Using artificial intelligence to predict events like harvest dates could give grain agencies an edge when it comes to marketing.
Scientists at Agriculture Canada’s Lethbridge Research Centre have been working with a sophisticated computer program known as neural network modeling to predict when the crop will be ripe.
This information can be passed onto the Canadian Wheat Board to more accurately predict when the agency needs cars and labor.
The wheat board wants harvest dates that are within a few days and grain analysts want the information by July, said researcher Bernie Hill.
Read Also
VIDEO: British company Antler Bio brings epigenetics to dairy farms
British company Antler Bio is bringing epigenetics to dairy farms using blood tests help tie how management is meeting the genetic potential of the animals.
This summer he not only predicted by July that this year’s harvest would be early, Hill forecasted within two days when the southern Alberta crop would be ready.
Neural network modeling or artificial intelligence programs are commercially available software that were originally used by the stock market to predict the price of stocks.
The software is designed to sift through numerous examples to find a solution to a given problem.
“They’re patterned after the way medical people think that the brain works,” said Hill.
The computer receives examples of past events similar to the way people solve a problem by comparing it to past experience.
Until now, computer models relied on expert systems to find answers by analyzing two or three different factors in a linear fashion.
“They found when they got into the real world of complex problems the expert systems were always coming up a little short,” said Hill.
To predict harvest dates, Hill’s team assembled 25 years worth of data from 30 locations in Western Canada. They looked at past harvest dates, seeding dates, weather patterns and soil zones.
These systems still require a person who recognizes what is or isn’t a practical answer.
“It’s an art, as well as a science,” said Hill, who has been working on the concept for five years.
Where artificial intelligence performs better than a human mind is in handling more information at one time to find an answer.
“With the power of computers, neural network modeling can handle 50, 60 or 100 factors simultaneously and can predict an outcome,” said Hill.
“Neural network modeling gives you the output but you don’t understand the process. It’s artificial intelligence,” he said.
The same type of modeling is used to identify cancer cells using image analysis, conduct target marketing, predict sports events, predict meat tenderness or adjust power grids for electrical load fluctuations.
Expert systems and neural networks were developed in the 1950s. Expert systems were used more because they didn’t require as much computing power.
Improvements in the power of computers makes this kind of system possible.
To make his predictions, Hill uses an ordinary desktop computer with 120 megahertz of power.
