CN104881707B - A kind of sintering energy consumption Forecasting Methodology based on integrated model - Google Patents
A kind of sintering energy consumption Forecasting Methodology based on integrated model Download PDFInfo
- Publication number
- CN104881707B CN104881707B CN201510225409.8A CN201510225409A CN104881707B CN 104881707 B CN104881707 B CN 104881707B CN 201510225409 A CN201510225409 A CN 201510225409A CN 104881707 B CN104881707 B CN 104881707B
- Authority
- CN
- China
- Prior art keywords
- model
- sample
- energy consumption
- integrated
- diffτ
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention relates to a kind of sintering energy consumption Forecasting Methodology based on integrated model, this method includes the following steps:1) feature selecting algorithm based on RReliefF carries out aspect of model selection;2) it according to the historical data of the selected aspect of model, establishes and improves extreme learning machine intelligent forecast model and Support Vector Machines for Regression intelligent forecast model;3) multiple single intelligent forecast models are weighted with integrated, the sintering energy consumption predicted value after being integrated.Compared with prior art, the present invention is based on the integrated predictive models of comentropy to have good precision of prediction and generalization ability, while have higher time efficiency, has application and popularization value in actual production.
Description
Technical field
The present invention relates to steel enterprise sintering process energy consumptions to predict field, more particularly, to a kind of burning based on integrated model
Tie energy consumption Forecasting Methodology.
Background technology
Sintering is one of the maximum process that consumes energy in iron and steel enterprise, although recent year Key Iron And Steel sintering circuit
Energy consumption reduces year by year, but it also has larger gap, and energy consumption level difference between domestic each iron and steel enterprise with external advanced value
Significantly, therefore research sintering circuit is energy-saving significant.Using reduce energy consumption and cost as be oriented to sintering ratio and
In process parameter optimizing research, sintering energy consumption prediction is one of critical issue therein.Sintering circuit energy consumption mainly includes electric power
Consumption and non-electricity consumption, wherein non-electricity energy consumption include solid fuel consumption, two major class of gas consumption in ignition, account for total energy consumption
80%~90%, be research sintering consumption reduction Main way.At present, domestic and international existing sintering energy consumption Forecasting Methodology all exists
Certain defect:Input parameter is difficult to determine, modeling method precision deficiency etc., all affect sintering energy consumption prediction precision and when
Effect.The difficult point of sintering energy consumption prediction is:1) feature is difficult to determine.The influence factor of energy consumption is intricate, it is difficult to simply by
Mechanism and empirically determined, needs to select useful information from many factors, rejects redundant variables, obtains more accurately mode input
Parameter.2) single method precision is insufficient.Energy consumption data fluctuation is larger in actual production, cannot be illustrated completely in its inherent mechanism
In the case of, Individual forecast method is difficult to obtain preferable precision of prediction, needs to carry out the Integrated research of a variety of methods and techniques.
By finding that the method predicted and proposed for sintering art energy consumption is seldom, for it to the retrieval of the prior art
The method that his field similar problems are proposed can be used for reference.Chinese patent " a kind of energy consumption Forecasting Methodology and device " (authorizes
Number:CN103544544A in), Yang Haidong et al. proposes a kind of energy consumption prediction based on SVR for enterprise's electricity needs forecast
Method.This method builds training sample set using history energy consumption data, and energy consumption data therein arranges in temporal sequence, passes through choosing
It selects best input exponent number and obtains best prediction model, verify that this method can preferably predict future electrical energy demand by contrast
Value.However, the precision of the Individual forecast model is still to be improved in terms of verification result.Chinese patent is " a kind of based on Bagging's
Sizing process rate of sizing flexible measurement method " (grant number:CN103018426A in), bagging integrated technologies are used for by Tian Huixin
In the prediction of the rate of sizing, it is proposed that the integrated predictive model based on SVR further improves the prediction accuracy of single SVR.No
It crosses, the invention is when determining mode input parameter, only with Analysis on Mechanism and the method for micro-judgment.For steel enterprise sintering
Process, wherein the factor complexity for influencing energy consumption variation is various, it is difficult to it determines to influence the notable feature of energy consumption by Analysis on Mechanism, and
Only judge to be theoretically unsound by rule of thumb, thus need more objective effective parameter selection method.
