Data-driven wide-load modeling and electricity-heat coordinated control for the supercritical combined heat and power unit

Improving the operational flexibility of combined heat and power (CHP) unit is an important means to promote the consumption of renewable energy and maintain grid stability. To this end, this paper integrates deep learning modeling, data-driven control and intelligent optimization algorithm to form a flexible scheme that adapts to the CHP unit operating in wide load and fast variable load conditions. Firstly, an improved crossformer structure is designed by integrating bidirectional gated recurrent unit and multi-receptive field convolution module to obtain the CHP unit model over a wide load range. Secondly, based on the established high-precision black box model, the model-free adaptive predictive control (MFAPC) strategy is applied to the load tracking control of the CHP unit to improve the response speed and accuracy of the unit operating in fast variable load conditions. Finally, the rime optimization algorithm is enhanced by combining the chaotic mechanism and the variational strategy for obtaining the optimal parameters of MFAPC. The objective function considering the dominant modes of electricity and heat is constructed, and the coordinated control of electricity and heat of the unit is achieved by minimizing the objective function. The simulation results show that the fitting accuracy of the improved crossformer neural network model is up to 99.18 %. When the electrical load command of automatic generation control changes frequently between 50 % and 90 % rated load, the proposed control strategy still has good tracking performance, which is favorable to the flexible operation of the unit.