This article introduced the principle and structure of on-line audio monitoring system for electrical equipment in non-attended power substation, then focused on data monitor and recognition. At first, the audio sensors collect the sound data of running electrical equipment of power substation, the audio data input to the monitoring host, and the monitoring host deals with the audio spectrum, analyzing the audio data and recognizing the fault by neural networks based on sine basis function, which can determine the operation status of electrical equipment and the type of fault quickly. Fundamentlly, this method can improve the previous artificial monitoring sound of electrical equipment to determine different type of error, and it also provides help for the timely maintainance and correcting the faults in power substation.