CN109657878A - A kind of Air-conditioning Load Prediction method and device - Google Patents
A kind of Air-conditioning Load Prediction method and device Download PDFInfo
- Publication number
- CN109657878A CN109657878A CN201910009088.6A CN201910009088A CN109657878A CN 109657878 A CN109657878 A CN 109657878A CN 201910009088 A CN201910009088 A CN 201910009088A CN 109657878 A CN109657878 A CN 109657878A
- Authority
- CN
- China
- Prior art keywords
- value
- prediction
- training
- data sequence
- load
- 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
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000004378 air conditioning Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 140
- 241001123248 Arma Species 0.000 claims abstract description 108
- 238000005070 sampling Methods 0.000 claims description 56
- 238000012545 processing Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 8
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000012731 temporal analysis Methods 0.000 abstract description 2
- 238000000700 time series analysis Methods 0.000 abstract description 2
- 238000005265 energy consumption Methods 0.000 description 4
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000009977 dual effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005096 rolling process Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 235000015170 shellfish Nutrition 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/08—Construction
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Air Conditioning Control Device (AREA)
Abstract
The present invention relates to time series analysis fields, disclose a kind of Air-conditioning Load Prediction method and device, which includes: that the multiple samples of acquisition form training set, and handle sample in the training set, obtain data sequence；Using data sequence training ARMA (p, q) model；It according to arma modeling order p, is successively inputted using the load value in data sequence as arma modeling, exports the estimated value of the load value of subsequent time corresponding with the load value of input；SVR model is trained with the difference of corresponding estimated value according to load value in data sequence and load value；According to arma modeling order p, using current slot internal loading value as the input of arma modeling, the initial prediction at moment to be predicted adjacent with current slot is exported；Using the load value in current slot as the input of SVR model, initial prediction corrected value is exported, the difference of initial prediction and corrected value is the actual prediction value at moment to be predicted.
Description
Technical field
The present invention relates to time series analysis field more particularly to a kind of Air-conditioning Load Prediction method and devices.
Background technique
With the rapid development of urban construction, building energy consumption amount is also growing at top speed.Operation of air conditioner energy consumption accounts for building especially
Large public building operation energy consumption specific gravity is very big, therefore the operation energy consumption for reducing air-conditioning system is always the emphasis of building energy conservation.
It is one of the Major Technology of raising air-conditioning system energy utilization rate using reasonable runing adjustment method, and the reality of the approach
Whether can accurately predict air conditioner load, it is seen that the method for predicting air conditioner load seems especially if now then needing to rely on
It is important.
Air-conditioning Load Prediction refers in the constructing operation stage, cold and hot amount required for being run to future time instance air-conditioning system into
Row short-term forecast, the purpose is to determine optimal fortune based on the power load distributing of prediction for air-conditioning system optimal control service
Row operating condition or set point specify optimal operation of air conditioner strategy, guarantee the comfort of air-conditioned room and the section of air-conditioning system operation
It can type.
Air conditioner load data are considered as a kind of time series, and linear autocorrelation is stronger, while again by a variety of outsides
The influence of enchancement factor, such as the use of electromechanical equipment in solar radiant heat (weather), outside air temperature, fresh air, architectural environment
Quantity etc..In addition to this, air conditioner load amount is related with working condition with its production, is transported again and again as unit of day
Turn, that is, has the characteristics that date periodicity.Therefore, traditional time series models return to the data with These characteristics quasi-
It closes and will receive limitation when predicting, prediction result is often not satisfactory.
Summary of the invention
The present invention provides a kind of Air-conditioning Load Prediction method and device, to improve precision of prediction.
The embodiment of the invention provides a kind of Air-conditioning Load Prediction method, which includes:
With air-conditioning system in intraday load value for a sample, acquires multiple samples and form training set；
For each of training set sample, a load value, multiple samples are acquired every the time of setting
Load value in this combines to form data sequence；
Using data sequence training ARMA (p, q) model；
According to the order p of the arma modeling, successively using the load value in the data sequence as the arma modeling
Input, the estimated value of output and the load value of the corresponding subsequent time of load value inputted；According in the data sequence
Load value and load value and corresponding estimated value difference training SVR model；
According to the order p of the arma modeling, using the load value in current slot as the input of the arma modeling,
Export the initial prediction at the to be predicted moment adjacent with the current slot；Made with the load value in the current slot
For the input of SVR model, the corrected value of the initial prediction is exported, the difference of the initial prediction and the corrected value is
The actual prediction value at the moment to be predicted.
