Abstract:

It is the basis of engine performance optimization and real-time monitoring to establish the prediction model of aeroengine performance parameters which can meet the requirements of both accuracy and real-time. Extreme learning machine (ELM) has good adaptability to the complex nonlinear aeroengine system. In this paper, a brain storm optimization (BSO) algorithm is proposed to optimize the network parameters of ELM for improving its performance. The acceleration process data of the engine on the bench test are used as training and verification samples, and the performance parameter prediction model of the turboshaft engine acceleration process is obtained by regression identification using BSO-ELM algorithm. The results show that the prediction parameters of output parameter gas generator speed n_g, gas generator outlet temperature T₄ and pressure ratio π_c are better than the prediction models obtained by backpropagation neural network optimized with BSO algorithm and ELM method optimized with particle swarm optimization, which indicates the feasibility and superiority of the BSO-ELM prediction model. In the same simulation environment, the BSO-ELM algorithm can greatly improve the computational efficiency and improve the real-time performance of the prediction model.

Key words: turboshaft engine, acceleration process, brain storm optimization, extreme learning machine, model identification, performance parameters prediction