Time sequence deep hashing method based on comparative learning

The invention provides a time sequence deep hashing method based on comparative learning. The method comprises the following steps: constructing a time sequence Hash network model based on a comparative learning architecture; preprocessing the training data and setting network hyper-parameters; sending the preprocessed training data into a time sequence hash network model according to a gradient updating method, optimizing the time sequence hash network model based on comparative learning, and storing the model with the highest precision; and loading the stored optimal model, sending to-be-processed time sequence data into the optimal model to obtain a corresponding hash code, calculating a Hamming distance in a historical time sequence hash database which is converted in advance by using the hash code to query similarity, and obtaining a similar query result. According to the time sequence deep hash method based on comparative learning, hash codes can be generated through end-to-end training, convergence of