A Research on High-precision Power Forecast Method for Photovoltaic Power Station
In recent years with the photovoltaic power station in the large-scale construction and combined to the grid,particularly of the mass concentration in northwest grid photovoltaic power station is now a growing proportion of the power grid load.And the northwest territories local absorptive ability is low, then it leads to a hard work of photovoltaic power stations load dispatching in several provinces and regions dispatching center, and to increase in the hot standby state of heat-engine plant, at the same time the photovoltaic power stations are also rationed the power supply, so it caused a lot of economic losses.The key to solve the above problem is to improve the short-term (24 hours) prediction accuracy of the photovoltaic power station, when the short-term prediction accuracy is up to90% or more, dispatching center can make an accurate and reasonable load distribution arrangement between the photovoltaic power station and the hot standby power plant, and then it is also to reduce the regulation frequency and the power grid wave.After years of development, the power prediction system for photovoltaic power station has formed a set of relatively complete and reasonable software and hardware system.But short-term power prediction accuracy is influenced by various factors such as the modeling principle and the numerical weather prediction precision, normally the prediction accuracy is only about 80%.So it needs a further study on the method of enhancing photovoltaic short-term prediction.From the aspects of forecasting system modeling principle this paper analyzes the neural network model,support vector machine (SVM) model and other commonly used forecasting system model method in selecting correlation parameters and sets up reasonable initial model parameters to improve the prediction precision.At the same time it analyzes the hybrid modeling and weighted average method to improve the precision of prediction system.And according to the characteristics of photovoltaic power station power prediction, this paper discusses the deviation correction method, according to the deviation between the prediction data and the station historical data corrects the new prediction results.Finally it uses the historical data to test the various improvement methods and to compare the prediction effect and accuracy.
Forecast Machine Neural Network Photovoltaic Predict PV Support Victor SVM
Kai Zhao Jingtao Han Shenqiang Wang Yiming Wang
Beijing Sifang Automation Co., Ltd., District Haidian,Beijing, China
国际会议
南京
英文
354-357
2015-09-21（万方平台首次上网日期，不代表论文的发表时间）