• The Sea:JOURNAL OF THE KOREAN SOCIETY OF OCEANOGRAPHY
• /
• v.23 no.1
• /
• pp.1-19
• /
• 2018

In order to compare significant wave height (SWH) data from multi-satellites (GFO, Jason-1, Envisat, Jason-2, Cryosat-2, SARAL) and SWH measurements from Ieodo Ocean Research Station (IORS), we constructed a 12 year matchup database between satellite and IORS measurements from December 2004 to May 2016. The satellite SWH showed a root mean square error (RMSE) of about 0.34 m and a positive bias of 0.17 m with respect to the IORS wave height. The satellite data and IORS wave height data did not show any specific seasonal variations or interannual variability, which confirmed the consistency of satellite data. The effect of the wind field on the difference of the SWH data between satellite and IORS was investigated. As a result, a similar result was observed in which a positive biases of about 0.17 m occurred on all satellites. In order to understand the effects of topography and the influence of the construction structures of IORS on the SWH differences, we investigated the directional dependency of differences of wave height, however, no statistically significant characteristics of the differences were revealed. As a result of analyzing the characteristics of the error as a function of the distance between the satellite and the IORS, the biases are almost constant about 0.14 m regardless of the distance. By contrast, the amplitude of the SWH differences, the maximum value minus the minimum value at a given distance range, was found to increase linearly as the distance was increased. On the other hand, as a result of the accuracy evaluation of the satellite SWH from the Donghae marine meteorological buoy of Korea Meteorological Administration, the satellite SWH presented a relatively small RMSE of about 0.27 m and no specific characteristics of bias such as the validation results at IORS. In this paper, we propose a conversion formula to correct the significant wave data of IORS with the satellite SWH data. In addition, this study emphasizes that the reliability of data should be prioritized to be extensively utilized and presents specific methods and strategies in order to upgrade the IORS as an international world-wide marine observation site.

• Journal of Intelligence and Information Systems
• /
• v.22 no.2
• /
• pp.127-142
• /
• 2016

Deep learning model is a kind of neural networks that allows multiple hidden layers. There are various deep learning architectures such as convolutional neural networks, deep belief networks and recurrent neural networks. Those have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Among those architectures, convolutional neural networks and recurrent neural networks are classified as the supervised learning model. And in recent years, those supervised learning models have gained more popularity than unsupervised learning models such as deep belief networks, because supervised learning models have shown fashionable applications in such fields mentioned above. Deep learning models can be trained with backpropagation algorithm. Backpropagation is an abbreviation for "backward propagation of errors" and a common method of training artificial neural networks used in conjunction with an optimization method such as gradient descent. The method calculates the gradient of an error function with respect to all the weights in the network. The gradient is fed to the optimization method which in turn uses it to update the weights, in an attempt to minimize the error function. Convolutional neural networks use a special architecture which is particularly well-adapted to classify images. Using this architecture makes convolutional networks fast to train. This, in turn, helps us train deep, muti-layer networks, which are very good at classifying images. These days, deep convolutional networks are used in most neural networks for image recognition. Convolutional neural networks use three basic ideas: local receptive fields, shared weights, and pooling. By local receptive fields, we mean that each neuron in the first(or any) hidden layer will be connected to a small region of the input(or previous layer's) neurons. Shared weights mean that we're going to use the same weights and bias for each of the local receptive field. This means that all the neurons in the hidden layer detect exactly the same feature, just at different locations in the input image. In addition to the convolutional layers just described, convolutional neural networks also contain pooling layers. Pooling layers are usually used immediately after convolutional layers. What the pooling layers do is to simplify the information in the output from the convolutional layer. Recent convolutional network architectures have 10 to 20 hidden layers and billions of connections between units. Training deep learning networks has taken weeks several years ago, but thanks to progress in GPU and algorithm enhancement, training time has reduced to several hours. Neural networks with time-varying behavior are known as recurrent neural networks or RNNs. A recurrent neural network is a class of artificial neural network where connections between units form a directed cycle. This creates an internal state of the network which allows it to exhibit dynamic temporal behavior. Unlike feedforward neural networks, RNNs can use their internal memory to process arbitrary sequences of inputs. Early RNN models turned out to be very difficult to train, harder even than deep feedforward networks. The reason is the unstable gradient problem such as vanishing gradient and exploding gradient. The gradient can get smaller and smaller as it is propagated back through layers. This makes learning in early layers extremely slow. The problem actually gets worse in RNNs, since gradients aren't just propagated backward through layers, they're propagated backward through time. If the network runs for a long time, that can make the gradient extremely unstable and hard to learn from. It has been possible to incorporate an idea known as long short-term memory units (LSTMs) into RNNs. LSTMs make it much easier to get good results when training RNNs, and many recent papers make use of LSTMs or related ideas.

