CN109389164A - Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model - Google Patents
Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model Download PDFInfo
- Publication number
- CN109389164A CN109389164A CN201811144451.7A CN201811144451A CN109389164A CN 109389164 A CN109389164 A CN 109389164A CN 201811144451 A CN201811144451 A CN 201811144451A CN 109389164 A CN109389164 A CN 109389164A
- Authority
- CN
- China
- Prior art keywords
- support vector
- energy consumption
- per unit
- vector regression
- regression model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The regional per Unit GDP Energy Consumption prediction technique based on support vector regression model that the present invention provides a kind of, the present invention establishes the regional per Unit GDP Energy Consumption prediction model based on support vector regression (SVR) of complete set, model covers the processes such as initial data PCA pretreatment, data extension, model optimization, model prediction Performance Evaluation, with modeling rule simple general-purpose, convenient for calculating the advantages that realizing and optimizing, preanalysis quickly flexibly can be carried out to regional per Unit GDP Energy Consumption according to specific needs.
Description
Technical field
The regional per Unit GDP Energy Consumption prediction technique based on support vector regression model that the present invention relates to a kind of.
Background technique
Per Unit GDP Energy Consumption is a kind of efficiency of energy utilization index, be able to reflect a country energy consumption level and
Energy-saving situation.The economic total volume in China has leapt to the second in the world, economic growth after the high speed development that have passed through decades
While energy-output ratio also gradually increasing.According to statistics, China's total energy consumption in 2000 is 1,469,640,000 tons of standard coals, until
Total energy consumption is 4,358,190,000 tons of standard coals within 2016, is 3 times in 2000.
The utilization efficiency of the energy decides the degree of sustainable development of a countries and regions, and China is same what is developed the economy
When also attach great importance to the utilization efficiency of the energy, state-owned unit's GDP energy consumption is down to 2016 from 1.33 tons of standard coal/ten thousand yuan in 2000
0.59 ton of standard coal/Wan Yuan in year, but due to the difference of the industrial structure between different regions to exist on per Unit GDP Energy Consumption
The per Unit GDP Energy Consumption of certain difference, the higher area of secondary industry specific gravity is higher, therefore to the pre- of regional per Unit GDP Energy Consumption
It surveys the raising to regional efficiency of energy utilization, industrial structure optimization adjustment and energy saving situation early warning etc. and all has important meaning
Justice.
The method predicted at present regional per Unit GDP Energy Consumption mainly has three classes:
The first kind: it is to carry out per Unit GDP Energy Consumption using gray scale prediction model to be predicted, generallys use GM (1,1) model.
This method is generate after cumulative pretreatment more having by original series using per Unit GDP Energy Consumption historical data as original series
Then the new data sequence of rule establishes the rule that the single order single argument differential equation finds new data sequence using new data sequence
Rule obtains data response sequence by the solution differential equation, carries out difference to calculated result and obtain original series match value, finally again
It is verified to ensure that the precision of prediction of model.This method has data requirements amount is few, calculates easy, initial data to be not necessarily to rule
The features such as distribution, but also with fault-tolerance, small, robustness is affected by parameter Estimation and is not suitable for carrying out Long-term analysis
The disadvantages of.
Second class: being the prediction that per Unit GDP Energy Consumption is carried out using time series method.Time series method is usually per GDP
The historical data of energy consumption is arranged to obtain a string of original series according to time sequencing, then carries out the outside forecast sequence to the sequence
The future developing trend of column, time series method are capable of the continuous-changing features of quick unit of analysis GDP energy consumption, reflect data pair
The unidirectional linearity of elephant contacts, but this method is higher to the dependency degree of historical data, is suitble to prediction to stablize the data continued, has ignored
The influence of other correlated variables factors.
Third class: being the per Unit GDP Energy Consumption prediction based on Multivariable regressive analysis model.This method assumes initially that one group of variable
Exist between factor and per Unit GDP Energy Consumption and be associated with and establish sample data set, data set is using per Unit GDP Energy Consumption value as dependent variable
Then other factors are estimated back as explanatory variable according to parameter of the observation of data set sample to regression model
Return equation and significance test is carried out to regression equation, per Unit GDP Energy Consumption is then carried out according to the regression equation and is predicted.It should
Method master has been able to reflect the linear relationship between per Unit GDP Energy Consumption and variable, but precision of prediction is by relevant explanation variable data
Quality is affected, and is difficult to reflect the non-linear relation of per Unit GDP Energy Consumption Yu each explanatory variable.
