Prediction of Aero-Engine Exhaust Gas Temperature Based on Chaotic Neural Network Model

Based on the research of complexity and non-linearity of aero-engine exhaust gas temperature (EGT) system, a regularization chaotic prediction model was proposed to build short time forecasting model of EGT. In this paper, in order to gain the best parameter to improve the accuracy of the forecasting model, a simple search algorithm arithmetic was adopted. The simulation analysis shows that the proposed forecasting model obviously exceeded the traditional chaotic forecasting model on prediction accuracy. Therefore, this arithmetic is efficient and feasible for a short-term prediction of aero-engine exhaust gas temperature

Download Full-text

Study on the correction model of aero-engine exhaust gas temperature

2011 Prognostics and System Health Managment Confernece ◽

10.1109/phm.2011.5939492 ◽

2011 ◽

Author(s):

Zhi-quan Cui ◽

Lin Lin ◽

Shi-sheng Zhong ◽

Ti-chun Wang

Keyword(s):

Exhaust Gas ◽

Gas Temperature ◽

Engine Exhaust ◽

Correction Model ◽

Aero Engine ◽

Exhaust Gas Temperature

Download Full-text

Aero-engine exhaust gas temperature prediction based on LightGBM optimized by improved bat algorithm

Thermal Science ◽

10.2298/tsci200520246l ◽

2020 ◽

pp. 246-246

Author(s):

Dingzhe Li ◽

Jingbo Peng ◽

Dawei He

Keyword(s):

Goodness Of Fit ◽

Bat Algorithm ◽

Percentage Error ◽

Exhaust Gas ◽

Gas Temperature ◽

Engine Exhaust ◽

Data Set ◽

Aero Engine ◽

The Mean ◽

Exhaust Gas Temperature

In this paper, an aero-engine exhaust gas temperature (EGT) prediction model based on LightGBM optimized by the chaotic rate bat algorithm (CRBA) is proposed to monitor aero-engine performance effectively. By introducing chaotic rate, the convergence speed and precision of bat algorithm are im-proved, which CRBA is obtained. LightGBM is optimized by CRBA and it is used to predict EGT. Taking a type of aero-engine for example, some relevant performance parameters from the flight data measured by airborne sensors were selected as input variables and EGT as output variables. The data set is divided into training and test sets, and the CRBA-LightGBM model is trained and tested, and compared with ensemble algorithms such as RF, XGBoost, GBDT, LightGBM and BA-LightGBM. The results show that the mean absolute error (MAE) of this method in the prediction of EGT (after normalization) is 0.0065, the mean absolute percentage error (MAPE) is 0.77% and goodness of fit R2 has reached to 0.9469. The prediction effect of CRBA-LightGBM is better than other comparison algorithms and it is suitable for aero-engine condition monitoring.

Download Full-text

Prediction of Aero-Engine Exhaust Gas Temperature Based on Autoregressive Integrated Moving Average Model

Proceedings of the 2017 2nd International Conference on Automation, Mechanical Control and Computational Engineering (AMCCE 2017) ◽

10.2991/amcce-17.2017.83 ◽

2017 ◽

Author(s):

Hong Zhang ◽

Keqiang Dong

Keyword(s):

Moving Average ◽

Exhaust Gas ◽

Gas Temperature ◽

Average Model ◽

Engine Exhaust ◽

Autoregressive Integrated Moving Average ◽

Aero Engine ◽

Moving Average Model ◽

Exhaust Gas Temperature

Download Full-text

Aeroengine Exhaust Gas Temperature Prediction Using Process Neural Network with Time-Varying Threshold Functions

Applied Mechanics and Materials ◽

10.4028/www.scientific.net/amm.423-426.2341 ◽

2013 ◽

Vol 423-426 ◽

pp. 2341-2346

Author(s):

Gang Ding ◽

Xiong Wei Wang ◽

Da Lei

Keyword(s):

Neural Network ◽

Neural Network Model ◽

Orthogonal Basis ◽

Basis Functions ◽

Exhaust Gas ◽

Time Varying ◽

Gas Temperature ◽

Threshold Functions ◽

Exhaust Gas Temperature ◽

Threshold Process

To predict the aeroengine exhaust gas temperature (EGT) more precisely, a process neuron with time-varying threshold function is proposed in this paper, and then the time-varying threshold process neural network model comprised of the presented process neurons is used for EGT prediction. By introducing a group of appropriate orthogonal basis functions, the input functions, the weight functions and the threshold functions of the time-varying threshold process neural network can be expanded as linear combinations of the given orthogonal basis functions, thus to eliminate the integration operation, then to simplify the time aggregation operation. The corresponding learning algorithm is also presented, and the effectiveness of the time-varying threshold process neural network model is evaluated through the prediction of EGT series from practical aeroengine condition monitoring.

Download Full-text