• Proceedings of the KOR-KST Conference
• /
• 1995.02a
• /
• pp.3-32
• /
• 1995

The Highway Capacity Manual specifies procedures for evaluating intersection performance in terms of delay per vehicle. What is lacking in the current methodology is a comparable quantitative procedure for ass~ssing the safety-based level of service provided to motorists. The objective of the research described herein was to develop a computational procedure for evaluating the safety-based level of service of signalized intersections based on the relative hazard of alternative intersection designs and signal timing plans. Conflict opportunity models were developed for those crossing, diverging, and stopping maneuvers which are associated with left-turn and rear-end accidents. Safety¬based level-of-service criteria were then developed based on the distribution of conflict opportunities computed from the developed models. A case study evaluation of the level of service analysis methodology revealed that the developed safety-based criteria were not as sensitive to changes in prevailing traffic, roadway, and signal timing conditions as the traditional delay-based measure. However, the methodology did permit a quantitative assessment of the trade-off between delay reduction and safety improvement. The Highway Capacity Manual (HCM) specifies procedures for evaluating intersection performance in terms of a wide variety of prevailing conditions such as traffic composition, intersection geometry, traffic volumes, and signal timing (1). At the present time, however, performance is only measured in terms of delay per vehicle. This is a parameter which is widely accepted as a meaningful and useful indicator of the efficiency with which an intersection is serving traffic needs. What is lacking in the current methodology is a comparable quantitative procedure for assessing the safety-based level of service provided to motorists. For example, it is well¬known that the change from permissive to protected left-turn phasing can reduce left-turn accident frequency. However, the HCM only permits a quantitative assessment of the impact of this alternative phasing arrangement on vehicle delay. It is left to the engineer or planner to subjectively judge the level of safety benefits, and to evaluate the trade-off between the efficiency and safety consequences of the alternative phasing plans. Numerous examples of other geometric design and signal timing improvements could also be given. At present, the principal methods available to the practitioner for evaluating the relative safety at signalized intersections are: a) the application of engineering judgement, b) accident analyses, and c) traffic conflicts analysis. Reliance on engineering judgement has obvious limitations, especially when placed in the context of the elaborate HCM procedures for calculating delay. Accident analyses generally require some type of before-after comparison, either for the case study intersection or for a large set of similar intersections. In e.ither situation, there are problems associated with compensating for regression-to-the-mean phenomena (2), as well as obtaining an adequate sample size. Research has also pointed to potential bias caused by the way in which exposure to accidents is measured (3, 4). Because of the problems associated with traditional accident analyses, some have promoted the use of tqe traffic conflicts technique (5). However, this procedure also has shortcomings in that it.requires extensive field data collection and trained observers to identify the different types of conflicts occurring in the field. The objective of the research described herein was to develop a computational procedure for evaluating the safety-based level of service of signalized intersections that would be compatible and consistent with that presently found in the HCM for evaluating efficiency-based level of service as measured by delay per vehicle (6). The intent was not to develop a new set of accident prediction models, but to design a methodology to quantitatively predict the relative hazard of alternative intersection designs and signal timing plans.