Abnormal Detection And Diagnosis Of Energy Efficiency Status Of Direct Air-Cooled Steam Turbine Unit

Under the background of the "dual carbon”strategy,the proportion of new energy installed capacity is increasing year by year.Steam turbine units are widely involved in peak regulation and frequency regulation,frequent load changes,and frequent occurrence of abnormal energy efficiency status,which seriously affects the economic and safe operation of units.Therefore,it is of great significance to carry out energy efficiency status monitoring and diagnostic research on steam turbine units for the sustainable and healthy economic development of coal-fired power units in my country.To this end,this paper takes a 300 MW direct air-cooled steam turbine unit as the research object,based on domain knowledge,combined with big data and artificial intelligence related technologies,carried out a research on the abnormal detection and diagnosis of the energy efficiency status of the steam turbine unit based on the dual drive of data and knowledge.Firstly,according to the hierarchical structure of "system-subsystem-equipment"combined with the analysis method of fault tree,the structure tree of the steam turbine unit and its auxiliary system is formed;The operational selection principle combed out the energy efficiency status indicators of the three subsystems of the steam turbine body system,heat recovery system,and cold end system from top to bottom,and formed the energy efficiency status index system of the steam turbine unit.Then,aiming at the problem that the historical operation data usually contains invalid,unsteady and a small amount of abnormal energy efficiency data,an automatic screening process for energy efficiency benchmark state samples is proposed.Using unit load and main steam parameters as steady-state characteristic variables,combined with sliding window detection technology to identify steady-state working conditions;taking unit load and ambient temperature as boundary conditions to divide the steady-state data samples into working conditions;using Gaussian mixture model combined with The AIC evaluation criterion performs multi-parameter synchronous clustering on the data samples under different working conditions,and selects the benchmark state samples according to certain screening criteria.Then,aiming at the diversity of working conditions of direct air-cooled steam turbine units and the complexity of energy efficiency status information,an abnormal detection method of energy efficiency status based on conditional variational autoencoder is proposed.Considering the impact of boundary conditions on the energy efficiency state of the unit,a multi-variable fusion energy efficiency benchmark model for variable operating conditions is constructed by using the filtered benchmark state samples combined with the conditional variational autoencoder,which realizes the characterization of the normal operating state of the unit under variable operating conditions;The reconstruction probability of the model is selected as the abnormal index of the energy efficiency state to detect the degree of deviation of the energy efficiency state of the unit,which improves the sensitivity of anomaly detection;combined with the principle of similarity,the concept of relative deviation degree of the index is proposed,and the abnormal energy efficiency state index is located according to the maximum deviation degree.Finally,aiming at the complex and difficult-to-trace reasons for the abnormality of the unit’s energy efficiency status,an in-depth analysis was conducted on the two types of reasons for abnormal operation and energy efficiency failures,and a diagnostic knowledge base for abnormal energy efficiency status of direct air-cooled steam turbine units was constructed in combination with domain knowledge and the actual situation of the unit;And on this basis,a general diagnostic model of energy efficiency status abnormality based on symbolic directed graph is constructed,which realizes the traceability and diagnosis of the subjective and objective reasons affecting the unit energy efficiency status.