Method and system for analyzing and predicting vehicle stay behavior based on multi-task learning

A method for analyzing and predicting a vehicle stay behavior based on multi-task learning, comprising:

• S1, acquiring vehicle Global Positioning System (GPS) data and vehicle On-Board Diagnostics (OBD) data comprising a vehicle ID, a travel start time, a start longitude, a start latitude, an end time, an end longitude and, an end latitude after desensitization of a vehicle;
• S2, preprocessing the vehicle GPS data and the vehicle OBD data to obtain vehicle stay behavior data comprising stay locations and stay durations, and extracting a spatial-temporal characteristic of the preprocessed vehicle stay behavior data by a deep recurrent neural network;
• S3, inputting the spatial-temporal characteristic into a multi-task learning and predicting network, and obtaining a correlation between a stay location prediction task and a stay duration prediction task through the multi-task learning and predicting network based on historical stay behaviors of the vehicle to predict the stay locations and the stay durations;
• wherein the step of obtaining a correlation between a stay location prediction task and a stay duration prediction task through the multi-task learning and predicting network based on historical stay behaviors of the vehicle to predict the stay locations and the stay durations comprises:
• inputting the spatial-temporal characteristic into the multi-task learning and predicting network, construing a stay location task model and a stay duration task model based on the historical stay behaviors and a long short-term memory network, respectively, allowing the stay location task model and the stay duration task model to learn at the same time through the multi-task learning and predicting network, and mining the correlation between the stay location prediction task and the stay duration prediction task to predict the stay locations and the stay durations in a future time period;
• wherein a process of multi-task learning comprises:
• using a relu function as an activation function, using a mean square error function as a loss function for predicting the stay durations, and using a cross entropy function as a loss function for predicting the stay locations;
• the loss function of a multi-task deep learning network is a sum of the mean square error function and the cross entropy function;
• the mean square error function and the cross entropy function are optimized at the same time, and the stay locations and the stay durations in the future time period are obtained by learning and prediction.