In above-described embodiment, by training arma modeling, the linear segment of data sequence is fitted, and with above-mentioned number
Input according to sequence as arma modeling exports the estimated value of the load value of subsequent time corresponding with the load value of input,
Again by the difference of the load value in the data sequence and load value and corresponding estimated value training SVR model, to data sequence
Non-linear partial be fitted；When being predicted, using the output result of SVR model as corrected value, for correcting ARMA
The prediction result of model, the algorithm that the two models combine improve the accuracy rate of single model, that is, improve precision of prediction.
Optionally, the training set is obtained especially by following manner:
It acquires multiple samples and forms sampling set；
Multiple stochastical sampling is carried out to the sampling set, obtains multiple training sets；
Further include:
For multiple training sets, final predicted value is obtained using strategy is combined according to corresponding actual prediction value.
Optionally, it is non-work that the sample in the training set, which is the sample in working day sample or the training set,
Make day sample.
Optionally, described to be specifically included using data sequence training ARMA (p, q) model:
Determine the order p and q of arma modeling respectively according to bayesian information criterion function.
Optionally, described to be directed to multiple training sets, it is obtained according to corresponding actual prediction value using strategy is combined
Final predicted value specifically:
Calculate the corresponding weight of each training set；
According to the corresponding weight of each training set and actual prediction value, obtained using weighted mean method described final
Predicted value.
It is optionally, described to calculate the corresponding weight of each training set specifically:
For the corresponding arma modeling of each training set and SVR model, other training in addition to itself are chosen
Test data of the data sequence of concentration as the arma modeling and the SVR model calculates each training set pair
The prediction error rate et answered are as follows:
Wherein, yiFor load value included in selected data sequence；It is described with selected data sequence
The input of arma modeling and the SVR model, hiFor the output valve of the output valve and SVR model of the arma modeling
Difference；N is data amount check included in selected data sequence；
According to the corresponding prediction error rate e of each training sett, weight λtAre as follows:
Optionally, described that multiple stochastical sampling is carried out to the sampling set, obtain multiple training sets specifically:
Multiple repairing weld is carried out to the sampling set using self-service sampling method, obtains multiple training sets.
Optionally, before using data sequence training arma modeling further include:
Stationary test is carried out to the data sequence, if meeting stationarity feature, is entered in next step；If being unsatisfactory for putting down
Stability feature then carries out difference processing to the data sequence, until meeting stationarity feature.
The embodiment of the invention also provides a kind of Air-conditioning Load Prediction device, which includes:
Sampling module forms training set for acquiring multiple samples, wherein be in intraday load value with air-conditioning system
One sample；
Data processing module, for for each of training set sample, the time acquisition every setting to be primary
Load value, and combine the load value in multiple samples to form data sequence；
Model training module, for using data sequence training ARMA (p, q) model；According to the arma modeling
Order p is exported corresponding with the load value of input using the load value in the data sequence as the input of the arma modeling
Subsequent time load value estimated value；According to the load value and load value and corresponding estimated value in the data sequence
Difference training SVR model；
Prediction module, for the order p according to the arma modeling, using the load value in current slot described in
The input of arma modeling exports the initial prediction at the to be predicted moment adjacent with the current slot；With it is described current when
Between input of the load value as SVR model in section, export the corrected value of the initial prediction, by the initial prediction with
Actual prediction value of the difference of the corrected value as the moment to be predicted.
In above-described embodiment, arma modeling and SVR model are respectively trained by model training module, and predicted
When, using the output result of SVR model as corrected value, for correcting the prediction result of arma modeling, the two models are combined
Algorithm improve the accuracy rate of single model, that is, improve precision of prediction.
Optionally, the sampling module is specifically used for acquiring multiple samples formation sampling sets, and carries out to the sampling set
Multiple stochastical sampling obtains multiple training sets；
The data processing module is also used to handle the sample in each training set, obtains corresponding data sequence
Column；
The model training module, be also used to using the corresponding data sequence training arma modeling of each training set and
SVR model；
The prediction module is also used to inputting the load value in current slot into the corresponding arma modeling of each training set
In SVR model, actual prediction value is obtained, and be directed to multiple training sets, according to corresponding actual prediction value, using combination
Strategy obtains final predicted value.
Optionally, the model training module is specifically used for determining ARMA mould respectively according to bayesian information criterion function
The order p and q of type.
Optionally, the prediction module is specifically used for according to the corresponding weight of each training set and actual prediction
Value, obtains the final predicted value using weighted mean method.
