基于滑动时窗策略自适应优化支持向量机的航空发动机性能参数在线预测

Adaptively Optimized Support Vector Machine Online Prediction of Aeroengine Performance Parameters Based on Sliding Time Window Strategy

关键词:  航空发动机  在线预测模型  滑动时窗策略  遗传算法  支持向量机

Key words:  Aero-engine  Online prediction model  Sliding time window  Genetic algorithm  Support vector machine

摘要:

针对传统航空发动机性能参数时间序列预测方法存在的不足，提出了基于滑动时窗策略自适应优化支持向量机（Support Vector Machine，SVM）在线预测模型。该方法解决了训练样本动态适应性差的特点和老旧数据信息影响预测模型精度的问题。在该方法中，滑动时窗策略实时更新时窗数据训练样本，最终误差预报准则（Final Prediction Error，FPE）自适应地确定嵌入维数，遗传算法（Genetic Algorithm，GA）则实时自适应优化SVM建模参数。应用航空发动机排气温度偏差值（Delta Exhaust Gas Temperature，DEGT）数据进行实例验证，结果表明基于滑动时窗策略的自适应GA优化的SVM（GASVM）在线预测模型比传统的GASVM预测模型预测精度有显著提高。进一步分析了预测模型不同时窗宽度对短期预测精度的影响，展示了1步~10步预测的效果，结果表明在线预测模型在不同时窗宽度下短中期（5步以内）预测效果良好且稳定。文中提出的在线预测模型可用于航空发动机性能参数的预测，实现对航空发动机未来性能变化的预警。

Abstract:

In order to improve the traditional time series prediction method of aeroengine performance parameters, an adaptively optimized Support Vector Machine (SVM) online prediction model based on sliding time window strategy is proposed. It solves the shortcoming of poor dynamic adaptability of training samples and increases the accuracy of prediction model affected by outdated data. The training samples of time window data are updated in real time based on sliding time window strategy. The embedding dimension is determined adaptively by using the Final Prediction Error (FPE) criterion. Genetic algorithm (GA) is applied to optimize the hyperparameters of SVM prediction model adaptively in real time. The aero-engine performance parameters Delta Exhaust Gas Temperature (DEGT) data are used to verify the validity of the proposed method, and the results show that the prediction accuracy of the adaptively GA-optimized SVM (GASVM) online prediction model based on sliding time window strategy is significantly better than that of the traditional GASVM prediction model. The influence of different time window widths on short-term prediction accuracy is further analyzed. The 1~10 steps prediction results of this model are presented. The results show that the online prediction model has good and stable short-medium term (less than 5 steps) prediction result under different time window widths. The online prediction model proposed in this paper can be used to predict the performance parameters of aero-engines. It realizes the early warning of the future performance changes of aero-engines.