Summary of the invention
The purpose of the present invention is to provide a kind of regional per Unit GDP Energy Consumption prediction side based on support vector regression model
Method.
To solve the above problems, to provide a kind of regional per Unit GDP Energy Consumption based on support vector regression model pre- by the present invention
Survey method, comprising:
Data set is established, the data set includes explanatory variable and dependent variable;
Carry out principal component analysis processing to the explanatory variable, the chief composition series data that obtain that treated;
Treated that chief composition series are extended accordingly to described, obtains the extension data of seasonal time series;
It is used as input vector based on the dependent variable and treated chief composition series data, building SVR model is instructed
Practice the support vector regression model completed;
Using the extension data of the seasonal time series as variable, the support vector regression mould that training is completed is inputted
Type carries out the season property forecast analysis of regional per Unit GDP Energy Consumption, obtains the prediction result of regional per Unit GDP Energy Consumption.
Further, in the above-mentioned methods, the support vector regression model is ε-support vector regression model.
Further, in the above-mentioned methods, the explanatory variable includes: regional fixed investment, social consumer goods
The above industrial added value of the total volume of retail sales, scale, society's electricity consumption amount, primary industry total output value, secondary industry total output value, third
Industry total output value, same period last year comprehensive energy consumption total amount.
Further, in the above-mentioned methods, the dependent variable includes: regional per Unit GDP Energy Consumption value.
Further, in the above-mentioned methods, in the carry out principal component analysis processing of the explanatory variable, dimension k value
It is 4.
Further, in the above-mentioned methods, the length of the extension data of the seasonal time series is unit GDP energy consumption
Target prediction length.
Further, in the above-mentioned methods, based on the dependent variable, chief composition series data are used as input with treated
Vector, construct support vector regression model, obtain training completion support vector regression model after, further includes:
Based on the per Unit GDP Energy Consumption in multiple areas, the support vector regression model that training is completed is tested and optimized
Training, the support vector regression model after being optimized, wherein the insensitive coefficient of ε is 0.1 in support vector regression model, is punished
Penalty factor C is 0.3, carries out model optimization using 10 folding cross validations in the training of support vector regression model optimization;
Using the extension data of the seasonal time series as variable, the support vector regression mould that training is completed is inputted
Type, comprising:
Support vector regression model using the extension data of the seasonal time series as variable, after input optimization.
Further, in the above-mentioned methods, the per Unit GDP Energy Consumption based on multiple areas, the supporting vector that training is completed
Training is tested and optimized to regression model, the support vector regression model after being optimized, comprising:
By predetermined period, based on the per Unit GDP Energy Consumption in multiple areas, the support vector regression model completed to training is automatic
Training is tested and is optimized offline, the support vector regression model after being optimized.
Further, in the above-mentioned methods, the kernel function of the support vector regression model is linear kernel function.
Compared with prior art, the present invention establishes the regional unit based on support vector regression (SVR) of complete set
GDP energy consumption prediction model, model covers initial data PCA pretreatment, data extension, model optimization, model prediction performance are commented
The processes such as estimate, have many advantages, such as modeling rule simple general-purpose, is realized and optimization convenient for calculating, it can be flexibly fast according to specific needs
Speed carries out preanalysis to regional per Unit GDP Energy Consumption.