Optionally, the sampling module carries out multiple stochastical sampling to the sampling set using self-service sampling method, obtains more
A training set.
Detailed description of the invention
Fig. 1 is the algorithm flow chart of prediction technique provided in an embodiment of the present invention；
Fig. 2 is the principle framework figure of prediction technique provided in an embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention is made into one with reference to the accompanying drawing
Step describes in detail, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts all
Other embodiments shall fall within the protection scope of the present invention.
The embodiment of the invention provides a kind of Air-conditioning Load Prediction methods, and the prediction technique is by using SVR model training
The residual error of arma modeling improves the accuracy rate of single model.The prediction technique specifically includes:
With air-conditioning system in intraday load value for a sample, acquires multiple samples and form training set；
For each of training set sample, a load value is acquired every the time of setting, it is negative in multiple samples
Charge values combine to form data sequence；
Using data sequence training ARMA (p, q) model；
According to the order p of arma modeling, successively using the load value in data sequence as the input of arma modeling, output with
The estimated value of the load value of the corresponding subsequent time of the load value of input；According to the load value and load value in data sequence
With the difference of corresponding estimated value training SVR model；
According to the order p of arma modeling, using the load value in current slot as the input of arma modeling, exports and work as
The initial prediction at adjacent moment to be predicted preceding period；Using the load value in current slot as the input of SVR model,
The difference of the corrected value of output initial prediction, initial prediction and corrected value is the actual prediction value at moment to be predicted.
In above-described embodiment, by training arma modeling, the linear segment of data sequence is fitted, and with above-mentioned number
Input according to sequence as arma modeling exports the estimated value of the load value of subsequent time corresponding with the load value of input,
Again by the difference of the load value in the data sequence and load value and corresponding estimated value training SVR model, to data sequence
Non-linear partial be fitted；When being predicted, using the output result of SVR model as corrected value, for correcting ARMA
The prediction result of model, the algorithm that the two models combine improve the accuracy rate of single model, that is, improve precision of prediction.
In order to more be apparent from the principle of Air-conditioning Load Prediction method provided in an embodiment of the present invention, now in conjunction with attached drawing
It is described in detail.
As shown in Figure 1, the prediction technique mainly comprises the steps that
Step S101: with air-conditioning system in intraday load value for a sample, multiple samples is acquired and form training set；
Such as unit of day, the load value of acquisition m days forms the training set；
Step S102: for each of training set sample, a load value, multiple samples are acquired every the time of setting
Load value in this combines to form data sequence；
Step S103: using data sequence training ARMA (p, q) model；According to the order p of arma modeling, successively with number
Input according to the load value in sequence as arma modeling exports the load value of subsequent time corresponding with the load value of input
Estimated value；According to the difference of the load value in data sequence and load value and corresponding estimated value training SVR model；
Step S104: according to the order p of arma modeling, using the load value in current slot as the defeated of arma modeling
Enter, exports the initial prediction at the to be predicted moment adjacent with current slot；
Step S105: using the load value in current slot as the input of SVR model, the correction of initial prediction is exported
The difference of value, initial prediction and corrected value is the actual prediction value at moment to be predicted.
It being influenced by festivals or holidays etc., air conditioner load will show different changing rules, therefore, to improve precision of prediction,
As in step s101, the sample in training set is working day sample, or is nonworkdays sample, and according to trained
To arma modeling and SVR model remove the load value at moment to be predicted prediction work day, or when prediction nonworkdays is to be predicted
The load value at quarter.
Before using data sequence training arma modeling, need to carry out stationary test, if meeting stationarity feature, into
Enter in next step；If being unsatisfactory for stationarity feature, need to carry out difference processing to data sequence, until meeting stationarity feature.For
Convenient for description, using d={ y1,y2…ykIndicating above-mentioned data sequence, k indicates the moment；After being detected by stationarity, adopt
With data sequence training arma modeling method particularly includes:
Firstly, the expression formula of ARMA model (Auto-Regressive Moving Average, ARMA)
Are as follows:
yt=φ1yt-1+φ2yt-2+……+φpyt-p+at-θ1at-1-θ2at-2-……-θqat-q
Wherein, φiIt is respectively the lag order of autoregressive coefficient and autoregression part, θ with pjIt is respectively rolling average with q
The lag order of coefficient and rolling average part, atIt is white noise sequence；
ARMA is determined respectively according to bayesian information criterion function (Bayesian Information Criteria, BIC)
The order p and q of model, after order determines, the parameter of arma modelingθjAnd white noise varianceMoments estimation meet following side
Journey:
Wherein, r0,r1,...rqFor the auto-covariance function of sequence, the solution of each parameter can be obtained using numerical solution.Root
According to different training sets, the order of arma modeling selects difference, so that the parameter of identification is also different, and then enhances individual study
The diversity of device.
