【作者】 陈达；

【导师】 王鹏； 张玮；

【作者基本信息】 大连理工大学 ， 土木工程（暖通）， 2025， 硕士

【摘要】 在“双碳”战略背景下,传统的内燃发动机汽车逐渐被电动汽车取代。增程式汽车兼顾传统油车长续航的特点,和电车行驶安静平顺的特点,因此深受市场认可。但是,汽车空调系统在冬天耗电量大,严重影响其实际续航里程,所以开发高效的增程器,以延长汽车续航至关重要。自由活塞直线发电机(FPLG)去除了曲柄连杆机构,具有可变压缩比,热效率高,清洁高效等特点,这些技术优势使其在车载增程器和分布式发电领域展现出独特的应用潜力。但是,也正因为去除了曲柄连杆机构,其活塞运行的稳定性受到了挑战。因此,本文针对FPLG活塞位移的稳定运行,构建了基于机器学习算法的模式识别系统和状态估计系统,并针对FPLG不同工况的运行特点,建立了多个模糊PID控制器以维持活塞稳定运行。首先,详细介绍了FPLG的运行机理,重点介绍了FPLG的三种运行阶段,包括振荡启动阶段,拖动燃烧阶段和稳定发电阶段,以及它们之间的联系。针对FPLG的运行数据,本研究进行了数据填充和归一化处理,并进行了可视化处理和数据集划分,以便后续机器学习算法的使用。其次,基于FPLG的活塞位移,活塞速度,活塞加速度,气缸压强和气缸温度这些历史数据,本文建立基于神经网络,决策树和SVM等多种机器学习算法的FPLG运行工况模式识别模型,用于实时识别FPLG的运行工况,以给出系统相应的控制策略。并利用贝叶斯算法优化各个模型的超参数,保障了方法的可靠性。最终,对FPLG不同频率下振荡启动阶段,拖动燃烧阶段和稳定发电阶段的识别准确率都能达到99%。基于以上数据,并结合多元线性回归,神经网络回归和决策树回归算法,本文还建立了FPLG状态估计模型,用于估计FPLG的运行状态,防止“失火”,“撞缸”等异常工况的发生。经过贝叶斯优化算法超参数后,FPLG各运行参数的R~2皆大于0.98。并建立了FPLG模式识别和状态估计的UI界面,帮助其理解本研究和做出更明智的决策。最后,针对FPLG活塞位移控制问题,本文设定了目标轨迹,建立了PID控制,多工况PID控制和模糊PID控制策略。针对振荡启动阶段,拖动燃烧阶段和稳定发电阶段FPLG具有不同的运行特点,本研究对应每个阶段分别建立PID控制器,并利用模糊算法实时优化PID的控制参数。研究结果显示,在模糊PID控制下,活塞位移控制的精确率能达到89%。控制策略实现了对FPLG活塞位移的精确控制,保障了FPLG稳定运行。更多还原

【Abstract】 Amid the global"Dual Carbon"strategy,traditional internal combustion engine vehicles are being progressively replaced by electric vehicles(EVs).Extended-range electric vehicles(EREVs),which combine the long-range advantages of conventional fuel-powered cars with the quiet,smooth operation of EVs,have gained significant market traction.However,the high energy consumption of automotive air conditioning systems during winter severely impacts the practical driving range of EVs,underscoring the urgency to develop efficient range extenders.The Free-Piston Linear Generator(FPLG),characterized by its elimination of the crankshaft linkage mechanism,variable compression ratio,high thermal efficiency,and clean energy conversion,has emerged as a promising technology for vehicular range extenders and distributed power generation.Nevertheless,the absence of the crankshaft mechanism poses challenges to piston motion stability.To address this,this study proposes a machine learning-based pattern recognition system and state estimation framework for FPLG operation,alongside multi-mode fuzzy PID controllers tailored to distinct operational phases,ensuring stable piston displacement.First,the operational mechanism of FPLG is comprehensively elucidated,focusing on its three critical phases:oscillation startup,drag combustion,and stable power generation,along with their interdependencies.Preprocessing of FPLG operational data—including data imputation,normalization,visualization,and dataset partitioning—was conducted to facilitate subsequent machine learning applications.Second,leveraging historical data on piston displacement,velocity,acceleration,cylinder pressure,and temperature,a multi-algorithm pattern recognition model was developed using neural networks,decision trees,and support vector machines(SVM).This model enables real-time identification of FPLG operational phases to guide adaptive control strategies.Bayesian optimization was employed to fine-tune hyperparameters,achieving 99%accuracy in phase recognition across varying frequencies.Furthermore,a state estimation model integrating multivariate linear regression,neural network regression,and decision tree regression was established to predict operational states and prevent anomalies such as misfire and piston collision.After Bayesian optimization,all operational parameters exhibited R~2values exceeding 0.98.An intuitive UI interface was developed to visualize pattern recognition and state estimation outcomes,enhancing decision-making clarity.Finally,for precise piston displacement control,target trajectories were defined,and three control strategies—PID,multi-mode PID,and fuzzy PID—were implemented.Distinct PID controllers were designed for oscillation startup,drag combustion,and stable power generation phases,with fuzzy logic dynamically optimizing PID parameters in real time.Experimental results demonstrate that the fuzzy PID controller achieves a displacement control precision rate of 89%,effectively stabilizing FPLG operation.更多还原