Research on Fault Diagnosis Method of Mobile Energy Storage Equipment Based on Deep Learning

A novel method based on CNN, which integrates the merits of self attention, is presented, which can be applied to the failure diagnosis of power storage devices. Aiming at long distance dependence in time sequence data, this method uses self-attention to enhance the precision of identification for complicated failure patterns. This paper carried out strict preprocessing on the failure data, such as de-noising, normalized and enhanced, so as to make sure that the input data is pure and usable. Secondly, this method utilizes CNN's strong ability to extract multi-dimension characteristics from raw data, and makes them more prominent by means of self attention. To validate the effectiveness of the proposed method, the paper chooses a dataset that includes many kinds of commonly used failures. Finally, it is proved by experiments that the new method is effective on the main parameters of failure diagnosis. The precision is 93.4%, the retrieval is 92.1%, and the F1 is 0.92. The results show that the performance of the system can be increased by over 8%. What's more, this model has proven to be able to reliably diagnose various kinds of failures, including abnormal temperature, cell aging and high frequency oscillation. Especially, it is more accurate and robust than the current techniques in detecting high frequency vibration and cell aging.