CN103617461A - Photovoltaic power station generated power predicting method - Google Patents
Photovoltaic power station generated power predicting method Download PDFInfo
- Publication number
- CN103617461A CN103617461A CN201310663379.XA CN201310663379A CN103617461A CN 103617461 A CN103617461 A CN 103617461A CN 201310663379 A CN201310663379 A CN 201310663379A CN 103617461 A CN103617461 A CN 103617461A
- Authority
- CN
- China
- Prior art keywords
- predicted
- day
- power
- photovoltaic
- data
- 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.)
- Withdrawn
Links
Images
Abstract
The invention relates to a photovoltaic power station generated power predicting method. According to the technical scheme, five historical days most similar to a day to be predicted are determined from data in the former years according to the situation of the weather forecast of the day to be predicted, eight sets of data, before the time point to be predicted, in the five similar days serve as a training sample, illumination intensity, temperature and the like which are predicted before the time point to be predicted serve as input data, and the predicted power of the time point to be predicted is worked out and obtained through a weighing support vector machine. By means of the similarity computing method and the similar day determining method, the short-term power prediction of a photovoltaic system is achieved through the weighing support vector machine, and the defect of the self-learning ability absence due to the fact that the prediction is carried out only through the similarity day method is overcome.
Description
Technical field
The present invention relates to the intelligent computation methods such as support vector machine, be specifically related to a kind of method of photovoltaic power station power generation power prediction.
Background technology
Photovoltaic power station power generation power often fluctuates larger, and next generated output and previous moment does not constantly have inevitable correlativity, and this will be unfavorable for that dispatching of power netwoks department arranges the cooperation of normal power supplies and photovoltaic generation.The access of large-scale photovoltaic power station, will produce important impact to the quality of power supply, power grid security.Photovoltaic plant output power is subject to the impact of intensity of illumination and temperature, and in physical environment, solar irradiation has great instability, make the generated energy of photovoltaic plant in certain hour section can have larger undulatory property [1-3], this can be unfavorable for that dispatching of power netwoks department arranges the cooperation of normal power supplies and photovoltaic generation very much.The power prediction based on intelligent computation model adopting as document [1-6], can be without knowing concrete environmental aspect, by artificial neural network, grey modeling, support vector machine etc., can predict the output power of photovoltaic plant, also can predict by environmental parameters such as illumination, temperature and wind speed the output power of photovoltaic plant.But neural network often needs relatively large training sample; with precision and the generalization ability obtaining; and for Small Sample Size; its estimated performance will reduce greatly; in addition, the structure and parameter of neural network is also difficult for definite, and existing training algorithm often can cause its parameter to be absorbed in local minimum; therefore, neural network has larger limitation in power prediction.Comparatively speaking, support vector machine but can, preferably for solving Small Sample Size, in the time of only need having a small amount of support vector, just can be determined the parameter of support vector machine, thereby obtain good estimated performance, so support vector machine is relatively suitable for the occasion of small sample.
Support vector machine (Support Vector Machine, be called for short SVM) be the new method [7] of the developed recently machine learning of getting up, consider empiric risk and put trade wind danger, preferably resolve the problems such as small sample, non-linear, high dimension, well having overcome conventional machines learning method crosses study and is easily absorbed in local minimum problem, there is very strong generalization ability, because it is a protruding double optimization algorithm, can guarantee that the minimax solution of trying to achieve by it is globally optimal solution [8 9] simultaneously.Weighted Support Vector is a kind of expansion of support vector machine, and it is mainly suitable for processing the data [10] that result had to Different Effects degree.Due in photovoltaic system power prediction, more higher to the significance level of future position close to the data of future position, therefore, the present invention adopts Weighted Support Vector to solve photovoltaic system short term power forecasting problem.
