Machine learning-based fuel flow rate prediction for Boeing 737-800 aircraft: A comprehensive approach across climb, cruise and descent flight phases

Abstract

This study presents a novel approach for unified Fuel Flow Rate (FFR) prediction across climb, cruise, and descent phases of a Boeing 737-800, using real flight data records, which is proven to provide superior accuracy compared to the existing models in the literature using the flight phase-limited approach. Unlike traditional mathematical models, the proposed framework employs machine learning techniques to accurately capture fuel consumption patterns. FFR is predicted based on 11 input features: altitude, Mach number, total air temperature, wind speed, rate of climb/descent, exhaust gas temperatures for engines 1 and 2, and engine power settings (N11C, N21C, N12C, N22C). Four models are implemented and compared: a two-layer Feed-Forward Network (FFN), a Nonlinear Autoregressive Exogenous (NARX), a Long-Short Term Memory (LSTM) model, and a Layer Recurrent Network (LRN). Performance is evaluated using mean absolute error, mean absolute percentage error, root mean square error, and R metrics, while generalizability is tested on eight completely unseen flight data. Among the tested models, the LRN delivers the most accurate results, proving to be highly effective for predicting fuel flow rate. Additionally, a comparative analysis with previous studies reveals that the proposed model achieves superior performance compared to existing methods across the considered flight phases.