Experimental Demonstration of Model Predictive Control in a - - PowerPoint PPT Presentation

Pengfei Li, Dapeng Li, Draguna Vrabie, Sorin Bengea, Stevo Mijanovic Presented by Pengfei Li United Technologies Research Center, USA July 14th, 2014

Experimental Demonstration of Model Predictive Control in a Medium-Sized Commercial Building

3rd International High Performance Buildings Conference at Purdue, West Lafayette, IN.

2

Advanced Building HVAC Controls Demonstration

Outline

• Motivation
• Overview of Building 101 HVAC system
• Control-oriented HVAC and zone model development
• Optimization-based controls design and problem formulation
• Advanced control deployment toolchain
• MPC demonstration results and analysis
• Summary

3

Advanced Building HVAC Controls Demonstration

Motivation

Specification Installation Startup Commissioning Building

• ccupation

System diagram

Project Mgmt & Installation 50% DDC Controls 12% Other Material & Warranty 18% Engineering & Commissioning 20%

• More than 70% cost of implementation of advanced

controls is involved in installation & commissioning

• Our goals:
• Reduce deployment and commissioning time

(reduce cost)

• Save energy
• Maintain occupant’s comfort

4

Advanced Building HVAC Controls Demonstration

Overview of Building 101 HVAC system

Primary system: Air-cooled chiller Secondary system: AHU w. DX evaporator Terminal system: VAVs w. reheat coil

Individual zone temp., RH and CO2 sensors (AHU3) Zones served by AHU3 AHU3 schematic

5

Advanced Building HVAC Controls Demonstration

Model Predictive Control for building HVAC supervisory control

MPC architecture Basic idea: receding prediction horizon

t+1 t+2 t+1+m t+1+N Predicted outputs Manipulated u(t+k) Inputs t t+1 t+m t+N

future past

Receding horizon

u*(t) u*(t+1)

Source: https://docs.google.com/viewer?a=v&pid=sites&srcid=ZGVmYXVsdGRvbWFpbnxtcGNsYWJvcmF0 b3J5fGd4OjYwMWNhMWE4OTBiOTYxYjI

Initialize thermal states Generate load forecasts Generate

• ptimized

set points Communicate set points to field controllers Sensor measurements Weather forecasts

• Supervisory (centralized): optimal control
• Local: set point tracking

Sequence of steps for on-line implementation

6

Advanced Building HVAC Controls Demonstration

System identification for dynamic thermal zone model

100 200 300 400 500 600 700 800 900 68 70 72 74 76 78 80 82 Samples (3 min.) Zone Temp. ( °F) Zone 5 Data Model 100 200 300 400 500 600 700 800 900 70 71 72 73 74 75 76 77 78 79 Samples (3 min.) Zone Temp. (°F) Zone 6 Data Model

Validation data vs. model (low-order state-space model)

1

7

Advanced Building HVAC Controls Demonstration

Data-driven control-oriented HVAC model

1

• , Δ

Δ

• AHU mixing box

DX compressor power AHU airflow AHU fan

5 10 15 20 2 4 6 8 10 12 14 16 18 Calculated Fan Power (kW) Measured Fan Power (kW) Fan Power (Validation Data)

Fan power validation Air flow validation

,

• VAV reheat coil

8

Advanced Building HVAC Controls Demonstration

Problem formulation

Slack variables 1

( , ) 1 ( ) ( , )

VAV

n cooling DA DA total fan sa reheating sa sai i DX boiler

P m T P aP m b c P m T COP 



  



   Objective: minimize energy + satisfy comfort

• Soft constraints on zone temperature lower and

upper bound

• Constraints on control authority

, min , max , ,

    

k k i z z k z i z

s s T T s T T Energy Comfort

9

Advanced Building HVAC Controls Demonstration

Control architecture and state estimation

( )

z

T k

ˆ( ) x k

z z

ˆ ˆ ( ) ... ( 1) T k T k N  

*( )

u k

A-priori state estimate : | | A-priori error covariance : | 1+Q A-priori output estimation error:

• |

Residual covariance:

|

Optimal Kalman gain:

|

• A-posteriori state estimation:

| A-posteriori estimation error covariance :

• |

Kalman Filter for state estimation at each time step

10

Advanced Building HVAC Controls Demonstration

Optimization-based control development & deployment toolchain

Optimization solver: IPOPT WebCTRL

Rapid advanced control deployment toolchain

11

Advanced Building HVAC Controls Demonstration

MPC demonstrates improved thermal comfort

09:00 12:00 15:00 70 72 74 76 78

Zone Temp. (Heuristic-Based Baseline)

Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 Z9 Z10 09:00 12:00 15:00 70 72 74 76 78

Zone Temp. (MPC) Time (occupied hours) ºF ºF

Cooling Setpoint Heating Setpoint Heating Setpoint Cooling Setpoint

Heuristic-based baseline MPC

50 100 150 200 65 70 75 80 85 Samples (3 min.)

Baseline (06/13/2013) MPC (09/13/2013)

Outdoor Air Temp. (ºF)

Similar OAT patterns

12

Advanced Building HVAC Controls Demonstration

MPC reduced compressor power with penalty of more fan power

09:00 12:00 15:00 50 100

Compressor Power (kW)

09:00 12:00 15:00 10 20

Fan power (kW)

09:00 12:00 15:00 50 100

Compressor + Fan power (kW)

Baseline MPC

Time (occupied hours)

Total power is reduced! More fan power Less compressor power

13

Advanced Building HVAC Controls Demonstration

MPC performance analysis based on similar OAT patterns

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

• 20
• 10

10 20 30 40 50 60 70 Test Days Energy Consumption Reductions (%) Energy Consumption Reductions from MPC (%)

50 100 150 200 65 70 75 80 85 Samples (3 min.)

Baseline (06/13/2013) MPC (09/13/2013)

Outdoor Air Temp. (ºF)

~33%

Similar OAT patterns

14

Advanced Building HVAC Controls Demonstration

Summary

• MPC implemented on a medium-size building HVAC system with

chiller, AHU and VAV boxes and demonstrated its benefits in energy saving and thermal comfort.

• Data-driven control-oriented HVAC model and low-order state space

model suitable for MPC execution and their effectiveness demonstrated

• Future work towards statistics-based performance analysis method

Acknowledgement

This work is funded by Consortium for Building Energy Innovation (formally known as Energy Efficient Buildings Hub), sponsored by the Department of Energy under Award Number DE-

• EE0004261. The authors are grateful to Hayden Reeve for managing the demonstration

activities as well as his technical insights and valuable inputs to improve the paper and Timothy Wagner for his project management and supervision. We also thank Ken Kozma from Radius Systems for his technical support on WebCTRL during our advanced controls

• demonstration. The authors also thank our collaborators from Purdue University and Virginia

Tech within the subtask 4.2.

• 2

15

Acknowledgement