Embed Neural Network in Optimization Model: An Application of Demand Response Aggregation Under Information Asymmetry

Due to information asymmetry, analytical model may fail to keep high performance when some necessary information are absent. This paper provides a novel perspective to embed neural network (data-driven model) in optimization model (analytical model). The new-style model is then formulated and solved by a hybrid method with dual neural network and successive linear programming. Here, one of the neural network is built up to accelerate the derivative calculation of another network. Demand response aggregation problem is analyzed as an application, where price response feature and incremental model overcome the problem of information asymmetry. Case studies verify the correctness and efficiency of the proposed method, showing the potential for wider applications.