10-04-2017, 08:56 PM
Prepared by:
Zhen Yu and Arthur Dexter
Abstract
Simulation based control schemes for a low-energy building system are introduced and compared in this paper. The simulation of a low-energy system is firstly constructed and a fast two-stage optimisation method is proposed to find the optimal control policy in short time. A Model Predictive Control (MPC) scheme and a Hierarchical Fuzzy Rule based Control (HFRC) scheme that is tuned online by a reinforcement learning (RL) agent are introduced. The MPC scheme runs the simulation online to predict the future behaviour in order to make longterm optimal decisions. On the other hand, the HFRC+RL scheme run the simulation offline to generate prior knowledge for the RL agent.
The performances of the different schemes are evaluated by comparing energy consumption, thermal comfort and computing time.
In the last few decades, computer simulation has enabled more detailed analysis of building energy system that could not be achieved with theoretical or experimental methods. Various information about the building, such as thermal response, energy consumption and even occupancy behaviour, can be acquired conveniently using simulation. This valuable information can be used for system design, equipment sizing, energy audit, fault identification, etc. Particularly, building simulation is very useful in terms of optimising the operation and optimal control of building energy systems. The use of the information about the controlled system is crucial for the controller design. Different ways of using this information exist. The building simulation can be used online as a model of the building energy system therefore a MPC scheme can be constructed based on it. Otherwise, it can be used offline, thus the optimisation results can be used to generate rules or performance maps to control the system online, or can provide good initial values or policies for online learning methods.
Introduction
Here, different ways of using the information provided by a building simulation are discussed and a low-energy building system is used as an example to demonstrate the properties of the different approaches. Firstly, a low-energy building system and its simulation are described. Then a fast twostage optimisation method is introduced which can find optimal control commands given an associated cost function. Two different ways of using the simulation are described and compared: the simulation and two-stage optimizer is used online as a MPC scheme; or the simulation is used offline to generate a HFRC controller and an accelerated RL method is introduced to tune the HFRC online. The discussion and conclusions are given at the end of this paper.
for more details, please visit
http://ibpsaproceedings/BS2009/BS09_1562_1568.pdf
Zhen Yu and Arthur Dexter
Abstract
Simulation based control schemes for a low-energy building system are introduced and compared in this paper. The simulation of a low-energy system is firstly constructed and a fast two-stage optimisation method is proposed to find the optimal control policy in short time. A Model Predictive Control (MPC) scheme and a Hierarchical Fuzzy Rule based Control (HFRC) scheme that is tuned online by a reinforcement learning (RL) agent are introduced. The MPC scheme runs the simulation online to predict the future behaviour in order to make longterm optimal decisions. On the other hand, the HFRC+RL scheme run the simulation offline to generate prior knowledge for the RL agent.
The performances of the different schemes are evaluated by comparing energy consumption, thermal comfort and computing time.
In the last few decades, computer simulation has enabled more detailed analysis of building energy system that could not be achieved with theoretical or experimental methods. Various information about the building, such as thermal response, energy consumption and even occupancy behaviour, can be acquired conveniently using simulation. This valuable information can be used for system design, equipment sizing, energy audit, fault identification, etc. Particularly, building simulation is very useful in terms of optimising the operation and optimal control of building energy systems. The use of the information about the controlled system is crucial for the controller design. Different ways of using this information exist. The building simulation can be used online as a model of the building energy system therefore a MPC scheme can be constructed based on it. Otherwise, it can be used offline, thus the optimisation results can be used to generate rules or performance maps to control the system online, or can provide good initial values or policies for online learning methods.
Introduction
Here, different ways of using the information provided by a building simulation are discussed and a low-energy building system is used as an example to demonstrate the properties of the different approaches. Firstly, a low-energy building system and its simulation are described. Then a fast twostage optimisation method is introduced which can find optimal control commands given an associated cost function. Two different ways of using the simulation are described and compared: the simulation and two-stage optimizer is used online as a MPC scheme; or the simulation is used offline to generate a HFRC controller and an accelerated RL method is introduced to tune the HFRC online. The discussion and conclusions are given at the end of this paper.
for more details, please visit
http://ibpsaproceedings/BS2009/BS09_1562_1568.pdf