Research On Haze Forecast Modeling Algorithms

In recent years, the frequent occurrence of haze weather has largely affected people’s health as well as the production and living problems of the people. Reducing the occurrence and impacts of haze have become an important task for governments at all levels and the relevant departments, The qualitative forecast and grade forecast of haze can provide reference for the relevant departments to make haze control and reduction measurements, which makes the haze forecast methods research become a hot research issue in the environment protection and weather forecast fields.In this paper, the statistics of multiple regression, the support vector regression and the related methods are used for the research of haze forecast algorithm.This paper researches the haze forecasts model based on the multiple stepwise regression and the probability mixed regression algorithm. First, a large sample data of the statistical prediction is used to determine the weights and the threshold value of each key factor. and then establish visibility forecast equation using the weather factors from numerical prediction according to the multiple stepwise regression method so as to reduce system error caused by numerical forecast system. Then, a haze forecasting model based on binary variables is established using probability regression combined with visibility, relative humidity and other physical parameters, and the probability value is obtained. Finally, the data of Beijing, Nanjing, Zhengzhou and other places are taken for example, compared with the existing haze forecasting system CUACE, the algorithm proposed in this paper is effective.This paper researches the haze forecasts model based on the combination of time series and support vector regression. The original visibility signal is decomposed by the Empirical Mode Decomposition(EMD), which results in that the different frequency of original signals are decomposed into smooth components. Then the support vector regression method is used to train and predict the decomposed components, and the components are integrated after forecasting. At last the haze is forecasted through the presence of precipitation and the threshold value of relative humidity. Beijing, Shijiazhuang, Nanjing, Zhengzhou, Hangzhou and other typical sites are taken as the training examples of the established model. The results show that the method proposed in this paper has improved in the aspect of visibility prediction compared with the multivariate stepwise regression algorithm.This paper proposed an improved online support vector regression algorithm for haze forecast. This method applies incremental learning algorithm on the new samples, and iteratively updates the boundary support vector with partitioned matrix. The atmospheric visibility prediction model is established combing with chaos particle swarm optimization algorithm for kernel function of support vector regression parameters optimization. With whether it will rain and the relative humidity threshold, the haze forecasting model is obtained. The results show that the running speed has significantly improved compared with the original online support vector regression, and the forecast accuracy has also improved compared with the haze forecasting system CUACE.