| The significance of quantum computing research has been recognized by manyscientists. In particular, the quantum neural network (QNN), being generated by thecombination of quantum computing and neural computing, has become a mainstreamdirection in the artificial intelligence area. After10-years rapid development, the QNNis considered as a powerful tool for information processing in many theoretical fields,because of its unique structure of knowledge representation and efficient performanceof information processing. It provides a new idea that some problems, being extremelydifficult to solve by the traditional neural network, can be addressed using the QNN.To improve the theory of the QNN and practice it, considerable effort has beenmade by many researchers, especially, by a large number of young doctors whoparticipate in this study. This paper provides some novel QNN models by introducingthe quantum principles and concepts into the ANN. Specifically, the work of this papercan be summarized as follows:①A quantum gates Elman neural network model is proposed. The new model iscomposed of qubit neurons and classic neurons. The law of quantum physics is appliedto the interaction between the qubit neurons and the classic neurons. For the newstructure, the quantum mapping layer is used to solve the mode inconsistent between thecontext-lyaer-output and hidden-layer-input. In the standard algorithm, it contains theupdating law of the parameters in the quantum gates. In the the gradient expansionalgorithm, the learning rate is adjusted by the time scheduling strategy. The context-layer-weights and the hidden-layer-weights are synchronized updated by gradientexpansion. In the adaptive dead vector algorithm, the learning parameters arecontrollable. The convergence of the algorithm in the sense of Lyapunov function isproved.②A quantum Hopfield neural network model with time delay is proposed. It is anew idea that this quantum Hopfield neural network can be explained in term ofprobability. The states of neurons and connection weights are described as quantumstates. The evolution of quantum states, described by the quantum superpositionprinciple and the measurement principle, is introduced into this model. Themeasurement matrix connected with the neurons includes the probability information ofthe quantum states. Through the calculation of the measurement matrix elements appearing at different times, we can obtain the probability of the quantum key inputwhich is evolved into the quantum memory prototype.③A quantum SOM neural network model with elastic radius is proposed. Thenew model objects the real data into the quantum initial states, then into the quantumexcited states. The weights connected with the quantum excited states are also quantized.The proposed quantum learning law make quantum excited states have an orderedtopology mapping. In addition, the elastic radius of neighborhood kernel function isdefined by the similarity degree and the distance between quantum excited states andquantized weights. It avoids dead zone to be formed so that neurons in competitive layercan not be trained.④The application of the proposed QNN models is described here. The quantumgate Elman neural network (QEN) is used in the short-term load forecasting. Theaccuracy of forecasting results clearly shows that the QEN has a rapid convergencespeed and a good generalization performance. The Hopfield neural network (HP) andthe quantum Hopfield neural network (QHP) are employed in the analog fault diagnosis.The fault features are extracted by the wavelet packet analysis. The energy of the faultresponses are calculated by a new energy function. The multiple fault diagnostic resultsimply that the fault diagnosis method based on QHP is much more useful than themethod based on HP. To verify the performance of the quantum SOM neural network(QSOM), the experiments about the sewage treatment effluent quality prediction arecarried out. The corresponding results offer a high forecasting accuracy and verifyeffectiveness of of the theory. |
PDF Full Text Request