Abstract: Aiming at the low prediction accuracy of the remaining life of high-pressure common rail injectors caused by improper selection of prediction model parameters,a Bidirectional Long Short Term Memory network (BiLSTM) based on Hierarchical Weighted Permutation Entropy (HWPE) and Improved Grasshopper Optimization Algorithm (IGOA) was proposed for the remaining life prediction method of high-pressure common rail injectors.Since HWPE can fully consider the high-frequency and low-frequency information of time series,the HWPE of the whole life cycle data was extracted to construct the Health Index (HI);to solve the problem of difficulty in selecting the optimal parameters of BiLSTM,an IGOA was proposed.By introducing the chaos strategy,the diversity and randomness of the initial population was enriched,the linear factor was reconstructed to enhance the ability of global search and local development,and the migration strategy was introduced to improve the quality of position update and adaptively obtain the optimal parameter combination of BiLSTM.Finally,the HWPE and the health index were used as the input and output of the IGOA-BiLSTM model for training and testing,and the output HI was fitted to the life degradation curve and the failure point was predicted to realize the remaining life prediction of the injector.Through comparative analysis with other commonly used methods,IGOA had better performance in parameter optimization,and the proposed remaining life prediction method had higher prediction accuracy.

Key words: residual life prediction, hierarchical weighted permutation entropy, improved grasshopper optimization algorithm, bi-directional long short term memory networks, high pressure common rail injectors

摘要： 针对预测模型参数选取不当而导致高压共轨喷油器剩余寿命预测精度较低的问题,提出一种基于层次加权排列熵(HWPE)与改进蝗虫优化算法(IGOA)优化双向长短时记忆网络(BiLSTM)的高压共轨喷油器剩余寿命预测方法。首先,由于HWPE能够充分考虑时间序列高频和低频信息,提取全寿命周期数据的HWPE构建健康度指标(HI);然后,针对BiLSTM最优参数选取困难的问题,提出了一种改进蝗虫优化算法,通过引入混沌策略以丰富初始种群的多样性和随机性,重构线性因子增强全局搜索和局部开发的能力,并引入迁徙策略提高位置更新的质量,自适应地获取BiLSTM最优参数组合;最后,将HWPE与健康度指标分别作为IGOA-BiLSTM模型的输入输出进行训练测试,将输出的HI拟合寿命退化曲线并预测失效点,实现喷油器的剩余寿命预测。通过与其它常用方法对比分析,IGOA在参数寻优方面性能更好,所提出剩余寿命预测方法具有更高的预测精度。

关键词: 剩余寿命预测, 层次加权排列熵, 改进蝗虫优化算法, 双向长短时记忆网络, 高压共轨喷油器