Design And Application Of Prediction Model Based On Time Series Analysis Technique

This thesis firstly studies the related technology of time series analysis including dimension reduction, time series pattern and similarity search on time series. In order to meet the demand of the dynamic pattern matching on real-time data, this thesis presents a dynamic pattern matching method of data mining. An optimize strategies is proposed to improve the efficiency of similarity search algorithm based on DTW. Based on all above, this thesis has designed and implemented a model for prediction, which can analysis multidimensional time series data and make a prediction.The main works and innovation of this thesis are as follows:1) A new time series pattern matching method is proposed as dynamic pattern matching, the pattern which is fond by this method, can be used to make a prediction directly, In order to obtain better matching effect of pattern matching, this method has used time series data normalization technology and DTW similarity search algorithm.2) This thesis takes some improvement and optimization to the DTW similarity search algorithm. First of all, we make the search algorithm can be used on multidimensional time series similarity search; Secondly we re-implement the algorithm by using SSE, This makes the algorithm to take more utilization percent of computing resources, and then improve the efficiency of algorithm in single threaded cases.3) We implemented a parallel operation framework. This framework uses early abandoning strategy, the sliding window technique and time series data segmentation technique. This framework improves the efficiency of the algorithm in a multithreaded computer system.4) By using all above, we have designed and implemented a general prediction model to analyze and predict the multidimensional time series. This model is applied to predict the ambient air quality, and obtain good prediction result.The main result of this thesis is that, implementing prediction model and well applying it to the early warning and forecasting system of ambient air quality. The model makes an analysis of the multidimensional time series which is comprised by the real air pollution index and the meteorological data of data, and then makes a prediction. This proved the availability of that prediction model based on time series analysis.