Minimizing Amortized Cost of the On

Minimizing Amortized Cost of the
On-demand Irrigation System in
Smart Farms
Tiantian Xie, Zhichuan Huang, Zicheng Chi, and Ting Zhu
(Tientien Hsieh)
Department of Computer Science and Electrical Engineering
University of Maryland, Baltimore County
Irrigation water is not abundant
• Only around 0.3% freshwater can be utilized by human and the ecosystem.
• Irrigation accounts for more than 70% of the freshwater withdraw Worldwide.
Global water outlook to 2025,International Water Management Institute;
Traditional control strategy
Trial-and-error (Manually turn on/off the irrigation).
• Requires extra efforts
• may suffer from over-irrigation or under-irrigation.
“I shall turn it off. An hour has passed.
It should last a few days”
State of the art
•
Soil moisture sensor based
“Dry? irrigate 50min.”
“Saturated? it’s good.”
State of the art
•
Evapotranspiration (ET) based
“Based on the local weather today
(no rain, high temperature and high
wind speed), irrigate 40 mins.”
State of the art
•
•
Soil moisture sensor based
Evapotranspiration(ET) based
Issue: Can’t prevent the over-irrigation before precipitation.
If numerical weather prediction (NWP) can be used, we might resolve it.
Irrigation also requires energy
In 2010, around 220 billion kWh is consumed on water for
agriculture in the U.S. (NCSL,National Conference of
State Legislatures).
Minimize both water and energy usage while fulfilling the
irrigation demand.
Architecture of a smart farm
Energy
Solar Panel
Water
Pump
Water Tank
Ground Water
Crops
Time-of-Use(TOU) model
Power Grid
Design goal: minimize the amortized cost of irrigation
Design Overview
We propose an on-demand irrigation system to control the energy
and water flow to meet the irrigation demand with minimal
amortized cost
NWP
TOU Model
Irrigation
Demand
Estimation
On-demand
Scheduling
Optimization
Solar
Energy
Prediction
Controller
Energy
Water
NWP: Numerical Weather Prediction
Irrigation demand estimation
Estimate the irrigation demand time and energy (solve overirrigation problem).
Energy demand time: the soil moisture level is first below the irrigation
starting level while no increase within the moisture tolerance.
Energy demand amount: the amount of energy to raise the soil moisture
level to the target moisture level. .
Soil moisture
Target level
Soil moisture
Target level
Start level
Start level
Tolerance level
Tolerance level
Predicted future
Current
hour
Predicted future
Current
hour
Irrigation demand estimation
Soil moisture estimation.
• Basic idea: The variance of the soil moisture level is affected
by the current weather and soil status.
Train an SVR (Support Vector
Regression) model that uses the
previous soil moisture level, the
current weather to estimate the
current soil moisture.
Estimation
Solar energy prediction
NWP
TOU Model
Irrigation
Demand
Estimation
On-demand
Scheduling
Optimization
Solar
Energy
Prediction
Controller
Energy
Water
NWP: Numerical weather prediction
Solar energy prediction
Predict the solar energy distribution for the scheduling
period in the future.
)
• SVR based on hourly NWP (e.g., cloud cover, dew point, and
precipitation) and historical solar energy outputs.
Solar energy prediction
Issue: Over-prediction will cause the scheduler shift less
energy to the power grid. Until at last, all energy demand
has to come from power grid probably with high TOU price.
Estimate an over prediction penalty parameter
using near historical
data by minimizing an error function:
predicted solar energy at time t
observation at time t
mean of observations
U(.) is a penalty function:
under-prediction
over-prediction
Solar energy prediction
2000
Prediction
Prediction-U
-
1500
-
Number of Over Prediction
Number of over-prediction over time:
1000
500
0
0
1000
2000
3000
Time(h)
4000
On-demand scheduling
NWP
TOU Model
Irrigation
Demand
Estimation
On-demand
Scheduling
Optimization
Solar
Energy
Prediction
Controller
Energy
Water
On-demand scheduling
optimization
Schedule the energy source in the near future.
The amortized energy cost based on TOU price model
The amortized solar energy cost.
The energy can’t exceed the energy
cache capacity C yet meet the
energy demand µ̃
The energy during each interval
should not exceed the system capacity
Linear Programming
Evaluation
•
Datasets: One year in Austin, TX.
•
•
•
•
Soil moisture data from NOAA
Numerical weather data from World Weather Online
Solar energy output from Pecan Street.
Baselines:
• Soil moisture based Irrigation (SMI)
• On-demand Irrigation (ODI)
• On-demand Irrigation without TOU (ODI-WOT)
• Oracle
Irrigation Performance
Origin
3000
4000
Hour
ODI
5000
SMI
6000
Our system saves water and energy just hours before precipitation around
hour 6,000.
Energy consumption and cost
Energy consumption: 7.97%
Water consumption:
Energy cost: 25.34%
Conclusion
•
Utilize future information to avoid over-irrigation before
precipitation.
•
On-demand irrigation with TOU to meet the irrigation
demand with minimal amortized cost
•
Reduce the amortized cost by 25.34% and save 7.97% water
and energy resources.
Thanks
Q&A