Stochastic Model Predictive Control: An Overview and Perspectives for Future Research
read more
Citations
Cautious Model Predictive Control Using Gaussian Process Regression
All you need to know about model predictive control for buildings
Reference and command governors for systems with constraints
Optimization under uncertainty in the era of big data and deep learning: When machine learning meets mathematical programming
Consensus control of stochastic multi-agent systems: a survey
References
Survey Constrained model predictive control: Stability and optimality
Monte Carlo Statistical Methods
Stochastic Finite Elements: A Spectral Approach
The Wiener--Askey Polynomial Chaos for Stochastic Differential Equations
Related Papers (5)
Survey Constrained model predictive control: Stability and optimality
Frequently Asked Questions (23)
Q2. What is the key challenge in SMPC of nonlinear systems?
A key challenge in SMPC of nonlinear systems is the efficient propagation of stochastic uncertainties through the system dynamics.
Q3. How do you guarantee the stability of a SMPC algorithm?
Closed-loop stability of these SMPC algorithms is commonly guaranteed by defining a negative drift condition via either a stability constraint or appropriate selection of the value function.
Q4. What is the affine parameterization of the feedback control law?
An affine parameterization of the feedback control law ( )i $r allows for obtaining a stochastic OCP that is convex in decision variables ih and .M ,i jIn [47], the value function is defined in terms of a linear function in disturbance-free states and control inputs, while polytopic constraints on inputs and state chance constraints are included in the stochastic OCP.
Q5. What is the simplest method of describing transition probability distributions of states?
For discrete-time nonlinear systems with additive disturbances, the Gaussian-mixture approximation [113] is used in [64] to describe the transition probability distributions of states in terms of weighted sums of a predetermined number of Gaussian distributions.
Q6. What are the disturbances and measurement noise that are unknown at the current and future time instants?
Rt nv! are disturbances and measurement noise that are unknown at the current and future time instants but have known probability distributions pw and pv , respectively; and f and h are (possibly nonlinear) Borel-measurable functions that describe the system dynamics and outputs, respectively.
Q7. What is the way to shape the distribution of states?
Complete characterization of probability distributions allows for shaping the distributions of states, as well as direct computation of joint chance constraints without conservative approximations.
Q8. What is the proposed approach for shaping the probability distribution of states?
For continuous-time stochastic nonlinear systems, a Lyapunov-based SNMPC approach is proposed in [35] for shaping the probability distribution of states.
Q9. What are the main schools of thought that are used to classify SMPC approaches?
SMPC approaches for linear systems are further categorized based on three main schools of thought: stochastictube approaches [42]–[46], approaches using affine parameterization of the control policy [47]–[56], and stochastic programming-based approaches [56]–[62].
Q10. What are the main problems of stochastic economic MPC?
Controlling periodic state trajectories typically observed in these control algorithms as well as establishing the closed-loop stability properties of stochastic economic MPC algorithms remain interesting open research problems.
Q11. What is the general treatment of the SMPC problem for linear systems with (addi?
The most general treatment of the SMPC problem for linear systems with (additive) unbounded stochastic disturbances, imperfect state information, and hard input bounds is provided in [52] where the SMPC algorithm of [51] is generalized to the case of output feedback control, while providing guarantees on recursive feasibility and stability.
Q12. What are the main shortcomings of the SMPC algorithm?
The main shortcomings of the SMPC algorithm presented in [73] are 1) an inability to consider saturation functions in the control policy to enable handling hard input bounds (as in [52]), 2) the conservatism associated with the Chebyshev–Cantelli inequality used for chance constraint approximation, and 3) nonconvexity of the algorithm.
Q13. What is the simplest way to solve the stochastic tube?
To achieve a computationally tractable formulation, the stochastic tube approaches use a feedback control law with a prestabilizing feedback gain.
Q14. What are the main features of the SMPC algorithm?
To summarize, affine-disturbance and affine-state parameterizations of a feedback control policy have been widely used to obtain convex SMPC algorithms.
Q15. What is the SMPC algorithm for bounded disturbances?
For the case of bounded disturbances, the recursive feasibility of an SMPC algorithm under an affine-disturbance feedback policy is established in [49] by using the concept of robust invariant sets (see [76]).
Q16. What is the reason for the no tion of feedback control laws?
The no - tion of affine-disturbance parameterization of feedback control laws originates from the fact that disturbance realizations and system states will be known at the future time instants.
Q17. What is the advantage of the offline computation of stochastic tubes?
The offline computation of stochastic tubes significantly improves the computational efficiency of the algorithm compared to stochastic tube approaches that use nested ellipsoidal sets [26] or nested layered tubes with variable polytopic cross sections [42] where the probability of transition between tubes and the probability of constraint violation within each tube are constrained.
Q18. What is the way to compute stochastic tubes?
These stochastic tubes can be computed offline with respect to the states that guarantee satisfaction of chance constraints and recursive feasibility.
Q19. What is the SMPC approach for linear systems with Gaussian additive disturbances?
Inspired by [16], [74], and [75], an SMPC approach is presented in [47] for linear systems with Gaussian additive disturbances [see (7a)], where the feedback control law ( )i $r is defined in terms of an affine function of past disturbances( , ) ,wx M w,i t i i j jij 11r h= + =- /with , MR R,i n i j n nu u w! !h # and : { , , }w ww i1 1f= - .
Q20. What is the need for systematic approaches for efficient and reliable uncertainty propagation through networked systems?
In addition, there is a need for systematic approaches for efficient and reliable uncertainty propagation through networked systems to address challenges associated with computational complexity of SMPC of integrated systems.
Q21. What is the main difference between the gPC framework and the nonlinear differential equations?
The gPC framework replaces the implicit mappings between uncertain variables/parameters and states (defined in termsA key challenge in SMPC of nonlinear systems is the efficient propagationof stochastic uncertainties through the system dynamics.40 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2016of nonlinear differential equations) with expansions of orthogonal polynomial basis functions; see [114] for a recent review on polynomial chaos.
Q22. What are the common approximations in SMPC?
various approximations are commonly made in most SMPC approaches (for example, approximations in uncertainty descriptions or the handling of chance constraints) to obtain tractable algorithms.
Q23. What are the open theoretical issues related to stochastic predictive control of interacting systems?
Stochastic predictive control of interacting systems gives rise to several open theoretical issues related to system-wide stability and control performance in the presence of probabilistic uncertainties.