Research On Distributed Energy Output Forecast Based On High Permeability Distribution Network

With the development of the times,China has shown a sharp increase in the consumption of fossil fuels for power generation,which will not only lead to a reduction in the reserves of fossil fuels in China,but also cause greater pollution.Therefore,China is currently vigorously developing new energy sources for power generation.Wind energy and solar energy are naturally expected as a clean,pollution-free and abundantly stored energy source.Therefore,they are included in China's energy development plan.Wind energy and solar energy are natural energy sources,which are characterized by randomness,volatility and uncontrollability.These characteristics increase the difficulty of being used.Similarly,its output also has great volatility,which will have a great impact on the safe and stable operation of the power grid.Therefore,improving the forecasting accuracy of the output of the scenery can optimize the scheduling of the power system and reduce the impact of the output volatility on the power grid.At the same time,reducing the difficulty of using wind and light energy and improving the competitiveness of scenery in the electricity market.In order to improve the prediction accuracy of wind power and photovoltaic,this paper proposes a method for predicting the output of a single photovoltaic power station and a wind farm,and a joint output prediction method for wind farms and photovoltaic power stations at the same time or near distance.In this paper,under the premise that distributed energy generation accounts for 30%,different methods are used to screen the data when establishing a single prediction model and a joint prediction model.Firstly,this paper describes the research status of power forecasting of wind farms and photovoltaic power plants,analyzes the advantages and disadvantages of various current forecasting methods,and explains and summarizes them to some extent,and analyzes the characteristics of wind and light,and summarizes certain rules.And stated the basis for data selection.Then combined with the analysis,the single prediction model and the joint prediction model are established,and the error analysis and comparison are carried out to obtain the optimal model.In the single PV prediction model,there is not much data processing problem which the data collected is relatively good,there is not much data processing problem.In the prediction process,the influence of weather type on PV forecasting is considered.Therefore,the weather type is first classified,the data of each type for several days is selected as the training sample and the test sample,and the data sequence is decomposed by modal decomposition to reduce the fluctuation.Interference,the amount of decomposition is used as the input of the prediction model.The prediction model uses the GRNN recurrent neural network with strong nonlinear regression ability.Finally,the prediction model is optimized and error analysis is performed.When establishing a single wind farm model prediction,the pretreatment method of wind power output anomaly data,according to the characteristics of the collected data,the spline interpolation method is used to complete the missing data and the neighboring value method is used to eliminate the abnormal data and the tick 0 method is used to process the collected data.Since the data sequence of the wind farm is more chaotic,support is added in the prediction.The vector machine(SVM)data series is further optimized to reduce the final prediction error,and then the optimized data sequence is used as an input sequence for prediction.After predicting the output of a single wind farm,a plurality of wind farms in the region are divided according to a clustering algorithm,and representative wind farms are selected in the sub-regions after the partitioning,and the wind power output prediction of the subregions is performed according to the temporal and spatial characteristics of the wind farm,and finally Predict the output of wind farm clusters in the area.When establishing the joint prediction model,Granger causality test method,maximum correlation and minimum redundancy algorithm are used to screen the selected data sequences,and the correlation between the largest and different attribute features is selected.A small set of attributes is used as a set of input variables for the power prediction model.Then,each landscape is used to construct the landscape prediction model independently,and its establishment is explained by the different kernel functions in the support vector regression model.Secondly,the prediction result of the univariate model is used as the input of the combined model for predicting the prediction of wind power.The multivariate prediction model is finally established and optimized.In the process of establishing each forecasting model,a wind farm,a photovoltaic power station and industrial data of the same field in Inner Mongolia were selected as data support.