Abstract: This paper used Differential Evolution (DE) algorithm with improved crossover strategy to study the generation of speed curve for energy-saving optimization of urban rail transit trains. The algorithm took the distance from the starting point to the target parking point and the condition value as individual genes. Based on the improved crossover strategy, combined with the principle of working condition conversion, the mutated individual genes were effectively processed. The paper established evaluation model of terminal speed, parking position error, running time and traction energy consumption, and generated the optimal Automatic Train Operation (ATO) speed curve under the requirements of safe operation and comfort. Through the data simulation test, the traction energy consumption is 28.8 kw·h. Under the premise of arriving on time, the generated speed curve has good energy-saving effect, which has certain reference value for the research on energy-saving optimization operation of urban rail transit.