Summary: The tides and imbalances of the travel demand lead to the imbalance of vehicle inventory among electric vehicle stations, which hinders the development of one-way electric car-sharing industry. Fully considering the randomness and timeliness of actual travel demand, a scheduling method for one-way electric car-sharing is proposed. The long short term memory neural network is applied to forecast the vehicle demands of the stations, and a multi-objective scheduling model with minimum scheduling cost, minimum inventory cost and minimum potential loss profit is built, then the hybrid genetic and simulated annealing algorithm is applied to find the optimal solution. The rationality and superiority of the method are verified by the empirical study.