Research On Power Prediction Method For Photovoltaic Power Plant Based On The Output Characteristic Under Typical Weather

China is rich in solar resources, and solar energy generation has technological and economic advantage in becoming the strategic energy resources in the long run. However, solar energy resource is a kind of random and uncontrolled power supply, and the grid connection of the large scale solar power station is a challenge to the safe and stable operation of power grid. The research on photovoltaic power forecasting technology is very important for the power department to adjust the scheduling plan, improve the reliability of power system and the access level of photovoltaic power plant.Understanding the output characteristics of photovoltaic power plant is the first step in the study of photovoltaic power prediction. The output of photovoltaic power plant is highly affected by solar radiation, temperature, humidity, wind speed and other meteorological factors.which shows clear trend and non-stationary characteristic. These features bring challenge to the convergence and generalization ability of ANN (Artificial Neural Networks) and make it hard to reach a good prediction performance.In this paper, the BP (Back Propagation) neural network theory is studied, and the experimental data and historical data of the photovoltaic power plant are set up to build a training set and a validation set, and two prediction models based on ANN are developed.According to the trend feature of photovoltaic power output, theoretical solar irradiance was introduced to de trend model target output. ANN model with the input of theoretical solar irradiance, forecast temperature, forecast humidity and forecast wind speed was established. Then, since the advantage of WD (Wavelet Decomposition) in dealing with the non-stationary signal, an optimized WD+ANN model is established and the input of the model was decomposed into approximation signal and detail signal by the wavelet decomposition. The results show that both ANN and WD+ANN models have good prediction performance, while WD+ANN model has higher prediction accuracy and faster convergence rate. The study of this paper is based on a large number of actual data, and has good generalization, which can provide reference for the researchers in the same field.