Detailed description of the invention
Fig. 1 is the PCA structure chart figure of one embodiment of the invention;
Fig. 2 is the SVR model structure of one embodiment of the invention;
Fig. 3 is the SVR model training procedure chart of one embodiment of the invention;
Fig. 4 is the per Unit GDP Energy Consumption prediction model overall structure figure of one embodiment of the invention;
Fig. 5 is certain province per Unit GDP Energy Consumption prediction effect figure of one embodiment of the invention.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
The present invention provides a kind of regional per Unit GDP Energy Consumption prediction technique based on SVR model, comprising:
Step S1, establishes data set, and the data set includes explanatory variable and dependent variable;
Step S2 handles the carry out principal component analysis (PCA) of the explanatory variable, the chief composition series that obtain that treated
Data;
Step S3, treated that chief composition series are extended accordingly to described, obtains prolonging for seasonal time series
Stretch data;
Step S4 is used as input vector, building SVR mould based on the dependent variable and treated chief composition series data
Type obtains support vector regression (SVR) model of training completion;
Step S5 inputs the supporting vector that training is completed using the extension data of the seasonal time series as variable
(SVR) model is returned, the season property forecast analysis of regional per Unit GDP Energy Consumption is carried out, obtains the prediction knot of regional per Unit GDP Energy Consumption
Fruit.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, the supporting vector is returned
Returning (SVR) model is ε-support vector regression (ε-SVR) model.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, the explanatory variable packet
It includes: the above industrial added value of regional fixed investment, the total retail sales of consumer goods, scale, society's electricity consumption amount.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, the dependent variable packet
It includes: regional per Unit GDP Energy Consumption value.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, step S2, to described
In carry out principal component analysis (PCA) processing of explanatory variable, dimension k value is 4.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, the seasonality time
The length of the extension data of sequence is unit GDP power dissipation obj ectives prediction length.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, step S4 is based on institute
Dependent variable and treated chief composition series data are stated as input vector, construct SVR model, obtain support that training is completed to
Amount returns after (SVR) model, further includes:
Based on the per Unit GDP Energy Consumption in multiple areas, to support vector regression (SVR) model that training is completed carry out test and
Optimization training, support vector regression (SVR) model after being optimized, wherein ε is unwise in support vector regression (SVR) model
Feeling coefficient is 0.1, and penalty coefficient C is 0.3, uses 10 folding cross validations in the training of support vector regression (SVR) model optimization
Carry out model optimization;
Step S5 inputs the supporting vector that training is completed using the extension data of the seasonal time series as variable
Return (SVR) model, comprising:
Support vector regression (SVR) using the extension data of the seasonal time series as variable, after input optimization
Model.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, the supporting vector is returned
The kernel function for returning (SVR) model is linear kernel function.
In regional one embodiment of per Unit GDP Energy Consumption prediction technique based on SVR model of the invention, based on multiple areas
Per Unit GDP Energy Consumption, training is tested and optimized to support vector regression (SVR) model completed to training, after being optimized
Support vector regression (SVR) model, comprising:
By predetermined period, based on the per Unit GDP Energy Consumption in multiple areas, support vector regression (SVR) mould completed to training
Training is tested offline automatically and optimized to type, support vector regression (SVR) model after being optimized.
Here, the invention proposes a kind of regional per Unit GDP Energy Consumption based on support vector regression (SVR) algorithm model is pre-
Survey method.
Regional per Unit GDP Energy Consumption is the ratio of this area's total output value and total energy consumption, and the present invention analyzes fixed assets
The regional GDPs phases such as the above industrial added value of investment, the total retail sales of consumer goods, scale, outer trade total import and export value
The Variable Factors of pass, and regional society relevant to regional per Unit GDP Energy Consumption electricity consumption, same period last year total energy consumption sequential value
Factor is established SVR model and is predicted current regional per Unit GDP Energy Consumption, pre- by carrying out to regional per Unit GDP Energy Consumption
It surveys, current this area's socio-economic development can be analyzed to the degree of dependence and trend of energy consumption, and be able to reflect the ground
The efficiency of energy utilization in area all has this area's energy consumption using the propulsion of horizontal, energy saving situation early warning and energy-saving work
Highly important meaning.
Support vector machines is a kind of algorithm based on statistical theory, it can be very good the classification applied to high dimensional data
And regression problem, and decision boundary is indicated using a subset of training data, which is known as supporting vector, supporting vector
Regression model (SVR) is a kind of method for solving regression problem with algorithm of support vector machine.SVR can be good at solving sample
The problems such as measuring small, non-linear, high-dimensional, local minimum also has the advantages such as calculating speed is fast, model stability, generalization ability are strong.
The present invention utilizes regional season fixed investment, the total retail sales of consumer goods, the above industrial added value of scale, outer trade
The variables such as total import and export value and area's season electricity demand, same period last year total energy consumption sequential value establish SVR model and carry out ground
The prediction of area's season per Unit GDP Energy Consumption.
