Reducing the computational effort of optimal process controllers for continuous state spaces by using incremental learning and post-decision state formulations

Multistage optimization problems that are represented by Markov Decision Processes (MDPs) can be solved by the approach of Dynamic Programming (DP). However, in process control problems involving continuous state spaces, the classical DP formulation leads to computational intractability known as the ‘curse of dimensionality’. This issue can be overcome by the approach of Approximate Dynamic Programming (ADP) using simulation-based sampling in combination with value function approximators replacing the traditional value tables. In this paper, we investigate different approaches of ADP in the context of a deep cup drawing process, which is simulated by a finite element model. In applying ADP to the problem, Artificial Neural Networks (ANNs) are created as global parametric function approximators to represent the value functions as well as the state transitions. For each time step of the finite time horizon, time-indexed function approximations are built. We compare a classical DP approach to a backward ADP approach with batch learning of the ANNs and a forward ADP approach with incremental learning of the ANNs. In the batch learning mode, the ANNs are trained from temporary value tables constructed by exhaustive search backwards in time. In the incremental learning mode, on the other hand, the ANNs are initialized and then improved continually using data obtained by stochastic sampling of the simulation moving forward in time. For both learning modes, we obtain value function approximations with good performance. The cup deep drawing process under consideration is of medium model complexity and therefore allows us to apply all three methods and to perform a comparison with respect to the achieved efficiency and the associated computational effort as well as the decision behavior of the controllers.     

基于马氏决策过程模型的动态系统学习控制:研究前沿与展望

基于马氏决策过程(Markov decision process, MDP)的动态系统学习控制是近年来一个涉及机器学习、控制理论和运筹学等多个学科的交叉研究方向, 其主要目标是实现系统在模型复杂或者不确定等条件下基于数据驱动的多阶段优化控制. 本文对基于MDP的动态系统学习控制理论、算法与应用的发展前沿进行综述,重点讨论增强学习(Reinforcement learning, RL)与近似动态规划(Approximate dynamic programming, ADP)理论与方法的研究进展,其中包括时域差值学习理论、求解连续状态与行为空间MDP的值函数逼近方法、 直接策略搜索与近似策略迭代、自适应评价设计算法等,最后对相关研究领域的应用及发展趋势进行分析和探讨.     

Continuous-action reinforcement learning with fast policy search and adaptive basis function selection

As an important approach to solving complex sequential decision problems, reinforcement learning (RL) has been widely studied in the community of artificial intelligence and machine learning. However, the generalization ability of RL is still an open problem and it is difficult for existing RL algorithms to solve Markov decision problems (MDPs) with both continuous state and action spaces. In this paper, a novel RL approach with fast policy search and adaptive basis function selection, which is called Continuous-action Approximate Policy Iteration (CAPI), is proposed for RL in MDPs with both continuous state and action spaces. In CAPI, based on the value functions estimated by temporal-difference learning, a fast policy search technique is suggested to search for optimal actions in continuous spaces, which is computationally efficient and easy to implement. To improve the generalization ability and learning efficiency of CAPI, two adaptive basis function selection methods are developed so that sparse approximation of value functions can be obtained efficiently both for linear function approximators and kernel machines. Simulation results on benchmark learning control tasks with continuous state and action spaces show that the proposed approach not only can converge to a near-optimal policy in a few iterations but also can obtain comparable or even better performance than Sarsa-learning, and previous approximate policy iteration methods such as LSPI and KLSPI.     

Kernel-based reinforcement learning in average-cost problems

Reinforcement learning (RL) is concerned with the identification of optimal controls in Markov decision processes (MDPs) where no explicit model of the transition probabilities is available. We propose a class of RL algorithms which always produces stable estimates of the value function. In detail, we use "local averaging" methods to construct an approximate dynamic programming (ADP) algorithm. Nearest-neighbor regression, grid-based approximations, and trees can all be used as the basis of this approximation. We provide a thorough theoretical analysis of this approach and we demonstrate that ADP converges to a unique approximation in continuous-state average-cost MDPs. In addition, we prove that our method is consistent in the sense that an optimal approximate strategy is identified asymptotically. With regard to a practical implementation, we suggest a reduction of ADP to standard dynamic programming in an artificial finite-state MDP.     

