Very Short-term Solar Irradiance Forecast Based On Sky Images

Photovoltaic power generation technology has the advantages of clean,safety,giant reserves and so on.It plays an important role in reducing environmental pollution and alleviating energy crisis.It has become an important direction for the development of renewable energy in the world.Ground horizon irradiance(GHI)is an important factor in power of photovoltaic power generation.Its randomness and periodicity are the main reasons for the fluctuation and instability of the photovoltaic power.The grid connection of large scale photovoltaic power stations will bring challenges to manage and dispatch power system as well operate power grid safely.Therefore,the prediction of ground horizon irradiance is of great significance to the prediction of photovoltaic power,as well the planning and operational control of photovoltaic power generation.Nowadays,the research method of very short-term ground horizon irradiance forecast is mainly based on historical data and weather forecast data modeling.However,it does not take the important influence factor of cloud cover into account.This dissertation used the all sky camera to shoot the sky images,got cloud data fully over the photovoltaic experimental platform by a wide angle fisheye lens,carried out a series of image processing on the sky images,and then predicted the ground horizon irradiance after five minutes according to the historical irradiance data.The specific experimental steps are as follows,Firstly,the pre-treatment work of distortion correction and obstacle removing was carried out on the collected ground nephogram.Secondly,the SIFT(Scale-invariant feature transform)algorithm optimized by RANSAC(RANdom SAmple Consensus)was used to match the continuous sky images in order to predict the velocity of cloud movement.Next,the dissertation used normalized blue red ratio threshold method and clear sky library method to detect the cloud cluster in different weather conditions,and got current cloud information about amount and distribution.Then,according to the predicted speed and distribution of cloud cluster,this dissertation predicted the cover degree of clouds to the sun in the future time,and analyzed the influence factor of ground horizon irradiance as the input of prediction model.Finally,the BP neural network optimized by genetic algorithm and support vector regression method were used to establish the ground horizon irradiance prediction model,and the prediction performances of the two methods were verified.The experimental results showed that the BP neural network prediction model optimized by genetic algorithm was of high precision and could be used for very short-term prediction of ground horizon irradiance.