Abstract: With the advancement of electricity market reform, industrial and commercial users will gradually enter the market, while the users not yet entering the market will purchase electricity through agents from grid companies. As key entities in agent-based electricity purchase, grid companies play a significant role in coordinating electricity purchases across multi-time-scale markets to reduce purchase costs, which is crucial for their operational efficiency. This study outlines the workflow of agent power purchase for power grid companies and establishes an annual power purchase decision review and optimization model that considers annual, monthly, and spot markets. The model is solved using the genetic algorithm and convolutional neural network (CNN) - long short-term memory network (LSTM) - attention mechanism (Attention), and the model's effectiveness is verified through case studies. The review results show that adjusting the power purchase ratio between the medium and long-term market and the spot market can effectively achieve cost reduction and efficiency improvement of power purchase strategies.