Abstract:We present a three-step method to perform system identification and optimal control of non-linear systems. Our approach is mainly data driven and does not require active excitation of the system to perform system identification. In particular, it is designed for systems for which only historical data under closed-loop control are available and where historical control commands exhibit low variability. In a first step, simple simulation models of the system are built and run under various conditions. In a second step, a neural network architecture is extensively trained on the simulation outputs to learn the system physics, and retrained with historical data from the real system with stopping rules. These constraints avoid overfitting that arise by fitting closed-loop controlled systems. By doing so, we obtain one (or many) system model(s), represented by this architecture, and whose behaviour can be chosen to match more or less the real system. Finally, state-of-the-art reinforcement learning with a variant of domain randomization and distributed learning is used for optimal control of the system. We first illustrate the model identification strategy with a simple example, the pendulum with external torque. We then apply our method to model and optimize the control of a large building facility located in Switzerland. Simulation results demonstrate that this approach generates stable functional controllers which outperform on comfort and energy benchmark rule-based controllers.

An Identification Method Based on the De and Nlj Hybrid Algorithm and Its Application

Identification of DFIG parameters based on improved PSO algorithm

Parameter identification of PEMFC model based on hybrid adaptive differential evolution algorithm

Hybrid optimization algorithm for load model parameter identification

Neural network system identification based on levenberg-marquardt algorithm

Identification of Time-Varying Delay Systems Using Particle Swarm Optimization

A Hybrid Modeling Method Based on Mechanism Analysis,Identification and Rbf Neural Networks

Improved genetic programming for system identification

An IMC-PID Controller Tuning Strategy Based on the DE and NLJ Hybrid Algorithm

A neural network-based approach to hybrid systems identification for control

An Identification Method for Sequence Impedance Model of DFIG Based on the Joint of Knowledge Driving and Data Driving

Hybridizing Differential Evolution and Nelder-Mead Simplex Algorithm for Global Optimization

A Novel Hybrid PSO Identification Method Simultaneously Estimate Model Parameter and the Structure

Parameter Identification of Chaotic Systems by Hybrid Nelder-Mead Simplex Search and Differential Evolution Algorithm

Hybrid identification method for fractional-order nonlinear systems based on the multi-innovation principle

Identification and Control of Nonlinear Systems by a Dissimilation Particle Swarm Optimization-Based Elman Neural Network

Identification of Nonlinear System Based on a New Hybrid Gradient-Based PSO Algorithm

A Mechanism-Data Hybrid-Driven Framework for Identifying Dynamic Characteristic of Actual Chemical Processes

A hybrid learning method for system identification and optimal control

Dynamic Extreme Learning Machine Identification for Nonlinear System with Long Time Delay

A New Robust Identification Algorithm for Hammerstein-Like System Using Identification Error Structure