Research On Wind Speed Probability Distribution Model Selection Method And Multi-scale Optimization Fittin

In recent years,with China’s clear goal of "carbon peaking" and "carbon neutrality",wind power has been developing rapidly,and the research and analysis of wind speed probability distribution is particularly important for the site selection,scheduling and better utilization of wind energy resources.Therefore,this paper focuses on the multi-scale optimal fitting of wind speed probability distribution and the selection of wind speed probability distribution fitting models for wind farms.The main method for fitting the wind speed probability distribution of a wind farm is to use a suitable distribution model to fit it.In this paper,we first study several wind speed probability distribution models that are widely used,and propose to use the T location-scale distribution model to fit the wind speed probability distribution.In order to verify the fitting effect of this model,the T location-scale distribution model is compared with four other traditional wind speed probability distribution models using five different goodness-of-fit tests,and the results show that the T location-scale distribution model has a better fitting effect.In order to further improve the fitting effect,this paper proposes to use the particle swarm intelligent optimization algorithm to optimize the parameters of the distribution model.At the same time,this paper proposes its own improvement measures to the traditional particle swarm algorithm,and the improved particle swarm algorithm obtains a faster and more stable finding speed compared with the traditional particle swarm algorithm.Based on the T location-scale distribution model,the improved particle swarm algorithm and the great likelihood method are used to estimate the model parameters and compare the goodness of fit respectively.The final results show that the T location-scale distribution model with parameter optimization using the improved particle swarm algorithm has a better fitting effect and higher fitting accuracy.In order to achieve the purpose of multi-scale optimal fitting,this paper uses different probability distribution models and different parameter estimation algorithms to compare the fit superiority tests simultaneously,and achieves the optimal fitting from both the distribution models and the parameter estimation algorithms.The final results show that the five goodness-of-fit tests for the T location-scale distribution model with optimized parameters using the improved particle swarm algorithm yield the best combined results.It is found that the selection of fitting models for wind speed probability distributions cannot be generalized,and different fitting models should be selected for different wind speed probability distributions according to local conditions.In order to make accurate predictions for different wind speed probability distribution fitting models,this paper proposes a random forest and logistic regression based wind speed probability distribution fitting model selection method.The final results show that the method can predict the fitting models of different wind speed probability distributions more accurately.