Research On Technologies Of Intelligent Diagnosis And Health Maintenance For Gas-Steam Combined Cycle Unit

Health diagnosis of gas-steam combined cycle unit is a complex system engineering involving multi-disciplinary knowledge due to its changeable operating conditions,coupling among parameters,and complex fault transmission and evolution rules.Based on the systematical analysis of typical fault modes which affect the health status of the unit,combined with knowledge engineering,big data,artificial intelligence and other related theories and technologies,the research on the intelligent diagnosis and health maintenance technology of the unit driven by data and knowledge driven was carried out.Firstly,aiming at the complex relationship among multi-source fault information and inefficient query and reasoning of gas-steam combined cycle unit,a structured representation method of domain knowledge and data based on ontology theory and semantic web technology was proposed.Guided by fault tree analysis and fault mode and effect analysis,the typical fault knowledge of each structure and function level of the gas-steam combined cycle unit was analyzed systematically.On this basis,the concept and hierarchy of the health maintenance ontology of gas-steam combined cycle unit were proposed and the corresponding semantic web including boundary conditions,fault knowledge,and monitoring data was constructed,which realized the multi-granularity semantic modeling of fault knowledge.In addition,the semantic reasoning and query method of fault knowledge and data based on ontology was studied to improve the management efficiency and application effect of fault knowledge and data.Secondly,the health anomaly detection method based on conditional variational auto-encoder was studied for the variability of operating conditions and the complexity of health status information of gas-steam combined cycle unit.According to the operating characteristics and monitoring data pattern of the unit,the historical operating data cleaning process was proposed,including steady-state detection,working condition division,benchmark sample screening.Considering the influence of load and other boundary conditions on monitoring parameters,a multi-parameter fusion benchmark model under variable conditions was established based on the conditional variational auto-encoder to represent health status under complex operating conditions.The reconstruction probability was selected as the characteristic index of anomaly detection to measure the deviation degree between the actual and reference state,which improved the accuracy and sensitivity of anomaly detection under variable conditions of gas-steam combined cycle unit.Thirdly,the diagnosis decision method based on counterfactual reasoning was studied for the diagnosis of gas-steam combined cycle unit under complex correlation between fault and symptom.The relationship between fault and symptom was redefined from the perspective of causality.The structural causal model of fault was constructed by introducing hidden variables to express the uncertainty of diagnosis and combined with the fault knowledge obtained by domain ontology search.On this basis,the expressions and calculation models of sufficient cause and necessary cause based on counterfactual reasoning were proposed to characterize the causal explanation of fault to evidence quantitatively.The robustness and engineering applicability of the proposed method were evaluated in terms of sensitivity,reliability and interpretability.Finally,the architecture of the intelligent diagnosis system of gas-steam combined cycle unit based on digital twin was studied,and the overall design idea,specific functions,and key technologies were introduced in detail.An intelligent diagnosis system based on the gas-steam combined cycle unit of Zhongshan power plant of Yuedian Group was developed,so as to promote the technical achievements transformation and engineering application of the research on intelligent diagnosis.