In this article, model predictive (MP) and deep belief net (DBN) are introduced to optimize and predict the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dq</i> -axis stator voltage of asynchronous motors for electric cars in the field-weakening region to achieve the optimal “speed–torque” (represents the optimization of speed and torque performance) control. First, the analytical model of the maximum torque output is established, and the effect of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dq</i> -axis voltage on the torque output is analyzed. Second, to reduce the influence of proportional–integral parameter tuning, optimize the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dq</i> -axis voltage, and achieve the accurate torque control, MP is introduced to establish an analytical model of MP controller tracking current loop for generating the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dq</i> -axis voltage. Third, the maximum torque vector control system based on the MP controller is built, and the influence of system parameters on control effect is analyzed. Fourth, the simulation data are collected, and the DBN voltage prediction model is established. The model is embedded in the field-oriented control to achieve optimal “speed–torque” control. Finally, the dynamometer experimental platform is established to collect the experimental data and establish the DBN voltage prediction model. The validity of the method is verified through the voltage data calibration, “speed–torque” characteristic calibration, motor efficiency calibration, “speed–torque” response, and ripple test.