Abstract: Five sets of results for different L⁹(3⁴) orthogonal tests were used as the training-study samples, the forecasting characters of L⁹(3⁴) orthogonal test were researched on the basis of the artificial neural network. The results showed that the self-contained orthogonal sample collection was the basic training-studying cell and was indivisible. Their forecasting results were tallied with the test results. When others samples were added into the self-contained orthogonal samples or the self-contained orthogonal samples were reduced, the forecasting results would be completely irresponsible. Under the same test conditions and with the same orthogonal test type, the self-contained orthogonal sample containing large information could forecast other self-contained orthogonal samples with small information. It provides a significant novel test-design approach for the orthogonal test.

Key words: orthogonal test, forecasting characters, artificial neural network, sample collection