It is crucial to measure dust concentration precisely, but it normally varies with changes of working conditions. To increase precision and on-line performance of coal dust sensor, an adaptive pattern recognition algorithm was presented. The signals of unitary angular spectrums were chosen as the adaptive eigenvector for pattern recognition and pattern bank was established in advance. Furthermore, the ratio of the sum of inner signals to that of outer signals about the diffraction angular was considered as the eigenvector of subclass pattern classification. After classification, pattern could be recognized easily and rapidly. Subsequently, number of detailed patterns within different pattern groups was increased reasonably. The errors of total coal dust and respiring coal dust decline from 6% to 2.5% and from 9% to 3%, respectively. As a result, the precision of sensor achieves 95% during the measurement. It can be concluded that the adaptive pattern recognition algorithm is effective to improve the precision and real-time performance of coal dust sensor.
Self-Adaptive Pattern Recognition Based on Multi-Agent
