Matching of External Criterion and Method of Sample Partitioning for Solving Problem of Structural-Parametric Identification by Group Method of Data Handling

摘要

We investigated the validity of the joint application of the data partitioning method and criterion of external additions in finding the most accurate models. Quasi-optimal partitions for all ratios of sizes of the initial matrices are described. GMDH external criteria are comprehensively investigated. Their use in certain types of optimal ways of partitioning the sample is theoretically justified. If the condition of proportionality of data is satisfied, it was theoretically substantiated and confirmed in numerical experiments that the criterion of parameters unbiasedness is not "adequate" to noise while minimizing the criterion of the sample division, it takes place only when it is maximized.