Invention content
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of prediction result is accurate,
The high sintering energy consumption Forecasting Methodology based on integrated model of time efficiency.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of sintering energy consumption Forecasting Methodology based on integrated model, this method include the following steps:
1) feature selecting algorithm based on RReliefF carries out aspect of model selection;
2) it according to the historical data of the selected aspect of model, establishes and improves extreme learning machine intelligent forecast model and regression branch
Hold vector machine intelligent forecast model;
3) multiple single intelligent forecast models are weighted with integrated, the sintering energy consumption predicted value after being integrated.
RReliefF algorithms are for regression problem of the processing target attribute for successive value, it is assumed that the scale of sample space S
For m, a sample D is randomly choosedi(i=1,2 ..., m), and calculate its nearest knearA neighbour's sample Sk, meanwhile, it is false
If the predicted value of sample is τ ().According to Bayes' theorem, the weight of characteristic attribute A can be obtained by following formula:
Here PdiffA=(diff (A, Di,Dj)|Dj∈Sk) it is sample DiWith its knearCharacteristic attribute A between a neighbour's sample
The probability of difference;Pdiffτ=(diff (τ (), Di,Dj)|Dj∈Sk) it is DiWith its knearPrediction index value between a neighbour's sample
The probability of τ () difference;Pdiffτ|diffA=(diff (τ (), Di,Dj)|diff(A,Di,Dj),Dj∈Sk) represent in known Di
With its knearBetween a neighbour's sample in the case of characteristic attribute A differences its prediction index value τ () difference conditional probability;
Pdiffτ&diffA=(diff (τ (), Di,Dj)·diff(A,Di,Dj)|Dj∈Sk) represent DiWith its knearBetween a neighbour's sample
The probability of characteristic attribute A differences and its prediction index value τ () difference.Here,
Wherein value (A, Di)、value(A,Dj) it is sample Di、DjThe value of middle characteristic attribute A;Max (A) and min (A) are
DiWith its knearThe maximum and minimum value of attribute A in a neighbour's sample.diff(τ(·),Di,Dj) therewith similarly.
The step 1) is specially:
Step101:Parameter initialization enables i=1, j=1;
Step102:Selection sample D in the sample space S for being m in scalei, selected from remaining m-1 sample from this
Sample DiClosest knearA sample, composition neighbour's sample set Sk;
Step103:From SkMiddle selection sample Dj, iterate to calculate the weight N under different predicted valuesdiffτ:
Ndiffτ=Ndiffτ+diff(τ(·),Di,Dj)/knear
Wherein, τ () is the predicted value of sample;
Step104:To each characteristic attribute A, the weight N of different characteristic attribute is iterated to calculatediffAAnd different predicted values
With the weight N of different characteristic attributediffτ&diffA:
NdiffA=NdiffA+diff(A,Di,Dj)/knear
Ndiffτ&diffA=Ndiffτ&diffA+diff(τ(·),Di,Dj)·diff(A,Di,Dj)/knear
Step105:I=i+1, j=j+1 are enabled, judges whether j meets j≤knear, if so, Step103 is returned, if it is not,
Then perform Step106;
Step106:Judge whether i meets i≤m, if so, Step101 is returned to, if it is not, then performing Step107;
Step107:For each attribute A, the final weight estimation of each characteristic attribute is calculated:
W [A]=Ndiffτ&diffA/Ndiffτ-(Ndiffτ-Ndiffτ&diffA)/(m-Ndiffτ)
Finally according to the characteristic attribute of weight preference pattern.
It is described improve extreme learning machine intelligent forecast model process of establishing be:
Stepa1:Determine the number M of sub- learning machine;
Stepa2:With Bootstrap methods from training dataset B sampling with replacement, obtain the instruction of every sub- learning machine
Practice data set Bk, k=1 ..., M, and BkIt is identical with B scales, all it is N;
Stepa3:B is used successivelykThe corresponding sub- learning machine of training, obtains M training result, and respectively with unified test
Data set C examines precision of prediction;
Stepa4:All sub- learning machine training results are integrated with the method for average, obtain final result, and verify model
Precision.
