基于高斯过程回归的高压共轨燃油系统多次喷射喷油量预测OA
Injection Quantity Prediction of High-Pressure Common Rail Systems under Multiple Injections Based on Gaussian Process Regression
高压共轨系统多次喷射下,预喷引发的压力波使得主喷油量产生波动,导致缸内燃烧效率降低、排放污染物增加.为实现多次喷射下喷油量的精确控制,本文提出一种基于高斯过程回归(Gaussian process regression,GPR)的多次喷射主喷油量数据驱动预测模型.首先,采用 D 最优设计和二阶响应面方法,以轨压、预喷脉宽、预-主喷间隔和主喷脉宽为因素建立主喷油量响应面模型,通过方差分析揭示 4个工况参数均属于极显著影响因素;然后,基于自主开发的多物理场耦合数字仿真平台,建立涵盖 528组工况的主喷油量样本集并进行训练模型;最后,系统对比零均值、常数、线性和二次多项式等不同均值函数以及 SEiso、RQard和 Matérn等不同核函数的组合形式,确定线性均值函数与二次有理核函数为最优配置.结果表明:在测试工况下,GPR模型所预测主喷油量的平均绝对百分比误差为 0.347％,决定系数 R2 为 0.999 6,不同主喷脉宽和预-主喷间隔下的预测结果均紧密分布于回归线附近;在非测试工况下,该模型仍能准确再现主喷油量随预-主喷间隔变化的波动规律,与前馈式神经网络(BP)、广义回归神经网络(GR)和支持向量机回归(SVR)模型相比具有更低的误差与更高的一致性.研究证明基于 GPR的多次喷射主喷油量数据驱动模型兼具较高预测精度与良好泛化能力,可为高压共轨系统多次喷射下的精确控制提供模型支撑.
In high-pressure common rail systems under multiple injections,the pressure waves induced by the pilot injection cause fluctuations in the main injection quantity,thereby reducing in-cylinder combustion efficiency and increasing pollutant emissions.A data-driven prediction model for the main injection quantity based on Gaussian process regression(GPR)was proposed to achieve the accurate control of injection quantity under multiple injections.First,D-optimal design and a second-order response surface method were employed to build a response surface model for the main injection quantity by utilizing rail pressure,pilot-injection pulse width,pilot-main injection interval,and main-injection pulse width as factors.Analysis of variance indicates that the four operating parameters all have extremely significant effects on the main injection quantity.Then,based on a self-developed multi-physics coupled digital simulation platform,a dataset containing 528 operating conditions was constructed,and the model was trained.On this basis,several combinations of mean functions(zero,constant,linear,and quadratic polynomials)and different kernel functions(SEiso,RQard,and Matérn)were systematically compared,and the linear mean function combined with the rational quadratic kernel function was identified as the optimal configuration.Results show that in test conditions,the mean absolute percentage error(MAPE)of main injection quantity predicted by the GPR-based model is 0.347％and the coefficient of determination R2 is 0.999 6,with predictions at different main-pulse widths and pilot-main intervals clustered closely around the regression line.In non-test conditions,the model can still accurately reproduce the fluctuation law of the main injection quantity with varying pilot-main intervals,and features lower error and higher consistency than BP,GR,and SVR models.The proposed GPR-based data-driven model under multiple injections is proved to have both high prediction accuracy and sound generalization capability,providing model support for the precise control of high-pressure common rail systems under multiple injections.
赵建辉;蓝中泽;卢相东;杨津韬
哈尔滨工程大学动力与能源工程学院,黑龙江 哈尔滨,150001哈尔滨工程大学动力与能源工程学院,黑龙江 哈尔滨,150001哈尔滨工程大学动力与能源工程学院,黑龙江 哈尔滨,150001哈尔滨工程大学动力与能源工程学院,黑龙江 哈尔滨,150001
能源科技
柴油机多次喷射GPR数据驱动模型喷油量高压共轨燃油系统
diesel enginemultiple injectionsGPR-based data-driven modelinjection quantityhigh-pressure common rail system
《西南交通大学学报》 2026 (2)
299-307,9
国家重点研发计划(2021YFE0114600)
评论