Ward Reduction In Unit-Commitment Problems

One of the most challenging aspects of unit commitment (UC) is dealing with the grid, which is commonly represented by thousands of nodes and branches, leading to large optimization models. Because the computational difficulty generally increases with the model size, managing the network representation size can lead to significant savings in running times. Thus, we explore the widely used Ward reduction to reduce the network’s size through a process that iteratively removes nodes from the network. We present how, under mild conditions, the resulting reduced network model is equivalent to the original one. We evaluate our approach on 20 UC instances, with up to 13,659 nodes and 18,625 branches. We are able to produce reduced models with as few as 6.8% of the original nodes. Moreover, we show that the network reduction provides average speed-ups from 10% to 36%.