Abstract: The vehicle shadows formed by fast-moving vehicles on the pavement PV array have complex dynamic random distribution characteristics, which will cause the P-V curve of the pavement PV array to exhibit dynamic multi-peak characteristics and bring challenges to the maximum power point tracking (MPPT) control of the pavement PV array. Therefore, a maximum power point voltage forecasting method based on Bayesian optimization (BO) convolutional neural network (CNN) is proposed. The images of environmental information of the pavement PV array are input into the maximum power point voltage forecasting model based on CNN for learning, and then this model is used for predicting the maximum power point operating voltage of the pavement PV array. Finally, simulation and experimental results show that this predicting model has good adaptability and can accurately predict the maximum power point operating voltage of the pavement PV array under different vehicle shadow conditions, especially in greatly improving the forecasting speed of the maximum power point voltage, which lays a foundation for MPPT control of pavement PV array under the shadows of dynamic random vehicles.