Summary of the invention
The uncertainty of the environmental factor such as intensity of illumination, environment temperature has determined that the output power of photovoltaic generating system and generated energy can not correctly be grasped, uncertainty for photovoltaic plant power in power generation process, select Weighted Support Vector as the prediction algorithm of photovoltaic system output power, from historical record, search the data of historical day higher with day similarity to be predicted Weighted Support Vector is trained, effectively predict that photovoltaic plant is at next power stage constantly.
In carrying out photovoltaic plant short term power forecasting process, if found out in historical data and day weather conditions to be predicted, weather pattern all more close one day, the generated output of this day is also more approaching with the generated output of to be predicted day so.Conventionally this close historical day, be called similar day.By finding, to carry out similar day power prediction be a lot of field staff conventional methods while carrying out power prediction.
, intensity of illumination similar with to be predicted day season by finding and environment temperature also more similar similar day, provided the concrete algorithm of choosing for similar day.Using similar day data as Weighted Support Vector training data, to estimate the output power of to be predicted day, the parameter that has provided Weighted Support Vector is selected, and has solved the shortcoming that there is no self-learning capability while purely utilizing similar day method prediction.
Calculate the object of similarity, mainly for based on objective method, from historical data, find and weather conditions in to be predicted day, especially intensity of illumination and the more similar time periods such as environment temperature, select these similar time period recorded data as learning sample, then based on these samples, build power prediction model.
In order to find out the historical day similar to be predicted day from historical data, need to calculate history day (m month n day) and the similarity in season of to be predicted day (i month j day), first we calculate the time gap of day j to be predicted and historical day n, is designated as
, have:
In formula:
be number of days corresponding to every month, the January is 31 days, is not or not February in leap year is 28 days, by that analogy.
According to formula (1), and consider that in 1 year, the similarity in Dec and January is higher, we provide to be predicted day as follows j and historical day n season similarity computing method:
Intensity of illumination similarity can be by comparing the weather forecast situation of to be predicted day and the weather forecast situation of historical day, and we provide suc as formula illumination similarity calculating method shown in (3):
In formula:
,
respectively the forecasting weather situation of i month j day, and the forecast situation of m month n day, by numeral, represent: 1 is fine day, and 2 is cloudy, 3 is the cloudy day, and 4 is light rain, and 5 is shower, and 6 is heavy rain, 7 is snow day.
Consideration is in somewhere, and the difference of the maximum temperature between any two days generally can not surpass 50 ℃, and therefore, for the similarity degree of maximum temperature, this chapter defines suc as formula similarity calculating method shown in (4):
Same reason, in somewhere, we have defined following lowest temperature similarity calculating method:
Defined each similarity of formula (2) to (5) is multiplied each other, we just can obtain the similarity of i month j day and m month n day, in this article the height of the data based similarity in a year is arranged, then find out to the highest 5 days of day similarity to be predicted as similar day, then select these similar day entry data as training sample.
Total similarity is:
From the data that the larger time period of similarity obtains, be designated as n, it should be more important to the effect of model, and therefore, training sample is corresponding
it is larger,
less; Otherwise, the data that similarity is less, it suitably reduces its effect in forecast model, and therefore, training sample is corresponding
it is less,
larger.Therefore, this chapter is by total similarity
determine the weights of each similar Japan-China training sample,
be shown below:
Can find out,
, and increase along with the increase of similarity.Further, by
, easily obtain
value.
Consider to utilize 8 groups of data (recorded one group of data every 15 minutes, therefore, record altogether 8 groups of data for two hours) of to be predicted day predicted time point the first two hour as the input of model, using the power of point to be predicted as output.Accordingly, to current the most similar 5 days of day of prediction, select corresponding predicted time point the first two hour data as the input of training sample, the power of the Japan-China corresponding predicted time point of similar history is as output.Model as shown in Figure 1, wherein
the time point i to be predicted photometric data of the first two hour constantly,
the temperature data of to be predicted some the first two hour,
it is the power stage of point to be predicted.
Beneficial effect of the present invention: utilize the computing method of similarity and definite method of similar day, adopt the output power to photovoltaic plant of Weighted Support Vector to predict.
