This paper develops a self-adaptive control strategy for a newly-proposed J-type air-based battery thermal management system (BTMS) for electric vehicles (EVs). The structure of the J-type BTMS is first optimized through surrogate-based optimization in conjunction with computational fluid dynamics (CFD) simulations, with the aim of minimizing temperature rise and maximizing temperature uniformity. Based on the optimized J-type BTMS, an artificial neural network (ANN)-based model predictive control (MPC) strategy is set up to perform real-time control of mass flow rate and BTMS mode switch among J-, Z-, and U-mode. The ANN-based MCP strategy is tested with the Urban Dynamometer Driving Schedule (UDDS) driving cycle. With a genetic algorithm optimizer, the control system is able to optimize the mass flow rate by considering several steps ahead. The results show that the ANN-based MPC strategy is able to constrain the battery temperature difference within a narrow range, and to satisfy light-duty daily operations like the UDDS driving cycle for EVs.