Study On Coal Gas Early-warning Technology Based On Support Vector Machine And Data Fusion

Coal mine gas disasters has become a main threat to safe production and the most important factor to constrain sustainable development of the coal industry in China. Understanding thoroughly the law and feature of the coal mine gas disaster, and realizing an accurate forecast and early warning of mine gas hazard are effective means of prevention and treatment of mine gas disaster, and also an important research fields of coal mine safety and intelligent information processing. To meet the present actual needs of the safe production in coal mine, this dissertation makes systematic study on the mine gas early warning technology based on support vector machine and data fusion. The main research contents are as follows:1. On the basis of the definition of related concept of gas warning, analysis of mine gas data flow, study of data fusion and support vector machine(SVM), the technology framework of coal mine gas data fusion based on support vector machine is constructed.2. By intensive study on the principle of particle swarm optimization(PSO) genetic algorithm(GA), the method for selection and optimization of parameter vectors of support vector machine based on chaotic particle swarm optimization-genetic algorithm (CPSO-GA) is proposed, which has created technical conditions for the research of gas early-warning technology based on data fusion combined with support vector machine.3. By using support vector machine, chaos theory, rough sets, clustering, non-linear combination forecast and other modern means of information processing, detailed data fusion technology of multi-source coal mine gas data in data-fusion level, feature-fusion and decision-fusion level are studied:(1) In the data-fusion level, the de-noiseing method for coal mine gas data based on support vector machine is proposed, which effectively eliminates the effects of ordinary noise data, ab-normal data and missing data existing in coal mine gas data, and provides an effective technical means to obtain more accurate and complete original coal mine gas data.(2) In the feature-fusion level, the forecasting methods of gas chaotic time series based on phase space reconstruction-clustering-multi support vector machine regression is proposed and the qualitatively and quantitatively forecasting methods of the types,intensity and coal amount of coal-and-gas outburst based on rough sets and support vector machine is put forward, which effectively help to extract the features information of mine gas data, and to forecast changes laws and development trends of mine gas.(3) In the decision-fusion level, the non-linear combination forecasting methods of gas emission amount is proposed, which reduces the risk of decision-making and improves the forecasting accuracy. The multi-variables decision forecasting method based on least squares support vector machine(LS-SVM) and same/different kinds sensors data fusion of CH4, CO, temperature and wind rate is also brought forward, by which a real-time evaluation model and forecasting model of coal mine safety states are constructed, effectively acquiring the feature vector of coal mine safety states, and realizing early-warning of gas disaster.The theoretical analysis and experimental results demonstrate that the methods proposed in this dissertation are effective and feasible, and can be applied to early-warning of gas disasters in coal mine.