The process of establishing of the Support Vector Machines for Regression intelligent forecast model is:
Stepb1:Initial data is normalized:
Wherein, QpFor p-th of value of each factor, p=1 ..., N, Qmax、QminMaximum value in respectively each factor and most
Small value, a, d be parameter, d=(1-a)/2;
Stepb2:Using RBF kernel functions, the parameter and training network of SVR models are determined;
Stepb3:With the precision of test data set C test networks, neural network forecast output valve is obtained, then carries out anti-normalizing
Change:
Wherein,For the network output valve of q-th of test sample, L is test data set number of samples,For q-th of survey
Predicted value after this renormalization of sample, q=1,2 ..., L;
Stepb4:Performance evaluation is carried out to SVR models.
In the step 3), multiple single intelligent forecast models are weighted using comentropy it is integrated, specially:
Step301:Calculate the degree of variation for the prediction result that each intelligent forecast model is obtained:
Wherein, euq、Represent the relative error magnitudes of u-th of model, q-th of sample and prediction output valve, yqIt represents q-th
Sample desired output, q=1,2 ..., L, L be test data set number of samples, u=1,2, respectively represent improve extreme learning machine
Intelligent forecast model and Support Vector Machines for Regression intelligent forecast model;
Step302:Calculate the entropy of each intelligent forecast model:
Wherein, PudThe Relative Error ratio of u-th of model for q-th of sample,
Step303:Calculate the weights of each intelligent forecast model:
Wherein, zuRepresent the weights of u-th of model;
Step304:The weighting for calculating each model integrates output:
Compared with prior art, the advantage of the invention is that:
1) present invention extracts sintering energy consumption characteristic factor with RReliefF algorithms, is only determined as compared with the past with artificial experience
With more theoretical foundation, the accuracy of prediction result is improved;
2) present invention is weighted integrated using a variety of intelligent prediction algorithms, can effectively realize the accurate pre- of sintering energy consumption
It surveys;
3) the addition Integrated based on comentropy merges the advantage of a variety of prediction models, further improves model essence
Degree, and with higher time efficiency.
Description of the drawings
Fig. 1 is the principle of the present invention schematic diagram;
Fig. 2 is the characteristic attributes weight row for each energy consumption index the present invention is based on the feature that RReliefF algorithms obtain
Sequence figure;
It is solid fuel consumption weight sequencing figure to scheme (2a);It is coal gas energy consumption weight sequencing figure to scheme (2b).
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to
Following embodiments.
As shown in Figure 1, the embodiment of the present invention provides a kind of sintering energy consumption Forecasting Methodology based on integrated model, this method packet
Include following steps:
Step 1), the feature selecting algorithm based on RReliefF carry out aspect of model selection.
The influence factor of energy consumption is numerous in sintering process, can be summarized as:1) dispensing parameter, 2) state parameter, 3) behaviour
Make parameter three classes.Dispensing parameter mainly includes major ingredient and auxiliary material additive amount and dielectric dissipation, these parameters are before sintering starts
Just with determining;State parameter mainly includes the production statuses value such as box temperature, pressure, igniting space gas flow, they are by scene
Sensor obtains;Operating parameter include sintering scene can manual adjustment parameter, as machine speed, thickness of feed layer, mixing machine add
Water etc., they are determined when being sintered beginning by operating personnel.By the analysis to sintering process, by sintering energy consumption influence factor
It summarizes as shown in table 1.
1 sintering energy consumption analysis of Influential Factors of table
The factor for influencing sintering energy consumption is varied, and the height of these factor significance levels is only with Analysis on Mechanism and artificial warp
It tests and is difficult to determine, need to obtain more objective rational conclusion by data analysis.
The present invention carries out aspect of model selection using a kind of improved Relief algorithms (RReliefF algorithms).Relief systems
Row algorithm is a kind of stochastic search methods that feature is typically selected according to weight, and main thought is:According to feature to low coverage
Separating capacity from sample assesses feature, and good feature should approach similar sample, and make inhomogeneous sample separate.
RReliefF algorithms are used for the regression problem that processing target attribute is successive value,
The detailed process that RReliefF algorithms carry out aspect of model selection is as follows:
Step101:Some sample D is selected from the sample set that scale is mi, selected from remaining m-1 sample from this
Sample DiClosest knearA sample, wherein 1≤i≤m, knearValue generally take and 10~20 be advisable.The distance refers to Euclidean
Distance.