Secondly, arma modeling determine after, according to the order p of arma modeling, successively using the load value in data sequence as
The input of above-mentioned arma modeling exports the estimated value of the load value of subsequent time corresponding with the load value of input；According to number
According to the difference of the load value in sequence and load value and corresponding estimated value training SVR model；
Specifically, the expression formula of support vector regression model (Support Vector Regression, SVR) are as follows:
F (x, w)=w φ (x)+b
I.e. with data setTraining SVR model parameter w and
b.The parametric solution of the model can be melted into following quadratic programming problems:
s.t.f(xi)-yi≤ε+ξi
yi-f(xi)≤ε+ξ′i
ξi≥0,ξ′i>=0, i=1,2...n
The problem is converted to a dual problem using Lagrange's equation:
It can be identified the parameter of SVR model by SMO algorithm.
Finally, being predicted by obtained arma modeling with air conditioner load of the SVR model to future time instance, that is, to work as
Input of the load value as arma modeling in the preceding period exports the initial pre- of the to be predicted moment adjacent with current slot
Measured value；Using the load value in current slot as the input of SVR model, the corrected value of initial prediction, initial predicted are exported
Value and the difference of corrected value are the actual prediction value at moment to be predicted.
In order to further increase precision of prediction, training set in available multiple step S101, to each training set according to
It is secondary using method described in step S102~step S104 operate, the load value at finally obtained moment to be predicted have it is multiple,
Final prediction result is obtained using strategy, such as the method for average, weighted mean method is combined according to these predicted values.
Specifically, step S201: acquiring multiple samples and form sampling set, and carry out multiple stochastical sampling to sampling set, obtain
To multiple above-mentioned training sets；
Wherein, the sample in sampling set is working day sample or nonworkdays sample, and by self-service sampling method to adopting
Sample collection is sampled, and the data difference in each training set can be made to maximize in this way, self-service sampling method specifically: given packet
Data set D containing m sample carries out sampling to it and generates data set D '；A sample is selected from D at random every time, is copied
Shellfish is put into D ', then again puts back to the sample in initial data set D, so that the sample is in sampling next time it is possible to by adopting
It arrives；This process repeat m times after to get having arrived the data set D ' comprising m sample.
Step S202: for each of each training set sample, acquiring a load value every the time of setting, and
It combines the load value in the multiple samples for belonging to same training set to form data sequence；
Step S203: being directed to each training set, and method training arma modeling and SVR mould in step S103 is respectively adopted
Type；
Step S204: being directed to each training set, and the method being respectively adopted in step S104 treats the load value of prediction time
It is predicted, obtains multiple actual prediction values；
Step S205: being directed to above-mentioned multiple training sets, is obtained most according to corresponding actual prediction value using strategy is combined
Whole predicted value.
In step S201~step S205, each training set can be regarded as the training set of a weak learner, by every
The arma modeling that the corresponding data sequence training of a training set obtains and the training pattern that SVR model is the weak learner, and
And the difference of the corrected value of the initial prediction and SVR model output of arma modeling output is the output of the weak learner
Value；Multiple such weak learners are by combining strategy to obtain the higher strong learner of a precision of prediction.
In a specific embodiment, the combination strategy of use are as follows: calculate the corresponding weight of each training set；According to every
The corresponding weight of a training set and actual prediction value, obtain final predicted value using weighted mean method.It can be to be each weak
Learner distributes a weight, wherein weight is determining according to the extensive error of weak learner, its power of the smaller weak learner of error
Value is then bigger, i.e. the effect in final strong learner is bigger.
Corresponding algorithm are as follows:
For the corresponding arma modeling of each training set and SVR model, choose in other training sets in addition to itself
Test data of the data sequence as the arma modeling and SVR model, calculate the pre- sniffing of corresponding each weak learner
Accidentally rate etAre as follows:
Wherein, yiFor load value included in selected data sequence；With selected data sequence for ARMA mould
The input of type and SVR model, hiFor the difference of the output valve of the output valve and SVR model of the arma modeling；N is selected
Data sequence included in data amount check；
According to the prediction error rate e of each weak learnert, weight λtAre as follows:
When being predicted, according to the output valve h of each weak learnert(actual prediction value) and corresponding weight λt?
