Search

Rear-end collision prediction has gained an increasing attention for safety improvement in smart cities. It is urgent to design efficient warning strategies for rear-end collisions which is one of the main causes of traffic accidents. The existing researches have been conducted to predict the collisions. Learning-based methods are proposed to solve this complicated issue, which the traditional methods are difficult to solve. However, due to some limitations in terms of the feature extraction and prediction performance, back-propagation learning methods are facing challenge. In this paper, we proposed a novel Rear-end Collision Prediction Mechanism with deep learning method (RCPM), in which a convolutional neural network model is established. In RCPM, the dataset is smoothed and expanded based on genetic theory to alleviate the class imbalance problem. The preprocessed dataset is divided into training and testing sets as the input to train our convolutional neural network model. The experimental results show that compared with the Honda, Berkeley and multi-layer perception neural-network-based algorithms, RCPM effectively improves performance to predict rear-end collisions. [ABSTRACT FROM AUTHOR]

A new deep learning (DL)-based parameter extraction method is presented in this brief; 50k training cases are generated by Monte Carlo simulations of these preselected parameters in Berkeley short-channel IGFET model (BSIM)-common multigate (CMG). DL models are trained using backward propagation with ${C} _{\text {gg}} - {V} _{g}$ and ${I} _{d} - {V} _{g}$ as the input and selected BSIM-CMG parameters as the output. A TCAD simulated FinFET device, calibrated to Intel 10-nm node, is used to test the DL models. The DL-based parameters extraction results show an excellent fit to capacitance and drain current data, with 0.16% rms error in ${C} _{\text {gg}} - {V} _{g}$ and 6.1% rms error in ${I} _{d} - {V} _{g}$ (0.69% rms error in above-threshold-voltage ${I} _{d} - {V} _{g}$), respectively. In addition, devices with a 10% variation in gate length and oxide thickness are successfully modeled with the trained DL model. The results show tremendous promise in using the DL-based models for parameter extraction. [ABSTRACT FROM AUTHOR]