Photovoltaic Power Prediction Based on the Error Probability Distribution Compensation

Accurate prediction of PV output is of great significance to the operation and scheduling of power system. In order to improve the prediction accuracy of photovoltaic power generation, this paper proposes to first fit the distribution law of the preliminary prediction error of photovoltaic power based on the Gauss mixed model (GMM), and then compensate the predicted value of the error to the original point prediction result to achieve the photovoltaic output prediction. A PV point prediction model based on long short-term memory network (LSTM) is constructed in the first place. Due to environmental factors such as light intensity, temperature and visibility, the prediction results of PV power generation output will be greatly affected. By using K-means clustering algorithm (K-means), the historical PV output data is divided into multiple weather scenarios, and then the distribution probability of error between the predicted value and the actual value is modeled and analyzed in different weather scenarios. Through the model fitting results under the specified confidence level interval, the predicted value of PV power point is modified and the final prediction interval is obtained. Finally, the fitting results of Gaussian mixture distribution are compared with those of normal distribution, t distribution and logistic distribution model. The results show that the probability distribution model is more accurate and reliable than the traditional distribution model.