CN106446440A - Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine - Google Patents
Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine Download PDFInfo
- Publication number
- CN106446440A CN106446440A CN201610888121.3A CN201610888121A CN106446440A CN 106446440 A CN106446440 A CN 106446440A CN 201610888121 A CN201610888121 A CN 201610888121A CN 106446440 A CN106446440 A CN 106446440A
- Authority
- CN
- China
- Prior art keywords
- power
- data
- learning machine
- prediction
- extreme learning
- 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
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The invention relates to a short-term photovoltaic generation power prediction method based on online sequential extreme learning machine; an online sequential extreme learning machine with forgetting mechanism is employed, and moment, temperature and illumination intensity are selected as inputs to a prediction model; the method comprises the steps of generating an initial hidden layer output matrix of the extreme learning machine according to input historical data, and calculating initial output weight; predicting photovoltaic generation power, and waiting for weather measurement data and power measurement data; after the arrival of the data awaited, calculating error evaluation indexes to obtain a prediction error, storing historical data, generating a new hidden layer output matrix, updating the output weight and photovoltaic generation predicted power, and continuing to read weather forecast data. The method of the invention can provide improved prediction precision.
Description
Technical field
The invention belongs to photovoltaic power generation power prediction field.
Background technology
The research of solar photovoltaic generation system, for energy problem and environmental problem is alleviated, improves energy consumption structure,
Distributed generation system performance is improved, photovoltaic generation industry is opened up, the meaning with great theory and reality.Although photovoltaic is sent out
Electricity have many, but photovoltaic generation is affected by various factors, its power output have undulatory property, randomness and
Intermittent.After photovoltaic generating system is incorporated into the power networks, its undulatory property can be impacted to stablizing for power system, threaten electrical network
Safety, substantially increase the difficulty of electric power system dispatching.
From the point of view of prediction mode, physical method and statistical method can be divided into.Physical method is first to directly affecting power
The factor of output is (such as:Radiant intensity, photovoltaic plate temperature) it is predicted, then it is input to the physics of photovoltaic system by predicting the outcome
In model, output is obtained.Statistical method need not then analyze specific physical model, by historical data, set up statistics
Model is learned, directly output is predicted.Conventional statistical method has at present:Support vector machine, artificial neural network, ash
Color-Markov Chain etc..
Content of the invention
It is an object of the invention to provide a kind of short-term photovoltaic power generation power prediction method, to reduce its impact to electrical network,
Be conducive to traffic department's arranged rational operation plan, adjust power system operation mode in time.Technical scheme is such as
Under:
A kind of short-term photovoltaic power generation power prediction method based on online online-sequential extreme learning machine, using with Forgetting Mechanism
Online online-sequential extreme learning machine, it is characterised in that from moment, temperature, intensity of illumination, the input quantity as forecast model.
Step is as follows:
Step 1:According to the historical data being input into, the initial hidden layer output matrix of extreme learning machine is generated, and is calculated
Initial output weight;
Step 2:Prediction photovoltaic generation power, waits meteorologic survey data and power measurements;
Step 3:After the data for being waited are reached, calculation error evaluation index, forecast error is obtained, stores history number
According to, new hidden layer output matrix H-matrix being generated, and output weight beta and photovoltaic generation prediction power P is updated, gas is read in continuation
As forecast data;
Step 4:Return to step 2.
Wherein, error assessment index can be as follows:
(1) standardized mean square error nRMSE is
(2) average absolute percent error MAPE is
Wherein, n is photovoltaic power station power generation period number of samples, PratedFor rated power, PpiFor the pre- power scale of i period,
PmiActual power for the i period.
The short-term photovoltaic work(based on the online online-sequential extreme learning machine (FOS-ELM) with Forgetting Mechanism that the present invention is provided
Rate prediction algorithm, continually introduces new data, and eliminates the impact of outdated data, by historical weather data, history photovoltaic work(
Rate data, data of weather forecast, carry out short-term forecast to photovoltaic generation power.Simulation example shows, the method has training speed
Degree is fast, the high feature of precision of prediction.Traffic department's arranged rational is conducive to adjust based on the short-term photovoltaic prediction algorithm of FOS-ELM
Degree plan is that multiple power sources space-time is complementary provides support with coordination control, to ensureing the safety and stability of system, promoting the excellent of electrical network
Change operation significant.