The invention proposes a kind of regional per Unit GDP Energy Consumption prediction techniques for being based on support vector regression (SVR).The present invention
The prediction of middle area season per Unit GDP Energy Consumption is based on regional season per Unit GDP Energy Consumption history value and correlated variables factor is established
SVR prediction model.It is history value as dependent variable using regional season per Unit GDP Energy Consumption, regional season fixed assets is thrown
The above industrial added value of money volume, the total retail sales of consumer goods, scale, outer trade total import and export value, enterprise of service industry battalion on rule
Industry income, primary industry GDP, secondary industry GDP, tertiary industry GDP, regional season electricity demand, same period last year energy consumption are total
Amount is used as explanatory variable to establish the regional season per GDP of SVR after carrying out principal component analysis (PCA) and seasonal time series processing
Energy consumption prediction model.
Regional per Unit GDP Energy Consumption prediction model establishment process based on support vector regression model is mainly as follows:
Dependent variable data pretreatment
1) dependent variable data PCA is handled
It establishes before model first to regional history GDP data and fixed investment, the total retail sales of consumer goods, scale
The explanatory variables such as the above industrial added value carry out pretreatment and increase the regularity of data to reduce the error of fitting of model.First will
Each variable carries out principal component analysis (PCA) processing, and PCA is that there may be the conversions of the variable of correlation by one group by orthogonal transformation
For one group, linearly incoherent variable, PCA algorithm can extract the main information of data and abandon redundancy.
The present invention is with fixed investment, the total retail sales of consumer goods, the above industrial added value of scale, season electricity consumption
The explanatory variables such as demand, primary industry GDP establish data matrix X:
Wherein, n is variable number, and m is variable data length, carries out 0-1 standardization to the matrix, standardization formula is such as
Under:
Wherein xijFor the value of variable,For the mean value of the variable, σiFor the standard deviation of the variable, after enabling standardization
Matrix is X*, then seeks the covariance matrix of matrix X*:
Solve characteristic equation | D (X*)-λ I |=0, I are unit matrix, solve eigenvalue λ1, λ2... λn, then solve feature
Value corresponding feature vector a1, a2 ..., an, then determine each components contribution degree:K principal components of selection calculate
Accumulate contribution degree:
K principal component G (k) value of selection should be greater than 85% or more.The value of k is 4 in the present invention.
Finally calculated using value of the ingredient expression formula to each ingredient:
The PCA processing structure of explanatory variable data is as shown in Figure 1.
2) extension of number of principal components evidence
For the demand of prediction data length, need that treated that chief composition series data are extended accordingly to PCA,
The present invention using seasonal time series carry out data extension, t node point to sequence data Y the estimated value of t+k node can
It is calculated by following formula:
Yt+k=at+kbt+ct+k, (k=1,2 ..., t)
A in formulatFor horizontal item, btFor slope term, ctFor season, it is assumed that K is to keep the minimum of t+k-s*K≤t establishment whole
Number, wherein s is cycle length, and the present invention is using season in year as the period, s=4, is enabled
ct+k=ct+k-sK, (k=1,2 ..., t)
Horizontal item calculates:
at=α (Yt-ct-s)+(1-α)(at-1+bt-1), (0 < α < 1)
Slope term calculates:
bt=β (at-at-1)+(1-β)bt-1, (0 < β < 1)
Season calculates:
ct=γ (Yt-at)+(1-γ)ct-s, (o≤γ≤1)
Utilize primary condition:
as+1=Ys+1
bs+1=(Ys+1-Y1)/s
ci=Yi-(Yi+bs+1(i-1)), (i=1,2 ..., s+1)
Solve ai, bi, ci, i=1,2 ..., s+1 iteration can obtain the estimation that t moment walks sequence k
Yt+k=at+kbt+ct+k-s
Parameter alpha, beta, gamma select under the conditions of mean absolute percentage error is the smallest.
(2) SVR model construction and training
The present invention is based on ε-SVR algorithms to construct model, and ε-SVR is to solve regression problem based on support vector machines (SVM)
A kind of method, what is sought is that an equation of linear regression goes to be fitted all sample points, and the optimal hyperlane that it seeks is to make sample
Point is minimum from hyperplane population variance, and structure is as shown in Figure 2.
