Hybrid neural network

About: Hybrid neural network is a research topic. Over the lifetime, 1305 publications have been published within this topic receiving 18223 citations.

G-OSNR estimation on dynamic PS-QAM channels using hybrid neural networks

Abstract: Aspects of the present disclosure describe systems, methods and structures in which a hybrid neural network combining a CNN and several ANNs are shown useful for predicting G-ONSR for Ps-256QAM raw data in deployed SSMF metro networks with 0.27 dB RMSE. As demonstrated, the CNN classifier is trained with 80.96% testing accuracy to identify channel shaping factor. Several ANN regression models are trained to estimate G-OSNR with 0.2 dB for channels with various constellation shaping.

Designing ADC by means of soft computing

Abstract: In this paper, the designing of ADC is studied by means of son computing. It is shown when fuzzy system has single input it can be multiplied in the premise rules by introducing proper fuzzy words and linguistic operator for amplifying knowledge base. Instinctive and learning behaviors, knowledge base and topology of circuit are found using fuzzy voronoi diagram in a POE model based program. This converter implements a set of processes during the analog to digital converting. It can be a fast ADC with more proper structure for hybrid neural network because its parameters are applied in training algorithms.

FSR vehicles classification system based on hybrid neural network with different data extraction methods

Abstract: This paper evaluates the performance of Forward Scatter Radar classification system using as so called “hybrid FSR classification techniques” based on three different data extraction methods which are manual, Principal Component Analysis (PCA) and z-score. By combining these data extraction methods with neural network, this FSR hybrid classification system should be able to classify vehicles into their category: small, medium and large vehicles. Vehicle signals for four different types of cars were collected for three different frequencies: 64 MHz, 151 MHz and 434 MHz. Data from the vehicle signal is extracted using above mentioned method and feed as the input to Neural Network. The performance of each method is evaluated by calculating the classification accuracy. The results suggest that the combination of z-score and neural network give the best classification performance compares to manual and PCA methods.

Multi-disciplinary design method for the matched response surface based on the hybrid neural network

Abstract: In order to improve the solving performance of Multi-disciplinary design optimization(MDO),a multi-disciplinary design method for the matched response surface based on the hybrid neural network is proposed.By combining the advantages of the back-propagation(BP) network with the adaptive resonance theory(ART) network and making full use of the target results sample through discipline-level optimization to adaptively change the traditional response surface structure,the proposed method improves the accuracy of the response surface and reduces the number of times of iteration of the discipline-level optimization,which leads to a better solving efficiency for the multi-disciplinary optimization methods.The optimization method is validated by a specific example.Comparison studies in terms of solving accuracy,discipline optimization times,average number of times of iteration and iterative occupancy hours of obtaining the optimization solution indicate that the method for the multi-disciplinary design of the matched response surface based on the hybrid neural network has a few advantages such as higher solution accuracy and higher computational efficiency.