To the output valve H (final predicted value) of strong learner are as follows:
H=λ1h1+λ2h2+……+λtht
In a specific embodiment, as shown in Fig. 2, sampling this number of days m=16, weak learner number T=12, ARMA
The order of model is limited in p≤5, in q≤5.
The present embodiment includes the following steps:
Step S301: middle on weekdays to acquire multiple samples with air-conditioning system in intraday load value for a sample
Working day sampling set is formed, multiple samples are acquired in nonworkdays and form nonworkdays sampling set, are respectively labeled as D1And D2,
In D1And D2It is middle to filter out 16 days data that missing values are few, sampled point is more complete respectively；
Step S302: to working day sampling set D obtained in the first step1, respectively by self-service sampling method, it is adopted
Sample is put back to former sampling set after stochastical sampling is primary specifically, being sampled as one day data every time as unit of day by sample, then
It carries out second of stochastical sampling, amounts to 16 times, obtain one and include 16 days, i.e., the training set of 16 samples；Repeat aforesaid operations
12 times, 12 training sets can be obtained, each training set is the training set of a weak learner；
Step S303: following operation is repeated to each weak learner:
1) for each of training set sample, invalid value is deleted, missing values is supplemented, air conditioner load curve is carried out flat
Sliding filtering, in the morning after 7 points, the data of interception in every ten minutes, pretreated data are about 60-70 a after obtaining one day；
2) load value in 16 samples combines to form data sequence, carries out stationary test to the data sequence, if full
3) sufficient stationarity feature, then enter step, if not satisfied, then carrying out difference processing to it, then carry out stationary test, until its
Meet weakly stationary feature；
3) parameter identification, the expression formula of arma modeling are carried out to arma modeling using corresponding data sequence are as follows:
yt=φ1yt-1+φ2yt-2+......+φpyt-p+at-θ1at-1-θ2at-2-......-θqat-q
Rank is determined using BIC criterion, after determining p value and q value, corresponding parameter Estimation formula is as follows:
Wherein, r0,r1,...rqFor the auto-covariance function of sequence.The solution of each parameter, ARMA mould are obtained using numerical solution
Shape parameter identification finishes；
4) after arma modeling determines, according to the order p of arma modeling, successively using the load value in data sequence as above-mentioned
The input of arma modeling exports the estimated value of the load value of subsequent time corresponding with the load value of input；According to data sequence
Load value and load value in column train SVR model with the difference of corresponding estimated value；
I.e. with data setTraining number as SVR model
According to solving following dual problem:
Kernel function K (xi,xj)=φ (xi)Tφ(xi) select gaussian kernel function, i.e. radial basis function (Radial Basis
Function, RBF):
Wherein, xcFor kernel function center, σ is the width parameter of function, controls the radial effect range of function；It is above-mentioned right
Even problem solving use SMO (Sequential Minimal Optimization) algorithm, finally obtain SVR model parameter w and
b；
5) air conditioner load is predicted with SVR model by obtained arma modeling, that is, according to the order of arma modeling
P exports the to be predicted moment adjacent with current slot using the load value in current slot as the input of arma modeling
Initial prediction；Using the load value in current slot as the input of SVR model, the corrected value of initial prediction is exported, just
Beginning predicted value and the difference of corrected value are the actual prediction value at moment to be predicted；
6) survey of the data sequence in other training sets in addition to itself as the arma modeling and SVR model is chosen
Data are tried, the corresponding prediction error rate et of each weak learner is calculated are as follows:
Wherein, yiFor load value included in selected data sequence；With selected data sequence for ARMA mould
The input of type and SVR model, hiFor the difference of the output valve of the output valve and SVR model of the arma modeling；N is selected
Data sequence included in data amount check；
According to the prediction error rate e of weak learnert, determine the weight λ of weak learnertAre as follows:
Step S304: average weighted combination strategy is used, weak learner is integrated into a strong learnerTo nonworkdays sampling set D obtained in the first step2, the operation of step S302 and step S304 are repeated, is obtained
To another strong learner H2；Using strong learner H1The air conditioner load at moment to be predicted prediction work day, using strong learner H2
Predict the air conditioner load at nonworkdays moment to be predicted.
