Adaptive Markov decision control of high-frequency drip irrigation systems

Peckett, Damian (2013) Adaptive Markov decision control of high-frequency drip irrigation systems. [USQ Project]

[img]
Preview
Text
Peckett_2013.pdf

Download (12Mb) | Preview

Abstract

Global demand for food increases every year, despite this increase, productive farming land is limited. Meeting the demands of the future are dependent on improving the
productivity of the land in use and a key factor behind this is intelligent agriculture. Fresh water remains in many situations the primary constraint on agriculture. This
paper seeks to investigate machine intelligence irrigation scheduling methods in order to increase yield and improve crop irrigation efficiency.

For the purposes of research the model organism Arabidopsis Thaliana was selected primarily due to its rapid life cycle and historical significance. Research began with the development of a water balance crop model for emphArabidopsis Thaliana. Upon completion of the model several experiments were performed to weakly verify its behaviour.

The final step looked at implementing an artificially intelligent control algorithm for maximizing crop yield in water constrained environments. Several possible approaches
were identified. The winning approach utilized a Markov decision process implementing a Radial Basis Function network as a value estimator. Performance was seen to exceed
the best hand coded algorithms by up to 10% and it successfully outperformed the normal irrigation schedule control by a factor of 6.5x far exceeding the initial goals of the project!

This solution was developed using a highly optimized, high performance computing MDP solver programmed by the author. Utilizing an Amazon Cluster Compute server possessing 32 Intel Xeon cores the author was able to outperform an initial naive, yet straight from the textbook, MATLAB MDP solver by a factor of 104!


Statistics for USQ ePrint 24664
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Faculty / Department / School: Current - Faculty of Health, Engineering and Sciences - School of Mechanical and Electrical Engineering
Supervisors: McCarthy, Alison
Date Deposited: 27 Feb 2014 05:08
Last Modified: 27 Feb 2014 05:08
Uncontrolled Keywords: markov decision control; high-frequency; drip irrigation systems
Fields of Research : 07 Agricultural and Veterinary Sciences > 0799 Other Agricultural and Veterinary Sciences > 079901 Agricultural Hydrology (Drainage, Flooding, Irrigation, Quality, etc.)
URI: http://eprints.usq.edu.au/id/eprint/24664

Actions (login required)

View Item Archive Repository Staff Only