From the historical datas such as the temperature that records for 1 year, intensity of illumination, 10 days have been selected at random as forecasting object.Using the output power of these days in a certain moment (such as noon 12:00) photovoltaic plant as value to be predicted, specifically arrange as shown in table 1.
Output power to these 10 these moment photovoltaic plants of to be measured day 12:00 is predicted, adopts 5 sample training WSVM of similar day, and its result as shown in Figure 3.From Fig. 3, can clearly find out, predict the outcome very approaching with measured result.
The result that error is calculated is as follows: maximum relative error is 12.23%; Minimum relative error is 0.16%; Mean value relative error is 5.61%.
Random to be predicted 10 days of selecting of table 1
Accompanying drawing explanation
Fig. 1 is model structure.
Fig. 2 is the method flow diagram of photovoltaic power station power generation power prediction of the present invention.
Fig. 3 is that the present invention predicts the outcome.
Embodiment
In program, read in 1 year or historical data for many years above, weather forecast situation according to be predicted day, from the data in former years, determine five with the highest historical day of to be predicted day similarity as similar day, using 8 groups of data before these five similar Japan and China time points to be predicted as training sample, using the intensity of illumination of having surveyed before time point to be predicted and temperature etc. as input data, by Weighted Support Vector, calculate the predicted power that obtains the predicted time point of wanting.
Above embodiment is only the preferred embodiment of this creation, not in order to limit this creation, any modification of making, is equal to replacement, improvement etc., within all should being included in the protection domain of this creation within all spirit in this creation and principle.
List of references
[1] Yona A., Senjyu T., Saber A.Y., Funabashi T., Sekine H., Chul-Hwan K.. Application of neural network to 24-hour-ahead generating power forecasting for PV system. Power and Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st Century, Pittsburgh, PA, 2008:1-6.
[2] Cai T., Duan S.X., Chen C.S.. Forecasting power output for grid-connected photovoltaic power system without using solar radiation measurement. The 2nd IEEE International Symposium on Power Electronics for Distributed Generation Systems, Hefei, China, 2010: 773-777.
[3] Wang S.X., Zhang N., Zhao Y.S., Zhan J.. Photovoltaic system power forecasting based on combined grey model and BP neural network[C]. International Conference on Electrical and Control Engineering (ICECE), Tianjin, China. 2011: 4623 – 4626.
[4] Sulaiman S.I., Abdul Rahman T.K., Musirin I., Shaari S., Optimizing three-layer neural network model for grid-connected photovoltaic system output prediction. Conference on Innovative Technologies in Intelligent Systems and Industrial Applications, Monash, 2009: 338-343.
[5] Ding M., Xu N.Z.. A method to forecast short-term output power of photovoltaic generation system based on Markov Chain[J]. Power System Technology, 2011, 35(1): 152-157.
[6] Zhu Yongqiang ，Tian army. the application of least square method supporting vector machine in photovoltaic power prediction. electric power network technique, 2011,35 (7): 54-59.
[7] Cortes C., Vapnik V.. Support-vector networks[J]. Machine Learning (S0885-6125), 1995, 20( 2): 273-297.
[8] Fortuna J., Capson D.. Improved support vector classification using PCA and ICA feature space modification[J]. Pattern Recognition(S0031-3203), 2004, 37(6):1117-1129.
[9] Chapelle O., Haffner P., Vapnik V.N.. Support vector machines for histogram-based image classification. IEEE Transactions on Neural Networks[J], 1999, 10(5):1055-1064.
[10] Adriano L. I. Oliveira, Carolina Baldisserotto, Julio Baldisserotto. A comparative study on support vector machine and constructive RBF neural network for prediction of success of dental implants[J]. Lecture Notes in Computer Science, 2005:1015-1026。
Claims (2)
1. a method for photovoltaic power station power generation power prediction, is characterized in that: adopt the output power to photovoltaic plant of Weighted Support Vector to predict.
