In order to reduce the system operation cost of predicting short-term load curves and improve the feasibility of artificial intelligence prediction, the FFT and GA composite algorithm is designed as an artificial intelligence algorithm with the goals of low time complexity, small required sample data, and high flexibility. The algorithm uses mature Fast Fourier Transform and Genetic Algorithm, and improves the flexibility of the prediction method by dynamically adding kernel functions. It can be used for load curve prediction of 96 data points or smaller granularity, and can also reuse the calculation results of big data prediction methods for larger data granularity. By improving the feasibility of individual low-reward data predictions, it can improve the accuracy, sensitivity, and adaptability of energy management plans, accelerate the high-quality development of the power grid, improve the level of digitization, and promote demand response.