Short-term traffic flow prediction is a core branch of intelligent traffic systems (ITS) and plays an important role in traffic management. The graph convolution network (GCN) is widely used in traffic prediction models to efficiently handle the graphical structural data of road networks. However, the influence weights among different road sections are usually distinct in real life and are difficult to analyze manually. The traditional GCN mechanism, which relies on a manually set adjacency matrix, is unable to dynamically learn such spatial patterns during training. To address this drawback, this study proposes a novel location graph convolutional network (location-GCN). The location-GCN solves this problem by adding a new learnable matrix to the GCN mechanism, using the absolute value of this matrix to represent the distinct influence levels among different nodes. Subsequently, long short-term memory (LSTM) is employed in the proposed traffic prediction model. Moreover, trigonometric function encoding was used in this study to enable the short-term input sequence to convey long-term periodic information. Finally, the proposed model was compared with the baseline models and evaluated on two real-world traffic flow datasets. The results show that our model is more accurate and robust than the other representative traffic prediction models.@en

The representation and discrimination of various traffic states play an essential role in solving traffic accidents and congestion as the foundation of traffic state prediction. However, the existing representation of the traffic state usually only considers the road congestion layer and divides the traffic state into congested and unblocked. Representation only at the congestion layer is difficult to reflect the road traffic state comprehensively. Therefore, we select three indicators from the layers of road congestion, road safety, and road stability, respectively, then utilizing K-means to cluster the traffic state. The clustering results can be regarded as a new type for the representation of a traffic state. As a result, the traffic states are divided into four classes, which comprehensively reflects the level of road congestion, safety, and stability. Using the four traffic states obtained from the clustering results as class labels, we applied a multi-layer perceptron (MLP) to classify the different traffic states, and the receiver operating characteristic (ROC) curve is assessed to verify the superiority of the classification results. Finally, a visual display of the real-time traffic state in a city’s central area was given.

@en

To date, automatic driving technology has become a hotspot in academia. It is necessary to provide a personalization of automatic driving decision for each passenger. The purpose of this paper is to propose a self-learning method for personalized driving decisions. First, collect and analyze driving data from different drivers to set learning goals. Then, Deep Deterministic Policy Gradient algorithm is utilized to design a driving decision system. Furthermore, personalized factors are introduced for some observed parameters to build a personalized driving decision model. Finally, compare the proposed method with classic Deep Reinforcement Learning algorithms. The results show that the performance of the personalized driving decision model is better than the classic algorithms, and it is similar to the manual driving situation. Therefore, the proposed model can effectively learn the human-like personalized driving decisions of different drivers for structured road. Based on this model, the smart car can accomplish personalized driving.

@en

End-to-end learning from crowds has recently been introduced as an EM-free approach to training deep neural networks directly from noisy crowdsourced annotations. It models the relationship between true labels and annotations with a specific type of neural layer, termed as the crowd layer, which can be trained using pure backpropagation. Parameters of the crowd layer, however, can hardly be interpreted as annotator reliability, as compared with the more principled probabilistic approach. The lack of probabilistic interpretation further prevents extensions of the approach to account for important factors of annotation processes, e.g., instance difficulty. This paper presents SpeeLFC, a structured probabilistic model that incorporates the constraints of probability axioms for parameters of the crowd layer, which allows to explicitly model annotator reliability while benefiting from the end-to-end training of neural networks. Moreover, we propose SpeeLFC-D, which further takes into account instance difficulty. Extensive validation on real-world datasets shows that our methods improve the state-of-the-art.@en

The natural environmental factors that typically affect the safe environment of maritime traffic are weather conditions and ocean conditions, specifically, wind, current, and waves. The environmental influence factors affect the ship's navigation and the crew's decisions by affecting the ship's maneuverability, along with the skill and mentality of the crew. In addition, autonomous ship maneuvering decisions are influenced by several factors, and it is especially important to distinguish the crucial influencing factors for efficient selection of the corresponding maneuvering decisions. At the same time, the autonomous ship maneuvering decision factors are a typical 'grey system which is suitable for the research by grey relational analysis. Therefore, this study mainly focuses on the concept of human-like maneuvering for the autonomous ship. Based on the actual crews' operational data from full-task handling simulation platform, we proposed a grey relational analysis model to prioritize the environmental influencing factors, thus to analyze the decision-making mechanisms for specific ship maneuvering scenario. This method can mine the key factors which affect maneuvering decisions and provide guidance for an autonomous ship-assisted or automatic maneuvering system and promotes the application of autonomous ships.

@en

To date, the increasing density of water traffic has caused the ship's navigation environment to deteriorate, resulting in frequent water traffic accidents. In addition, a majority of maritime accidents are caused by human factors, and one of the important ways to solve the ship accidents caused by human factors is to utilize intelligent maneuvering of ships. Based on the actual crews' operational data from full-task handling simulation platform, this study combines a 30,000-ton bulk carrier inbound navigation scenario and uses the decision tree method to propose a knowledge learning model under multiple environmental constraints to give intelligent ships the ability to make decisions like a human: An intelligent ship Human-like Decision-making Maneuvering Decision Recognition (HDMDR) model. The decision-making mechanism for the maneuvering behavior of Officer On Watch (OOW) under the influence of the specific water traffic environment in the inbound scenario is analyzed, and the OOW's decision-making knowledge is automatically acquired and represented. The validation tests and the comparative analysis with the classic classification algorithms of k-Nearest Neighbours (k-NN) and Support Vector Machine (SVM) are performed to demonstrate the accuracy of the proposed HDMDR model. This paper provides a feasible basis for the human-like decision-making analysis of intelligent ships.

@en

With the further development of marine and information technologies, ship intelligence, green policies and automation will become mainstream with global cargo ships. Ship labor costs increase every year, so for the foreseeable future, the number of experienced crew members will be greatly reduced as smart ship emergence accelerates. At present, there is no mature research system for the human-like piloting of smart ships. In this paper, we use an improved decision tree, which could address problems of fuzziness and uncertainty. This will allow us to study the decision mechanisms of different piloting behaviors in order to realize the automatic acquisition and representation of the pilot's decision-making knowledge in inbound ship analysis as well as the simulated reproduction of the pilot's behavior. The simulation results show that the piloting decision recognition model, based on the fuzzy Iterative Dichotomiser 3 (ID3) decision tree, possesses a high reasoning speed and can accurately identify current piloting behavior. This provides theoretical guidance and a feasibility basis for research into human-like piloting behavior and the realization of automatic smart ship piloting systems.

@en