Photovoltaic Generation Power Prediction Based On Extreme Learning Machine And Long Short-term Memory Neural Network

Solar photovoltaic power generation technology has become a key research focus in the field of renewable energy technology because of its advantages of zero pollution,safety,reliability and noise-free.The artificial neural network has been widely employed to predict photovoltaic generation power.However,because of cumbersome parameter setting and unstable prediction performance,the photovoltaic generation power prediction models based on traditional neural network algorithms are not applicable for different prediction occasions.To tackle this issue,two photovoltaic generation power prediction models are designed based on extreme learning machine,long short-term neural network and assistant optimization algorithm in this dissertation.The specific contents are as follows.To achieve an accurate and stable photovoltaic generation power prediction,a day-ahead photovoltaic generation power prediction scheme is designed based on extreme learning machine.In the proposal,a similar day analysis method is designed to improve the quality of training samples and reduce the time consumption of the training process.To improve the performance of the prediction model,the weight values and the bias values of the extreme learning machine are optimized with the genetic algorithm.The forecast accuracy of the scheme in four seasons is analyzed with the real dataset of Alice Springs in Australia,and the results show that the predicted values are consistent with the true values.The accuracy and the stability of the prediction scheme are validated by comparative experiments with other prediction models.By considering the influence of noise data on the prediction results,a photovoltaic generation power prediction model is studied based on empirical mode decomposition,sine cosine algorithm and long short-term memory neural network.In this scheme,the original climate signal is processed with the de-noised method based on empirical mode decomposition,which avoids the influence of noise data on the prediction results to some extent.To further improve the accuracy and the stability of the prediction model,the parameters of long short-term memory neural network are optimized with the sine cosine algorithm.The influence on the prediction results is analyzed in terms of input climate variables,dataset division and the parameters setting of the sine cosine algorithm,and the prediction results obtained from two typical months are compared and tested.Compared with other prediction schemes,the proposed prediction model has higher accuracy in the day-ahead photovoltaic generation power prediction.