Abstract: Constructing a new type of power system is an important means to promote the transformation and development of modern power systems and achieve the dual-carbon goal. Accurate load forecasting results are crucial for optimizing the balance of power supply and demand and enhancing energy utilization efficiency, and artificial intelligence (AI) technology represented by deep learning can effectively optimize the balance of power supply and demand and enhance energy utilization efficiency. AI technology represented by deep learning can effectively optimize the balance of power supply and demand and improve energy utilization efficiency. Based on this, the paper firstly analyzes the current status of load forecasting research from the perspectives of scene objects, data types, evaluation methods, forecasting methods, etc., and systematically evaluates and summarizes the development history, advantages and disadvantages of the existing deep learning-based load forecasting methods for power systems. Finally, in view of the challenges of load forecasting under the new type of power system, the research outlook of the future technology is made from the model and scenario levels, respectively.