Abstract: Aiming at the development goal of green civil aviation, it is of great significance to accurately calculate aircraft fuel consumption. The problem of aircraft fuel consumption was studied and analyzed. Based on Tensorflow deep learning framework, the prediction model of aircraft fuel consumption was established by using convolutional neural network. The prediction results were compared with the existing flight plan and multi-layer perceptron neural network models to verify the accuracy of the convolution neural network model, and the 10-fold cross validation was carried out to verify the robustness of the proposed model. The results show that the average error rate of the prediction results of the convolution neural network model is 5.26%, which is significantly better than 17.67% of the existing flight plan and 7.69% of the multi-layer perceptron model. The error rate variance of 10-fold cross validation of the proposed model is 0.16%. Therefore, the aircraft fuel consumption prediction model based on convolutional neural network has good accuracy and robustness, which can provide airlines with more accurate fuel weight carried by aircraft, avoid more fuel consumption, effectively reduce operating costs and achieve the energy-saving goal of green civil aviation.

Key words: traffic and transportation engineering; green civil aviation; aircraft fuel prediction; convolutional neural network; cross validation

摘要： 针对绿色民航的发展目标，准确预测飞机燃油消耗具有重要意义。对飞机油耗问题进行研究分析，基于TensorFlow深度学习框架，使用卷积神经网络，建立飞机燃油消耗预测模型。将预测结果分别与现有的飞行计划、多层感知机神经网络模型对比，验证卷积神经网络模型的准确性，并进行10折交叉验证，验证模型的鲁棒性。结果表明：卷积神经网络模型预测结果的平均误差率为5.26%，明显优于现有的飞行计划的17.67%和多层感知机模型的7.69%，模型10折交叉验证的误差率方差为0.16%；基于卷积神经网络的飞机燃油消耗预测模型具有很好的准确性和鲁棒性，能够为航空公司提供更加准确的飞机携带燃油量，避免产生更大燃油消耗，可有效降低运营成本，实现绿色民航的节能目标。

关键词: 交通运输工程；绿色民航；飞机燃油预测；卷积神经网络；交叉验证