Shaving monthly peak load by overestimating future load in microgrid model predictive control

Microgrid energy management is essential to improve economic performance, reduce carbon footprint, and enhance grid stability, for which model predictive control (MPC) is considered a promising strategy. The operating expenditure (OPEX) of a microgrid typically includes both the electricity usage charge and the demand charge. The former is based on the total consumption, while the latter is proportional to the peak imported power within a billing cycle.

In the practice of deploying MPC strategies, we need to forecast future loads and track the peak power. In this work, we evaluate the value of load forecasts for microgrid energy management under MPC. That is, we assess how a prediction error on load may influence the OPEX of the control system. We find that, in scenarios without considering the demand charge, the performance of MPC is robust to prediction errors.

However, when the demand charge is introduced, even minor prediction errors result in substantially suboptimal control outcomes. Moreover, an intriguing observation is that the value of information is asymmetric: load underestimates lead to much worse MPC performance than overestimates.

The underlying reason is that tracking the actual peak power can be misleading to the optimal control problem when it is unintentionally inflated in previous steps due to load underestimates. Accordingly, we propose a simple approach of overestimating the load forecast to filter the asymmetric impact of uncertainty. Numerical experiments demonstrate that this approach effectively alleviates the asymmetry issue and relieves the reliance on highly accurate forecasting models.