A steady-state detection method based on Gaussian discriminant analysis for the on-line gas turbine process

In the unsteady-state process, the relationship among the state variables of the gas turbine cannot reflect the current performance level accurately. In order to monitor the condition deterioration of the gas turbine in the peak-shaving timely, an on-line steady-state detection model for gas turbine is proposed in this paper. First, the change rate of the out-power is presented as the key indicator of the difference between the steady-state and unsteady-state process, and the difference value of out-power at adjacent time is available as the characteristic test statistic. Second, the interval estimation of the statistical mean value is used to classify the historical data into the steady-state samples and the unsteady-state samples. Then, these classified samples are employed to train the detection model based on Gaussian discriminant analysis. Because the change rate of the out-power in the unsteady-state process is not a fixed value, the Gaussian mixture model is built for the unsteady-state process, and the initial value of the parameters and the number of sub-models are optimized by k-means++ and AIC evaluation criteria respectively. Finally, the detection model was verified by using the data of a real gas turbine for 10 days.