Experimental Comparison of the Randomized and Fully Connected Neural Networks for Anomaly Detection in the Industrial Systems

Nowadays the challenge of anomaly detection in industrial systems in the conditions of limited resources is highly relevant. The randomized neural networks are claimed to provide the anomaly detection efficiency while the limited resources. In the paper the results of the experimental comparison of the randomized neural networks on the example of the stochastic configuration networks and the fully connected neural networks on the example of the perceptron are described. The models are compared considering the accuracy, root mean square deviation and training time. The industrial dataset containing data on the dumpers movements and the database of the handwritten digits are selected for the experiments. The results of the experiments are analysed and the architecture for the further research is selected.