自动驾驶汽车控制系统参数辨识与学习（英文）

针对自动驾驶汽车在行驶过程中会遇到随时间和交通环境变化的不确定性,须对自动驾驶系统参数进行辨识和学习。通过获取闭环状态下的系统行为和学习数据驱动下的系统参数,可极大提升系统辨识可靠性和稳定性。本文对自动驾驶车辆跟车的典型场景,综合考虑车辆控制和动力学特性,构建了基于车辆运行数据和置信椭圆的系统参数辨识的学习算法,提升系统辨识的可靠性和鲁棒性。结果表明提出的参数辨识和学习方法可准确估计车辆参数。

Parameter estimation and reliable fault detection of electric motors

Accurate model identification and fault detection are necessary for reliable motor control. Motor-characterizing parameters experience substantial changes due to aging, motor operating conditions, and faults. Consequently, motor parameters must be estimated accurately and reliably during operation. Based on enhanced model structures of electric motors that accommodate both normal and faulty modes, this paper introduces bias-corrected least-squares （LS） estimation algorithms that incorporate functions for correcting estimation bias, forgetting factors for capturing sudden faults, and recursive structures for efficient real-time implementation. Permanent magnet motors are used as a benchmark type for concrete algorithm development and evaluation. Algorithms are presented, their properties are established, and their accuracy and robustness are evaluated by simulation case studies under both normal operations and inter-turn winding faults. Implementation issues from different motor control schemes are also discussed.

Hybrid competitive Lotka–Volterra ecosystems with a hidden Markov chain

This work concerns Lotka–Volterra models that are formulated using stochastic differential equations with regime-switching.Distinct from the existing formulations,the Markov chain that models random environments is unobservable.For such partially observed systems,we use Wonham’s filter to estimate the Markov chain from the observable evolution of the population,and convert the original system to a completely observable one.We then show that the positive solution of our model does not explode in finite time with probability 1.Several properties including stochastic boundedness,finite moments,sample path continuity and large-time asymptotic behaviour are also obtained.Moreover,stochastic permanence,extinction and feedback controls are also investigated.

State of health based battery reconfiguration for improved energy efficiency

This paper analyzes the system-level state of health(SOH)and its dependence on the SOHs of its component battery modules.Due to stochastic natures of battery aging processes and their dependence on charge/discharge rate and depth,operating temperature,and environment conditions,capacities of battery modules decay unevenly and randomly.Based on estimated SOHs of battery modules during battery operation,we analyze how the SOH of the entire system deteriorates when battery modules age and become increasingly diverse in their capacities.A rigorous mathematical analysis of system-level capacity utilization is conducted.It is shown that for large battery strings with uniformly distributed capacities,the average string capacity approaches the minimum,implying an asymptotically near worst-case capacity utility without reorganization.It is demonstrated that the overall battery usable capacities can be more efficiently utilized to achieve extended operational ranges by using battery reconfiguration.An optimal regrouping algorithm is introduced.Analysis methods,simulation examples,and a case study using real-world battery data are presented.

Capital injections with negative surplus and delays: models and analysis

This work develops a new model to deal with the scenario that some companies can still run business even the surplus falls below zero temporarily. With such a scenario in mind, we allow the surplus process to continue in this negative-surplus period, during which capital injections will be ordered to assist in the stabilization of financial structure, until the financial status becomes severe enough to file bankruptcy. The capital injections will be modeled as impulse controls. By introducing the capital injections with time delays, optimal dividend payment and capital injection policies are considered. Using the dynamic programming approach, the value function obeys a quasi-variational inequality. With delays in capital injections, the company will be exposed to the risk of bankruptcy during the delay period. In addition, the optimal dividend payment and capital injection strategies should balance the expected cost of the possible capital injections and the time value of the delay periods. This gives rise to a stochastic control problem with mixed singular and delayed impulse controls. Under general assumptions, the lower capital injection barrier is determined, where bankruptcy occurs. The closed-form solution to the value function and corresponding optimal policies are obtained.

Gene regulatory networks driven by intrinsic noise with two-time scales： a stochastic averaging approach

This work focuses on gene regulatory networks driven by intrinsic noise with two-time scales. It uses a stochastic averaging approach for these systems to reduce complexity. Comparing with the traditional quasi-steady- state hypothesis （QSSH）, our approach uses stochastic averaging principle to treat the intrinsic noise coming from both the fast-changing variables and the slow-changing variables, which yields a more precise description of the underlying systems. To provide further insight, this paper also investigates a prototypical two-component activator-repressor genetic circuit model as an example. If all the protein productions were linear, these two methods would yield the same reduction result. However, if one of the protein productions is nonlinear, the stochastic averaging principle leads to a different reduction result from that of the traditional QSSH.