Vehicle speed prediction provides important information for the energy management strategy of new energy vehicles, but accurate vehicle speed prediction is challenging. In order to overcome the interference of deterministic or stochastic factors, e.g., the driving condition, driver’s intention and vehicle type, in this paper, a multi-condition speed predictor is proposed based on a DK (DTW-based K-means) clustering model. The speed predictor splits the vehicle speed sequences into different driving conditions by the DK clustering model, and the future vehicle speeds under different driving conditions are predicted by the sub-predictor which combines one-dimensional convolutional neural network (conv1D) and long short-term memory neural network (LSTM). Based on the proposed predictor, the effects of different input-sequence lengths and the number of clusters on the predictor are discussed. Moreover, the performance of the proposed predictor is compared with other commonly used models. The results show that the proposed predictor has better adaptability to multiple driving conditions, and the prediction accuracy is higher than other models.