| CPC H01M 10/486 (2013.01) [G06N 20/00 (2019.01); H01M 2200/10 (2013.01)] | 7 Claims | 

| 
               1. A method for managing a lithium battery, comprising: 
            obtaining a battery sound signal sequence of a to-be-detected lithium battery; 
                performing a level-1 warning of thermal runaway of the to-be-detected lithium battery, comprises: 
                carrying out an identification of data points with abnormal data on the obtained battery sound signal sequence, if a data point with an anomaly score in the obtained battery sound signal sequence is identified, providing an alarm that the to-be-detected lithium battery has a risk of thermal runaway; then go to a level-2 warning of thermal runaway of the to-be-detected lithium battery; 
                performing the level-2 warning of thermal runaway of the to-be-detected lithium battery, comprises: 
                extracting a time-domain feature and a frequency-domain feature of the identified abnormal data point, and detecting whether there is a sound of a expansion caused by the thermal runaway of the to-be-detected lithium battery in the time-domain feature and the frequency-domain feature; if yes, cutting off charging and discharging circuits of the to-be-detected lithium battery, performing a cooling treatment on the to-be-detected lithium battery with ventilation or carrying out a replacement on the to-be-detected lithium battery; 
                wherein, 
                in the level-1 warning, using an isolation forest algorithm to identify the data points with the abnormal data in the obtained battery sound signal sequence, comprises using a Mel-spectrogram feature as an identification feature to calculate an average path length of each data point of the obtained battery sound signal sequence in an isolation tree of the isolation forest algorithm, then calculating an anomaly score of the each data point based on the average path length, and then identifying a data point of which the calculated abnormal data is less than a set threshold value as the abnormal data point; and 
                in the level-2 warning, after extracting the time-domain feature and the frequency-domain feature of the identified abnormal data point, using a sparrow search algorithm-optimized eXtreme Gradient Boosting (SSA-XGBoost) algorithm to detect, in the time-domain feature and the frequency-domain feature, whether there is the sound of the expansion caused by the thermal runaway of the to-be-detected lithium battery; wherein, 
                in the SSA-XGBoost algorithm, performing an optimal parameter adjustment on a number of iterations, a learning rate, and a decision tree depth of the eXtreme Gradient Boosting (XGBoost) algorithm through the sparrow search algorithm (SSA); and 
                the extracted time-domain feature is a root mean square (RMS) energy, and the extracted frequency-domain feature comprises Mel-frequency cepstral coefficients (MFCCs), formant energy, and a spectral centroid. 
               |