Research On Short-term Wind And Photovoltaic Power Prediction Based On Combination Deep Learning Model

At present,the proportion of wind and photovoltaic power generation to the total electricity consumption in society is gradually increasing,and the uncertainty and volatility of wind and photovoltaic power generation poses challenges for peak shaving and scheduling of the power system.The accurate prediction of short-term wind and photovoltaic power plays an important role in the power grid’s safe and stable operation and economic dispatch.In response to the current difficulties in short-term power generation prediction of wind and photovoltaic power,this article conducts research on short-term wind and photovoltaic power prediction based on three aspects: wind and photovoltaic anomaly data processing,wind,and photovoltaic time series decomposition,and combined deep learning models.The specific content includes:In response to the problems of missing and abnormal data during the operation of wind power and photovoltaic power systems,research was conducted on data feature analysis and preprocessing techniques.Proposed anomaly data detection methods based on Z-score,missing data filling methods based on cubic spline interpolation.A correlation analysis was conducted on environmental factors using the Pearson correlation coefficient.Through experimental analysis,the data quality has been improved,and the generalization and interpretability of the prediction model have been improved.The signal decomposition technique can be used to decompose the original wind power and photovoltaic sequences into a number of subsequences with different frequencies,which can effectively alleviate the impact of sequence volatility on the prediction results.A complete ensemble empirical mode decomposition method based on adaptive noise is proposed,which can decompose the time series into multiple curve-smoothed subcomponents.To reduce the number of subsequences and improve the speed and accuracy of model prediction,the sample entropy is used to calculate the subsequence energy,and the reconstructed sequence is divided into low frequency,intermediate frequency,and high frequency sequences based on the energy magnitude and the reconstructed sequence is used as the input of the prediction model.In establishing a short-term photovoltaic power prediction model,the paper adopts a dual-input prediction method of irradiation intensity and historical photovoltaic power.A photovoltaic power prediction model based on Sparrow Search Algorithm(SSA)and Long Short-Term Memory(LSTM)network model is proposed.This model utilizes the SSA algorithm to optimize the number of hidden layer neurons,learning rate,and maximum training frequency of the LSTM model,thereby improving the calculation speed and prediction accuracy of the LSTM.The prediction accuracy of the combined model of SSA-LSTM and LSTM,combined with data processing and signal decomposition techniques under sunny,cloudy,and rainy weather conditions,was analyzed.The results show that the prediction accuracy of the models proposed in this article is higher than that of other models.In establishing a short-term wind power prediction model,this article adopts a single input prediction method for wind power.A short-term wind power prediction model based on Convolutional Neural Network(CNN)and Bidirectional Long Short-Term Memory(Bi LSTM)is proposed.This model uses Bi LSTM to establish connections between historical data in a single wind power sequence,extracts input features through the CNN algorithm,improves CNN structure,and further extracts data change features to improve model prediction accuracy.Comparing the prediction accuracy of models such as CNN-Bi LSTM,CNN-LSTM,and Bi LSTM,the results show that the prediction accuracy of the proposed model is higher than that of other models.