Research On PV Power Prediction Based On Neural Network

With the increasingly serious environmental problems such as energy depletion and rising global temperature,countries around the world have realized the importance of developing and utilizing renewable energy.As an inexhaustible clean energy,solar energy is valued and developed by more and more countries.One of the channels for utilizing solar energy is photovoltaic power generation,but photovoltaic power generation is prone to fluctuations due to changes in external factors such as solar irradiance and ambient temperature.The direct integration of fluctuating photovoltaics into the power grid will affect the safe operation of the power grid,resulting in a large number of "abandoned light" phenomena.High-precision prediction of fluctuating photovoltaic power generation is an important measure to improve photovoltaic consumption rate.In this thesis,LSTM neural network,K-means clustering and GWO optimization algorithm are used to improve the prediction accuracy and carry out research on photovoltaic power prediction.A K-means clustering and GWO-LSTM-QRRF prediction model is proposed.The main research contents of this thesis are as follows:(1)Considering the different characteristics of photovoltaic power generation under different weather types,it may be difficult for a single LSTM neural network model to achieve an ideal prediction effect.The historical data is clustered according to the weather type by the K-means clustering algorithm,and the LSTM neural network model is trained according to the clustering results.(2)The gray wolf optimization algorithm is used to automatically optimize the structural parameters of the neural network model and automatically adjust the structural parameters of the network.Compared with manually adjusting parameters,it saves time and improves the prediction accuracy of the model.(3)The quantile regression random forest method is used to construct the prediction interval,which realizes the prediction interval of photovoltaic power generation and solves the problem that the uncertainty of photovoltaic power generation cannot be quantified in certain point prediction.(4)A simulation experiment is carried out on a data set published by a photovoltaic power station in Australia,which verifies the feasibility of the K-means clustering and GWO-LSTM-QRRF method proposed in this thesis.This method can obtain high-precision photovoltaic power generation deterministic point prediction and a prediction interval that meets the requirements.Compared with GRU,SVM,and BPNN models,this method is the best in all indicators,which verifies the effectiveness of this method and provides a guarantee for the safe operation of photovoltaic grid-connected.