دانلود رایگان مقاله اصل چند جایگزینی در سیستم های هوشمند. مدل شبکه عصبی فعال

Abstract

The article deals with intelligent systems that contain artificial neural networks. After a close comparison of artificial and biological neural networks the authors reveal some fundamental flaws of artificial neural networks. It is shown that the reason for those disadvantages is the constancy of structure or the so called passivity of the neural network. To avoid this problem it is proposed to simulate the information processes in the neural network instead of simulating neurons themselves. The following consideration involves several evolutionary principles of multi-alternativity, such as multilevel approach, diversity and modularity. Those principles find their implementation in facet memory organization that is characterized by the reconfigurable structure and therefore close to its biological prototype. The advantage of the suggested approach is demonstrated by the example of an intellectual system based on an active neural network. The system applied to control an electrical supply network under critical events, such as breaks and overloads. In case of a critical event neural network takes the blocking decision that prevents breakage or accident conditions in the electrical network.

4. Conclusion

Using passive models of neural network management systems and decision-making faces significant difficulties for their implementation due to the propensity of these models to retrain and low extrapolating opportunities.

Designing neural systems based on evolutionary principles can create multiple-active neural models with a reconfigurable structure, in its properties to a much greater degree approaching their biological prototypes:

multi-level hierarchical scheme of internal connections in the network provides high generalizing ability of the system when decision-making in situations not encountered during training;

modular structure allows to build the structure of the new system of ensembles of neurons, without meeting the restrictions “curse of dimensionality” and retraining;

facet memory organization according to the rule “one event – one ensemble” enables unlimited selectively increasing the number of events in the system and the practical implementation of the principle of requisite variety of information.