[论文解读] Energy-Environment evaluation and Forecast of a Novel Regenerative turboshaft engine combine cycle with DNN application

Mahdi Alibeigi, Mohammadreza Sabzehali|arXiv (Cornell University)|Sep 24, 2022

Vehicle emissions and performance被引用 5

一句话总结

本研究提出了一种新型再生涡轮轴发动机联合循环，采用进气冷却技术，并以氢气为燃料，评估其能量与环境性能。采用每层625个神经元的两隐藏层深度神经网络（DNN），精确预测热效率与NOx排放（NO + NO2），在训练集与测试集上均实现R² > 0.99，且MSE、MAE与RMSD值极低。

ABSTRACT

In this integrated study, a turboshaft engine was evaluated by adding inlet air cooling and regenerative cooling based on energy-environment analysis. First, impacts of flight-Mach number, flight altitude, the compression ratio of compressor-1 in the main cycle, the turbine inlet temperature of turbine-1 in the main cycle, temperature fraction of turbine-2, the compression ratio of the accessory cycle, and inlet air temperature variation in inlet air cooling system on some functional performance parameters of Regenerative turboshaft engine cycle equipped with inlet air cooling system such as power-specific fuel consumption, Power output, thermal efficiency, and mass flow rate of Nitride oxides (NOx) including NO and NO2 has been investigated via using hydrogen as fuel working. Consequently, based on the analysis, a model was developed to predict the energy-environment performance of the Regenerative turboshaft engine cycle equipped with a cooling air cooling system based on a deep neural network (DNN) with 2 hidden layers with 625 neurons for each hidden layer. The model proposed to predict the amount of thermal efficiency and the mass flow rate of nitride oxide (NOx) containing NO and NO2. The results demonstrated the accuracy of the integrated DNN model with the proper amount of the MSE, MAE, and RMSD cost function for both predicted outputs to validate both testing and training data. Also, R and R^2 are noticeably calculated very close to 1 for both thermal Efficiency and NOx emission mass flow rate for both validations of thermal efficiency and NOx emission mass flow rate prediction values with its training and its testing data.

研究动机与目标

• 评估一种新型再生涡轮轴发动机循环（配备进气冷却）的能量与环境性能。
• 分析关键运行参数（如飞行马赫数、高度、压气机压比及涡轮进口温度）对发动机性能的影响。
• 开发深度神经网络（DNN）模型，用于预测发动机循环中的热效率与NOx排放（NO + NO2）。
• 通过多种统计指标在训练集与测试集上验证DNN模型的预测准确性。
• 通过集成能源-环境分析与机器学习预测，支持可持续航空发动机设计。

提出的方法

• 对配备进气冷却与氢燃料的再生涡轮轴发动机循环开展能源-环境分析。
• 改变关键设计与运行参数：飞行马赫数、飞行高度、压气机压比（压气机-1）、涡轮进口温度（涡轮-1）、涡轮-2的温度分数、附属循环压比及进气温度。
• 构建一个两隐藏层DNN，每层含625个神经元，用于预测热效率与NOx质量流量（NO + NO2）。
• 利用热力学分析生成的仿真数据对DNN进行训练与测试。
• 采用均方误差（MSE）、平均绝对误差（MAE）与均方根偏差（RMSD）作为损失函数，以优化模型性能。
• 通过决定系数（R²）与相关系数（R）评估模型在训练集与测试集上的泛化能力。

实验结果

研究问题

• RQ1 飞行马赫数与高度如何影响再生涡轮轴发动机循环的热效率与NOx排放？
• RQ2 压气机压比与涡轮进口温度对功率输出与燃油消耗有何影响？
• RQ3 进气冷却如何影响发动机循环的整体能源与环境性能？
• RQ4 深度神经网络能否在多种运行条件下准确预测热效率与NOx排放（NO + NO2）？
• RQ5 DNN模型在未见测试数据上的预测准确度与泛化能力如何？

主要发现

• DNN模型在训练集与测试集上对热效率与NOx排放质量流量预测的R²值均非常接近1。
• 两种输出的皮尔逊相关系数（R）也均非常接近1，表明预测值与实际值之间存在强烈的线性一致性。
• 模型在两种输出上均表现出高精度，MSE、MAE与RMSD值均极低。
• DNN有效捕捉了发动机运行参数与性能指标之间的复杂非线性关系。
• 模型在测试数据上的表现证实其在训练集之外的鲁棒性与泛化能力。
• 热力学分析与深度学习的结合，可实现对先进涡轮轴发动机能量与环境性能的精确预测。