Description of the drawings
Fig. 1 extreme learning machine structure
Fig. 2 FOS-ELM forecast model flow chart
Predictive value when Fig. 3 (a) (b) (c) is respectively the fine day of three kinds of methods is compared with actual value
Fig. 4 (a) (b) (c) be respectively three kinds of methods overcast and rainy when predictive value compare with actual value
Specific embodiment
(1) extreme learning machine
The network structure of extreme learning machine (extrem learning machine, ELM) as shown in figure 1, rudimentary algorithm such as
Under.
Assume that BP network model has L hidden layer node, activation primitive is G ().For N number of different study
Sample (x, y), x ∈ Rd*N,y∈RN,ai∈R1*d, bi∈ R, ai、biIt is all the matrix and vector for randomly generating, ELM expression formula is such as
Shown in formula.
G () is activation primitive, is the weight vector for connecting i-th node and output node, is the quantity of hidden node,
G () can be arbitrarily infinite differentiable function.If Sigmoid function is as shown in formula.
Formula (1) is rewritten into matrix form as shown in formula,
H β=Y * MERGEFORMAT (3)
Wherein,
β=[β1β2... βL]TY=[y1y2... yN]T
The least square solution of modus ponens (5) is as shown in formula.
Matrix H is the hidden layer output matrix of ELM, and its i-th row h is with respect to the hidden layer output vector of input, exports
If weights β is unique parameter for needing training to determine. the ELM model containing individual hidden node can learn the training that quantity is
Sample, and no residual error is present.
In order to the stability of result and generalization ability is improved, increase regularization parameter C, as shown in formula:
(2) the online online-sequential extreme learning machine with Forgetting Mechanism
In many practical applications, training data is not only a collection of (fixed or changed batch volume) or one
One arrival, and generally have ageing, that is to say, that data are only effective within a period of time, therefore, in online sequence
Practise in the learning process of algorithm, the out-of-date training data for failing after several unit interval should be abandoned, here it is losing
Forget the thought of effect.Only the ageing of online training data can not be reflected with OS-ELM, in this section, we add the machine of forgetting
System, gradual exclusion may cause the outdated data of error message.In photovoltaic prognoses system, due to illumination with temperature with season
Change can change, and training data is only effective within a season, and therefore have Forgetting Mechanism passes through order limit study online
Machine (Online sequential Extreme Learning machine with forgotten mechanism, FOS-
ELM) compared with ELM, it is contemplated that data ageing, be more applicable for photovoltaic prognoses system.FOS-ELM algorithm is as follows.
The first step, initialization.
Step 1, using a small quantities of training dataAs primary data
Randomly generate aj、bjJ=1,2..., L
A) initial hidden layer output matrix H is calculated0
B) estimate initial output weight as shown in formula.
β0=P0H0 TY0\*MERGEFORMAT(6)
Wherein
C) k=0 is set;
Second step, with the on-line study for forgeing effect
Assume that+1 batch data of kth arrives
A) local hidden layer output matrix H is calculatedk+1
B) with calculating output weight beta according to formula(k+1)、Pk+1
(3) photovoltaic forecast model
Unit area photovoltaic array output is as shown in formula.
Ps=η SI [1-0.005 (t0-25)] \*MERGEFORMAT(11)
Wherein, η is photovoltaic array conversion efficiency;S is array area;I is intensity of illumination;t0For atmospheric temperature.
By formula as can be seen that the power output of photovoltaic array and photovoltaic array conversion efficiency, area, intensity of illumination, air
Temperature is relevant.
For given photovoltaic array, its conversion efficiency and area are all fixing, and its numerical value all lies in and go through
Among history data, and Intensity of the sunlight, it is periodically variable with the time, therefore we select, moment, temperature, illumination
Intensity, the historical power in the first two moment are used as the input quantity of forecast model, and the input vector for obtaining is as shown in formula.
xi=[time tem I]T\*MERGEFORMAT(12)
Wherein, time is moment value, such as 06:00, then variable time should be 0600;As 06:15, then variable time should be
0625.Tem is atmospheric temperature, degrees Celsius.
As, in the input vector shown in formula, the dimension of each data is not fully identical, it is therefore desirable to be normalized
Process, method is as shown in formula.
Wherein, xiFor input or output data, xmax、xminThe respectively maximum of data variation scope and minima.
Whole prediction algorithm is carried out according to the flow process shown in Fig. 2.
