随着发动机的使用，发动机的性能逐步下滑，为了判断发动机是否能够继续保证正常工作状态，避免故障的发生，需要对发动机性能的衰退程度进行预测评估；在发动机故障产生后，也需要判断发动机故障发生的原因并给出具体的解决方法。发动机性能趋势预测是发动机退化和故障诊断研究的基础。实现精确的预测不仅需要算法的精度，也需要多种类、高精度的大量数据。目前，用于航空发动机健康状态评估、剩余寿命预测、部件故障诊断与维修决策等的主要数据来源是传感器测量参数。近年来航空业飞速发展，海量化的航空发动机数据使得预测越来越依赖大数据技术。因此，国内外对大数据技术越来越重视。本文在大数据相关技术的基础上对航空发动机重要性能参数进行了预测。本文首先配置了3个节点的硬件环境，采用Hadoop架构搭建了大数据运维环境，采用Ambari顶层监控软件来管理、配置和监控Hadoop集群，选择基于HDFS的列式数据库Hbase存储飞行数据，并测试了Hbase的存储和查询性能。为了便于数据分析模块的集成和二次开发，结合软件的整体功能架构，用python语言开发了web前端显示界面和数据可视化模块：发动机大数据管理系统。其次，确定了航空发动机在起飞状态中一个稳定的点，将所有飞行参数数据中该点的发动机性能参数提取出来，分别按照时间排序建立新的数据集，并对数据进行去噪处理。先使用3 准则检测异常点并对缺失值进行处理，再使用二次指数方法对数据进行平滑。接着在对航空发动机性能参数进行预处理的基础上，提出使用多输入的LSTM神经网络对航空发动机性能参数进行预测，并研究了不同的超参数对LSTM性能的影响。在航空发动机性能参数数据集上的实验结果显示，LSTM算法比BP神经网络算法的预测效果更好。最后在SVR模型的基础上，先使用PSO算法进行参数寻优，再应用到SVR算法中，提出了一种基于PSO-SVR的航空发动机性能参数预测方法，并研究了不同训练样本对于预测结果的影响。在航空发动机性能参数数据集上的实验结果显示，PSO-SVR模型比标准SVR模型的预测效果更好。

With the use of the engine, the performance of the engine gradually declines. In order to judge whether the engine can continue to ensure the normal working state and avoid the occurrence of failures, it is necessary to predict and evaluate the decline of the engine performance; after the engine failure occurs, it is also necessary to determine the engine failure Causes and specific solutions. The prediction of engine performance trend is the basis of engine degradation and fault diagnosis research. Realizing accurate prediction requires not only the accuracy of the algorithm, but also a large amount of data of various types and high precision. At present, the main data sources for aero-engine health assessment, remaining life prediction, component fault diagnosis and maintenance decisions are sensor measurement parameters. In recent years, the aviation industry has developed rapidly, and the quantified aero-engine data makes prediction increasingly dependent on big data technology. Therefore, more and more attention is paid to big data technology at home and abroad. This paper predicts important performance parameters of aeroengines based on big data related technologies.This article first configured the hardware environment of 3 nodes, built a big data operation and maintenance environment using Hadoop architecture, used Ambari top-level monitoring software to manage, configure and monitor the Hadoop cluster, selected the HDFS-based column database Hbase to store flight data, tested Hbase storage and query performance. In order to facilitate the integration and secondary development of the data analysis module, combined with the overall functional architecture of the software, the web front-end display interface and data visualization module are developed in Python language: engine big data management system.Secondly, a stable point of the aero engine in the take-off state is determined, the engine performance parameters at that point in all flight parameter data are extracted, and a new data set is established according to time sorting respectively, and the data is denoised, first used 3 The criterion detects abnormal points and processes missing values, and then uses the quadratic index method to smooth the data.Then, on the basis of preprocessing the aeroengine performance parameters, it is proposed to use multi-input LSTM neural network to predict the aeroengine performance parameters, and the effects of different hyperparameters on LSTM performance are studied. Experimental results on the aeroengine performance parameter data set show that the LSTM algorithm has better prediction effect than the BP neural network algorithm.Finally, based on the SVR model, the PSO algorithm is used to optimize the parameters, and then applied to the SVR algorithm. A PSO-SVR-based aeroengine performance parameter prediction method is proposed, and the prediction results of different training samples for the prediction results are studied. influences. Experimental results on the aeroengine performance parameter data set show that the PSO-SVR model has a better prediction effect than the standard SVR model.