Research On Time Series Similarity Search Algorithms For Gas Monitoring Data

Analyzing gas emission law by taking advantage of the gas monitoring time series data sampled from coal mine gas monitoring systems is an important and challenging research field. Discovering hidden patterns in gas time series are helpful to realize the rule and trend of gas data's change, which means a lot to safety of coal mine. This thesis aims to explore a new way to study gas emission based on gas monitoring dataset by using time series similarity search technology. With careful study and comparison of several methods, time series similarity search algorithms were introduced to adapt to characteristic of gas monitoring data, and to develop software of analysis gas emission. The main contributions are included as follows:A piecewise polynomial representation (PPR) based similarity search approach for coal well gas monitoring time series data is proposed in this thesis. PPR is a linear polynomial representation and an orthogonal transform. Proposed approach was compared with Discrete Fourier Transform (DFT) based approach and Discrete Wavelet Transform (DWT) based approach respectively. Experimental data is real gas monitoring data sampled from Yuhua Coal Mine, Shaanxi Province, China. The Evaluation criterions are both information loss and mean search efficiency. Experimental results show that in the condition of same compress rate, values of information loss of PPR, DFT and DWT based approaches are very close to each other; Whereas, mean search efficiency of PPR based approach is 32% higher than that of DFT based approach, and 34% higher than that of DWT based approach respectively. So PPR based approach is more suitable for similarity search in gas monitoring data.A kind of multivariate time series similarity search algorithm of gas monitoring data based on two-dimensional wavelet transform was proposed. Firstly, Multivariate time series (MTS) data is stored as data matrix. Those matrixes are dimensional reduced with two dimensional wavelet transform. Then, each data matrix is displayed clearly with visual methods. Euclidean distance and Extended Frobenius norm (Eros) distance of the corresponding covariance matrix are used to measure similarity respectively. The results show that it is feasible to represent MTS data with gray pictures. Based on two-dimensional wavelet, the query efficient of Euclidean distance is better than that of Eros distance.On the basis of above research, a gas monitoring data similarity search sub-system was developed which is used as one part of gas emission analysis system based on time series data mining. The sub-system employs PPR dimensionality reduction technology, R-tree multi-dimensional index, which eliminates zone overlapping and increases search speed. Given a query series, after setting a threshold value, similar sub-series are found and effect of scaling and shifting similar search are carried out. The test results show that the sub-system functions are correct, and the desired goal is achieved.