Stable convolutional neural network for economy applications

A convolutional neural network does not require to be stable when it is used for economy applications being related with the offline learning. Nevertheless, a convolutional neural network requires to be stable when it is used for economy applications being related with the online learning. Therefore, it would be interesting to ensure the stability of a convolutional neural network for economy applications being related with the online learning. In this investigation, a stable algorithm considering a time varying learning rate is proposed to adapt the weights of a stable convolutional neural network. The stable algorithm considering a time varying learning rate is used to improve the learning and to ensure the stability and robustness of the stable convolutional neural network, where the time varying learning rate will obtain big steps when the minimum of the cost function is far, and the time varying learning rate will obtain small steps when the minimum of the cost function is near. The stable convolutional neural network is compared with the principal component analysis neural network, non-negative matrix factorization neural network, and convolutional neural network for economy applications being related with the online learning considering the electrical energy consumption modeling and hybrid chiller modeling.