2. the method for a kind of photovoltaic power station power generation power prediction as claimed in claim 1, is characterized in that: using similar day data as Weighted Support Vector training data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310663379.XA CN103617461A (en) | 2013-12-10 | 2013-12-10 | Photovoltaic power station generated power predicting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310663379.XA CN103617461A (en) | 2013-12-10 | 2013-12-10 | Photovoltaic power station generated power predicting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103617461A true CN103617461A (en) | 2014-03-05 |
Family
ID=50168165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310663379.XA Withdrawn CN103617461A (en) | 2013-12-10 | 2013-12-10 | Photovoltaic power station generated power predicting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103617461A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103942622A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Wind power short-term prediction method using composite data source based on self-learning Sigmoid kernel function support vector machine |
| CN103942618A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Photovoltaic power generation power short-term prediction method using composite data source based on self-learning polynomial kernel function support vector machine |
| CN103942619A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Photovoltaic power generation power short-term prediction method using composite data source based on self-learning Sigmoid kernel function support vector machine |
| CN104732296A (en) * | 2015-04-01 | 2015-06-24 | 贵州电力试验研究院 | Modeling method for distributed photovoltaic output power short-term prediction model |
| CN105260800A (en) * | 2015-10-26 | 2016-01-20 | 国网浙江省电力公司电力科学研究院 | Photovoltaic module temperature prediction method and device |
| CN105426989A (en) * | 2015-11-03 | 2016-03-23 | 河海大学 | EEMD and combined kernel RVM-based photovoltaic power short-term prediction method |
| CN105512775A (en) * | 2016-02-01 | 2016-04-20 | 北京交通大学 | Method for power prediction of photovoltaic power generation system |
| CN105678397A (en) * | 2015-11-18 | 2016-06-15 | 江苏省电力公司泰州市姜堰区供电公司 | Short-term photovoltaic power prediction method based on improved EMD algorithm and Elman algorithm |
| CN106529814A (en) * | 2016-11-21 | 2017-03-22 | 武汉大学 | Distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chain |
| CN106709159A (en) * | 2016-12-05 | 2017-05-24 | 华北电力大学 | Photovoltaic power generation scheduling rule considering reward and punishment system |
| CN107403015A (en) * | 2017-08-07 | 2017-11-28 | 南京金水尚阳信息技术有限公司 | A kind of short-term luminous power Forecasting Methodology based on Time Series Similarity |
| CN111754026A (en) * | 2020-05-28 | 2020-10-09 | 国网冀北电力有限公司 | Photovoltaic power station group power prediction method and device, computer equipment and storage medium |
-
2013
- 2013-12-10 CN CN201310663379.XA patent/CN103617461A/en not_active Withdrawn
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103942618A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Photovoltaic power generation power short-term prediction method using composite data source based on self-learning polynomial kernel function support vector machine |
| CN103942619A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Photovoltaic power generation power short-term prediction method using composite data source based on self-learning Sigmoid kernel function support vector machine |
| CN103942622A (en) * | 2014-04-18 | 2014-07-23 | 国家电网公司 | Wind power short-term prediction method using composite data source based on self-learning Sigmoid kernel function support vector machine |
| CN104732296A (en) * | 2015-04-01 | 2015-06-24 | 贵州电力试验研究院 | Modeling method for distributed photovoltaic output power short-term prediction model |
| CN105260800A (en) * | 2015-10-26 | 2016-01-20 | 国网浙江省电力公司电力科学研究院 | Photovoltaic module temperature prediction method and device |
| CN105426989A (en) * | 2015-11-03 | 2016-03-23 | 河海大学 | EEMD and combined kernel RVM-based photovoltaic power short-term prediction method |
