Forecasting And Optimization Method For Generation Power Of Hydro-wind-solar Energy System In Microgrids Based On Deep Learning

In order to alleviate the energy crisis and respond to climate change,the large-scale application of renewable energy has become a trend.Distributed power supplies consume electricity locally,meet local power supply needs,effectively save power generation costs,and realize power transmission by connecting to the large power grid as a supplement and support for the large power grid.However,the shortcomings of distributed power sources,such as intermittentness and uncertainty,gradually manifest with the increase of the penetration rate of distributed power sources in the power system,which affects the planning and adjustment of the power system and easily causes the stability of the power system to decrease.Accurate generation power prediction can not only ensure the smooth connection of distributed power sources,but also reduce the impact on the power system when the distributed power sources are connected to the grid,and improve the power quality and reliability of the power system.This paper focuses on the prediction of hydro-wind-solar power generation in a microgrid environment.The main work is summarized as follows:The power generation prediction problem of small hydro power units is studied.Aiming at the problems of small hydro-power stations that are greatly affected by environmental factors,short station construction time,and limited recorded data,a method for power generation prediction of small hydro-power stations based on an improved particle swarm optimization algorithm for BP neural network is proposed.According to the actual situation,select the appropriate influencing factors,analyze the optimization capability of the particle swarm optimization algorithm,and use the improved particle swarm algorithm to optimize the BP neural network to build a power generation prediction model for the small hydro-power station.The accuracy of the proposed prediction model is also shown.(2)The wind speed and wind turbine power generation prediction problem is studied.Aiming at the intermittentness and uncertainty of wind speed,a combined wind power prediction method based on the combination of variational mode decomposition,sample entropy reconstruction and deep belief network is proposed.First,the variational modal decomposition is used to decompose the wind speed sequence,and the optimal number of decomposition modes is determined according to the modal aliasing degree of the modal components and the corresponding change in the center frequency value.Second,the reduction of modeling workload is based on The sample entropy of the modal component after decomposition is used to reconstruct the modal component.Finally,the vertical and horizontal crossover algorithm is used to optimize the deep belief network prediction modelfor the reconstructed modal component.The predicted values of the modal components are accumulated to obtain the final prediction result.The wind speed is obtained by the wind speed-wind power conversion formula.The simulation analysis using the measured data of the National Renewable Energy Laboratory Wind Energy Center validates the feasibility and effectiveness of the proposed combined forecasting method.(3)This paper studies the power generation prediction of photovoltaic power generation units.Aiming at the influential factors of photovoltaic power generation prediction,the complexity of model construction,and the low prediction accuracy,a method for predicting photovoltaic power generation based on improved grey correlation analysis and vertical and horizontal crossover algorithm to optimize deep belief networks is proposed.First,the external factors affecting photovoltaic output were analyzed,and the mutual information and minimum error methods were used to determine the photovoltaic output influencing factors.Secondly,in view of the shortcomings of traditional gray correlation analysis,a distance similarity and trend similarity composition based on improved gray correlation analysis was established.A similar day selection method based on comprehensive similarity;finally,based on the randomization of the initial weight of the deep belief network,a vertical and horizontal crossover algorithm was used to optimize the deep belief network to predict the photovoltaic power generation,and a case simulation analysis was performed using solar data in Australia to verify The accuracy and effectiveness of the proposed method in typical weather types.