Duck shaped load curve supervision using demand response program with LSTM based load forecast

Abstract

A large volume of solar energy dissemination in a supply grid originates extreme variations in the load, resulting in a duck-form load arc that can cause stability issues. Also, the cost of energy consumption is found to vary between the off-peak and peak loads observed in the duck-shaped load curve. Accurate load forecast and demand response program is a key task for duck curve management in a distribution structure. Hence, this work proposes a Demand Response (DR) program using deep learning neural networks namely Long Short-Term Memory (LSTM). The proposed DR program is implemented in a modified 12 bus radial distribution network for duck curve management, where voltage stability is taken care of simultaneously minimizing the electricity cost in an energetic pricing environment. LSTM is used for forecasting the load and linear programming is used for load shedding. Therefore, this paper resolves the dual aims of flattening the duck-shaped load arc and minimizing electricity costs by combining them into a single objective function.