Integration Scheme for Economic Load Dispatching and Optimization Control in Coal-Fired Plants Based on Sparse Gaussian Process Model and Deep Reinforcement Learning

The economic load dispatching and optimization control of coal-fired plants are of utmost importance to achieve carbon peaking and carbon neutrality for China. This article proposes an integrated scheme for combining the economic load dispatching and optimization control of power plants as a whole. In this scheme, both dynamic cost and static cost are employed to develop the model of plant-level economic load dispatching compared with the traditional method in that only static cost is considered. Specifically, the average coal consumption rate under steady-state operation conditions is applied to represent the static cost, and the sparse Gaussian process model that can effectively balance the modeling accuracy and the modeling cost is employed to establish the coal consumption rate model, faced with the big data characteristics of the historical operation data of coal-fired units. To properly evaluate dynamic cost, the Q function that comes from the process knowledge-assisted deep reinforcement learning control method is applied, according to the definition of the Q function and the designed reward function in the reinforcement learning. Moreover, different optimization strategies are designed for operators to choose according to actual dispatching requirements. Finally, the simulation cases that are similar to the actual dispatching scenarios are designed to verify the feasibility and effectiveness of the proposed integrated scheme. The simulation results show that compared with the traditional dispatching method, the proposed integration load dispatching and optimization control method in this paper can improve both the dynamic performance of transient conditions and the economic benefits of steady-state conditions and enhance the safety and economy of power plants.