Research On Short-term Photovoltaic Power Prediciton Based On Long Short Memory Neural Network

The penetration rate of photovoltaic power sources in the power grid is continuously increasing.Accurate short-term photovoltaic power generation forecasts are conducive to ensuring the safe and stable operation of the grid with a high ratio of photovoltaic power sources.Further improving the prediction accuracy of photovoltaic power is necessary in the process of power development.Photovoltaic output power will be affected by weather conditions and seasons,and will easily change with changes in external factors,showing volatility.Therefore,with the introduction of multiple influencing factors and the increasing amount of reference data,accurate and efficient short-term photovoltaic power generation forecasts are conducive to ensuring the safe and stable operation of the grid with a high rate of photovoltaic power access.This paper proposes and optimizes a long short-term memory neural network(LSTM)model suitable for photovoltaic power prediction,and explores ways to improve the efficiency of photovoltaic power prediction and ensure the accuracy of prediction.The main research contents are as follows:First of all,various factors affecting photovoltaic power generation are studied to establish a foundation for constructing an LSTM prediction model.This paper discusses the applicability of photovoltaic systems and deep learning neural networks to photovoltaic power prediction and the prediction process,and uses KMeans weather clustering to divide the historical data of existing photovoltaic power stations into three types: sunny,cloudy and rainy.Secondly,the historical data of photovoltaic power was preprocessed.Then the data is unified for abnormal data detection,correction and data standardization.Through gray correlation analysis,the meteorological characteristics that mainly affect photovoltaic power generation are selected,and finally a new data set is divided.Then,the structure and principle of Long Short Memory Neural Network(LSTM)are explained and summarized,and the principle of support vector regression(SVR)and recurrent neural network(RNN),two commonly used algorithms for photovoltaic forecasting,are introduced.The influencing factors selected through the correlation analysis were used as the input of the model,and the photovoltaic power generation prediction model based on the long and short memory neural network(LSTM)was established,and the training process of the neural network was optimized by a small batch of gradient descent algorithm.Using the power load data actually measured in the photovoltaic power station,the proposed LSTM algorithm model and the other two common algorithm models are used to complete the prediction simulation experiment.Through the comparison of prediction results,the optimized LSTM neural network can not only have higher training and learning efficiency;it can also predict the day-a-day photovoltaic power generation capacity of the power station more accurately under different weather types and different seasons.Finally,taking into account the further improvement of the prediction accuracy of short-term power generation,the long and short memory neural network(LSTM)prediction model optimized by the small batch of gradient algorithms is combined with the convolutional neural network(CNN)that can extract spatial feature information to establish Algorithm model of CNN-LSTM hybrid neural network.At the same time,the experiment is compared with the proposed long and short memory neural network(LSTM)model and a separate convolutional neural network(CNN)model to verify that the mixed CNN-LSTM neural network model has a better prediction effect.