| CN105678397A (en) * | 2015-11-18 | 2016-06-15 | 江苏省电力公司泰州市姜堰区供电公司 | Short-term photovoltaic power prediction method based on improved EMD algorithm and Elman algorithm |
| CN105512775A (en) * | 2016-02-01 | 2016-04-20 | 北京交通大学 | Method for power prediction of photovoltaic power generation system |
| CN106529814A (en) * | 2016-11-21 | 2017-03-22 | 武汉大学 | Distributed photovoltaic ultra-short-term forecasting method based on Adaboost clustering and Markov chain |
| CN106529814B (en) * | 2016-11-21 | 2020-01-07 | 武汉大学 | Distributed photovoltaic ultra-short term prediction method based on Adaboost clustering and Markov chain |
| CN106709159A (en) * | 2016-12-05 | 2017-05-24 | 华北电力大学 | Photovoltaic power generation scheduling rule considering reward and punishment system |
| CN106709159B (en) * | 2016-12-05 | 2021-03-09 | 华北电力大学 | Photovoltaic power generation dispatching method considering reward and punishment system |
| CN107403015A (en) * | 2017-08-07 | 2017-11-28 | 南京金水尚阳信息技术有限公司 | A kind of short-term luminous power Forecasting Methodology based on Time Series Similarity |
| CN111754026A (en) * | 2020-05-28 | 2020-10-09 | 国网冀北电力有限公司 | Photovoltaic power station group power prediction method and device, computer equipment and storage medium |
| CN111754026B (en) * | 2020-05-28 | 2024-03-15 | 国网冀北电力有限公司 | Photovoltaic power station group power prediction method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103617461A (en) | Photovoltaic power station generated power predicting method | |
| Wu et al. | A novel hybrid model for short-term forecasting in PV power generation | |
| Xiao et al. | Combined forecasting models for wind energy forecasting: A case study in China | |
| Hong et al. | Optimal sizing of hybrid wind/PV/diesel generation in a stand-alone power system using Markov-based genetic algorithm | |
| Nabavi et al. | Deep learning in energy modeling: Application in smart buildings with distributed energy generation | |
| CN106446494B (en) | Honourable power forecasting method based on wavelet packet-neural network | |
| CN102930358B (en) | A kind of neural net prediction method of photovoltaic power station power generation power | |
| CN104361406B (en) | A kind of photovoltaic plant can utilize solar power generation amount Forecasting Methodology | |
| Chandra et al. | A detailed literature review on wind forecasting | |
| CN103218674A (en) | Method for predicating output power of photovoltaic power generation system based on BP (Back Propagation) neural network model | |
| Xu et al. | Short-term photovoltaic power forecasting with weighted support vector machine | |
| CN109934395B (en) | Multi-hydropower-region short-term power load prediction method based on time-sharing and regional meteorological data | |
| Kim et al. | Designing solar power generation output forecasting methods using time series algorithms | |
| CN105184399B (en) | A kind of predicting power of photovoltaic plant method | |
| Mukhoty et al. | Sequence to sequence deep learning models for solar irradiation forecasting | |
| Michael et al. | Short-term solar irradiance forecasting based on a novel Bayesian optimized deep Long Short-Term Memory neural network | |
| CN104200274A (en) | Power prediction method for photovoltaic devices | |
| Ayoub et al. | ANN model for energy demand and supply forecasting in a hybrid energy supply system | |
| CN111525552A (en) | Three-stage short-term wind power plant group power prediction method based on characteristic information | |
| Munir et al. | Solar pv generation forecast model based on the most effective weather parameters | |
| Chen et al. | Research on wind power prediction method based on convolutional neural network and genetic algorithm | |
| Luo et al. | Short-term photovoltaic generation forecasting based on similar day selection and extreme learning machine | |
| Liu et al. | A weight-varying ensemble method for short-term forecasting PV power output | |
| Mellit | Sizing of photovoltaic systems: a review | |
| Zhang et al. | Joint forecasting of regional wind and solar power based on attention neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C04 | Withdrawal of patent application after publication (patent law 2001) | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20140305 |