Thermal management system (TMS) plays a crucial role in improving driving range and passenger comfort for electric vehicles (EVs) in winter. However, EVTMS still faces unsolved challenges such as accurate modeling of time-delay, compensating for temperature variations, and achieving efficient control strategies, which are important for enhancing performance, safety, and energy efficiency of EVs. To address above issues, this study builds an integrated EVTMS model and corresponding controllers, aiming to obtain superior battery temperature and cabin comfort control performance through co-simulations. Uniquely, neural networks are set as control feedforward to predict the temperature changes because of their strong nonlinear mapping ability. The results reveal that, NN feedforward can well predict and compensate for the system time-delay and nonlinearity, significantly improving the performance of PI controller and MPC, reducing control overshoot, saving energy and extending component lifespan. The battery temperature and PMV fluctuations can be effectively reduced up to 32.50 % and 30.82 %, and the cabin and battery PTC consumption can be reduced by up to 6.03 % and 6.74 %. The results can provide valuable insights for TMS engineers, aiding in the selection of advanced control strategies.