The embodiment of the invention also provides a kind of Air-conditioning Load Prediction device, which includes:
Sampling module forms training set for acquiring multiple samples, wherein be in intraday load value with air-conditioning system
One sample；
Data processing module, for acquiring a load every the time of setting for each of training set sample
Value, and combine the load value in multiple samples to form data sequence；
Model training module, for using data sequence training ARMA (p, q) model；According to the order p of arma modeling, with
Input of the load value as arma modeling in data sequence exports the load of subsequent time corresponding with the load value of input
The estimated value of value；According to the difference of the load value in data sequence and load value and corresponding estimated value training SVR model；
Prediction module, for the order p according to arma modeling, using the load value in current slot as arma modeling
Input exports the initial prediction at the to be predicted moment adjacent with current slot；Using the load value in current slot as
The input of SVR model exports the corrected value of initial prediction, using the difference of initial prediction and corrected value as the moment to be predicted
Actual prediction value.
In order to further increase precision of prediction, available multiple such training sets, and pass through data processing module pair
Sample in each training set is handled, and one group of data sequence is obtained；According to corresponding data sequence, pass through model training mould
Block training arma modeling and SVR model, the load value at moment to be predicted are exported eventually by prediction module, according to these respectively
Predicted value obtains final prediction result using strategy, such as the algorithm method of average, weighted mean method is combined.
Specifically, sampling module, is specifically used for acquiring multiple samples formation sampling sets, and carry out sampling set repeatedly random
Sampling, obtains multiple training sets, generallys use self-service sampling method and carries out multiple stochastical sampling to sampling set；
Data processing module is also used to handle the sample in each training set, obtains corresponding data sequence；
Model training module is also used to using the corresponding data sequence training arma modeling of each training set and SVR model；
Wherein it is possible to determine the order p and q of arma modeling respectively according to bayesian information criterion function；
Prediction module, be also used to input the load value in current slot the corresponding arma modeling of each training set and
In SVR model, actual prediction value is obtained, and be directed to multiple training sets, it is tactful using combining according to corresponding actual prediction value,
Such as arithmetic mean method, weighted mean method, final predicted value is obtained；In a specific embodiment, it is specifically used for according to every
The corresponding weight of a training set and actual prediction value, obtain final predicted value using weighted mean method.
By above description as can be seen that in the embodiment of the present invention, using self-service sampling method and algorithm parameter method of perturbation
Increase the otherness between weak learner；In each weak learner, using the residual error of SVR model training arma modeling, to
The prediction result of arma modeling is corrected, the combination of both models improves the accuracy rate of single model；In conjunction with multiple weak study
Device obtains strong learner using weighted mean method, improves precision of prediction.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (13)
1. a kind of Air-conditioning Load Prediction method characterized by comprising
With air-conditioning system in intraday load value for a sample, acquires multiple samples and form training set；
For each of training set sample, every in load value of time acquisition for setting, multiple samples
Load value combine to form data sequence；
Using data sequence training ARMA (p, q) model；
According to the order p of the arma modeling, successively using the load value in the data sequence as the defeated of the arma modeling
Enter, exports the estimated value of the load value of subsequent time corresponding with the load value of input；According to negative in the data sequence
Charge values and load value and the difference of corresponding estimated value training SVR model；
According to the order p of the arma modeling, using the load value in current slot as the input of the arma modeling, output
The initial prediction at the to be predicted moment adjacent with the current slot；Using the load value in the current slot as
The input of SVR model exports the corrected value of the initial prediction, and the difference of the initial prediction and the corrected value is institute
State the actual prediction value at moment to be predicted.
2. Air-conditioning Load Prediction method as described in claim 1, which is characterized in that the training set is especially by following manner
It obtains:
It acquires multiple samples and forms sampling set；
Multiple stochastical sampling is carried out to the sampling set, obtains multiple training sets；
Further include:
For multiple training sets, final predicted value is obtained using strategy is combined according to corresponding actual prediction value.
3. Air-conditioning Load Prediction method as described in claim 1, which is characterized in that the sample in the training set is work
Sample in day sample or the training set is nonworkdays sample.
4. Air-conditioning Load Prediction method as described in claim 1, which is characterized in that described using data sequence training
ARMA (p, q) model specifically includes:
Determine the order p and q of arma modeling respectively according to bayesian information criterion function.
5. Air-conditioning Load Prediction method as claimed in claim 2, which is characterized in that described to be directed to multiple training sets, root
Final predicted value is obtained using strategy is combined according to corresponding actual prediction value specifically:
Calculate the corresponding weight of each training set；
According to the corresponding weight of each training set and actual prediction value, the final prediction is obtained using weighted mean method
Value.
