Research Of Short-term Photovoltaic Power Forecasting Based On Cloud Computing And Machine Learning

With the application of energy change in the whole world,the proportion of renewable energy in the energy resource structure is increasing rapidly.Photovoltaic power generation,as an efficient and clean energy,is becoming a new growth point in renewable energy based power generation.In recent years,PV industry market grows strong,countrys' new installed capacity increased rapidly.The photovoltaic power is influenced by solar irradiance,temperature,humidity and other meteorological factors.It has the characteristics of intermittent,volatility and periodicity.Large scale photovoltaic access will increase the peak-valley gap in power grid,resulting in difficulties in peak regulation,affecting the power quality and the safe and stable operation of power grid.Therefore,combined with the historical data and meteorological data to effectively predict the future output of the photovoltaic power can help planning power grid scheduling,managing the operation of the system,has very important significance for the safe and stable operation of power system.This has a very important significance for the safe and stable operation of power system.This paper choose short-term(one day)photovoltaic power generation forecast as the main research content.Due to the complexity of the factors affecting photovoltaic power generation,the Pearson correlation coefficient and Spearman rank correlation coefficient are used to analyze them,and design similar date clustering based on power.Then,for each cluster,we propose a prediction model of radial basis function neural network based on adaptive fireworks algorithm to improve the accuracy of photovoltaic power prediction.It takes the advantage of population cooperative search in adaptive fireworks algorithm to optimize the network parameters to achieve more accurate PV output prediction.At the same time,after a long period of operation,the photovoltaic power station has accumulated a large amount of historical data.With the operation of power plants,the amount of data will also grow fast.The calculation of these data consumes a lot of time under single computer.It affects the rapid scheduling of power grids.In this paper,a Spark cloud computing platform which bases on memory is built,and implement parallelization of the proposed algorithm.The algorithm is run on the Spark platform to improve the computational efficiency.Experiments are carried out on single machine and multi node Spark cloud platform respectively compared with the traditional single RBFNN and RBFNN optimized by particle swarm optimization algorithm,it shows that the proposed algorithm improves the prediction accuracy and reduces the computation time.