Abstract: The complexity of the compressor structure and the nonlinear characteristics of its operation pose challenges in predicting faults in gas turbine compressors. To enhance the fault prediction capability of gas turbine compressor, a novel approach is proposed using an improved lion swarm optimization (ILSO) to optimize deep extreme learning machine (DELM) for fault prediction. Through Pearson correlation analysis, the measurement points with high correlation with the early warning parameters are obtained, the ILSO-DELM prediction model is constructed, the absolute value of the early warning parameters under normal conditions is obtained, the threshold is determined by parameter estimation, and the operation of the compressor is indirectly judged according to whether the absolute value of the residual exceeds the early warning line. Based on the analysis of the operation data of a gas turbine power plant in Shanghai, the verification shows that the proposed method can give early warning of compressor faults, and the prediction accuracy is higher than that of the DELM model.

    Key words: compressor; deep extreme learning machine; lion swarm optimization algorithm; fault warning