Step102:Iterative calculation is in the sample DiOutput-index value P0Under the conditions of weight sets ndC:
Wherein, P0Represent the sample DiOutput-index P value, Pi(1≤i≤knear) it is the knearIn a sample
The output-index value of i-th of sample, PmaxAnd PminIt is the maximum value and minimum value of output-index in m sample respectively;
It calculates in the sample DiWeight sets n under the conditions of input feature vector AdA[A] is iterated calculating as the following formula:
Wherein, A0Sample D belonging to expressioniInput feature vector A value, Ai(1≤i≤knear) it is the knearIn a sample
The value of input feature vector A, A in i-th of samplemaxAnd AminIt is the maximum value and minimum value of input feature vector A in m sample respectively;
It calculates in the sample DiOutput-index value P0With the weight sets n under the conditions of input feature vector AdC&dA[A]:
Step104:Calculate the weighted value W [A] of input feature vector A:
Step105:According to weight preference pattern feature.
Step 2) according to the historical data of the selected aspect of model, is established and improves extreme learning machine intelligent forecast model and return
Return type support vector machines intelligent forecast model.
On the basis of influence sintering energy consumption important feature is obtained, using the improvement ELM based on bagging, (limit learns
Machine, Extreme Learning Machine) algorithm improve original ELM algorithms stability and generalization ability, using ε-SVR (time
Return type support vector machines) algorithm improves Generalization Capability under condition of small sample.
(1) modeling process for improving ELM sintering energy consumption prediction models B-ELM based on Bagging is as follows:
Stepa1:Determine the number M of sub- learning machine.
Stepa2:Determine sub- learning machine training dataset.Bagging data sets are obtained by data scrubbing and feature extractionEach (xi,yi) there are n input and 1 output, i.e. x ∈ Rn, y ∈ R。2/3 is selected as training data
Collect B, remaining is as test data set C.
With Bootstrap methods from training dataset B sampling with replacement, obtain the training dataset of every sub- learning machine
Bk, k=1 ..., M, and BkIt is identical with B scales, all it is N, enables k=1.
Stepa3:Use BkThe corresponding sub- learning machine of training obtains training result, and is examined with unified test data set C
Precision of prediction, k=k+1.
It is random to obtain initial input weight w if K is hidden node numberlWith biasing bl, l=1 ..., K.It is defeated to calculate hidden layer
Go out matrix H { hnl(n=1 ..., N, l=1 ..., K).The generalized inverse matrix H of H is obtained+, according to β=H+T calculates output weights β.
Stepa4:Judge whether k meets k≤M, if so, return to step Stepa3, if it is not, then with the method for average to all
Sub- learning machine training result is integrated, and obtains final result, and verify model accuracy.
(2) process based on the prediction modeling of ε-SVR sintering energy consumptions is as follows:
Stepb1:Initial data is normalized:
Wherein, QpFor p-th of value of each factor, p=1 ..., N, Qmax、QminMaximum value in respectively each factor and most
Small value, a, d be parameter, d=(1-a)/2;
In the present embodiment, by initial data specification to [0.2,0.8] section, a=0.6.
Stepb2:Determine the parameter and training network of SVR models.This model uses RBF kernel functions, penalty factor c and RBF
Variance g in kernel function is obtained by cross-validation method.It comprises the concrete steps that and c and g is respectively set between -10~10 with 0.01
It is changed for step-length, calculates the precision of SVR models respectively under the combination of different c and g, find optimal c and g combinations
Value, in this, as the parameter training network one by one of SVR models.
Stepb3:With the precision of test data set C test networks, neural network forecast output valve is obtained, then carries out anti-normalizing
Change:
Wherein,For the network output valve of q-th of test sample, L is test data set number of samples,For q-th of survey
Predicted value after this renormalization of sample, q=1,2 ..., L.
Stepb4:Performance evaluation, judgement schematics are carried out to SVR models:
Wherein, MeanRe represents average relative error, yqThe actual value of (q=1,2 ..., L) for q-th of test sample.
Step 3) is weighted multiple single intelligent forecast models integrated, the sintering energy consumption predicted value after being integrated.
