Short term data stream prediction method based on long short term memory (LSTM) network model

The invention discloses a short term data stream prediction method based on a long short term memory (LSTM) network model. The method comprises steps: multiple training samples in an observation pointare firstly obtained, features of the training samples are then extracted, the training samples are classified according to the features of the training samples, and two classes of a violet data stream value change trend and a mild data stream value change trend or two classes of a rising change trend and a falling change trend are obtained; all training samples are adopted for training for the LSTM model, a main model after training is obtained, the two classes of training samples are adopted to train the main model respectively, and a first-class sub model and a second-class sub model are obtained respectively. Testing samples in the observation point are acquired, the testing samples are classified by a classifier, the testing samples are inputted to the first-class sub model or the second-class sub model acco