Research On Automatic Division Of Operating Condition In Thermal Power Process Based On Structured ODE-Net Modeling

Thermal process is an important link in thermal power production.In view of the multi-operating condition characteristics of thermal production process of thermal power units,this thesis studies automatic division of operating conditions and effective modeling of the subsystems to which operating conditions belong.Firstly,a structured ordinary differential equation network(ODE-Net)modeling method with the ability to represent the structure of the state equation is designed.Secondly,on this basis,an automatic division method of operating conditions based on the feedback of modeling results is proposed,which innovatively integrates the accurate division of system operating conditions and the accurate solution of the model.And it is verified in the automatic division of operating conditions and the accurate solution of the model in the thermal production process of thermal power units.The specific work and innovations are as follows:(1)The current research status of thermal process operating condition division,data-driven modeling and ordinary differential equation network are summarized.The deficiencies of the existing research methods and the problems to be solved urgently are pointed out.And the necessity of combining neural network with prior structural knowledge is analyzed.(2)Aiming at the problems that ODE-Net is not suitable for modeling of systems with external input and the data models established with deep learning tools are not solvable,the structured ODE-Net is designed by combining the system state space model and introducing prior structural knowledge,which expands the application range of ordinary differential equation networks.And the model output parameters are consistent with the linear part parameters of the mechanism model,which improves the interpretability of ODE-Net.Finally,the modeling ability of the structure is proved by the numerical theoretical experiments of linear system and affine nonlinear system as well as the verification application experiment of thermal power unit system.(3)Aiming at the automatic division of operating conditions,an automatic operating condition division method based on feedback of modeling results is proposed,which avoids the problem of inaccurate division caused by different clustering algorithms and modal types in traditional clusteringbased methods.The designed method combines the division of operating conditions and the solution of sub-models into one.The automatic division of the operating conditions of the system is completed,and the accurate solution of the corresponding sub-model is also realized.This method uses structured ODE-Net to solve the model of the pre-cut data segment.The mean squared error between the steady state output of the model and the true value within the sliding window is calculated.The deviation between the error value and the loss function value after model training is used as feedback to guide the division of operating conditions.Finally,in the constructed linear switching system and the superheated steam temperature system of thermal power units containing multiple typical loads,the effectiveness of the automatic division method of operating conditions with the introduction of feedback mechanism and the solution based on the structured ODE-Net modeling are verified.