After B-ELM the and ε-SVR models of sintering energy consumption are established, collection is weighted to above-mentioned submodel using entropy assessment
Into accurately to predict sintering energy consumption.Integrated predictive model idiographic flow based on comentropy is:
Step301:Calculate the degree of variation for the prediction result that each intelligent forecast model is obtained:
Wherein, euq、Represent the relative error magnitudes of u-th of model, q-th of sample and prediction output valve, yqIt represents q-th
Sample desired output, q=1,2 ..., L, L be test data set number of samples, u=1,2, represent B-ELM and ε-SVR moulds respectively
Type.
Step302:Calculate the entropy of each intelligent forecast model:
Wherein, PudThe Relative Error ratio of u-th of model for q-th of sample,
Step303:Calculate the weights of each intelligent forecast model:
Wherein, zuRepresent the weights of u-th of model.
Step304:The weighting for calculating each model integrates output:
By taking the integrated iron and steel works of certain 6,500,000 tons of steel scale of annual output as an example, 2 × 380m2 scales sintered production line is produced per year
8,360,000 tons of finished product sinter, operating rate 94%, usage factor 1.40t/m2h.Choose 311 groups of burnings in 1~December in 2010
It ties production history data and carries out analysis modeling.Every group of sample includes two energy consumption indexs of solid fuel consumption and coal gas energy consumption and 73 spies
Sign, characteristic attribute is as shown in table 1, wherein dispensing parameter 12, state parameter 50, operating parameter 11.Carrying out, data are clear
On the basis of reason, weight sequencing of the feature for each energy consumption index is obtained based on RReliefF algorithms, as shown in Figure 2.
For solid fuel consumption, coming the feature of first 5 is respectively:Coke powder addition, cloud powder addition, mass flow PD-
1st, firing temperature T3, blending ore addition;For coal gas burnup, coming the feature of first 5 is respectively:Firing temperature T3, manifold
Air pressure, blending ore addition, mass flow PD-1, coke powder addition.It can be seen that these feature orderings and actual conditions
It substantially conforms to.
For verification and comparison model precision and performance, the present embodiment chooses following 4 evaluation indexes:
1) average relative error MeanRe:
2) residual error average value e:
3) worst error Emax:
4) precision Pr:
For the validity for the integrated model that comprehensive verification is proposed, first in the case of no progress feature selecting,
Three kinds of models of the prediction result of this model and ELM, B-ELM, ε-SVR are compared, as shown in table 2 and table 3.It is selected by feature
Model accuracy after selecting compares as shown in table 4 and table 5.It should be noted that for each single algorithm (son) model, it is all logical
It crosses cross-validation method and determines optimal model parameters;Meanwhile each model is all run 5 times, takes the average value of each index as most
Terminate fruit.Bagging submodel numbers are set as 10, and characteristic factor retains number and is set as 30.
By table 2 and table 3 it is found that the average relative error of the integrated predictive model proposed by the present invention based on comentropy is excellent
In other four kinds of models, other indexs also show different degrees of advantage, and model is established so as to demonstrate the present invention
Superiority.Here, since integrated model is the normalization weighted sum of 2 submodels, thus its worst error between submodel most
It is reasonable between big error.By table 4 and table 5 it is found that the model accuracy after feature selecting has by a relatively large margin compared with table 2 and table 3
It improves, so as to demonstrate the validity of feature selection approach in the present invention.
Further, since the advantages of ELM and ε-SVR, which have, need not adjust network weight, and training speed is fast;RReliefF is calculated
Method equally has the characteristics that time efficiency is high, therefore integrated model proposed by the present invention has well as a kind of heuritic approach
Time efficiency.As shown in table 6, integrated predictive model total time-consuming about 88s.
Solid fuel consumption prediction index of the table 2 without feature selecting compares
Coal gas energy consumption prediction index of the table 3 without feature selecting compares
The solid fuel consumption prediction index that table 4 has feature selecting compares
The coal gas energy consumption prediction index that table 5 has feature selecting compares
The time efficiency of the different models of table 6 compares
In conclusion the integrated predictive model proposed by the present invention based on comentropy has good precision of prediction and extensive
Ability, while there is higher time efficiency, there is application and popularization value in actual production.
