Research and application of quantum-inspired double parallel feed-forward neural network

A novel artificial neural network called fast learning network (FLN) was proposed by Li et al. in 2014, which showed better performance than some famous traditional artificial neural network algorithms. Inspired by the conventional FLN and quantum mechanics, this work proposes a kind of quantum-inspired double parallel feed-forward neural network (QIDPFNN). Compared with the FLN, the proposed algorithm presents two obvious differentials. Firstly, the input weights between hidden layer and input layer are generated by quantum computing. Secondly, a new hidden layer activation function is introduced. In order to verify the QIDPFNN validity, it is applied to 13 regression applications. The experimental results show that the QIDPFNN owns better generalization ability and stronger stability than FLN on most applications. Simultaneously, the QIDPFNN is applied to build NOx emission concentration model and thermal efficiency model of a 330 MW circulating fluidized bed boiler. The experiment results demonstrate that the proposed method has high regression precision, strong stability and generalization ability.