First, initial historical data is input into, and the mode that is write according to formula is initialized, and generates initial H0Square
Battle array, calculates initial output weight beta0And P0.Data of weather forecast of the input comprising intensity of illumination and the subsequent time of temperature, meter
Output result is calculated, that is, is predicted the outcome.Wait comprising meteorologic survey data and power measurements.After data are reached, according to prediction
The computational methods of error assessment index calculate forecast error.And store historical data.Judge whether the time expires a hour, if not
A full hour, then continue reading data of weather forecast and be predicted;If a full hour, by a hour of storage
Historical data carries out pretreatment, generates new H-matrix, and updates output weight beta and P, and continues to read data of weather forecast and
Row prediction.
For prediction accuracy is weighed, following error assessment index is introduced.
Standardized mean square error (normalized Root Mean Square Error, nRMSE) is as shown in formula.
Average absolute percent error (Mean Absolute Percent Error, MAPE) is as shown in formula.
Wherein, n is photovoltaic power station power generation period number of samples, PratedFor rated power, PpiFor the pre- power scale of i period,
PmiActual power for the i period.
Using Oregon, America university, photovoltaic monitoring experiment is published on its website the 5kW photovoltaic power positioned at Ashland
Data and weather data, carry out the foundation based on FOS-ELM photovoltaic generating system short term power forecast model and validation verification.
Model 1 (FOS-ELM model):From sigmond function as activation primitive, with 24 hours before the predicted moment
6 are only taken (:00-18:00 data) every the data of 15 minutes as training data, predict that the photovoltaic in the moment is exerted oneself, each
Individual hour updates a training data, and the data before eliminating 24 hours, that is, carry out once having the on-line study for forgeing effect
Calculating.
Model 2 (OS-ELM model):From sigmoid function as activation primitive, with 24 hours before the predicted moment (only
Take 6:00-18:00 data) every the data of 15 minutes as training data, predict that the photovoltaic in the moment is exerted oneself, each is little
Training data of Shi Gengxin, that is, carry out the calculating of an on-line study.
Model 3 (ELM model):From sigmoid function as activation primitive, to be located predicted day, month is upper one month
(only take 6 within last 24 hours:00-18:00 data) every the data of 15 minutes as training data, predict the moon each when
The photovoltaic at quarter is exerted oneself.That is ELM of re -training every other month.
Regularization parameter in model takes 1000, and hidden layer nodes take 200.
Predicting the outcome for a certain three kinds of models of fine day of in January, 2015 is taken, pre- power scale is as shown in Figure 3 with actual power.
Wherein horizontal axis representing time, 600 represent 6:00.
From the figure 3, it may be seen that during fine day, precision of prediction is higher, error is less, and prediction effect is good.Calculate the in a few days model 1
It is 9.707 that prediction nRMSE is 0.023, MAPE;It is 10.893 that the prediction nRMSE of model 2 is 0.035, MAPE;The prediction of model 3
It is 12.706 that nRMSE is 0.054, MAPE.
Predicting the outcome for a certain overcast and rainy three kinds of models of in January, 2015 is taken, pre- power scale is as shown in Figure 4 with actual power.
As shown in Figure 4, overcast and rainy, Cloud amount is larger, brings more uncertainties, but forecast model still is able to
Accurately predicted.Calculate this in a few days model 1 prediction nRMSE be 0.067, MAPE be 13.833;Model 2 pre-
It is 14.303 that survey nRMSE is 0.074, MAPE;It is 15.112 that the prediction RMSE of model 3 is 0.082, MAPE.
The further precision of parser, below takes in April, 2015 (spring), July (summer), October (autumn), January (winter), and four
The data of individual month carry out three kinds of test of heuristics, and it is as shown in table 1 to compare nRMSE, RMSE, MAPE comparative result.
1 three kinds of model prediction accuracy of table compare
As shown in Table 1, from the point of view of index nRMSE, the prediction accuracy in summer in winter is higher than spring and autumn, be due to spring and autumn Changes in weather
More violent, and the weather conditions in summer in winter are relatively stable;The accuracy of model 1 is higher than model 3 higher than model 2.From index MAPE
From the point of view of due to the generated output in winter lower than summer, the accuracy of summer is higher than winter;The accuracy of model 1 is high higher than model 2
In model 3.All in all, the precision of prediction of FOS-ELM is higher than OS-ELM model, and the precision of prediction of OS-ELM is higher than ELM.
