Multidimensional analysis and performance prediction of heavy-duty gas turbine based on actual operational data

This study aims to analyze and predict the degradation, safety and operational performance of a heavy-duty gas turbine based on actual operational data after running for a long time. The dataset consists of 288,039 operational records from the Ban Shan Power Plant. The data is screened using Pearson correlation coefficients and 2means clustering. Then, efficiency degradation of compressor and gas turbine is conducted by decoupling operation state from degradation. Next, Long short-term memory(LSTM) is employed to build data-driven Tmodel based on algorithm structure similarity. Meanwhile, the safety requirements containing the compressor surge and turbine inlet temperature boundary are determined, which are considered in operational performance analysis. Results show that abnormal operation data is identified during operational times 37, 38, 39, 40, 41, 42, and 248. After 254,341 minutes, the compressor efficiency degrades nearly 1%. The surge margin of a compressor is a binary function of the corrected air mass flow and inlet guide vane (IGV). The pressure ratio of the working margin moves downward by 0.11 for each 1 degrees increase in IGV. What's more, the T-model demonstrates high accuracy and can predict operational performance at high environmental temperatures, with the accuracy of turbine outlet temperature within 0.995. The unit is particularly sensitive to IGV, natural gas flow mass, compressor inlet temperature, and turbine outlet gauge pressure. When the relative humidity is 68%, the electrical power is the highest and overall efficiency is the best. Overall, this study's approach has significant potential for multidimensional analysis and performance prediction.