SVR model makes structure risk replace empiric risk as expected risk by introducing ε insensitive loss function, and
Keep the sparse property of support vector machines;Insensitive loss function is as follows:
C (x, y, f (x))=| y-f (x) |ε
|y-f(x)|ε=max 0, | and y-f (x) |-ε }
Here ε is to take a fixed positive number in advance, when the difference of the observed value y and predicted value f (x) of x point are no more than in advance
When given ε, then it is assumed that the predicted value f (x) to change the time is free of losses.
The present invention by regional season fixed investment, the total retail sales of consumer goods, the above industrial added value of scale,
It is data set X, regional season per Unit GDP Energy Consumption that the explanatory variables such as sales Value of Industry, which carry out the pretreated data definition of PCA,
Data definition is data set R, to establish data sample training set D as input vector:
D={ (x1, y1), (x2, y2) ..., (xn, yn), xi∈ X, yi∈ R, (i=1,2 ..., n)
First with nonlinear mapping function Φ by sample training collection D Nonlinear Mapping a to space higher-dimension Hilbert
H, then linear regression is carried out in higher dimensional space H, under kernel function mapping, D corresponds to the training set in the space H are as follows:
D '={ (z1, y1), (z2, y2) ..., (zn, yn)={ (φ (x1), y1), (φ (x2), y2) ..., (φ (x3),
yn)}
Wherein z is the Nonlinear Mapping of x;Then linear regression is carried out in high-dimensional feature space H, is tieed up to obtain in original
The effect of linear regression, estimation function are carried out on degree are as follows:
F (x)=(w, φ (x))+b
Wherein w and b is model parameter;
The present invention uses soft ε-support vector regression method (ε-SVR), when using linear function to return sample point, big portion
Divide sample point in ε-band, has a small amount of sample point outer in ε-band, the sample point fallen in outside ε-band is punished using penalty coefficient C
It penalizes, the original optimization problem (i.e. minimum structure risk) being expressed as:
Wherein ξi,For slack variable.
To solve above-mentioned original optimization problem, it is converted into antithesis using Lagrange (Lagrange) multiplier method and asks
Topic.Introduce Lagrange function:
Wherein, α(*)≥0;η(*)>=0 is Lagrange multiplier.
Lagrange function is first solved about w, b, ξ(*)Minimum.Extremum conditions has:
Above formula is substituted into Lagrange function, then original optimization problem is converted into following dual problem are as follows:
Above-mentioned dual problem is solved, α is obtained(*).Then parameter w, b can be calculated by following formula:
Select some αj> 0 orCalculate b:
B=yi-(w·zj)-ε
W is substituted into estimation function equation, obtains regression function:
Wherein xiSupporting vector in referred to as training set D, K (x, xi) it is kernel function, K (x, xi) rely only on Nonlinear Mapping
The inner product of function phi, it is desirable that meet Mercer theorem.The present invention is in regional season per Unit GDP Energy Consumption prediction model using linear
Kernel function is as kernel function:
K(xi, xj)=xi·xj
The training process of SVR model is as shown in Figure 3 in the present invention.
(3) per Unit GDP Energy Consumption is predicted
The extension data of seasonal time series is inputted into trained SVR model, regional per Unit GDP Energy Consumption can be carried out
The forecast analysis of season property.Regional per Unit GDP Energy Consumption prediction model overall structure is as shown in Figure 4.
(4) quality evaluation is modeled
The present invention by the public data of local area statistics office and State Statistics Bureau to provinces such as Zhejiang, Jiangsu consolidating in recent years
Determine the season data of the variables such as the above industrial added value of assets investment volume, the total retail sales of consumer goods, scale, society's electricity consumption amount
It is collected, and regional season per Unit GDP Energy Consumption is calculated according to regional season GDP and energy-consuming data.It will collect
Arrive province data building data set and using partial data as training set carry out model training, rest part as verifying collect with
Assessment modeling quality.
