Intelligent predictive cooling strategy for liquid-cooled lithium-ion batteries under dynamic operating conditions

To manage the temperature of liquid-cooled lithium-ion batteries under complex operating conditions, an adaptive Long Short-Term Memory–Model Predictive Control (LSTM–MPC) collaborative control framework is proposed. The LSTM network performs multi-horizon short-term prediction of temperature rise based on historical current, voltage, and temperature profiles, while predictive uncertainty is quantified using Monte Carlo (MC) Dropout. An interval score–based weighting scheme is employed to fuse multi-horizon forecasts and provide reliable look-ahead information for the MPC controller, which optimizes coolant flow under thermal safety and pump power constraints. Under the US06 driving cycle, the maximum temperature overrun is reduced from 1.335 °C to 0.352 °C, while the over-temperature duration is shortened from 631 s to 202 s. For composite driving cycles at ambient temperatures of 30 °C, 35 °C, and 40 °C, pump energy consumption is reduced by 52%, 58%, and 37%, respectively, compared with constant-flow control, while maintaining comparable peak temperature. The results demonstrate that the proposed LSTM–MPC framework supports anticipatory pre-cooling and improved energy efficiency under thermal safety constraints, indicating promising potential for practical battery thermal management applications.