Photovoltaic Short-term Power Prediction Based On Dual Clustering And Deep Learning

As a typical representative of new energy generation,the installed capacity of PV power generation has been increasing year by year in recent years.However,due to the uncertainty of numerical weather forecasting,the PV output power has significant randomness,fluctuation and uncertainty,which makes it difficult to connect PV to the grid.Therefore,the improvement of PV power prediction accuracy is crucial for new energy consumption.In addition,since grid dispatchers need more reliable prediction information to make dispatching decisions,interval prediction of PV power is important to promote PV grid connection and improve the economic operation of the power system.Based on the data of a PV power plant in Jilin Province,this paper conducts the following studies from data analysis and processing,clustering means,short-term prediction and short-term interval prediction,respectively.First,the raw data provided by PV power plants are susceptible to abnormalities or missing data due to climate change,component aging,and recording errors,resulting in unsatisfactory prediction accuracy.Therefore,in this paper,basic processing such as abnormal data identification and missing data completion are carried out on the provided PV power data.The reasons for the fluctuation of PV power are also analyzed,and the correlation degree is obtained by calculating the correlation coefficient between meteorological factors and PV power,which provides the basis for feature selection for prediction.Secondly,this paper adopts double clustering to achieve fine clustering of the raw data.Compared with the conventional clustering means,the double clustering mode takes into account both the temporal characteristics of PV power and the numerical differences of weather information,and progressively realizes the dynamic characteristics of power curve clustering and numerical clustering of weather information.Secondly,considering the temporal uncertainty of PV power and the high-dimensional characteristics of numerical weather forecasting,the discrete wavelet transform and the deep self-encoder are used to process the raw data separately to avoid the limitation of single data processing means and improve the data utilization rate.Once again,the paper uses a deep learning model combined with double clustering means to achieve short-term deterministic prediction of PV power.Convolutional neural networks have powerful feature extraction ability and weight sharing mechanism,and the weather element data after dimensionality reduction is used as the input variable of the network,and the denoised PV power data is used as the target variable to establish the short-term prediction model,and the traditional machine learning model and other deep learning models are used as comparison models to verify the effectiveness of the proposed model through the error index of four seasons.Finally,the error of deterministic point prediction is analyzed theoretically,and the probability distribution statistics as well as curve fitting of the error results are performed by the nonparametric kernel density estimation method,and the upper and lower bounds of its interval are calculated under the specified 80%,85%,and 90% confidence levels,and the interval prediction results of PV power day are finally obtained,which avoids the limitations of other estimation methods that require prior assumption of parameters.The analysis of the simulation results shows that the interval prediction results obtained by this method can meet the actual engineering requirements and enhance the prediction application value.