EFFICIENT ABSTRACTION SELECTION IN REINFORCEMENT LEARNING

This article addresses reinforcement learning problems based on factored Markov decision processes (MDPs) in which the agent must choose among a set of candidate abstractions, each build up from a different combination of state components. We present and evaluate a new approach that can perform effective abstraction selection that is more resource‐efficient and/or more general than existing approaches. The core of the approach is to make selection of an abstraction part of the learning agent's decision‐making process by augmenting the agent's action space with internal actions that select the abstraction it uses. We prove that under certain conditions this approach results in a derived MDP whose solution yields both the optimal abstraction for the original MDP and the optimal policy under that abstraction. We examine our approach in three domains of increasing complexity: contextual bandit problems, episodic MDPs, and general MDPs with context‐specific structure. © 2013 Wiley Periodicals, Inc.     

Policy iteration based feedback control

It is well known that stochastic control systems can be viewed as Markov decision processes (MDPs) with continuous state spaces. In this paper, we propose to apply the policy iteration approach in MDPs to the optimal control problem of stochastic systems. We first provide an optimality equation based on performance potentials and develop a policy iteration procedure. Then we apply policy iteration to the jump linear quadratic problem and obtain the coupled Riccati equations for their optimal solutions. The approach is applicable to linear as well as nonlinear systems and can be implemented on-line on real world systems without identifying all the system structure and parameters.     

Reducing reinforcement learning to KWIK online regression

One of the key problems in reinforcement learning (RL) is balancing exploration and exploitation. Another is learning and acting in large Markov decision processes (MDPs) where compact function approximation has to be used. This paper introduces REKWIRE, a provably efficient, model-free algorithm for finite-horizon RL problems with value function approximation (VFA) that addresses the exploration-exploitation tradeoff in a principled way. The crucial element of this algorithm is a reduction of RL to online regression in the recently proposed KWIK learning model. We show that, if the KWIK online regression problem can be solved efficiently, then the sample complexity of exploration of REKWIRE is polynomial. Therefore, the reduction suggests a new and sound direction to tackle general RL problems. The efficiency of our algorithm is verified on a set of proof-of-concept experiments where popular, ad hoc exploration approaches fail.     

REINFORCEMENT LEARNING FOR POMDP USING STATE CLASSIFICATION

Reinforcement learning (RL) has been widely used to solve problems with a little feedback from environment. Q learning can solve Markov decision processes (MDPs) quite well. For partially observable Markov decision processes (POMDPs), a recurrent neural network (RNN) can be used to approximate Q values. However, learning time for these problems is typically very long. We present a new combination of RL and RNN to find a good policy for POMDPs in a shorter learning time. This method contains two phases: firstly, state space is divided into two groups (fully observable state group and hidden state group); secondly, a Q value table is used to store values of fully observable states and an RNN is used to approximate values for hidden states. Results of experiments in two grid world problems show that the proposed method enables an agent to acquire a policy with better learning performance compared to the method using only a RNN.     

An empirical study of policy convergence in Markov decision process value iteration

The value iteration algorithm is a well-known technique for generating solutions to discounted Markov decision process (MDP) models. Although simple to implement, the approach is nevertheless limited in situations where many Markov decision processes must be solved, such as in real-time state-based control problems or in simulation/optimization problems, because of the potentially large number of iterations required for the value function to converge to an ε-optimal solution. Experimental results suggest, however, that the sequence of solution policies associated with each iteration of the algorithm converges much more rapidly than does the value function. This behavior has significant implications for designing solution approaches for MDPs, yet it has not been explicitly characterized in the literature nor generated significant discussion. This paper seeks to generate such discussion by providing comparative empirical convergence results and exploring several predictors that allow estimation of policy convergence speed based on existing MDP parameters.     

Recursive Learning Automata Approach to Markov Decision Processes

In this note, we present a sampling algorithm, called recursive automata sampling algorithm (RASA), for control of finite-horizon Markov decision processes (MDPs). By extending in a recursive manner Sastry's learning automata pursuit algorithm designed for solving nonsequential stochastic optimization problems, RASA returns an estimate of both the optimal action from a given state and the corresponding optimal value. Based on the finite-time analysis of the pursuit algorithm by Rajaraman and Sastry, we provide an analysis for the finite-time behavior of RASA. Specifically, for a given initial state, we derive the following probability bounds as a function of the number of samples: 1) a lower bound on the probability that RASA will sample the optimal action and 2) an upper bound on the probability that the deviation between the true optimal value and the RASA estimate exceeds a given error.     

