Information Technology Reference
In-Depth Information
5.11.5
Experiments
The system architecture is sketched in Fig. 5.6. The historical data are stored
in the case library, after a series of pretreatments such as appropriate
clustering, variation. In order to improve the predication precision, we employ
multi-strategy retrieval and utilize different similar computing technologies to
find out the array of neighboring similar fishing grounds. Then the prediction
results will be revised in accordance to the rules already stored in the system.
The joint employment of machine learning and expertise will further improve
the predication precision and system robustness.
While predicting the fishing grounds in winter, the historical data in
summer are not considered, and due to the fact that remote fishing grounds
have little effects on the formation of central fishing grounds, we only choose
those sea situations closer to central fishing grounds. In our experiments, we
choose 150 out of 600 pieces of sea situations dynamically. In this way,
calculation can be reduced greatly through filtrations in terms of time and
position while keeping prediction precision to a certain degree; the system
respond time is less than 7 seconds. Visual human-computer interaction
module offers friendly interface for input, output, inspection, and revision
(see Fig. 5.7).
Fig. 5.8. The sea situation and fishing grounds
