Study And Application Of Multisensor Correlation Analysis Methods

The thesis mainly focuses on methods of synchronous multisensor correlation analysis and estimating asynchronous correlation delay time, and employs these methods to solve primary feature selecting of classifiers and detecting abnormal sample-data and estimating missing values of gas sensors in coal mines. At the same time, a prototype software system of gas emission analysis based on time series data mining was developed. The main contents are as follows:The multisensor correlation combination was defined and its characters were analyzed by using of correlation information entropy. A fast algorithm for finding all the correlative sensors combinations from a multisensor system was then proposed. The proposed algorithm is named SCC (Super Correlation Combination) algorithm. SCC algorithm is able to find out all the correlation combinations effectively.A method of estimating asynchronous correlation delay time based on Particle Swarm Optimization is proposed. Two fitness functions, i.e. correlation information entropy based fitness function and correlation coefficient based fitness function, are respectively employed by proposed method. The experimental results show that proposed method is efficient.A method of primary feature selecting based on multisensor correlation analysis is proposed. The concepts of primary feature and primary feature subset are presented to describe key features of a classified object. Then, calculation methods of primary feature and primary feature subset were proposed. Primary feature and primary feature subset of a classified object are both independent with classifiers used to classify the object. The experimental results of proposed method were given in this thesis.For a gas monitor system of coal mines, how to detect abnormal sample-data of gas sensors and how to estimate missing sample values of gas sensors are two important problems. Previous research was that employs sample data of all gas sensors of a gas monitor system to solve the two problems. However, it has disadvantages, which are larger amount of calculation, and lower accuracy rate for detecting abnormal sample-data and lower precision for estimating missing sample values. The methods which employ correlation combination sensors to solve the two problems were provided respectively. The simulation results show it overtakes shortcomings of previous research.Based on above research results, gas emission analysis software for gas monitoring of coal mines was developed. Two modules of the software, i.e. multisensor correlation analysis module and gas concentration predicting module are described. The test results show that the software runs well.