Fig. 5 is that certain province season per Unit GDP Energy Consumption models effect, per Unit GDP Energy Consumption actual value dot lines table in figure
Show, training match value and predictive estimation value are indicated with triangulation point lines, and non-shaded portion is that training set per Unit GDP Energy Consumption data are real
Actual value and model training match value, dash area are verifying collection per Unit GDP Energy Consumption actual value and model prediction data, SVR season
The mean absolute error MAE=0.007 of per Unit GDP Energy Consumption model training collection fitting, training set mean absolute percentage error
MAPE=1.55%, prediction verifying collection MAE=0.0069, MAPE=1.76%.Area per Unit GDP Energy Consumption provincial to part
SVR modeling and forecasting the result shows that, model-fitting degree it is higher and be not present multicollinearity, over-fitting the problems such as, to relevant explanation
SVR model is established after variable progress PCA pretreatment to carry out rationally effective forecast analysis to regional per Unit GDP Energy Consumption.
(6) Optimization Mechanism of per Unit GDP Energy Consumption prediction model
Due to the influence of the factors such as economic law, Regional Industrial Structure adjustment, fixed investment, social consumption are explained
A degree of change can occur for the explanatory variables such as the product total volume of retail sales, society's electricity consumption amount and the relationship of dependent variable per Unit GDP Energy Consumption,
Therefore, the present invention formulates the Continuous optimization mechanism of a set of model.
Area per Unit GDP Energy Consumption prediction model established by the present invention carries out the automatic offline excellent of model by some cycles
Change.According to the actual situation, every 1 year after model foundation and investment application, system will choose relevant explanation variable nearly 3 again
Year season data and per Unit GDP Energy Consumption data establish training set, learning training again is carried out to model, so that optimization is current
Regional per Unit GDP Energy Consumption prediction model.
In conclusion 1) present invention is by carrying out regional per GDP energy using soft ε-support vector regression (ε-SVR) model
Consume the method for prediction.The present invention is increased using more than regional fixed investment, the total retail sales of consumer goods, scale industry
The data such as the variables such as value, society's electricity consumption amount establish SVR model as dependent variable as explanatory variable, regional per Unit GDP Energy Consumption value,
Learning training based on model realizes the forecast analysis to regional per Unit GDP Energy Consumption value.
2) in model of the invention in the PCA preprocess method of explanatory variable, model leads the explanatory variable of input
Constituent analysis (PCA) reduces the input dimension of model, reduces the Problems of Multiple Synteny between variable outlier and variable
SVR model bring is influenced, dimension k value is 4 when PCA of the present invention handles explanatory variable.
3) data extension is carried out using seasonal time series in model of the invention.Using the seasonal time in the present invention
Sequence extends the pretreated sequence data of PCA, and the data of extension input the SVR model trained as variable
With the prediction purpose of implementation model.The length of the extension data of the seasonal time series is that per Unit GDP Energy Consumption target is pre-
Survey length.
4) in the overall model flowage structure of regional per Unit GDP Energy Consumption of the invention prediction, regional per GDP in the present invention
The prediction model structure of energy consumption includes the foundation of data set, the pretreatment of explanatory variable, the extension of preprocessed data, SVR training
Model tuning and data in the process predicts process.
5) in regional per Unit GDP Energy Consumption SVR prediction model of the invention in the optimization of relevant parameter, the present invention is based on to more
The per Unit GDP Energy Consumption in a area is established SVR prediction model and is tested, and the insensitive coefficient of ε is 0.1 in model, and penalty coefficient C is
0.3, model optimization is carried out using 10 folding cross validations in model training.
6) in the selection of per Unit GDP Energy Consumption SVR prediction model Kernel Function of the invention, the present invention is based on to multiple areas
Per Unit GDP Energy Consumption and various kernel function establish SVR model carry out test comparison after, select linear kernel function as unit
Kernel function in GDP energy consumption SVR prediction model.
7) in the Optimization Mechanism of per Unit GDP Energy Consumption SVR prediction model of the invention, the present invention has formulated holding for a set of model
Continuous Optimization Mechanism, the per Unit GDP Energy Consumption prediction mould that the present invention establishes carry out automatic offline optimization by some cycles.
Compared with prior art, the present invention establishes the regional unit based on support vector regression (SVR) of complete set
GDP energy consumption prediction model, model covers initial data PCA pretreatment, data extension, model optimization, model prediction performance are commented
The processes such as estimate, have many advantages, such as modeling rule simple general-purpose, is realized and optimization convenient for calculating, it can be flexibly fast according to specific needs
Speed carries out preanalysis to regional per Unit GDP Energy Consumption.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, the specific application and design constraint depending on technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
Obviously, those skilled in the art can carry out various modification and variations without departing from spirit of the invention to invention
And range.If in this way, these modifications and changes of the present invention belong to the claims in the present invention and its equivalent technologies range it
Interior, then the invention is also intended to include including these modification and variations.
