基于随机学习的履带车辆多模混合驱动系配置与控制优化设计

Hybrid powertrains have been increasingly applied to tracked vehicles due to the high energy efficiency, emission reduction and powertrain electrification trend. The tracked vehicle's hybrid powertrain has developed as a typical complex and comprehensive mechatronic system and equipment. The increasing complexity is challenging the sizing and control design for tracked vehicle's hybrid powertrains and form the basic theoretical and design problem restricting the performance improvement. Aiming at the solution to the sizing and control design for the multi-mode hybrid powertrain of the tracked vehicle, this project explores the stochastic learning theory-based application with the focuses of markov decision process-based powertrain sizing and control combined optimal design and reinforcement learning-based multi-object and adaptive optimal control.The proposed content consists of: 1) Disclosing the multi-scale time/frequency model-based dynamic process and mechanism during the multi-mode transition and shifting under the stochastic loading; 2)Building the markov decision process-based powertrain sizing and control combined optimal design theory and methodology and disclosing the interaction between powertrain dynamic characteristic and control design under the stochastic driving cycles; 3)Exploring the reinforcement learning-based adaptive and multi-object optimal control design theory and applicable method; 4) Verifying and validating all through the high-performance numerical simulation and high-transient dynamic bench test.The proposal is helpful to improve the design theory and is valuable for the control application and extented adaptivity.

节能减排以及驱动系统电气化趋势使得混合驱动系在地面履带车辆上应用日益增多。该类混合驱动系已经发展成为一种典型的复杂机电综合系统和装备。其复杂性对混合驱动系配置与控制设计形成了新挑战，是制约我国该类装备性能发展的基础理论和关键技术。本项目以履带式多模混合动力驱动系配置与控制设计为突破点，从履带车辆功率负载随性刻画入手，建立基于随机学习的动态系统理论框架，重点解决：1）探明随机负载下、履带车辆多模混合驱动系时频多尺度模型及其动态切换和变迁机理；2）建立基于马尔科夫决策过程的多模混合驱动系配置与控制耦合优化设计理论，揭示随机工况下驱动系特性与系统控制的相互作用及规律；3）探索适用于履带车辆多模混合驱动系、基于强化学习的、多目标最优控制设计理论和应用方法，4）并通过高性能数值模拟和台架试验完成其理论及应用验证。对地面车辆复杂混合驱动系及其控制设计具有理论意义和应用价值。

摘要：本项目以履带式多模混合动力驱动系为研究对象，重点从履带车辆功率负载随性刻画、履带车驱动系配置与控制协同优化和混合驱动系自适应、多目标最优控制角度开展研究，主要解决履带式复杂混合动力驱动系配置优选和最优控制设计问题。项目重点研究了履带车辆功率需求随机性刻画和在线更新方法、履带车辆混合驱动系统自动化建模方法、履带车辆驱动系配置与控制协同优化算法和基于强化学习的能量管理策略，并完成了实验应用验证。履带车辆需求功率采用马尔科夫链模型进行建模，利用转移概率矩阵实现其随机性描述，通过引入Kullback Leibher偏异率对转移概率矩阵间的差异进行量化，并基于此设计了需求功率在线更新算法。通过建立混联式混合动力驱动系统耦合装置和汇流装置数学模型，探明了混合动力驱动系统自动化建模规律和步骤，并基于此进行编程实现，形成了混合驱动系统自动化建模程序。履带车辆驱动系配置与控制协同优化算法采用双层被控对象-控制器协同优化架构，通过采用DP算法迭代实现驱动系参数匹配和控制策略求解。基于强化学习的能量管理策略以需求功率马尔科夫链为基础，应用Q-Learning和Dyna两种强化学习算法实现了履带车辆能量管理最优控制，并基于转移概率矩阵在线更新算法设计了强化学习实时能量管理策略。