平均和折扣准则MDP基于TD(0)学习的统一NDP方法

为适应实际大规模M arkov系统的需要,讨论M arkov决策过程(MDP)基于仿真的学习优化问题.根据定义式,建立性能势在平均和折扣性能准则下统一的即时差分公式,并利用一个神经元网络来表示性能势的估计值,导出参数TD(0)学习公式和算法,进行逼近策略评估;然后,根据性能势的逼近值,通过逼近策略迭代来实现两种准则下统一的神经元动态规划(neuro-dynam ic programm ing,NDP)优化方法.研究结果适用于半M arkov决策过程,并通过一个数值例子,说明了文中的神经元策略迭代算法对两种准则都适用,验证了平均问题是折扣问题当折扣因子趋近于零时的极限情况.     

Interactive visualization for testing Markov Decision Processes: MDPVIS

Markov Decision Processes (MDPs) are a formulation for optimization problems in sequential decision making. Solving MDPs often requires implementing a simulator for optimization algorithms to invoke when updating decision making rules known as policies. The combination of simulator and optimizer are subject to failures of specification, implementation, integration, and optimization that may produce invalid policies. We present these failures as queries for a visual analytic system (MDPVIS). MDPVIS addresses three visualization research gaps. First, the data acquisition gap is addressed through a general simulator-visualization interface. Second, the data analysis gap is addressed through a generalized MDP information visualization. Finally, the cognition gap is addressed by exposing model components to the user. MDPVIS generalizes a visualization for wildfire management. We use that problem to illustrate MDPVIS and show the visualization's generality by connecting it to two reinforcement learning frameworks that implement many different MDPs of interest in the research community.     

The control of a two-level Markov decision process by time aggregation

The solution of Markov Decision Processes (MDPs) often relies on special properties of the processes. For two-level MDPs, the difference in the rates of state changes of the upper and lower levels has led to limiting or approximate solutions of such problems. In this paper, we solve a two-level MDP without making any assumption on the rates of state changes of the two levels. We first show that such a two-level MDP is a non-standard one where the optimal actions of different states can be related to each other. Then we give assumptions (conditions) under which such a specially constrained MDP can be solved by policy iteration. We further show that the computational effort can be reduced by decomposing the MDP. A two-level MDP with M upper-level states can be decomposed into one MDP for the upper level and M to M(M-1) MDPs for the lower level, depending on the structure of the two-level MDP. The upper-level MDP is solved by time aggregation, a technique introduced in a recent paper [Cao, X.-R., Ren, Z. Y., Bhatnagar, S., Fu, M., & Marcus, S. (2002). A time aggregation approach to Markov decision processes. Automatica, 38(6), 929-943.], and the lower-level MDPs are solved by embedded Markov chains.     

Least squares approximate policy iteration for learning bid prices in choice-based revenue management

We consider the revenue management problem of capacity control under customer choice behavior. An exact solution of the underlying stochastic dynamic program is difficult because of the multi-dimensional state space and, thus, approximate dynamic programming (ADP) techniques are widely used. The key idea of ADP is to encode the multi-dimensional state space by a small number of basis functions, often leading to a parametric approximation of the dynamic program’s value function. In general, two classes of ADP techniques for learning value function approximations exist: mathematical programming and simulation. So far, the literature on capacity control largely focuses on the first class.In this paper, we develop a least squares approximate policy iteration (API) approach which belongs to the second class. Thereby, we suggest value function approximations that are linear in the parameters, and we estimate the parameters via linear least squares regression. Exploiting both exact and heuristic knowledge from the value function, we enforce structural constraints on the parameters to facilitate learning a good policy. We perform an extensive simulation study to investigate the performance of our approach. The results show that it is able to obtain competitive revenues compared to and often outperforms state-of-the-art capacity control methods in reasonable computational time. Depending on the scarcity of capacity and the point in time, revenue improvements of around 1% or more can be observed. Furthermore, the proposed approach contributes to simulation-based ADP, bringing forth research on numerically estimating piecewise linear value function approximations and their application in revenue management environments.     

A unified approach to Markov decision problems and performance sensitivity analysis with discounted and average criteria: multichain cases

We propose a unified framework to Markov decision problems and performance sensitivity analysis for multichain Markov processes with both discounted and average-cost performance criteria. With the fundamental concept of performance potentials, we derive both performance-gradient and performance-difference formulas, which play the central role in performance optimization. The standard policy iteration algorithms for both discounted- and average-reward MDPs can be established using the performance-difference formulas in a simple and intuitive way; and the performance-gradient formulas together with stochastic approximation may lead to new optimization schemes. This sensitivity-based point of view of performance optimization provides some insights that link perturbation analysis, Markov decision processes, and reinforcement learning together. The research is an extension of the previous work on ergodic Markov chains (Cao, Automatica 36 (2000) 771).     

