零维预测燃烧模型建模方法
Modeling method of the zero-dimensional predictive combustion model
胡登 1王贺春 1王彬彬 1王银燕 1杨传雷 1史明伟1
作者信息
- 1. 哈尔滨工程大学 动力与能源工程学院,黑龙江 哈尔滨 150001
- 折叠
摘要
为了解决神经网络建立的柴油机零维燃烧模型对稳态和动态工况预测能力不稳定问题,本文采用遗传算法对神经网络的初始权值、阈值进行综合优化,提出了遗传算法-神经网络算法.基于TBD620 型柴油机,通过稳态和瞬态试验获得运行参数和缸压数据,通过代数分析法结合遗传算法获得对应燃烧参数,最后分别利用遗传算法-神经网络算法和神经网络算法对燃烧模型进行构建并对比辨识结果.结果表明:与神经网络算法相比,遗传算法-神经网络算法构建的零维燃烧模型对应ϕ50 和IMEP预测值平均误差分别降低了 43.84%和 42.73%,遗传算法具有高效的权值、阈值寻优能力,模型具有更高的预测精度,泛化性更好,适用于柴油机零维燃烧模型研究.
Abstract
To solve the problem that the zero-dimensional predictive combustion model of diesel engines established using the neural network algorithm is unstable in predicting steady-state and dynamic operating conditions,this study uses the genetic algorithm to comprehensively optimize the initial weights and thresholds of the neural network and proposes a genetic-neural network algorithm.First,based on the TBD620 diesel engine,the operating parameters and cylinder pressure data were obtained by steady-state and transient tests.Then,the corresponding combustion parame-ters were obtained by algebraic analysis combined with the genetic algorithm.Finally,the combustion models were constructed using the genetic-neural network algorithm and the neural network algorithm,and the identification results were compared.Results showed that,compared with the neural network algorithm,the zero-dimensional predictive combustion model constructed by the genetic algorithm-neural network algorithm reduces the average error of the pre-dicted values of θ50 and IMEP by 43.84%and 42.73%,respectively.The genetic algorithm has a high-efficiency ca-pability to optimize the weight and threshold.The model has higher prediction accuracy and better generalization,which is suitable for the study of the zero-dimensional predictive combustion model of diesel engines.
关键词
柴油机/韦伯方程/零维燃烧模型/神经网络/遗传算法/生物柴油/代数分析法/遗传算法-神经网络算法Key words
diesel engine/Weber function/zero-dimensional predictive combustion model/neural network/genetic algorithm/biodiesel/algebraic analysis method/genetic-neural network algorithm引用本文复制引用
基金项目
国家自然科学基金项目(52171298)
博士研究生科研创新基金项目(3072023GIP0303)
出版年
2024
国家科技期刊平台
NSTL
万方数据