The invention provides an electromechanical equipment fault early warning method, which comprises the following steps: constructing an edge server, collecting electromechanical equipment state data, and transmitting the electromechanical equipment state data to a cloud big data platform after the edge server performs fault classification; periodically preprocessing the accumulated state data, andsegmenting the state data into a training set and a test set; training a GRU recurrent neural network model of fault early warning classification by using the sample data of the training set, and operating the trained model detection accuracy on the test set; and performing fault early warning classification calculation on electromechanical equipment data in real time by using the verified GRU recurrent neural network model, and issuing fault warning and fault early warning messages. According to the electromechanical equipment fault early warning method, the GRU recurrent neural network modelis adopted to carry out electromechanical equipment fault real-time prediction, and the accuracy and response speed of early warning are improved, and operation and maintenance personnel are assistedto discover possible fault risks of the electromechanical equipment in advance.