The Intelligent Evaluation Of DTS Based On Self-Organizing Feature Map

In this paper, the theory of the Dispatcher Training Simulator(DTS) intelligent evaluation based on the Self-Organizing Feature Map(SOFM) is discussed on the basis of the systematic analysis to the SOFM and comprehensive evaluation method. In the first place, this paper discusses the general method of the comprehensive evaluation, such as deducting-mark method, fuzzy evaluation, expert system, decision tree, points out their merits and defects. Then, the basic principal of the neural network is described, and some practical questions about neural network are discussed. The SOFM, a specific neural network discussed detailedly. The visualized implementation algorithm is given. In the second place, aiming at the evaluation of DTS, according to the idea of pattern classification and the fact that the SOFM obtains the excellent classifying ability by presenting the structures of the input data using the weight vector of the neuron puts forward the intelligent evaluation method based SOFM, which make the result of the simulation training evaluation object and automatic. The practical verifications of the DTS in Nanyang power system have demonstrated the feasibility and effectiveness of the proposed method. In the third place, a feature-selected method based on the rough set theory is put forward, which solve the misclassifying difficulty led by the redundancy (correlation) of the input data. The principal of the harm of the redundancy is discussed. Some key questions of the evaluation are discussed deeply. Besides, this paper brings forward an idea of evaluation based on section. Some problems in the non-section evaluation are expatiated and analyzed, which detail is discussed detailedly. And the possibility that the evaluation based section is carried out in the present is discussed. A simplified algorithm on real-time evaluation is discussed simply. Finally, the modification of the inherence default and structure limitation when used on evaluation of the neural network is prospected.