The operation of the power grid is becoming increasingly intricate, primarily owing to the integration of renewable energy sources such as solar and wind. These sources are essential components of the Energy Green Transition. To ensure the safe and efficient operation of the grid in this evolving landscape, artificial intelligence (AI) and strategic planning play pivotal roles. 

Grid2Op offers a simulation platform that facilitates decision-making in power grid operations. This platform allows intelligent agents to propose actions based on real-time observations and forecasts. Nevertheless, certain critical actions, such as generation redispatching and battery operations, require multi-step planning. This is where the AIPlan4Grid project comes into play, aiming to address this challenge. It combines the modeling capabilities of Grid2Op with the planning features of the AIPlan4EU Unified Planning Library to create a hybrid grid planning agent.

The proposed solution takes the form of a planning agent with the ability to propose an optimized operational plan for the transmission power grid in the upcoming minutes. This plan takes into account actions on the grid, such as generation redispatching (reallocating energy from one node to another while maintaining the system balance) and managing the state of charge of storage units. To visualize the agent's plan, the Grid2Viz tool can be employed. Importantly, this solution is an open-source Python package available on GitHub.

Experiments reveal that incorporating a lookahead mechanism enables the agent to delay blackouts. While this is promising, it is crucial to acknowledge that the lookahead agent still falls short of preventing blackouts entirely by the end of the simulation. To reach better planning performance and realism, we believe additional constraints and secondary objectives in the problem modeling would help. These are avenues for further developments. These ideas are detailed in the notebooks presenting our solution.