Research On PV Power Forecasting Based On Improved Neural Network

With the rapid development of the global economy,the consumption of energy by mankind is also increasing.Traditional energy is becoming less and less and facing the danger of exhaustion.At the same time,the environmental pollution caused by the burning of traditional fossil energy has also attracted people’s attention.Actively explore various renewable resources to solve the shortage of fossil energy and environmental pollution.Compared with other new energy sources,solar energy has the advantages of wide distribution,high utilization rate,and no pollution.However,due to the randomness and intermittent nature of photovoltaic power generation,the safe and stable operation of large-scale photovoltaic power plants is facing severe challenges.Therefore,accurate prediction of photovoltaic power generation output power has important research value and practical significance.This thesis is dedicated to optimizing the neural network model and exploring ways to improve the accuracy of the photovoltaic power generation prediction model.The main research contents are as follows:(1)The Conv LSTM algorithm is used to predict photovoltaic power generation.At the same time,in order to further improve the prediction accuracy,a photovoltaic power generation prediction model based on EMD-Conv LSTM is proposed and established.First,the EMD algorithm is used to decompose the original power data into a number of smooth sub-sequences,And then use the Conv LSTM algorithm to predict each sub-sequence of the EMD decomposition,superimpose the prediction results to obtain the final prediction result,and finally verify the feasibility of the model through experiments.Simulation experiments were carried out on the LSTM and EMD-LSTM algorithm models.The results showed that the predicted value curve of the EMD-Conv LSTM neural network photovoltaic power generation power prediction model is in better agreement with the actual power curve,the error is smaller,and the prediction accuracy is higher.(2)Proposes and establishes a photovoltaic power generation prediction model based on the firework algorithm(FWA)optimized BP neural network.Taking into account the randomness of the initial weights and thresholds selected by the BP neural network algorithm,it is easy to fall into local minimums and the convergence speed is slow.Disadvantages,the firework algorithm is used to optimize the initial weights and thresholds of the BP neural network,and the FWA-BP neural network photovoltaic power generation prediction model is established.At the same time,a single BP neural network and a genetic algorithm optimized BP neural network(GA-BP)model were established,and the three models were simulated by examples,and the prediction curves of different models and the corresponding error index data were obtained.The comparative analysis of the data verifies the feasibility and accuracy of the proposed algorithm model.(3)The visual display of the power prediction model is realized,including data processing module,prediction module and GUI interface module.Users only need to click a button to complete data import,data processing and power prediction when using it,and understand the prediction intuitively and clearly through the GUI interface result.