Claims (5)
1. a kind of sintering energy consumption Forecasting Methodology based on integrated model, which is characterized in that this method includes the following steps:
1) feature selecting algorithm based on RReliefF carries out aspect of model selection;
2) according to the historical data of the selected aspect of model, establish improve extreme learning machine intelligent forecast model and regression support to
Amount machine intelligent forecast model;
3) multiple single intelligent forecast models are weighted with integrated, the sintering energy consumption predicted value after being integrated, specially:
Step301:Calculate the degree of variation for the prediction result that each intelligent forecast model is obtained:
Wherein, euq、Represent the relative error magnitudes of u-th of model, q-th of sample and prediction output valve, yqRepresent q-th of sample
Desired output, q=1,2 ..., L, L be test data set number of samples, u=1,2, respectively represent improve extreme learning machine intelligence
Prediction model and Support Vector Machines for Regression intelligent forecast model;
Step302:Calculate the entropy of each intelligent forecast model:
Wherein, PudThe Relative Error ratio of u-th of model for q-th of sample,
Step303:Calculate the weights of each intelligent forecast model:
Wherein, zuRepresent the weights of u-th of model;
Step304:The weighting for calculating each model integrates output:
2. the sintering energy consumption Forecasting Methodology according to claim 1 based on integrated model, which is characterized in that the step 1)
Specially:
Step101:Parameter initialization enables i=1, j=1;
Step102:Selection sample D in the sample space S for being m in scalei, selected from remaining m-1 sample from sample Di
Closest knearA sample, composition neighbour's sample set Sk;
Step103:From SkMiddle selection sample Dj, calculate the weight N under different predicted valuesdiffτ:
Ndiffτ=Ndiffτ+diff(τ(·),Di,Dj)/knear
Wherein, τ () is the predicted value of sample;
Step104:To each characteristic attribute A, the weight N of different characteristic attribute is calculateddiffAAnd different predicted values and different spies
Levy the weight N of attributediffτ&diffA:
NdiffA=NdiffA+diff(A,Di,Dj)/knear
Ndiffτ&diffA=Ndiffτ&diffA+diff(τ(·),Di,Dj)·diff(A,Di,Dj)/knear
Step105:I=i+1, j=j+1 are enabled, judges whether j meets j≤knear, if so, Step103 is returned to, if it is not, then holding
Row Step106;
Step106:Judge whether i meets i≤m, if so, Step101 is returned to, if it is not, then performing Step107;
Step107:For each attribute A, the final weight estimation of each characteristic attribute is calculated:
W [A]=Ndiffτ&diffA/Ndiffτ-(Ndiffτ-Ndiffτ&diffA)/(m-Ndiffτ)
Finally according to the characteristic attribute of weight preference pattern.
3. the sintering energy consumption Forecasting Methodology according to claim 2 based on integrated model, which is characterized in that the knear's
Be worth is 10~20.
4. the sintering energy consumption Forecasting Methodology according to claim 1 based on integrated model, which is characterized in that the improvement pole
Limit learning machine intelligent forecast model process of establishing be:
Stepa1:Determine the number M of sub- learning machine;
Stepa2:With Bootstrap methods from training dataset B sampling with replacement, obtain the training number of every sub- learning machine
According to collection Bk, k=1 ..., M, and BkIt is identical with B scales, all it is N;
Stepa3:B is used successivelykThe corresponding sub- learning machine of training, obtains M training result, and respectively with unified test data set
C examines precision of prediction;
Stepa4:All sub- learning machine training results are integrated with the method for average, obtain final result, and verify model essence
Degree.