The run time of three kinds of each steps of model is calculated in MATLAB, obtains result:Model 1, required for initialization
Time be about 0.095s, the time of each on-line study is about 0.052s;Model 2 initializes the time and 1 phase of model for needing
With the time of each on-line study is about 0.049s;Model 3 trains once the required time to be about 0.076s.FOS-ELM exists
Line study saves for about 30% time than each re -training, and OS-ELM each training time is saved about than FOS-ELM
20% time.
The present invention proposes the short-term photovoltaic power prediction algorithm based on FOS-ELM, and enters with classical ELM and OS-ELM
Go and compared.Theory analysis Simulation Example is proved, either fine day, cloudy day or the monthly data of Various Seasonal, just add
The FOS-ELM for then changing parameter has training speed soon, and generalization ability is strong, the advantage of high precision.
Claims (2)
1. a kind of short-term photovoltaic power generation power prediction method based on online online-sequential extreme learning machine, adopts with Forgetting Mechanism
Online online-sequential extreme learning machine, it is characterised in that from moment, temperature, intensity of illumination, the input quantity as forecast model.Step
Rapid as follows:
Step 1:According to the historical data being input into, the initial hidden layer output matrix of extreme learning machine is generated, and is calculated initial
Output weight;
Step 2:Prediction photovoltaic generation power, waits meteorologic survey data and power measurements;
Step 3:After the data for being waited are reached, calculation error evaluation index, obtain forecast error, store historical data, life
The hidden layer output matrix H-matrix of Cheng Xin, and output weight beta and photovoltaic generation prediction power P is updated, continue to read weather forecast
Data;
Step 4:Return to step 2.
2. short-term photovoltaic power generation power prediction method according to claim 1, it is characterised in that error assessment index is such as
Under:
(1) standardized mean square error nRMSE is
(2) average absolute percent error MAPE is
Wherein, n is photovoltaic power station power generation period number of samples, PratedFor rated power, PpiFor the pre- power scale of i period, PmiFor
The actual power of i period.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610888121.3A CN106446440A (en) | 2016-10-11 | 2016-10-11 | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610888121.3A CN106446440A (en) | 2016-10-11 | 2016-10-11 | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106446440A true CN106446440A (en) | 2017-02-22 |
Family
ID=58174002
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610888121.3A Pending CN106446440A (en) | 2016-10-11 | 2016-10-11 | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106446440A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108376261A (en) * | 2018-02-06 | 2018-08-07 | 南京信息工程大学 | One kind being based on density and online semi-supervised learning tobacco sorting technique |
CN108764548A (en) * | 2018-05-18 | 2018-11-06 | 杭州电子科技大学 | The online short term prediction method of photovoltaic generation dynamically associated based on sky brightness information |
CN109034464A (en) * | 2018-07-11 | 2018-12-18 | 南京联迪信息系统股份有限公司 | A kind of method that short-term photovoltaic generating system power prediction and result are checked |
CN109284882A (en) * | 2017-07-21 | 2019-01-29 | 中国电力科学研究院 | A kind of method and system that photovoltaic module performance determines |
CN109376951A (en) * | 2018-11-21 | 2019-02-22 | 华中科技大学 | A kind of photovoltaic probability forecasting method |
CN111025041A (en) * | 2019-11-07 | 2020-04-17 | 深圳供电局有限公司 | Electric vehicle charging pile monitoring method and system, computer equipment and medium |
CN111242371A (en) * | 2020-01-10 | 2020-06-05 | 华北电力大学 | Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942461A (en) * | 2014-05-14 | 2014-07-23 | 浙江师范大学 | Water quality parameter prediction method based on online sequential extreme learning machine |
CN107591844A (en) * | 2017-09-22 | 2018-01-16 | 东南大学 | Consider the probabilistic active distribution network robust reconstructing method of node injecting power |
-
2016
- 2016-10-11 CN CN201610888121.3A patent/CN106446440A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942461A (en) * | 2014-05-14 | 2014-07-23 | 浙江师范大学 | Water quality parameter prediction method based on online sequential extreme learning machine |
CN107591844A (en) * | 2017-09-22 | 2018-01-16 | 东南大学 | Consider the probabilistic active distribution network robust reconstructing method of node injecting power |
Non-Patent Citations (4)
Title |
---|
FARANAK GOLESTANEH: "Very Short-term Nonparametric Probabilistic", 《《IEEE TRANSACTIONS ON POWER SYSTEMS》》 * |
