Research On Short-term Photovoltaic Power Prediction Based On Feature Learnin

With the increasing demand for energy and the increasingly prominent environmental problems,the development and utilization of solar energy is gradually becoming a major trend of new energy development.Solar energy is a kind of renewable energy,compared with other new energy,solar energy has the advantages of wide distribution,high efficiency,no pollution and energy source can be used as power generation,photovoltaic power has strong randomness and volatility,large-scale grid connection brings great challenges to real-time dispatching and safe operation of power grid,Therefore,improving the prediction level of photovoltaic power generation is of great significance to the safety of photovoltaic energy grid connection.In this paper,the influence of meteorological characteristics such as short-wave radiation,temperature and cloud cover on the output power of photovoltaic power generation is analyzed firstly.K-means++ clustering is used for weather classification to improve the similarity of training data set,and the measured data of photovoltaic power station and numerical weather forecast data are normalized.It provides a good data base for building prediction model.Aiming at the problem that there are a lot of redundant and irrelevant features in multi-dimensional numerical weather forecast data,which will affect the accuracy of prediction and increase the complexity of the model,a direct prediction model of short-term photovoltaic power based on factor analysis was proposed.First,structure difference features more effectively,at the same time have the effect of filter,then the existing meteorological features are analyzed by factor analysis,The common factors far less than the number of features are taken as the input data of the prediction model,and the XGBoost algorithm is used to predict the photovoltaic power.The experiment shows that the factor analysis can extract effective features and help improve the prediction accuracy.Considering the strong randomness of meteorological conditions in cloudy and rainy weather,the prediction error fluctuates greatly,and it is difficult to fit the abnormal fluctuation value in direct prediction.Therefore,an indirect short-term photovoltaic power prediction model based on the standard clear sky set without climbing event definition is proposed.The standard clear sky set was defined with no climbing event,the difference between historical actual power and the standard clear sky set was taken as the target variable,the deviation characteristics were constructed,the long and short-term memory model was used to predict the deviation value,then the predicted deviation value and the standard clear sky set were superimposed,and then the actual power value was obtained indirectly.Experiments show that the model can indirectly improve the prediction accuracy of actual power by controlling the error fluctuation range.In order to provide more complete photovoltaic power output information to the power grid dispatching department,a probabilistic prediction model based on long and short-term memory quantile regression is established,which can accurately quantify the uncertainty of photovoltaic power and provide comprehensive prediction information for power system operation decision.The model is used to predict the photovoltaic output at different fractile,and the kernel density estimation method is used to estimate the probability,and then the confidence intervals under different confidence levels are obtained.The prediction results show that the prediction method can provide the fluctuation of photovoltaic power well,and has certain reference value for practical application.