Dynamic Grid Tariffs for Power Peak Reduction Using Reinforcement Learning

Increasing power flows due to the growing share of distributed renewable energy resources and electrification of transportation and heating may lead to significant reinforcement needs in distribution grids. Demand response (DR) can reduce or defer grid reinforcements by shifting flexible demand through direct control or time-variable tariffs. However, determining the optimal price signals is challenging as the decision-making of cost-minimizing customers needs to be considered, while the information on customers' devices may be limited. Therefore, this work uses reinforcement learning to determine real-time price signals that reduce power peaks. The conducted case study shows that the learned pricing policies can reduce peaks compared to the benchmarks of no control and a price signal which is proportional to the inflexible load. However, the case study results also indicate that other pricing schemes may be even more effective for reducing peaks in consumption.