (en) * | 2021-12-21 | 2023-06-22 | International Business Machines Corporation | Carbon emission bounded machine learning |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824128A (en) * | 2014-02-24 | 2014-05-28 | 华南理工大学 | Method for scale prediction of regional distributed type comprehensive energy-supply system |
-
2015
- 2015-05-07 CN CN201510228035.5A patent/CN104809522B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103824128A (en) * | 2014-02-24 | 2014-05-28 | 华南理工大学 | Method for scale prediction of regional distributed type comprehensive energy-supply system |
Also Published As
Publication number | Publication date |
---|---|
CN104809522A (en) | 2015-07-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104809522B (en) | A kind of comprehensive energy prediction technique | |
He et al. | Mining transition rules of cellular automata for simulating urban expansion by using the deep learning techniques | |
Abualigah et al. | Salp swarm algorithm: a comprehensive survey | |
Roy et al. | Forecasting heating and cooling loads of buildings: A comparative performance analysis | |
Zhai et al. | Simulating urban land use change by integrating a convolutional neural network with vector-based cellular automata | |
He et al. | Comprehensive evaluation of regional clean energy development levels based on principal component analysis and rough set theory | |
Wei et al. | Comprehensive evaluation model for water environment carrying capacity based on VPOSRM framework: A case study in Wuhan, China | |
Luo et al. | Data-driven fuzzy analysis in quantitative mineral resource assessment | |
Shang et al. | Short-term load forecasting based on PSO-KFCM daily load curve clustering and CNN-LSTM model | |
Liao et al. | A neighbor decay cellular automata approach for simulating urban expansion based on particle swarm intelligence | |
Jiang et al. | Prediction of ecological pressure on resource-based cities based on an RBF neural network optimized by an improved ABC algorithm | |
Adeleke et al. | Prediction of municipal solid waste generation: an investigation of the effect of clustering techniques and parameters on ANFIS model performance | |
Gao et al. | Comparing deep learning models for multi energy vectors prediction on multiple types of building | |
Wu et al. | A surrogate-assisted optimization framework for microclimate-sensitive urban design practice | |
Huang et al. | Research on urban modern architectural art based on artificial intelligence and GIS image recognition system | |
Xia et al. | Exploring the effects of partitioned transition rules upon urban growth simulation in a megacity region: A comparative study of cellular automata-based models in the Greater Wuhan Area | |
Ren et al. | A novel hybrid approach for water resources carrying capacity assessment by integrating fuzzy comprehensive evaluation and analytical hierarchy process methods with the cloud model | |
Liu et al. | Using the ART-MMAP neural network to model and predict urban growth: a spatiotemporal data mining approach | |
Gil et al. | Spatial statistical methods applied to the 2015 Brazilian energy distribution benchmarking model: Accounting for unobserved determinants of inefficiencies | |
Dong et al. | Uncertainty-based multi-objective decision making with hierarchical reliability analysis under water resources and environmental constraints | |
CN102708298B (en) | A kind of Vehicular communication system electromagnetic compatibility index distribution method | |
Zhang et al. | Estimating the outdoor environment of workers’ villages in East China using machine learning | |
Jiang et al. | Rule-based expert system to assess caving output ratio in top coal caving | |
Ren et al. | Time series prediction based on echo state network tuned by divided adaptive multi-objective differential evolution algorithm | |
Gu et al. | Quantifying the effects of different data streams on the calibration of building energy simulation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
EXSB | Decision made by sipo to initiate substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |