A Simple Machine Learning Technique for Model Predictive Control
read more
Citations
Supervised Imitation Learning of Finite-Set Model Predictive Control Systems for Power Electronics
Modeling, diagnostics, optimization, and control of internal combustion engines via modern machine learning techniques: A review and future directions
Neural Network Based Model Predictive Controllers for Modular Multilevel Converters
Verification of Neural Networks Meets PLC Code: An LHC Cooling Tower Control System at CERN
Inductive biases and Self Supervised Learning in modelling a physical heating system.
References
Approximation by superpositions of a sigmoidal function
Survey Constrained model predictive control: Stability and optimality
Fundamentals of Power Electronics
Iterative Methods for Optimization
Approximation by Superpositions of a Sigmoidal Function
Related Papers (5)
Performance and safety of Bayesian model predictive control: Scalable model-based RL with guarantees
Frequently Asked Questions (9)
Q2. What future works have the authors mentioned in the paper "A simple machine learning technique for model predictive control" ?
Further studies are still needed to evaluate medium-scale nonlinear problems and investigate optimal state estimation.
Q3. What is the potential limitation of the here-proposed approach?
A potential limitationto this approach is ill-conditioning and loss of continuities of the characteristic equations when the value of the control horizon is too large.
Q4. What is the main advantage of the method?
Since the method intrinsically solves an optimal control problem with a final state constraint, it can be used to solve other regular optimal control problems (with a final time cost for instance).
Q5. What is the average model of a boost converter?
The average model of a boost converter is given bydi dt = −(1− u)v/L+ E/Ldv dt = (1− u)i/C − v/(RC) (18)where E is the input voltage, i is the inductance current, v is the output voltage, and u is the duty cycle of the converter acting as control input.
Q6. What is the drawback of the Levenberg-Marquardt algorithm?
In order to overcome this drawback, a low-discrepancy sequence of M samples {pi(T )}i=1,M , each of them belonging to a domain Vd of Rn, is generated.
Q7. What is the estimate of the MPC approach?
demonstrates the effectiveness of the proposed approach to deal with medium-scale MPC problems at least when the dynamics is linear and the cost function is convex and nonlinear.
Q8. What is the potential limitation of the proposed approach?
Small control horizons potentially can lead to large optimal controls which could be sometimes incompatible with physical constraints.
Q9. What is the solution to the MPC problem?
Here the MPC problem is defined by the following convex but nonlinear optimal problem with control horizon T = 0.25 defined as what follows:min u12 ∫ t+0.25 t (log(1 + 1 2 ‖x(τ)‖2) + ‖u(τ)‖2)dτ. (28)Supervised learning was performed from a Sobol sequence of only 1000 samples pi(t + T ) defined in hypercube [−0.15, 0.15]50.