Abstract: With the widespread use of distributed photovoltaic large-scale power generation, the net load ′duck′ curve has become more apparent. However, electric vehicle charging during the day is unable to fully utilize the new energy, and charging at night only adds to the already existing load peak. To address the issue of net load ′peak-to-peak′ exacerbation, the charging load transfer process is facililated with the objective of minimizing the peak-valley difference of net load. To achieve this, statistical data on fast and slow charging behaviors are used to predict future charging load distribution through Monte Carlo simulations. Fast and slow charging load constraints are then established based on the network access characteristics of slow charging and the delayed charging characteristics of fast charging. Then, the load transfer rate is calculated using the gradient descent method, and the charging load price response model is constructed based on user consumption psychology. Finally, economic analysis of power grid peak shaving limits the constraint of electricity price change, and a charging guidance model is constructed with the goal of minimizing the peak-valley difference of net load. Deep reinforcement learning is used to solve the model and solution strategy. The simulation results show that the proposed model and solution strategy can effectively guide the charging load to avoid the peak period of the net load, determine a reasonable price, and reduce the peak-valley difference of the power grid.