Research on Investment Decision-making Optimization of Power Grid Companies Based on Deep Reinforcement Learning

Grid companies face multiple challenges in investment decisions under the background of new power system transformation, which requires balancing short-term economic benefits and long-term grid resilience, as well as coping with the dynamic complexity of new energy fluctuations and load growth. This study innovatively introduces deep reinforcement learning theory to construct an intelligent decision-making framework based on deep Q network (DQN), and models power grid infrastructure investment as a time series optimization problem. By accurately defining a high-dimensional state space containing real-time load, equipment status and new energy forecasts, a compound reward function integrating economy, reliability and sustainability is designed to drive agents to learn optimal investment strategies autonomously in a simulated environment. Empirical analysis shows that the model significantly improves investment efficiency in backbone upgrade and regional intelligent transformation scenarios, in which regional intelligent upgrade achieves $\mathbf{3 8 \%}$ increase in distribution network elasticity and 87% reduction in economic loss in peak load response scenarios. The research results provide a theoretical breakthrough and practical paradigm for power grid companies to achieve intelligent and forward-looking investment decisions.