JIANWEI ZHAO: "Online sequential extreme learning machine with forgetting mechanism", 《NEUROCOMPUTING》 * |
刘念: "基于核函数极限学习机的分布式光伏短期功率预测", 《农业工程学报》 * |
李多: "基于EMD与ELM的光伏电站短期功率预测", 《可再生能源》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109284882A (en) * | 2017-07-21 | 2019-01-29 | 中国电力科学研究院 | A kind of method and system that photovoltaic module performance determines |
CN108376261A (en) * | 2018-02-06 | 2018-08-07 | 南京信息工程大学 | One kind being based on density and online semi-supervised learning tobacco sorting technique |
CN108376261B (en) * | 2018-02-06 | 2022-03-15 | 南京信息工程大学 | Tobacco classification method based on density and online semi-supervised learning |
CN108764548A (en) * | 2018-05-18 | 2018-11-06 | 杭州电子科技大学 | The online short term prediction method of photovoltaic generation dynamically associated based on sky brightness information |
CN108764548B (en) * | 2018-05-18 | 2021-06-29 | 杭州电子科技大学 | Photovoltaic power generation online short-term prediction method based on sky brightness information dynamic correlation |
CN109034464A (en) * | 2018-07-11 | 2018-12-18 | 南京联迪信息系统股份有限公司 | A kind of method that short-term photovoltaic generating system power prediction and result are checked |
CN109376951A (en) * | 2018-11-21 | 2019-02-22 | 华中科技大学 | A kind of photovoltaic probability forecasting method |
CN111025041A (en) * | 2019-11-07 | 2020-04-17 | 深圳供电局有限公司 | Electric vehicle charging pile monitoring method and system, computer equipment and medium |
CN111242371A (en) * | 2020-01-10 | 2020-06-05 | 华北电力大学 | Photovoltaic power generation short-term prediction correction method based on non-iterative multi-model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106446440A (en) | Short-term photovoltaic generation power prediction method based on online sequential extreme learning machine | |
Liu et al. | Random forest solar power forecast based on classification optimization | |
CN102999786B (en) | Photovoltaic generation power short-term earthquake prediction method based on similar day tagsort Yu extreme learning machine | |
CN111008728B (en) | Prediction method for short-term output of distributed photovoltaic power generation system | |
CN103390116B (en) | Use the photovoltaic power station power generation power forecasting method of stepping mode | |
CN103996082B (en) | A kind of intensity of solar radiation Forecasting Methodology theoretical based on dual random | |
CN109858673A (en) | A kind of photovoltaic generating system power forecasting method | |
CN111260126B (en) | Short-term photovoltaic power generation prediction method considering correlation degree of weather and meteorological factors | |
CN104978611A (en) | Neural network photovoltaic power generation output prediction method based on grey correlation analysis | |
CN103218674A (en) | Method for predicating output power of photovoltaic power generation system based on BP (Back Propagation) neural network model | |
CN103049798A (en) | Short-period electric generation power forecasting method applied to photovoltaic electric generation system | |
CN113537582B (en) | Photovoltaic power ultra-short-term prediction method based on short-wave radiation correction | |
CN114021848A (en) | Generating capacity demand prediction method based on LSTM deep learning | |
CN113516271A (en) | Wind power cluster power day-ahead prediction method based on space-time neural network | |
Luo et al. | Short-term photovoltaic generation forecasting based on similar day selection and extreme learning machine | |
CN116451821A (en) | Deep learning-based large-area new energy generation power prediction method and system | |
CN107256437A (en) | A kind of photovoltaic plant ultra-short term irradiation level Forecasting Methodology and system | |
CN108710966B (en) | Photovoltaic power generation power prediction method based on multi-cluster ESN neural network | |
Dong et al. | Day-ahead prediction of wind power based on conditional generative adversarial network | |
Bamisile et al. | Long-Term Prediction of Solar Radiation Using XGboost, LSTM, and Machine Learning Algorithms | |
Chen et al. | Probabilistic Prediction of Photovoltaic Power Using Bayesian Neural Network-LSTM Model | |
CN113034018B (en) | Power grid supply and demand balance adjustment method based on power moment analysis | |
CN116029419A (en) | Deep learning-based long-term new energy daily average generation power prediction method and system | |
Wang et al. | A short-term forecasting method for photovoltaic power based on ensemble adaptive boosting random forests | |
CN112581311B (en) | Method and system for predicting long-term output fluctuation characteristics of aggregated multiple wind power plants |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination |