This study introduces a model for optimizing backpropagation with improved cuckoo search algo-rithm(ICS-BP),aimed at augmenting the accuracy and generalization capability of predicting compressor char-acteristic curves under boundary operating conditions.Focusing on the flow-pressure ratio predictions for an axial flow compressor,this paper evaluates and contrasts the performance of the proposed ICS-BP model against tradi-tional BP,BP optimized by genetic algorithm(GA-BP),cuckoo search optimized BP(CS-BP),radial basis function networks(RBF),extreme learning machines(ELM),and self-optimized support vector machines(MS-VM).The comparative analysis reveals that the ICS-BP model achieves the lowest relative prediction error,con-sistently under±1%,showcasing superior precision and robustness,along with the best generalization across var-ious conditions.This optimized model effectively addresses the common pitfalls of BP algorithms.While ELM and RBF models maintain commendable accuracy at high operational speeds,their performance deteriorates at the boundary operating points along the speed line,making them more suited for time-sensitive applications.Specifi-cally,for the 7F heavy-duty gas turbine and NASA74A compressor characteristic curve,the ICS-BP model's predictions of flow-pressure ratio characteristics exhibit high fidelity,and the average absolute percentage error of the overall prediction results is 1.129%and 0.590%,respectively,thereby affirming its superiority in charac-teristic curve prediction.
NETL
NSTL
万方数据
