The increasing electrification of the transport and heating sectors through the use of electric vehicles and heat pumps is placing a growing load on low-voltage distribution grids, which were not originally designed to handle such demand. On sunny days, high PV feed-in can additionally cause reverse-direction congestion. Therefore, new control mechanisms are needed to enable the integration of sustainable transport solutions and renewable energy into the everyday lives of end users without overloading the distribution grid.

To tackle this problem, this work presents a decentralized, agent-based approach that predicts potential grid congestion using forecasts. The aim is to prevent congestion proactively while minimizing costs incurred by the grid operator.

The basis for this work is provided by existing legal frameworks which, to avoid grid congestion and under compensation obligations, allow grid operators to curtail or disconnect specific consumption and generation units for limited periods. In addition to this legally mandated control of loads, this work further considers a mechanism in which households voluntarily grant the grid operator access to their battery storage systems in exchange for financial compensation. These batteries can then be pre-charged ahead of predicted congestion events, thereby shifting part of the predicted load forward in time to actively manage grid load. This mechanism presents an alternative to curtailing consumption when a congestion situation is already in progress and thus reduces the impact on end user comfort.

This work implements a multi-agent system in which the grid operator manages household agents. At regular intervals, new consumption and generation forecasts are generated, and the system evaluates whether a congestion situation is expected based on these forecasts. In case of a predicted congestion situation, the agents execute a cooperative optimization to compute a cost-minimizing schedule for the entire day, coordinating the curtailment of flexible loads and the anticipatory use of battery storage. The optimization is performed using an adapted form of the Alternating Direction Method of Multipliers (ADMM), structured to be executed in parallel by distributed agents. This decentralized structure is characterized by high scalability and robustness against failures of individual agents.

Simulations on an aggregated low-voltage distribution grid model based on the IEEE-33-bus system with standard load and generation profiles show that the implemented agent-based approach reliably detects and mitigates predicted congestion events through coordinated household behavior. The optimization algorithm demonstrates stable convergence within timeframes short enough for real-time operation. Furthermore, the integration of voluntarily provided battery storage capacity reduces restrictions on user comfort by mitigating the need for curtailment in congestion events. The system is also capable of forecasting and mitigating congestion caused by high PV feed-in. The presented approach thus enables cost-effective congestion mitigation while preserving user comfort.