Self-adaptive deep learning for multimode process monitoring

•A novel self-adaptive deep learning method named LAS-VB is proposed for multimode process monitoring.•Local adaptive standardization (LAS) method is used to standardize and preprocess the local moving window data.•Variational auto-encoder bidirectional long short-term memory (VAE-BiLSTM) monitoring model is used to detect the unstable deviation in the local moving data.•Compared with other multimode methods, LAS-VB has outstanding performance on fault detection for new operating modes that never occurred in history.