The Degradation Judgment Criterion And Data-Drive Model Optimization of Heavy-Duty Gas Turbine Based on Algorithm Structure Similarity

Due to the lack of specific structure and size of gas turbine, the characteristic maps of components cannot be obtained, which causes many difficulties on building an accurate gas turbine model and judging degradation with the increase of running time. In this research, the field data of Ban shan Power Plant from August 2019 to October 2021 is collected, which accounts 381233 sets of running data. Firstly, the steady data sorted by running time. Secondly, according to algorithm structure similarity, a data-drive T-model with running time is built based on the Elman neural network. Degradation of gas turbine is judged based on the degradation judgment criterion. Then, the field data is further selected and the S-model without running time is built, which has more accuracy than T-model. Turbine outlet temperature of S-model has the highest accuracy within 0.01. Furthermore, a novel method based on the field data restrictions and thermodynamic requirements is proposed to optimize S-model. After the optimization, the range of operation efficiency calculated by data-drive model is more reasonable. This method of building data-drive model and judging the degradation can be applied to other types of gas turbine, which can timely make operation adjustment or overhaul the unit