Research On Forecast Of Photovoltaic Power Generation Based On Elman Neural Network

Photovoltaic power generation has been recognized by the market due to its environmental protection and renewable advantages,and has been vigorously promoted by the country,so it has achieved rapid development.In practice,the output power of the photovoltaic power generation system is closely related to the weather conditions and equipment operating conditions at the time,and fluctuates with time and weather conditions.With the integration of large-scale photovoltaic power generation into the grid,the safe operation of the power system has been greatly challenged.Therefore,achieving accurate prediction of photovoltaic power generation is of great significance for optimizing the start-up and shutdown plan of the unit and ensuring the safe and stable operation of the power grid.A hybrid model of photovoltaic power generation forecast based on Elman neural network、Rough set feature selection and Fuzzy C-means clustering is proposesed in this paper.Firstly,to address the problem of abnormal values in the original data series,an outlier processing method that combines hampel filtering to remove extreme outliers and threshold comparison method to complete the missing measurement data is proposed in this paper,first using hampel filtering to process the extreme outliers for samples with power data,then using threshold comparison method to complete the missing index data,thus obtaining a sample data set with complete,accurate and reasonable data,thereby reducing the influence of outliers on model parameter estimation.Secondly,to address the problem of model input parameter selection,by analyzing the characteristics of power changes in different time periods,a temporal index,which is proposed to replace the weather type index to avoid the wrong indexes caused by unclear weather type division,thereby eliminating the impact of inaccurate weather class division on prediction accuracy;At the same time,for the problem of many interferences in the sequence of influence factors in the original data,the key influence indicators of the data set,which are extracted by using the rough set attribute simplification theory to eliminate the redundant indicators and data in the original data,thereby reducing the dimensionality of the model input.Finally,dividing the data into similar time periods by using Fuzzy C-means clustering,so that the data with similar power generation characteristics are divided into one category,then Elman neural network is used for classification modeling to achieve short-term prediction of PV power generation.The simulation results show that,compared with the traditional weather-based prediction model,the prediction method proposed in this paper has significantly higher accuracy than the weather-based prediction model in clear weather conditions and slightly better accuracy than the traditional weather-based prediction model in non-clear weather conditions.Overall,the proposed model has the advantages of high prediction accuracy and good stability.