Research On Wind Power Combination Prediction Method Based On Physical Data Fusion

The development and deployment of clean energy,such as wind power,will contribute to the optimization of energy architecture and the reduction of the greenhouse impact.Wind power is unstable owing to the unpredictability of the wind;wind power prediction is a dependable method of ensuring effective wind power utilization.Wind power prediction is complicated by the diverse topography conditions of wind farms and the unreliability of numerical weather prediction data.The typical single forecast approach struggles to capture the variance trend of wind power data.Wind speed varies widely,and prediction accuracy is poor.The combined prediction approach can reduce the inaccuracy of combined prediction by developing a linear accumulation model of diverse forecasting outcomes and allowing each model to fully use its benefits.The single forecasting technique and the combination prediction approach are explored in this work,and the key contents are as follows:（1）Physical prediction methods based on NWP data are being researched.Firstly,the data preprocessing method is used to process the historical data and extract the measured power curve.Secondly,a physical prediction model is established based on the local effect of the wind farm.According to the results of numerical simulation of the wind farm flow field,the mesoscale NWP wind speed and wind direction data are downscaled to each location of the wind turbine.Combined with the wind power curve,the output power of the wind farm is predicted.The factors affecting the prediction accuracy are summarized through the prediction results.The physical prediction method does not need to consider the internal relationship of wind power time series,and is more suitable for new or incomplete wind farms.（2）Investigate statistical prediction methods based on historical data.The data preprocessing approach is performed first to process the historical data and extract the observed power curve.Second,a physical prediction model based on the wind farm’s local influence is developed.The mesoscale NWP wind speed and wind direction data are downscaled to each wind turbine site based on the findings of the wind farm flow field numerical simulation.The wind farm’s output power is projected when combined with the wind power curve.The prediction results describe the parameters influencing prediction accuracy.The physical prediction approach is better suited for new or unfinished wind farms since it does not need consideration of the internal connection of wind power time series.（3）Investigation of combined prediction methods based on physical and statistical data.The goal of the combination prediction method is to improve the physical and statistical prediction results.The fixed weights are predicted using the arithmetic mean method,the reciprocal variance method,and the simple weighted average method.The performance of a single prediction method is optimized using the Generalized Induced Weighted Average（GIOWA）variable weight combination prediction method.The weight of the variable weight combination prediction method changes dynamically at each time point on the time scale,and the prediction error is lower than that of the single prediction method and the fixed weight combination prediction method.It is critical to relieve the power grid’s peak regulating pressure and determine the reserve capacity of traditional power supply.