Optimization on DRenabled Thermostat Control

1
Optimization on DR-enabled Thermostat Control

• Xue Chen
• Therese Peffer
• Jaehwi Jang
• Prof. David Auslander
• Prof. Edward Arens

February 12th, 2007
2
Overview

• DR Thermostat/Control Group overview
• Analysis on standard thermostat control
• System structure
• Decision making process
• Learning of Occupancy schedule pattern
• Demonstration

3
Goal Demand Response-Enabled Technology
Utility
Temperature sensors
Price signal
Power sensor
Price Indicator
Electricity used
Power actuators
Occupancy sensors
4
DREAM Goals

• Adaptive adaptive comfort model considering
  users schedule, outdoor and indoor temperature
  and RH
• Optimizing tradeoff between cost and comfort
• Learning thermal preference, schedule, house
  identification
• Educating Informing occupant of price and energy
  consumption

5
Tstat Group Tasks

• Hardware
• Design and code various sensor motes with plug-in
  sensors temperature, RH, motion, solar radiation
  and etc.
• Design and code actuator of HVAC system and
  indicator
• Hardware and network reliability test (lab test
  and field test)
• Software
• Analyze standard thermostat control
• Design and implement layered controller structure
• Communicate with users, wireless network,
  database, internet and the following simulations
• Simulate price signals, occupancy activities and
  house with HVAC systems for testing purpose.

6
Tstat Group Task
7
Tstat Group Task
8
Overview

• DR Thermostat/Control Group overview
• Analysis on standard thermostat control
• System structure
• Decision making process
• Learning of occupancy schedule pattern
• Demonstration

9
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control
• Minimum off control

10
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control
• Minimum off control

Power
Power on
Power off
Temperature
Tset
11
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control
• Minimum off control

Power
Power on
Hysteresis band
Power off
Temperature
Tset
12
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control
• Minimum off control

Power
Power on
Hysteresis band
Power off
Temperature
Tset
13
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control Sets a maximum limit on the
  cycle rate
• Minimum off control

14
Analysis of thermostat control

• On-off control
• On-off control with hysteresis
• Anticipation asymmetric hysteresis
• Cycle rate control sets a maximum limit on the
  cycle rate
• Minimum off control set a minimum limit on the
  off time

15
Two Hypotheses

• Three parameters are sufficient for thermostat
  control
• temperature setpoint
• hysteresis band
• hysteresis offset

16
Two Hypotheses

• Temperature Setpoint ? Average Temperature

On/Off-Hysteresis Control with Anticipator and
Cycle Rate Control
17
Overview

• DR Thermostat/Control Group overview
• Analysis on standard thermostat control
• System structure
• Decision making process
• Learning of occupancy schedule pattern
• Demonstration

18
System Structure
19
Functions of Learning
DB
Learning Modules
Learning modules
20
Hierarchical Control Structure
Goal Seeking Layer
Information Flow
Query Flow
21
Decision Making Goal Seeking Layer

• Goal
• Optimize the cost of energy and thermal comfort
  of resident
• Functionality
• Design control strategies to decide temperature
  setpoint
• Control strategies transition

Goal Seeking Layer
22
Hierarchical Decision Making Process
Cooling mode? heating mode?
Strategies? (Normal Precool/preheat
Precondition)
Setpoints Start/end timing
23
Decision Making Cooling mode vs. Heating mode

• Depends on outdoor temperature averaged over day
  8 through day 2 prior to the current day.
• gt 60F cooling mode
• lt 60F heating mode
• (2005 California Energy Commission
  Residential ACM Manual )

24
Decision Making Control Strategy
Table Control Mode Design
25
Decision Making Modes Transition
Departure Preparation
Occupied/ Normal
Precool/Preheat
Combination

• Event based state transition.

26
Normal Mode
Utility function Us(T) (1-e) cost e
discomfort (1-e) cost e (1
comfort) Tset Tset Us(Tset) minUs(T)

Comfort
Cost (scaled)
1.0
1.0
0.6
0.6
0
80
0
76
72
64
88
76
T F
T F
27
Economics Index e

• A user specified value to show his economics
  preference.
• Small value of e less cost but less comfort
• Big value of e more cost and better comfort

Value of e 0
1
28
Energy Cost and Comfort Conversion

• To add comfort and energy cost, a common currency
  is required. A conversion factor is desired.
• Scale comfort level to dollars. (office
  building)
• comfort level ? productivity (p) ? dollar
• 0 100 0 100
  hourly payp
• Scale energy cost and comfort in percentages.

0
100
Full AC at peak price
All off
Energy cost
comfort
uncomfortable
comfortable
29
Precool/Preheat Mode

• Choose a setpoint schedule
• Utility function is defined on the pre-cool
  period

Price

• price signal
• temp setpoint
• Indoor temp

Time
Temperature
Pre-cool
Time
Temperature
Time
30
Departure/Arrival Preparation Mode

• Decision with uncertain information
• Choose setpoint based on prediction probability

Define Prob probability of prediction
succeeds (known) 1- Prob probability of the
prediction fails (known) Us(T) utility in
the case prediction succeeds Uf(T) utility
in the case prediction fails Then Tset T
min Prob Us (1-Prob) Uf
31
Combination Mode

• Consider both future price increase and predicted
  occupancy changes
• Find the balance when two future events push
  temperature setpoint in different directions

32
Thermostat Optimization Hypotheses

• Optimization is adjustable by user selected
  economic index e (01).

Temperature
Tset
1
Time
0.5
0.85
0. 6
comfort
Tset
Time
33
Thermostat Optimization Hypotheses

• Precool/preheat might save money. Savings depend
  on the price difference, with economic index
  fixed.

76
71
67
Temperature setpoint
34
Thermostat Optimization Hypotheses

• Better information (estimates) results in better
  optimization less cost and more comfort.
• Comfort preference is transmitted from User
  Interface Layer.
• Power/Energy information is estimated by
  supervisory layer.

Goal Seeking Layer
35
Decision Making Control Strategy
Table Control Mode Design
36
Overview

• DR Thermostat/Control Group overview
• Analysis on standard thermostat control
• System structure
• Decision making process
• Learning of occupancy schedule pattern
• Demonstration

37
Learning Occupancy Schedule

• Task
• To estimate the probability at which occupants
  are leaving or coming in the future
• To identify occupancy schedule pattern.
• Input
• Time of the day, day of the week
• Current and historical occupancy status
• Output
• Probability of arrival/departure at a time
  instance.

38
Learning Method

• Discrete time
• Look up table
• Encode the possible schedules into time bins. The
  table shows default output values without any
  learning. It is also a start point of pattern
  learning.
• Neural network
• Continually adapt occupancy schedule by
  learning.
• Network structure

39
Learning Test

• Occupancy simulation
• Simulate occupants activity with a certain
  randomness.
• Five modes of activity are defined based on
  different requirements of temperature
• Probability of transferring from one mode to
  another depends on time of the day.

• Real data
• Gathered from real-house test by a occupancy
  switch sensor.

40
Overview

• DR Thermostat/Control Group overview
• Analysis on standard thermostat control
• System structure
• Decision making process
• Learning of occupancy schedule pattern
• Demonstration

41
Future Work

• Implement all the learning modules.
• Integrate the learning modules with the control
  system.
• Test the learning modules with simulation.
• Real-house testing this March July