The invention discloses a method for predicting the energy consumption of a hybrid truck based on a variable time domain model, and the method comprises the steps: repeatedly operating the hybrid truck in an urban working condition at a test stage, obtaining the original operation data of the truck, carrying out the preprocessing, carrying out the classification through a k-means clustering algorithm, and generating a plurality of state segments; establishing a Makarov working condition prediction model, and providing a plurality of prediction time domains for each state segment to obtain predicted vehicle speeds in the prediction time domains; calculating and recording the actual vehicle speed of the vehicle, and selecting the prediction time domain with high prediction precision as the parameter of the state segment; in the application stage, according to the characteristic parameters of normal running of the truck, the prediction time domain corresponding to the working condition with the high matching degree in the state fragments under the cyclic working condition is found to serve as the prediction time domain under the running working condition, and energy consumption is obtained according to the Makarov working condition prediction model. According to the prediction method, timely adjustment of the prediction time domain can be realized, and the calculation burden is not increased too much while the vehicle speed in the time domain is predicted more accurately.