6. Air-conditioning Load Prediction method as claimed in claim 5, which is characterized in that described to calculate the corresponding power of each training set
Value specifically:
For the corresponding arma modeling of each training set and SVR model, choose in other training sets in addition to itself
Test data of the data sequence as the arma modeling and the SVR model, it is corresponding to calculate each training set
Prediction error rate etAre as follows:
Wherein, yiFor load value included in selected data sequence；With selected data sequence for the arma modeling
And the input of the SVR model, hiFor the difference of the output valve of the output valve and SVR model of the arma modeling；N is
Data amount check included in selected data sequence；
According to the corresponding prediction error rate e of each training sett, weight λtAre as follows:
7. Air-conditioning Load Prediction method as claimed in claim 2, which is characterized in that it is described to the sampling set carry out repeatedly with
Machine sampling, obtains multiple training sets specifically:
Multiple repairing weld is carried out to the sampling set using self-service sampling method, obtains multiple training sets.
8. Air-conditioning Load Prediction method as described in any one of claims 1 to 7, which is characterized in that use the data sequence
Before training arma modeling further include:
Stationary test is carried out to the data sequence, if meeting stationarity feature, is entered in next step；If being unsatisfactory for stationarity
Feature then carries out difference processing to the data sequence, until meeting stationarity feature.
9. a kind of Air-conditioning Load Prediction device characterized by comprising
Sampling module forms training set for acquiring multiple samples, wherein with air-conditioning system in intraday load value for one
Sample；
Data processing module, for acquiring a load every the time of setting for each of training set sample
Value, and combine the load value in multiple samples to form data sequence；
Model training module, for using data sequence training ARMA (p, q) model；According to the order of the arma modeling
P, using the load value in the data sequence as the input of the arma modeling, under output is corresponding with the load value of input
The estimated value of the load value at one moment；According to the difference of load value and load value and corresponding estimated value in the data sequence
It is worth training SVR model；
Prediction module, for the order p according to the arma modeling, using the load value in current slot as the ARMA mould
The input of type exports the initial prediction at the to be predicted moment adjacent with the current slot；In the current slot
Input of the load value as SVR model, the corrected value of the initial prediction is exported, by the initial prediction and the school
Actual prediction value of the difference of positive value as the moment to be predicted.
10. Air-conditioning Load Prediction device as claimed in claim 9, which is characterized in that
The sampling module is specifically used for acquiring multiple samples formation sampling sets, and repeatedly adopts at random to sampling set progress
Sample obtains multiple training sets；
The data processing module is also used to handle the sample in each training set, obtains corresponding data sequence；
The model training module is also used to use the corresponding data sequence training arma modeling of each training set and SVR
Model；
The prediction module, be also used to input the load value in current slot the corresponding arma modeling of each training set and
In SVR model, actual prediction value is obtained, and be directed to multiple training sets, according to corresponding actual prediction value, using in conjunction with plan
Slightly, final predicted value is obtained.
11. Air-conditioning Load Prediction device as claimed in claim 9, which is characterized in that the model training module is specifically used for
Determine the order p and q of arma modeling respectively according to bayesian information criterion function.
12. Air-conditioning Load Prediction device as claimed in claim 10, which is characterized in that the prediction module is specifically used for root
According to the corresponding weight of each training set and actual prediction value, the final predicted value is obtained using weighted mean method.
13. Air-conditioning Load Prediction device as claimed in claim 10, which is characterized in that the sampling module is adopted using self-service
Sample method carries out multiple stochastical sampling to the sampling set, obtains multiple training sets.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910009088.6A CN109657878A (en) | 2019-01-04 | 2019-01-04 | A kind of Air-conditioning Load Prediction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910009088.6A CN109657878A (en) | 2019-01-04 | 2019-01-04 | A kind of Air-conditioning Load Prediction method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109657878A true CN109657878A (en) | 2019-04-19 |
Family
ID=66118409
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910009088.6A Pending CN109657878A (en) | 2019-01-04 | 2019-01-04 | A kind of Air-conditioning Load Prediction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109657878A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598923A (en) * | 2019-09-03 | 2019-12-20 | 深圳市得益节能科技股份有限公司 | Air conditioner load prediction method based on support vector regression optimization and error correction |
