Electric bus energy consumption prediction is conducive to bus route scheduling and charging optimization management. However, existing prediction models do not comprehensively consider vehicle parameters, auxiliary power, speed profile and external factors. In particular, there is a lack of in-depth research on generating speed profile based on low-resolution data for electric bus. The aim of this study involves predicting bus trip-level energy consumption. Thus, the temperature effect on powertrain efficiency and auxiliary power was first analyzed, and then a Markov-based speed profile generation method considering the speed state transition spatiotemporal characteristics and driving style was developed. Reliable process-related variables extracted from the aforementioned model development in conjunction with external-related variables are used as the input of Markov-based Gaussian processing regression model (M-GPR). Finally, 7919 regular bus driving samples on the same route were extracted from the raw bus operation dataset to verify the model effectiveness in improving the prediction accuracy of trip-level energy consumption. The results indicate that M-GPR exhibits a better prediction performance with an average prediction error of 8.69%, which decreases by 0.62%–5.41% compared with other models. Furthermore, M-GPR presents wide applicability, with average errors between 7.01% and 9.90% in different operating scenarios.