From Perturbation Analysis to Markov Decision Processes and Reinforcement Learning

The goals of perturbation analysis (PA), Markov decision processes (MDPs), and reinforcement learning (RL) are common: to make decisions to improve the system performance based on the information obtained by analyzing the current system behavior. In this paper, we study the relations among these closely related fields. We show that MDP solutions can be derived naturally from performance sensitivity analysis provided by PA. Performance potential plays an important role in both PA and MDPs; it also offers a clear intuitive interpretation for many results. Reinforcement learning, TD(), neuro-dynamic programming, etc., are efficient ways of estimating the performance potentials and related quantities based on sample paths. The sensitivity point of view of PA, MDP, and RL brings in some new insight to the area of learning and optimization. In particular, gradient-based optimization can be applied to parameterized systems with large state spaces, and gradient-based policy iteration can be applied to some nonstandard MDPs such as systems with correlated actions, etc. Potential-based on-line approaches and their advantages are also discussed.     

Semi-Markov decision problems and performance sensitivity analysis

Recent research indicates that Markov decision processes (MDPs) can be viewed from a sensitivity point of view; and the perturbation analysis (PA), MDPs, and reinforcement learning (RL) are three closely related areas in optimization of discrete-event dynamic systems that can be modeled as Markov processes. The goal of this paper is two-fold. First, we develop the PA theory for semi-Markov processes (SMPs); and then we extend the aforementioned results about the relation among PA, MDP, and RL to SMPs. In particular, we show that performance sensitivity formulas and policy iteration algorithms of semi-Markov decision processes can be derived based on the performance potential and realization matrix. Both the long-run average and discounted-cost problems are considered. This approach provides a unified framework for both problems, and the long-run average problem corresponds to the discounted factor being zero. The results indicate that performance sensitivities and optimization depend only on first-order statistics. Single sample path-based implementations are discussed.     

Incremental Value Iteration for Time-Aggregated Markov-Decision Processes

A value iteration algorithm for time-aggregated Markov-decision processes (MDPs) is developed to solve problems with large state spaces. The algorithm is based on a novel approach which solves a time aggregated MDP by incrementally solving a set of standard MDPs. Therefore, the algorithm converges under the same assumption as standard value iteration. Such assumption is much weaker than that required by the existing time aggregated value iteration algorithm. The algorithms developed in this paper are also applicable to MDPs with fractional costs.     

The $n$th-Order Bias Optimality for Multichain Markov Decision Processes

In this paper, we propose a new approach to the theory of finite multichain Markov decision processes (MDPs) with different performance optimization criteria. We first propose the concept of nth-order bias; then, using the average reward and bias difference formulas derived in this paper, we develop an optimization theory for finite MDPs that covers a complete spectrum from average optimality, bias optimality, to all high-order bias optimality, in a unified way. The approach is simple, direct, natural, and intuitive; it depends neither on Laurent series expansion nor on discounted MDPs. We also propose one-phase policy iteration algorithms for bias and high-order bias optimal policies, which are more efficient than the two-phase algorithms in the literature. Furthermore, we derive high-order bias optimality equations. This research is a part of our effort in developing sensitivity-based learning and optimization theory.     

A robust approach to robot team learning

The paper achieves two outcomes. First, it summarizes previous work on concurrent Markov decision processes (CMDPs) currently demonstrated for use with multi-agent foraging problems. When using CMDPs, each agent models the environment using two Markov decision process (MDP). The two MDPs characterize a multi-agent foraging problem by modeling both a single-agent foraging problem, and multi-agent task allocation problem, for each agent. Second, the paper studies the effects of state uncertainty on a heterogeneous robot team that utilizes the aforementioned CMDP modelling approach. Furthermore, the paper presents a method to maintain performance despite state uncertainty. The resulting robust concurrent individual and social learning (RCISL) mechanism leads to an enhanced team learning behaviour despite state uncertainty. The paper analyzes the performance of the concurrent individual and social learning mechanism with and without a particle filter for a heterogeneous foraging scenario. The RCISL mechanism confers statistically significant performance improvements over the CISL mechanism.