| CN110991774A (en) * | 2019-12-31 | 2020-04-10 | 新奥数能科技有限公司 | Electric quantity load prediction method and device |
| CN111322716A (en) * | 2020-02-24 | 2020-06-23 | 青岛海尔工业智能研究院有限公司 | Air conditioner temperature automatic setting method, air conditioner, equipment and storage medium |
| CN112747416A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Energy consumption prediction method and device for air conditioning system |
| CN113111419A (en) * | 2021-04-16 | 2021-07-13 | 西安建筑科技大学 | Method and system for establishing and predicting air-conditioning load prediction model in office building |
| CN115018188A (en) * | 2022-06-29 | 2022-09-06 | 西安热工研究院有限公司 | Intake valve jam prediction method based on ARMA algorithm |
| CN118428550A (en) * | 2024-05-22 | 2024-08-02 | 北京云庐科技有限公司 | Prediction method, training method and prediction device for building energy consumption load |
-
2019
- 2019-01-04 CN CN201910009088.6A patent/CN109657878A/en active Pending
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598923A (en) * | 2019-09-03 | 2019-12-20 | 深圳市得益节能科技股份有限公司 | Air conditioner load prediction method based on support vector regression optimization and error correction |
| CN112747416A (en) * | 2019-10-31 | 2021-05-04 | 北京国双科技有限公司 | Energy consumption prediction method and device for air conditioning system |
| CN112747416B (en) * | 2019-10-31 | 2022-04-05 | 北京国双科技有限公司 | Energy consumption prediction method and device for air conditioning system |
| CN110991774A (en) * | 2019-12-31 | 2020-04-10 | 新奥数能科技有限公司 | Electric quantity load prediction method and device |
| CN111322716A (en) * | 2020-02-24 | 2020-06-23 | 青岛海尔工业智能研究院有限公司 | Air conditioner temperature automatic setting method, air conditioner, equipment and storage medium |
| CN111322716B (en) * | 2020-02-24 | 2021-08-03 | 青岛海尔工业智能研究院有限公司 | Air conditioner temperature automatic setting method, air conditioner, equipment and storage medium |
| CN113111419A (en) * | 2021-04-16 | 2021-07-13 | 西安建筑科技大学 | Method and system for establishing and predicting air-conditioning load prediction model in office building |
| CN115018188A (en) * | 2022-06-29 | 2022-09-06 | 西安热工研究院有限公司 | Intake valve jam prediction method based on ARMA algorithm |
| CN118428550A (en) * | 2024-05-22 | 2024-08-02 | 北京云庐科技有限公司 | Prediction method, training method and prediction device for building energy consumption load |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109657878A (en) | A kind of Air-conditioning Load Prediction method and device | |
| CN110555561B (en) | Medium-and-long-term runoff ensemble forecasting method | |
| CN102779228B (en) | Method and system for online prediction on cooling load of central air conditioner in marketplace buildings | |
| CN102183621B (en) | Aquaculture dissolved oxygen concentration online forecasting method and system | |
| CN103912966B (en) | A kind of earth source heat pump refrigeration system optimal control method | |
| CN107730031B (en) | Ultra-short-term peak load prediction method and system | |
| CN101480143B (en) | Method for predicating single yield of crops in irrigated area | |
| CN110705743A (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
| CN103942461A (en) | Water quality parameter prediction method based on online sequential extreme learning machine | |
| CN105069525A (en) | All-weather 96-point daily load curve prediction and optimization correction system | |
| CN103489039B (en) | There is the freeway traffic flow amount fusion forecasting method of online self-tuning optimization ability | |
| CN102183802B (en) | Short-term climate forecast method based on Kalman filtering and evolution modeling | |
| CN104217258B (en) | A kind of electric load sigma-t Forecasting Methodology | |
| CN106600050A (en) | BP neural network-based ultra-short load prediction method | |
| Sülo et al. | Energy efficient smart buildings: LSTM neural networks for time series prediction | |
| CN104517162A (en) | On-line hardness forecasting method of continuous annealing product by means of integrated learning | |
| CN104679989A (en) | Hydrogen atom clock error predicting method based on modified BP (back propagation) neural network | |
| CN109242265A (en) | Based on the smallest Urban Water Demand combination forecasting method of error sum of squares | |
| CN114572229B (en) | Vehicle speed prediction method, device, medium and equipment based on graph neural network | |
| CN111461466A (en) | Heating household valve adjusting method, system and equipment based on L STM time sequence | |
| CN115170006B (en) | Dispatching method, device, equipment and storage medium | |
| Kofinas et al. | Daily multivariate forecasting of water demand in a touristic island with the use of artificial neural network and adaptive neuro-fuzzy inference system | |
| CN109993354A (en) | A method of it is predicted for energy consumption | |
| CN107316096A (en) | A kind of track traffic one-ticket pass passenger amount of entering the station Forecasting Methodology | |
| JPH04372046A (en) | Method and device for predicting demand amount |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190419 |
|
| RJ01 | Rejection of invention patent application after publication |