UTILIZATION OF GRID AND ARTIFICIAL INTELLIGENCE TECHNOLOGIES IN AGRICULTURE

1
UTILIZATION OF GRID AND ARTIFICIAL INTELLIGENCE
TECHNOLOGIES IN AGRICULTURE

• Salga Péter, salga_at_thor.agr.unideb.hu
• University of Debrecen, FAERD,
• Department of Business and Agriinformatics

2
Problems

• Decision Support Systems
• Data access to resources
• User right policies
• Sharing instruments
• Modelling the effect of climate change

3
Technologies

• Neural networks
• Fuzzy logic
• Genetic alogorithm
• Distributed computing
• Clusters
• GRID

4
Artificial Neural Networks
Warren McCulloch and Walter Pitts biological
inspiration
5
General properties of neural modells

• Determination of weighted sum of input data,
• Processing by treshold logic,
• Neurons change theirs behavior based on input
  signals.

6
Application e.g. pattern recognition
7
Java Object Oriented Neural Engine (Joone)
www.joone.org
8
Fuzzy logic

• Lotfi Zadeh
• Based on set theory but solves real-world
  problems
• In matematics no exact definiton for not too
  much or beautiful

9
Multi-processor systems

• Symmetrical Multiprocessing (SMP)
• symmetrical each processor can working in each
  task
• Alternative dedicated processor wating for the
  another one
• They use common RAM are any other processor
  using this block?
• There is only one memory bus toward processors

10
MPP - Massively Parallel Systems

• Direct access to local data
• Hundreds of processors have their own RAM and
  operating system. We call them node-s.
• Scientific applications, data-warehouses, DSS,
  meteorology
• Supercomputers

11
Top 10 supercomputer
FLOP Floating point operations per second
12
cluster

• Group of separeted PC-s which use common
  resources and share processing power
• Connection among nodes GBit or higher
• Clustering, scheduling software (PBS- Portable
  Batch System, Open Mosix)
• There are different type of clusters

13
grid

• Cluster aggregate computers to one system
• Grid systems share theirs resources
• the constituent systems are administered
  separately and are physically distinct from each
  other

14
NorduGrid

• Innovative middleware solutions are key to the
  NorduGrid testbed, which spans academic
  institutes and supercomputing centres throughout
  Scandinavia and Finland and provides continuous
  grid services to its users.
• Launched in May 2001.

15
Decision Support System in Agriculture
16
DSS using GRID and AI
17
AgModel

• MetBroker provides consistent, timely access to
  Internet-based weather databases in eight
  countries.
• DemBroker provides consistent access to digital
  elevation models (DEMs) with differing formats.
  Currently DEMBroker provides global coverage at
  1km through the GTOPO30 DEM and coverage of Japan
  at 50m.
• ResourceServer Web-based system for handling
  program text such as button labels and menu
  items. It lets a software application quickly
  download the text items it requires in the
  national language of the user. Translators can
  add and update the text provided through a
  Web-based editor.
• ChizuBroker provides applications with
  consistent dynamic access to rasterized maps from
  various online sources. Currently maps are
  available for Japan, New Zealand, US and Europe.

18
Global surface mean temperatures
Jones et al., 1998
19
Global CO2 emission scenariosIntergovernmental
Panel on Climate Change (IPCC)
Higher emissions
Lower emissions
1000 ppm
550 ppm
20
Extrem weather events
Hurricane Ivan
21
European models

• LARSSIRIUS - M. Semenov (UK)
• LAPS - D. Mihailovic (Serbia)
• STICS F. Ruget (Fr)
• WOFOST (Cz)
• PERUN (Cz)
• 4M (Hu)

19 22 August 2006 Summer Univerity in
Debrecen http//odin.agr.unideb.hu/su2006
In case of GCM (general circulation model)
teraflop computing capacity needed!
22
The atmosphere-ocean-biosphere system is not
linear
Linear process
?
? aiet-i (t0, 1, 2, )
i-?
where et is white-noise process (independent
standard normal sequence)
23
Problems of application linear methodes

• GCM-s (general circulation model) are good for
  global forecasts, but not good for regional scale
• Long-term forecasts needs enormous compuing power
• Traditional statistical methods not applicable
  for clustering of past weather data

24
Observed climatic problems

• Climate-analogy - clustering
• Effects on farming pattern recognition using
  neural networks
• Measurement of frequency of extreme events fuzzy
  clustering
• Modell verifikation forecast using pattern
  recognition

25
Filtering extreme events from past data

• Neural network can group the extreme events based
  its classification property (? needs fuzzy
  too!!)
• Forecasts based on pattern of extereme events

26
Complex solution distributed neural network

• Running on java cluster
• Torque scheduling system
• JBoss server
• Joone on cluster
• Distributed neural network
• More parameters, larger storage, powered
  processing