When a large number of electric vehicles randomly access the power system, the peak-to-valley difference and power fluctuation will become more serious, and even exceed the power system's power supply capacity and affordability, which will affect the stable operation of the power grid, make the charging cost too high and hinder the development of electric vehicle industrialisation. Therefore, analysing and optimizing the charging scheduling method for electric vehicles to reduce the adverse effect of charging load on the power grid is very significant in practical applications. In the optimisation of the above problems, satisfying the constraints of peak-to-valley difference, daily load variance, and electric vehicle charging power, a charging model is constructed with the goal of minimizing charging costs. But using a standard genetic algorithm to solve the objective function will cause problems such as inhomogeneous spatial distribution leading to local optimum solutions, premature convergence, and slow operation speeds. Therefore, an improved genetic algorithm is proposed in this paper. Through the introduction of orthogonal operators in population initialisation and crossover operations, the problems such as non-uniform distribution of solution space are overcome, the diversity of species is strengthened, a more optimised charging scheme is obtained, and a more accurate solution set is gained. The experimental simulations compared the improved genetic algorithm with the standard genetic algorithm, showing that the charging cost obtained by the improved genetic algorithm is reduced by 17.1%, the peak-to-valley difference is reduced by 252.6 kW, the variance is reduced by 0.3121*10⁵kW² and the convergence speed has nearly doubled. The experimental results verify the effectiveness of the improved genetic algorithm in solving the EV charging model.