The Classirication Forecasting Of Multifactor Time Series Based On Quotient Space Theory

With the development of computer software and network technology, the information technology is developing more and more rapidly in modern society. A large amount of information is produced every day and stored by the form of time series data. Facing such huge amount of information, how to use these data series is becoming a research focus in many research fields. Classification and prediction, as two important analysis forms of time series, also attracts a number of researchers’attention. As the research is progressing, nonlinear and uncertain factors are too complex, so that the time series prediction of traditional statistics or static model can not meet the requirements. Neural network, because of its special character, is suitable for processing complex models and abstract useful information. However, traditional neural network has some disadvantages, which Limits its scope of application. Based on multivariate time series, this paper adopts the model of structural neural network combined with GM (1,1) theoretical model to complete numerical prediction on time series prediction and classification.In order to make prediction for multivariate time series more effectively, the paper introduces the theory of quotient space granular computing. The paper does analysis to time series from various levels and perspective, and combines qualitative and quantitative analysis effectively, which reduces the complexity of the problem, highlights the characteristic properties, which impact the development trend of the time series. This is favorable for increasing the efficiency and accuracy of predictionThe main work of this paper is as follows:(1) The theory of quotient space granular computing base on granular computing is adopted to solve the problem of large amount of time series data by selecting different granular of data in size. With the quotient space theory, use partitioning or combine the different granular in size to make the characteristics of data more apparent to obtain more comprehensive information and reduce the complexity of problem solving.(2) In the time series forecasting model, the forecasting models base on neural network can process multiple variables well. These models are often used to solve the problem of many prime time series prediction. Compared with the traditional neural networks, construct a structural neural network is simple and well structured. Besides, it also has strong readability and high calculating speed. But the learning order of covering algorithm is randomly selected. Experiments show that the learning sequence has a significant impact on the network performance. This paper proposes a new kind covering algorithm based on competition. In this algorithm, sphere domains can be adjusted gradually, which are created them alternately by randomly selected or selected from the covered nodes. Experiments show that this algorithm can effectively reduce the number of sphere domains, decrease the number of rejected samples and improve the recognition accuracy.(3) In the multivariate time series prediction, GM (1,1) combined with the structure neural network prediction model is adopted. Constructive neural network is an excellent classification network model. The paper adopts the model to complete classification prediction, and make qualitative analysis to time series, that is classification. Because time series having a certain tendency over time, the time series’prediction is divided into values and status values, and predict the two parts respectively. GM(1,1) model is used to complete forecasting time series trend values and the status value is completed by structural neural net. At last, colligate the values of the two predictions as the final result of time series prediction.(4) The experiment. Adopt the monitoring data from coal mine as the experiments’data. Use the classification prediction model based on structural neural net to forecast the trend of gas concentration. And use the combined prediction model of GM(1,1) and structural neural net to forecast the value of the gas concentration. The experimental results show that, the classification prediction model accurately forecast the trend of gas concentration, and the forecasting values of gas concentration, which is forecasted by the combined model of GM(1,1) and structural neural net is more precise than that forecasted by other prediction model. Compared with other model based on neural net, the combined prediction model also has a small amount of computation and high accuracy.