Artificial Neural Networks: Approximation and Learning Theory
Artificial Neural Networks: Approximation and Learning Theory
Hi-index | 0.00 |
A novel structure of dynamic BP neural network (NDBP) with quick-time-variable real-time training-algorithm based on the modern dynamical control theory is developed in the CAD-grid 2# ("China-Austria-Data-Grid"-cooperation-project) and CEFSP 20080 5# (Chinese Education Foundation Science Project) and proposed in this paper. This NDBP suites to real-time modeling for tracking the process the characters of which are dynamic and strongly time-variable, owing to its memory-power that every output on time k is strongly related to the input on time k, k-1, k-2, and its forgetting ability that only the newer output errors effect the weight-correction strongly. So, both advantages, robust process-function approach and high speed process-character tracking, are possessed by the NDBP facing the urgent requirement of the research fields such as bio-chemical-process analysis, on-line diagnosis medical measurement, industrial adaptive process control, nature synusiologic development monitoring, and so on. The basic principle, algorithm composition and description, execution effect and typical application in medical measurement are introduced.