Neural-network-based optimization for economic dispatch of combined heat and power systems

Abstract One of the major research areas in combined heat and power (CHP) systems is optimal dispatch, which involves the minimization of the operating cost. In economic dispatch, it is important to use a model that accurately simulates the performance of the power and heat generation equipment. However, physics-based characteristic models require considerable time for the analysis, so it is hard to apply them to the optimization of dispatch schedules. This study introduced a neural network model, which was built based upon the simulation results of a physics-based model, to optimize a CHP system. The novel method was used to optimize the operation schedule of a system consisting of a gas turbine, steam turbine bottoming cycle, compressed air energy storage, and a boiler. The schedule was optimized to minimize the operation cost per day and according to the power and heating demand of users. The results showed that the introduction of the neural network reduced the time required for the system analysis by more than 7000 times. Furthermore, the optimization results confirmed the importance of accurately predicting the performance of each device using the physics-based model. This study contributes to the reduction in computation time and improvement of optimization accuracy.