The Research Of Fog-haze Prdiction Model Based On DAP-SVDD Algorithm Of Changchun In The Next 24 Hours

Fog-haze prediction is becoming a new research field with the rising attention of fog-haze weather. But no complete theoretical model has been established until now. Though most of the weather prediction model use neural network method, it will be caught in“dimension disaster”when building fog-haze prediction model.This paper proposes a new clustering method — — DAP algorithm which is using dichotomy to search preference according to the characteristics of the fog sample data. Then combines the DAP algorithm with SVDD model to predict the fog-haze weather in the 24 hours later. For the number of “fog-haze day” is quite smaller than normal day, that leads to the unbalance of the data and a low accuracy. So in this paper,we chose SVDD algorithm to build the prediction model to solve the unbalance problem. And we use AP algorithm as a pretreatment method to rise the accuracy.Details are as follows:Introduce the research background、purpose、significance and the research status of the papers.Fog-haze weather receives more and more attention in recent years, it has a great negative impact on people’s health、traffic、produce and so on. So if we could build a fog-haze prediction model to forcast this weather. It will be very convenience for people to take measures to reduce the unnecessary harm.But there has no complete model for people to use.Because of the imbalance of data, traditional SVM machine learning can not predict accurately. So ater we study the algorithm of SVDD and SVM, find that the SVDD algorithm do well in predicting and classing the small sample and imbalance data. What’s more, AP clustering will turn the big sample data into some Convex type of small sample data set. From what has been discussed above, this article put these two methods together and improve it could Give full play to the advantages of both.Proposed DAP-SVDD fog-haze prediction model which is based dichotomy to search preference.Before building the Classification model with SVDD algorithm, we should use APclustering algorithm to get some mall data cluster to improve the prediction accuracy.Clustering results have much to do with preference, traditionally, we set the preference as half of the similarity matrix’s mid-value. But as we know, every sample has it’s own trait, choose the fixed value will lead to the uncertain results of clustering, we could not sure if this is the best clustering results.In order to solve this problem, this article introduce dichotomy on the basis of AP clustering tofind the best preference in order to get the best clustering results. We named it DAP algorithm, then we choose SVDD algorithm as Downstream processors to build the prediction model.Design and implement DAP-SVDD fog-haze prediction model of Changchun.Firstly, dividing the air quality data and the atmosphere pollution data of Changchun area into test and training two sets. Then training DAP-SVDD model with SPSS data analysis software and MATLAB software to establish the prediction model. Finally we get the model prediction accuracy with the set of test data.The analysis and compare of the result of prediction model.After making the simulation and getting the accuracy of DAP-SVDD prediction model,we use the method of comparison to valid the fesibility of this model. Using the data to establish a AP-SVDD model、SVDD model、C-SVM model, then using the test data set to get the accuracy of each model. From the results we can easy sea that DAP-SVDD in dealing with fog-haze prediction problem have the best accuracy and the fastest efficiency.