Unsupervised learning battery production process abnormal fluctuation detection method

The invention discloses an unsupervised learning battery production process abnormal fluctuation detection method. In a feature fusion reconstruction network provided by the invention, an encoder network performs spatial feature extraction on an input multi-channel feature matrix through multilayer convolution operation. The ConvLSTM network extracts time features of an input multi-channel feature matrix sequence in different time steps to complete feature capture of data, and an attention mechanism added on this basis can complete weight allocation, allocate more attention weights to key features and reduce noise interference. The feature mapping obtained in the previous step can be decoded through a decoder network, and meanwhile, a feature extraction matrix is constructed by utilizing the asymmetric capability of feature matrix information, so that the feature reusability among layers is enhanced. The hierarchical feature fusion model increases feature interaction between layers, so that the model can sens