The invention discloses a Markov chain microstroke time interval prediction method based on space-time learning, and the method comprises the steps that a work condition data collection module collects a large amount of speed condition data as a training sample, collects the speed condition data of a vehicle in real time, determines an effective speed time sequence, analyzes the collected work condition data, judges the effectiveness and reliability of data, and removes the noise data; an idling time duration prediction module determines a training data subset with the higher correlation degree with the real-time collected data according to the time information and the space information, learns and transfers the matrix through the subset, and predicts the type of idling time duration through the transferred matrix; an effective speed time sequence determining module removes the data nose, caused by the fluctuation of equipment, in the collected speed data and longitude and latitude data through a median filtering method, and determines the length of the optimal effective speed time sequence according to the correlation of the speed data. According to the invention, the method employs a data training transfer matrix with the higher correlation in time and space, and predicts the possible duration under the next idling work condition.