Hierarchical Model Predictive Control Approach for CHP Unit Toward Flexible Load Services

With the wide application of renewable energy sources in power system, the advantages of flexible combined heat and power (CHP) unit on the stable generation are becoming prominent. In order to fully explore the operational flexibility of CHP unit to provide flexible load services, a practical hierarchical model predictive control (HMPC) approach is proposed in this article. First, to facilitate the CHP unit response to variable load demand, the advanced HMPC algorithm integrating the multiobjective model predictive control (MPC)-based guidance layer and reinforcement learning (RL) agent-based optimization layer is designed. Then, the initial control law reaching the Pareto optimization between the carbon emission and generation cost is attained in the guidance layer. As the control law is employed to accelerate the training of underlying RL agent, the economic and flexible load services can be provided by the unit. Moreover, with the continuous learning feature of RL agent, the control law is adjusted in real time to match the load demand uncertainties and model perturbations. Finally, the economic and flexible load services are successfully offered in wide load range by 350 MW CHP unit combined with the proposed HMPC algorithm. Especially, both the energy-saving and carbon trading indices of CHP unit have been improved at low-rated load, which confirms the effectiveness of the proposed HMPC approach.