Abstract: Aviation fuel cost is one of the most important cost expenditures for airline companies, exerting huge impact on the profit margin. Trip fuel accounts for the highest proportion of aviation fuel; therefore, reasonable prediction and optimization of the trip fuel volume can effectively reduce the extra fuel carrying volume of the flight, thereby improving the fuel economy and effective commercial consumption of the fleet. However, some limitations exist in previous fuel prediction studies. For example, the data source, the applicable operation scenarios and aircraft models are relatively limited, making them difficult to popularize. More importantly, previous studies have not solved the problem of flight safety risk preference. Therefore, combined with the problem characteristics of flight trip fuel volume prediction, a model for flight trip fuel volume prediction based on the random forest algorithm considering risk preference adjustment is proposed. The model selects the characteristic items according to the flight plan, operation environment, aircraft configuration and performance, and adds the model evaluation index reflecting economy and safety on the basis of mathematical characteristics. Then, the model is fitted and tested with the actual operation data from the routes of an airline company. The experimental results show that the model can complete the prediction with expected accuracy and practical significance, and has better performance than the previous commonly used models. The research results have been applied in an airline company, providing important reference for flight plan making, dispatcher refueling, as well as energy saving and emission reduction analyses.

Key words: prediction, machine learning, trip fuel volume, risk preference, air transportation, aviation big data