Research on real-time improvement method of on-board physical mechanism model of turboshaft engine based on neural network

Abstract On-board model is the basis of aero-engine advanced control algorithm. Physical mechanism model has high modeling accuracy, but its poor real-time performance limits its on-board application. Iterative solving algorithm is an important factor leading to the poor real-time performance of the physical mechanism model. Therefore, this paper proposes an improved real-time performance method of the on-board physical mechanism model of turboshaft engine based on neural network. This method establishes a solver model of engine flow balance and pressure balance equations based on neural network. It does not need to repeatedly calculate the inlet, compressor and other component models when solving the balance equation, thus improving the real-time performance of the model. Meanwhile, the powerful nonlinear fitting ability of neural network ensures the modeling accuracy. The simulation results show that compared with the conventional physical mechanism model, the calculation time of the proposed method is reduced by 70.8 %. The maximum modeling error of the key performance parameters at the steady-state condition is no more than 0.6 %, the relative error of the key parameters at the ground point is no more than 0.3 % under the large step of fuel and the variable flight altitude, and the relative error of the key parameters at the variable flight speed is no more than 0.2 %. The effectiveness of the proposed method is verified. The maximum modeling error is less than 0.5 %, which verifies the effectiveness of the proposed method.