An LSTM-Driven Efficient Solution Algorithm of Economic Dispatch with Integral Constraints for Energy-Intensive Enterprise Microgrids

Accurate and efficient economic dispatch (ED) is important for the operation of energy-intensive microgrids. The traditional discrete modeling method cannot meet the actual needs of enterprises. The exact representation of device-specific operational dynamics requires the solution of continuous-time differential equations to determine their respective performance curves, which is generally computationally intractable. Motivated by these challenges, this paper proposes a data-driven efficient solution algorithm of ED with integral constraints. Specifically, first, to better capture the real-time characteristics of power generation, this paper formulates an energy-balanced-constrained ED model by introducing integral constraints into the dispatch formulation. Second, a pair of boundary constraints is applied to reformulate the continuous-time integral constraints into quadratic equations. Then, a data-driven efficient solution algorithm is developed, which requires only an initial point satisfying all inequality constraints. To enhance the quality of the initial point, a long-short-term memory (LSTM) neural network is trained to predict the warm-start point. The numerical results implemented in an energy-intensive microgrid system validate the feasibility and optimality of the proposed method.