The invention discloses a modal identification method based on LSTM and EMD. The modal identification method comprises the following steps: acquiring signal data; predicting and extending data at theleft end and the right end of the signal by using an LSTM algorithm to obtain expanded signal data; decomposing the expanded signal data by using an EMD algorithm to obtain each order of modal signalsof the signal; and cutting off the modal signals of each order at the left end point and the right end point of the original signal to form final modal signals of each order. According to the method,the endpoint effect and the modal aliasing effect of a traditional EMD (empirical mode decomposition) method during signal decomposition can be effectively solved, so that the result of damage identification and mechanical fault diagnosis is more accurate, and the robustness is stronger.