Pacific whiting, with a maximum sustainable yield between 150,000 and
250,000 metric tons, is the largest stock of fish found off Oregon. The majority of the
fish are processed into surimi. Hundreds of variables could potentially affect surimi
quality (gel strength). Alternative harvesting and processing input combinations, as
well as product quality attributes and their influences, were collected for the 1992-94
Pacific whiting seasons. This data was combined with other research on Pacific
whiting quality to develop a comprehensive model of the Pacific whiting fishery.
Neural network and induction modeling methods were used to isolate the importance
of each input variable and document its interactive effects on other variables. Neural
network modeling does not have the limitations of standard modeling techniques. A
neural network model can "learn" and adjust weights among inputs and interactions as
situations change. This allows for development of models which assign weights to all
inputs, yet is easily maintained and updated.
Another modeling method, known as induction, divides the information into
smaller, more defined, subgroups which are analyzed separately using regression. This
strategy reduces complications due to discontinuities in the data. A hybrid model was
developed by combining results of the two modeling methods.
These methods were compared to multiple regression for their effectiveness in
prediction. The hybrid model provided the most accurate predictions (96% of
predictions within 10% of actual value), followed by neural networks (92%), induction
(84%), and regression (74%).
Of the 88 variables examined, only ten and their interactions were significantly
related to final product quality. These variables include the time it takes to process
the fish from capture, the temperature the fish are stored until processing, the salinity,
moisture content, and pH of the fish, the length and weight of the fish, the date and
place where the fish were captured, and the water:meat wash ratio of the various
surimi washes during processing. Most of the variables were highly interactive and
nonlinear.
The information derived from these models can be used to optimize production
decisions and maximize profit. Quality influences of Pacific whiting are crucial for
long term production and can be used to benefit the entire industry. / Graduation date: 1996
| Identifer | oai:union.ndltd.org:ORGSU/oai:ir.library.oregonstate.edu:1957/34581 |
| Date | 08 December 1995 |
| Creators | Peters, Gregory J. |
| Contributors | Bolte, John, Morrissey, Michael T. |
| Source Sets | Oregon State University |
| Language | en_US |
| Detected Language | English |
| Type | Thesis/Dissertation |
Page generated in 0.0022 seconds