Claims (9)
1. a kind of regional per Unit GDP Energy Consumption prediction technique based on support vector regression model characterized by comprising
Data set is established, the data set includes explanatory variable and dependent variable;
Carry out principal component analysis processing to the explanatory variable, the chief composition series data that obtain that treated;
Treated that chief composition series are extended accordingly to described, obtains the extension data of seasonal time series;
It is used as input vector based on the dependent variable and treated chief composition series data, building SVR model obtains having trained
At support vector regression model;
Using the extension data of the seasonal time series as variable, the support vector regression model that training is completed is inputted, into
The season property forecast analysis of row area per Unit GDP Energy Consumption, obtains the prediction result of regional per Unit GDP Energy Consumption.
2. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In the support vector regression model is ε-support vector regression model.
3. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In, the explanatory variable include: regional fixed investment, the total retail sales of consumer goods, the above industrial added value of scale,
Society's electricity consumption amount, primary industry total output value, secondary industry total output value, tertiary industry total output value, same period last year comprehensive energy
Consume total amount.
4. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In the dependent variable includes: regional per Unit GDP Energy Consumption value.
5. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In in the carry out principal component analysis processing of the explanatory variable, dimension k value is 4.
6. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In the length of the extension data of the seasonal time series is unit GDP power dissipation obj ectives prediction length.
7. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In being used as input vector based on the dependent variable and treated chief composition series data, building support vector regression model obtains
After the support vector regression model completed to training, further includes:
Based on the per Unit GDP Energy Consumption in multiple areas, training is tested and optimized to the support vector regression model completed to training,
Support vector regression model after being optimized, wherein the insensitive coefficient of ε is 0.1 in support vector regression model, penalty coefficient
C is 0.3, carries out model optimization using 10 folding cross validations in the training of support vector regression model optimization;
Using the extension data of the seasonal time series as variable, the support vector regression model that training is completed is inputted, packet
It includes:
Support vector regression model using the extension data of the seasonal time series as variable, after input optimization.
8. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as claimed in claim 7
In, based on the per Unit GDP Energy Consumption in multiple areas, training is tested and optimized to the support vector regression model completed to training,
Support vector regression model after being optimized, comprising:
By predetermined period, based on the per Unit GDP Energy Consumption in multiple areas, the support vector regression model completed to training is automatically offline
Training is tested and is optimized, the support vector regression model after being optimized.
9. the regional per Unit GDP Energy Consumption prediction technique based on support vector regression model, feature exist as described in claim 1
In the kernel function of the support vector regression model is linear kernel function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811144451.7A CN109389164A (en) | 2018-09-28 | 2018-09-28 | Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811144451.7A CN109389164A (en) | 2018-09-28 | 2018-09-28 | Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109389164A true CN109389164A (en) | 2019-02-26 |
Family
ID=65418393
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811144451.7A Pending CN109389164A (en) | 2018-09-28 | 2018-09-28 | Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109389164A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112232608A (en) * | 2020-12-16 | 2021-01-15 | 浙江大学 | Regional building air conditioner short-term power consumption prediction method based on data driving |
CN114403877A (en) * | 2022-01-21 | 2022-04-29 | 中山大学 | Multi-physiological-signal emotion quantitative evaluation method based on two-dimensional continuous model |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2004326240A (en) * | 2003-04-22 | 2004-11-18 | Ntt Power & Building Facilities Inc | Energy consumption management method, energy consumption managerial system, and program for building |