5. the sintering energy consumption Forecasting Methodology according to claim 1 based on integrated model, which is characterized in that the regression
The process of establishing of support vector machines intelligent forecast model is:
Stepb1:Initial data is normalized:
Wherein, QpFor p-th of value of each factor, p=1 ..., N, Qmax、QminMaximum value and minimum value in respectively each factor,
A, d be parameter, d=(1-a)/2;
Stepb2:Using RBF kernel functions, the parameter and training network of SVR models are determined;
Stepb3:With the precision of test data set C test networks, neural network forecast output valve is obtained, then carries out renormalization:
Wherein,For the network output valve of q-th of test sample, L is test data set number of samples,For q-th of test sample
Predicted value after renormalization, q=1,2 ..., L;
Stepb4:Performance evaluation is carried out to SVR models.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510225409.8A CN104881707B (en) | 2015-05-04 | 2015-05-04 | A kind of sintering energy consumption Forecasting Methodology based on integrated model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510225409.8A CN104881707B (en) | 2015-05-04 | 2015-05-04 | A kind of sintering energy consumption Forecasting Methodology based on integrated model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104881707A CN104881707A (en) | 2015-09-02 |
CN104881707B true CN104881707B (en) | 2018-07-03 |
Family
ID=53949194
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510225409.8A Active CN104881707B (en) | 2015-05-04 | 2015-05-04 | A kind of sintering energy consumption Forecasting Methodology based on integrated model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104881707B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106802977B (en) * | 2016-12-14 | 2020-10-02 | 同济大学 | Method for predicting performance index of sinter and evaluating comprehensive quality |
CN106709640A (en) * | 2016-12-15 | 2017-05-24 | 华南理工大学 | Method for predicting ship energy consumption based on deep learning and support vector regression |
CN109376934A (en) * | 2018-10-31 | 2019-02-22 | 广东兴发铝业有限公司 | A kind of energy consumption prediction technique of the extruder based on big data |
CN109740683A (en) * | 2019-01-13 | 2019-05-10 | 胡燕祝 | A kind of trend forecasting method based on RReliefF algorithm and support vector regression |
CN110363229B (en) * | 2019-06-27 | 2021-07-27 | 岭南师范学院 | Human body characteristic parameter selection method based on combination of improved RReliefF and mRMR |
CN110443413B (en) * | 2019-07-23 | 2021-08-27 | 华南理工大学 | Electric power material demand prediction system and construction method of electric power material demand model |
-
2015
- 2015-05-04 CN CN201510225409.8A patent/CN104881707B/en active Active
Non-Patent Citations (1)
Title |
---|
基于支持向量机的烧结能耗及性能指标预测模型;王俊凯等;《同济大学学报(自然科学版)》;20140831;第42卷(第8期);第1256-1260页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104881707A (en) | 2015-09-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104881707B (en) | A kind of sintering energy consumption Forecasting Methodology based on integrated model | |
CN109272176B (en) | Calculation method is predicted to platform area line loss per unit using K-means clustering algorithm | |
CN106094744A (en) | The determination method of thermoelectricity factory owner's operational factor desired value based on association rule mining | |
CN108051660A (en) | A kind of transformer fault combined diagnosis method for establishing model and diagnostic method | |
CN105809349B (en) | Dispatching method for step hydropower station group considering incoming water correlation | |
CN110399928A (en) | Voltage of solid oxide fuel cell prediction technique, terminal device and storage medium | |
CN107909208A (en) | Damage method drops in a kind of taiwan area distribution | |
CN107437135B (en) | Novel energy storage type selection method | |
CN111525587B (en) | Reactive load situation-based power grid reactive voltage control method and system | |
CN110400231A (en) | A kind of electric energy measuring equipment crash rate predictor method weighting Nonlinear Bayesian | |
Zhu et al. | Time series data-driven batch assessment of power system short-term voltage security | |
CN109239603A (en) | A kind of extreme learning machine under manifold regularization frame predicts power battery SOC method | |
CN110263834A (en) | A kind of detection method of new energy power quality exceptional value | |
CN102184328A (en) | Method for optimizing land use evolution CA model transformation rules | |
CN107292489A (en) | A kind of dispatching of power netwoks runs lean evaluation system | |
CN112305441A (en) | Power battery health state assessment method under integrated clustering | |
CN112329269B (en) | Sintering ignition temperature modeling prediction method based on working condition identification | |
CN103489007A (en) | Method for selecting kernel function of support vector machine based on sample prior information and application | |
CN108694475A (en) | Short-term time scale photovoltaic cell capable of generating power amount prediction technique based on mixed model | |
CN104850612A (en) | Enhanced cohesion hierarchical clustering-based distribution network user load feature classifying method | |
Yong et al. | Short-term building load forecasting based on similar day selection and LSTM network | |
Cheng et al. | Study of the comprehensive evaluation of energy efficiency of an oilfield heating furnace based on combination weighting method | |
CN109886332A (en) | Improvement DPC clustering algorithm and system based on symmetrical neighborhood | |
Huang et al. | SOC prediction of Lithium battery based on fuzzy information granulation and support vector regression | |
CN108520087A (en) | A kind of the robustness measurement and balance optimizing design method of mechanical structure foreign peoples multiple target performance |
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 |