Using of Artificial Neural Networks to Recognize the Noisy Accidents Patterns of Nuclear Research Reactors

In this paper, an approach based on neural networks for recognizing the nuclear research reactor accidents (patterns) is presented. A neural network is designed and trained, initially without noise, to recognize the nuclear research reactors accidents patterns (using MATLAB's Neural Network Toolbox). When the neural network response is simulated, the 9x9 simulation output values of the matrix's diagonal is larger than 0.9, (approximately equal 1), this means the outputs is approximately equal the targets and the network is well trained. A new copy of the neural network was made, to train it with noisy accident's patterns. When this network was trained on this noisy input vectors (patterns), it is greatly reduces its errors and its output is approximately equal the output as when it is trained without noise input vectors. This new copy was trained also on accidents patterns without noise to gain the maximum performance and the high reliability of the network. Experiments have shown excellent results; where the network did not make any errors for input vectors (patterns) with noise level from 0.00 up to 0.14. When the noise level larger than 0.15 was added to the vectors (patterns); both neural networks began making errors.