A dynamic and steady-state aero-engine airborne model building method

Abstract

The invention discloses a dynamic and steady-state aero-engine airborne model building method, which belongs to the technical field of aero-engine control. First, the similarity criterion and the Taylor expansion principle are used to compress the sampled data, which greatly reduces the amount of sampled data and time; then, the dynamic data and steady-state data in the compressed sampled data are used to train the dynamic airborne model based on the sparse autoencoder and Based on the steady-state airborne model of the BP neural network, the corresponding quasi-steady-state judgment logic is finally set, and the sparse autoencoder dynamic model is used in the dynamic process, and the BP network steady-state model is used in the steady-state process. Compared with the prior art, the aero-engine airborne model constructed by the present invention has higher accuracy under dynamic and steady-state conditions, better real-time performance, and lower requirements on data storage capacity.

Description

technical field [0001] The invention relates to the technical field of aero-engine control, in particular to a method for constructing an airborne model of an aero-engine in a dynamic and steady state. Background technique [0002] The United States carried out the Integrated High Performance Turbo Engine Technology (IHPTET, Integrated High Performance Turbo Engine Technology) program between 1988 and 2005, aiming to improve engine performance, reduce engine weight, and increase engine thrust-to-weight ratio. Among them, the advanced turbine engine control adopts the model-based intelligent engine control (IEC, Intelligent Engine Control) technology, which changes the traditional sensor-based control mode and reflects the actual working state of the engine online through the onboard adaptive model , calculate engine performance parameters such as thrust, power, surge margin, etc. as feedback quantities, and form a control loop of direct performance parameters to fully tap th...

AI Technical Summary

Problems solved by technology

For example, using the neural network method to model the airborne model, first use the engine component-level model to sample points, and use the engine component-level model calculated by the N-R method to calculate a sample point for about 200-300 iterations, which takes about 5 seconds, and the sample input The quantity is 10 dimensions. If 10 points are required for each dimension to ensure that the sample is dense, it will take a total of 1.38×107 hours, which is obviously difficult to sample and even more difficult to train.

Images