| description abstract | Obtaining accurate components' characteristic maps has great significant for gas-turbine operating optimization and gas-path fault diagnosis. A common approach is to modify the original components' characteristic maps by introducing correction factors, which is known as performance adaptation. Among the existing methods, total average prediction error of measurable parameters (MPTAPE) at specified conditions is used to evaluate the adaptation accuracy. However, when a gas turbine undergoes a field operation, the performance parameters of each component are zonally distributed under the operating conditions. Under such circumstances, randomly selecting a few data points as the error control points (ECPs) for performance adaptation may lead to an inappropriate correction of the characteristic maps, further lowering the prediction accuracy of the simulation model. In this paper, a genetic-algorithm-based improved performance adaptation method is proposed, which provides improvements in two aspects. In one aspect, similarity between the components' predicted performance curves and the performance regression curves is used as the criterion with which to evaluate the adaptation accuracy. In the other aspect, in the process of off-design performance adaptation, the performance parameters at the design point are recalibrated. The improved method has been verified by using rig test data and applied to field data of a GE LM2500+SAC gas turbine. The comparison results show that the improved method can obtain more accurate and stable adaptation results, while the computational load can be significantly reduced. | |
