Research On Abnormal Acoustic Detection Algorithm For Train Operation Safety Monitoring System In Station

As the acoustic signal is susceptible to the influence of environmental noise,the effective noise reduction methods and abnormal feature extraction methods are very important in abnormal acoustic detection.Due to the change of driving environment,vehicle weight and speed,there are great differences in traffic noise,which makes it impossible to effectively model the traffic noise.As a result the real-time estimation and updating algorithm for train noise needs to be further studied.At present,train abnormal acoustic detection mainly analyses single abnormal type,and the abnormal feature extraction method is only applicable to specific abnormal type,thus effective recognition of various abnormal traffic types needs further research.The main contents of this dissertation are as follows:(1)Aiming at the problem that multi-noise sources and traffic noise can not be effectively modeled,an improved minima controlled recursive averaging(MCRA)algorithm is proposed to estimate traffic noise,which effectively solves the problems of noise estimation delay and inaccurate noise power spectrum estimation in MCRA algorithm.The algorithm can estimate the environmental noise in real time and effectively solve the problem that the noise model can not be effectively modeled in different environments.(2)Aiming at the characteristics of many types of abnormal driving types,a new method of train abnormal acoustic detection based on improved energy entropy ratio is proposed.Unlike the traditional energy entropy ratio method,which can only detect energy mutation,the improved method can effectively identify four types of abnormal driving conditions by selecting specific abnormal frequency bands for energy entropy ratio detection according to different abnormal types.(3)The train operation safety monitoring system is built,which consists of noise reduction module,headstock and carriages separation module and abnormal type detection module.The functions of locating abnormal carriages and distinguishing abnormal types are achieved,and the speed of the train can be determined.Through the detection of 5000 segments of driving sound collected from Yangliuqing,Jinghai and Tangguantun stations,the abnormal traffic conditions are effectively distinguished,and the detection accuracy is 91%.The experimental results show that the improved MCRA algorithm can effectively reduce the impact of traffic noise on abnormal acoustic detection.The combination of the improved energy entropy ratio detection method can effectively identify the types of abnormal driving,which is valuable in practical.