Research On Short-term Prediction Of Photovoltaic Power Generation Based On Deep Learning Combined Model

Photovoltaic(PV)power generation has been a strategic emerging industry for China to respond to the energy crisis and the new economic development situation,and its share in the total energy structure has risen sharply.Due to the influence of weather and other factors,the characteristics of high volatility and intermittency of PV power generation,however,have had an impact on ensuring the stability,safety,and economy of power system operation.Highprecision PV power generation forecasting is essential to reduce the impact on power dispatching and help dispatching department formulate efficient power generation plans.For the short-term prediction of PV power generation,this paper relying on the powerful feature learning capabilities of deep learning,conducts in-depth research from the perspectives of full mining of meteorological factors,construction of combined prediction models,and the realization of different prediction forms of point prediction and interval prediction.The specific contents are as follows:(1)Aiming at the problem of low prediction accuracy due to incomplete extraction of meteorological factors and inaccuracy,and in order to effectively utilize the advantages of deep learning technology in nonlinear fitting,a PSO-LSTM combination algorithm based on full mining of meteorological factors is proposed,achieving high-precision prediction of PV output.The K-Nearest Neighbor(KNN)is introduced to fully explore the key factors that affect PV output from many meteorological factors,and the Particle Swarm Optimization(PSO)is used to optimize the hyperparameters of the Long and Short-Term Memory(LSTM)automatically.The impact of meteorological factors on PV output has been fully explored,and the accuracy of prediction has been improved.The experimental results show that this method selects LSTM as the core algorithm,makes full use of meteorological factors under the cooperation of KNN algorithm,and considers the timing of data.It has better prediction performance,smaller errors and lays a foundation for further research.(2)Aiming at the short-term point prediction research of PV power,on the basis of analyzing the internal variation law of PV output power,a DBN-XGBoost combination model based on PV sequence decomposition and reconstruction is proposed.Firstly,the original PV power generation data is decomposed into a series of modal functions using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN).Their characteristics are more obvious in the time domain,which can play a role in extracting the characteristics of the time series of PV power generation on the time scale.Then,the Deep Belief Network(DBN)and Extreme Gradient Boosting(XGBoost)are used to predict the reconstructed modal components.The experimental results obtained are compared with a single DBN,XGBoost,and BP algorithm,which proves the effectiveness of the proposed algorithm.(3)Aiming at the short-term interval prediction of PV power,considering the linear and nonlinear characteristics of PV power sequence,a short-term PV power interval prediction based on ARIMA-Seq2 Seq is proposed.The basic idea of the algorithm is to use the Auto Regressive Integrated Moving Average(ARIMA)to model and predict the linear components of the original PV power generation data based on the weather types divided into cloud,sunny and rainy States,then introduce the main meteorological factors affecting PV power generation,and use the Sequence to Sequence model(Seq2Seq)to get the prediction value of ARIMA residual sequence,and the ARIMA prediction results can be revised in this way.Finally,the proportional coefficient method based on particle swarm optimization is used to determine the upper and lower limits of the interval prediction for the point prediction results,and the final prediction interval is obtained.Through the comparative experiments of different weather conditions,it is proved that the proposed algorithm can efficiently mine the potential nonlinear relationship between PV data,and has the superiority in PV power generation interval prediction.