CN104573854A (en) * | 2014-12-23 | 2015-04-29 | 国家电网公司 | Iron steel electricity consumption forecasting method and device |
CN107909182A (en) * | 2017-10-09 | 2018-04-13 | 中国电子科技集团公司第二十八研究所 | A kind of prediction of economic indexes method and system based on regression analysis |
CN108009668A (en) * | 2017-10-31 | 2018-05-08 | 中国南方电网有限责任公司 | A kind of tune load forecasting method on a large scale using machine learning |
CN108470227A (en) * | 2018-03-29 | 2018-08-31 | 郑州工程技术学院 | Based on the Regional Macro economic System prediction modeling method of assessment GDP growth figures and application |
-
2018
- 2018-09-28 CN CN201811144451.7A patent/CN109389164A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2004326240A (en) * | 2003-04-22 | 2004-11-18 | Ntt Power & Building Facilities Inc | Energy consumption management method, energy consumption managerial system, and program for building |
CN104573854A (en) * | 2014-12-23 | 2015-04-29 | 国家电网公司 | Iron steel electricity consumption forecasting method and device |
CN107909182A (en) * | 2017-10-09 | 2018-04-13 | 中国电子科技集团公司第二十八研究所 | A kind of prediction of economic indexes method and system based on regression analysis |
CN108009668A (en) * | 2017-10-31 | 2018-05-08 | 中国南方电网有限责任公司 | A kind of tune load forecasting method on a large scale using machine learning |
CN108470227A (en) * | 2018-03-29 | 2018-08-31 | 郑州工程技术学院 | Based on the Regional Macro economic System prediction modeling method of assessment GDP growth figures and application |
Non-Patent Citations (2)
Title |
---|
MUHAMMAD SHAKEEL 等: "Energy consumption and GDP with the role of trade in South Asia", 《IEEE》 * |
魏亮 等: "影响我国单位GDP能耗的因素分析", 《产业与科技论坛》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112232608A (en) * | 2020-12-16 | 2021-01-15 | 浙江大学 | Regional building air conditioner short-term power consumption prediction method based on data driving |
CN114403877A (en) * | 2022-01-21 | 2022-04-29 | 中山大学 | Multi-physiological-signal emotion quantitative evaluation method based on two-dimensional continuous model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhong et al. | Super efficiency SBM-DEA and neural network for performance evaluation | |
Li et al. | Innovation efficiency of semiconductor industry in China: A new framework based on generalized three-stage DEA analysis | |
Lee et al. | Forecasting energy consumption using a grey model improved by incorporating genetic programming | |
Lee et al. | Forecasting nonlinear time series of energy consumption using a hybrid dynamic model | |
Mingming et al. | A multiple adaptive wavelet recurrent neural network model to analyze crude oil prices | |
Shao et al. | Density prediction and dimensionality reduction of mid-term electricity demand in China: A new semiparametric-based additive model | |
CN109389164A (en) | Regional per Unit GDP Energy Consumption prediction technique based on support vector regression model | |
Zhu et al. | Energy efficiency evaluation and prediction of large-scale chemical plants using partial least squares analysis integrated with Gaussian process models | |
Wang et al. | A new seasonal cycle GM (1, 1) model and its application in railway passenger volume forecasting | |
Chou et al. | Comparison of AHP and fuzzy AHP methods for human resources in science technology (HRST) performance index selection | |
CN112330030B (en) | System and method for predicting requirements of expansion materials | |
CN109993354A (en) | A method of it is predicted for energy consumption | |
CN116957640A (en) | Product sales prediction method and system | |
He et al. | Combined forecasting of regional logistics demand optimized by genetic algorithm | |
Wei et al. | A time-varying diffusion index forecasting model | |
CN116703644A (en) | Attention-RNN-based short-term power load prediction method | |
Ye et al. | Novel weight-adaptive fusion grey prediction model based on interval sequences and its applications | |
CN106408118A (en) | GRNN (generalized regression neural network) combination model-based urban daily water supply prediction method | |
CN107730025A (en) | A kind of power supply enterprise's Cash flow prospects model changed based on electricity | |
Jiang et al. | A Transformer Based Method with Wide Attention Range for Enhanced Short-term Load Forecasting | |
Hu | Demand forecasting of green metal materials using non-equidistant grey prediction with robust nonlinear interval regression analysis | |
Novaria et al. | Predicted demand for 3 kg LPG gas in each provinces area in Indonesia | |
Bekiros et al. | Forecasting the US economy with a factor-augmented vector Autoregressive DSGE model | |
Shi et al. | A study of sales forecasting in multinational retail companies: a feature extraction-machine learning-classification based forecasting framework | |
Wang et al. | Grey prediction based software stage-effort estimation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190226 |