IN-24-C008 - Fast Demand Response Based on Model Predictive Control for a Net Zero Energy Building in Cold Climate

Demand response is one of the key demand-side flexibilities in the recent situation of expanding power generation systems derived from renewable energy sources. In particular, fast demand response (Fast-DR) with control response in less than 10 minutes is considered effective for photovoltaic power generation systems, which are affected by weather conditions and have uncertainty in generation output. However, there are some control complexities in its utilization, to which model predictive control can be adapted. In this study, a Fast-DR mechanism based on model predictive control is constructed for facilities utilizing renewable energy and the results of its verification. The forecasting method using artificial neural network (ANN) was changed from a sequential prediction type to a period batch prediction type, and location encoding information of time was added to the input parameters. The introduction of this new method reduces the inference time to less than 25% of the conventional method, while maintaining the prediction accuracy. In addition, Lower Confidence Bound (LCB) likelihood comparison method, was introduced to escape the local optima of the e-constrained differential evolution (eDE) optimization algorithm to derive a heuristic solution within the control response time required by Fast-DR. Furthermore, in eDE-LCB, we proposed a decision process that can automatically adjust the balance between exploration and exploitation in setting the confidence parameter. This method makes it possible to obtain an optimal solution at an early number of generations by changing the behavior of the upper and lower individuals of the objective function.