Research Of Fuzzy Time Series Prediction Based On Hybrid Model

In many areas, improving the forecasting accuracy especially improve the accuracy ofprediction of time series is an important and difficult problem. The study results show that,the mixing of different model can effectively improve their predictive performance, improvethe precision of prediction. In this thesis, designing a kind of hybrid system based on fuzzytime series forecasting which combines autoregressive integrated moving average model,subtractive clustering, neural network, fuzzy logic system. Then using BP algorithm adjuststhe system parameters, and a simulation study is given. The simulation results show that thedesigned fuzzy time series forecasting hybrid model is applied to practical problems isfeasible and effective. The specific work is as follows:(1)Introduce the development and application of time series prediction and moving selfregression model and subtractive clustering, and thire revevant knowledge. Introduce thestructure of type-1fuzzy logic system. And introduce the structure, the development andapplication of interval type-2fuzzy logic system and neural network.(2)Design a kind of type-1hybrid system based on fuzzy time series forecasting whichcombines autoregressive integrated moving average model, subtractive clustering, neuralnetwork, fuzzy logic system. Use self regression model to determine the number of inputlayer nodes, and use the subtractive clustering to extract rules. Integrate type-1fuzzy logicsystem into the neural network, which determines the structure of type-1hybrid system, usingBP algorithm to adjust system parameters. Given the application forecast the price of CPI, andsimulation with MATLAB. The simulation results show that the design of the type-1hybridsystem is effective and feasible.(3)Design a kind of interval type-2hybrid system based on fuzzy time series forecastingwhich combines autoregressive integrated moving average model, subtractive clustering,neural network, fuzzy logic system. Use self regression model to determine the number ofinput layer nodes, and use the subtractive clustering to extract rules. Then using BP algorithmto adjust system parameters, and KM algorithm is applied to type-reduction. Integrate intervaltype-2fuzzy logic system into the neural network, which determine the structure of intervaltype-2hybrid system.(4)The MATLAB simulation results can be seen the design of the fuzzy hybrid model isfeasibility and effectiveness which was applied to fuzzy time series prediction problem of CPIand the Australian thermal coal spot price and total social consumption. Though the trackingdiagrams and the RMSE that can be seen the interval type-2fuzzy hybrid model has smallererror and better prediction performance than type-1fuzzy hybrid system.