The Coal Multi-sensor Data Fusion Prediction Of Chaotic Time Series

Coal mine safe production issue has become one of the problems to be solved in China, coal mine gas disaster as the number one killer threat to China’s coal mine safety. Therefore, to ensure the validity and accuracy of the warning, now in mine safe production process,is imminent. With the improvement of mine automation technology, most of mine has established a variety of automation systems, such as the monitoring of underground production environment (gas, wind speed, temperature, etc.), gas safety monitoring and control system, monitoring system of underground personnel activities, up to20kinds of underground automation system. These systems played a crucial role in the mine safe production.This topic firstly learn and organize the predict techniques used in the coal mine warning field at home and abroad, and analysis inadequacies of the related system. Learning knowledge and algorithm of data fusion, for one-sidedness of prediction of the single sensor coal mine data, a data fusion method of thinking applied to underground warning has been explored, multi-sensor data fusion technology forecast coal mine data has been proposed. Based on phase space reconstruction and information fusion technology, prediction models of multi-sensor mine data has been built. A variety of sensors on the underground, including the gas concentration, wind speed, temperature sensor fusion forecast.The main findings are summarized as follows:1.The data is from WuYi coal mine and Yangquan Coal Mine in Shanxi Province,nearly20G data were collected. First, taking the gas sensor for the study, using the idea of chaos theory, phase space reconstruction and weighted first-order local prediction method, achieve the prediction of single sensor. Respectively,taking1,5,10,15,20minutes for the time interval, the experimental results obtains the minimum error which is0.035.2. It takes Gas concentration, wind speed, temperature, three types of sensors, time-series data for the study Multi-sensor time-series data for the study, use multivariate phase space reconstruction and various types of sensor data fusion.By using the weighted first order local the idea of on the basis of the K-Means clustering, it achieves the prediction of multi-sensor data fusion prediction. The experimental results obtains the minimum error which is0.003, and the error declined significantly.