Three years after its 2010 enactment, the French government proposed a timeline for the development of a new metro network providing new rapid transit lines in the Ile de France region. Implemented by the Société du Grand Paris (SGP), the Grand Paris Express (GPE) thus became the largest transport project in Europe. As in any new railway project, a safety record must be established by the SGP. Among all criteria that had to be investigated, the ability of the network manager to prevent and detect broken rails was a particularly sensitive point. Indeed, beyond the obvious consequences induced by a broken rail impacting the safety of passengers, such an event has a very strong impact on the availability of the infrastructure, which is a key point for all metro line automation projects. The SGP therefore required some decision support tools to enable us to evaluate the consequences of a broken rail on the network operating conditions and, moreover, to determine the best monitoring strategy and the adequacy of the maintenance policy to prevent broken rails. Based on a former study commissioned by RATP (the historic operator and infrastructure manager of the Paris metro network) dealing with the evaluation and optimization of detection and prevention of broken rails in a metro line automation context, the SGP wanted to possess an equivalent decision support tool customised to the needs and the characteristics of the GPE. Nevertheless, in the GPE context, the network that was considered does not yet exist. Moreover, an equivalent system does not exist. Therefore, no feedback databases are available to estimate reliability parameters necessary for such decision support tools. Therefore, the present paper proposes an original approach combining physical modelling of rail deterioration with statistics to overcome this limitation.

As an efficient large-scale grinding and drying equipment, the vertical roller mill is widely used in many fields, especially in the fields of grinding cement raw meal, clinker, and blast furnace slag. In this thesis, the key components of the vertical roller mill are preferred according to the design requirements, including the determination of general parameters (clamp angle, mill size, grinding roller size, pressure of grinding roller, speed of millstone, power and selection of the main motor, and selection of the main reducer), and the determination of process parameters (material thickness, material granularity, drying capacity, and air volume of the mill). Additionally, the mill production capacity is verified by the above-preferred parameters, and it is examined whether it meets the design requirements. Finally, the stress at the core position is obtained through the analysis of the loading system of the vertical roller mill.

The centralized control has been played a very important role in the control structure of microgrids. With the multi-source and multi-inverter system in microgrids, accompanying a large number of unbalanced and nonlinear loads, the coordination between system models and data will be delayed or blocked; this will not only affect the data communication, but also seriously affect the power quality. In this paper, a novel networked hierarchical control approach for voltage unbalance and harmonics compensation is proposed based on the communication technology and hierarchical control theory and combined with a distributed consensus algorithm. The networked hierarchical control includes primary and secondary control. In the primary controller, some basic controls are proposed and modified, with the secondary controller to maintain the stable operation of the system. In the secondary controller, using the distributed consistency algorithm, the distributed secondary control (DSC) method is proposed to eliminate the deviations of voltage and frequency and also improve the nonlinear load power sharing among distributed generators (DGs). In this control structure, primary control and secondary control are combined, each DG is controlled independently, and the entire controller system is connected through a communication network. Compared with centralized secondary control, it reduces the system's requirements for high communication bandwidth and improves the system's reliability and expansion flexibility. The proposed approach not only compensates for the voltage unbalance and harmonic at sensitive load bus (SLB), but also improves the precision of power distribution and load current sharing. Finally, simulation and experimental results validate the effectiveness and feasibility of the proposed approach.

Aiming at the problem of low efficiency and low accuracy in the classification of multiple unknown protocol data frames in an unknown network environment, this paper proposes a k-means clustering algorithm based on information entropy and density. First, according to the characteristics of the protocol data frame in the bitstream form, the Euclidean distance between the data frames is weighted by using information entropy. Then, the high-density data frame set is determined by the statistics of the density of each data frame, and the cluster center point is determined in this set by the maximum and minimum distance criterion. Finally, the ratio of the distance in the protocol cluster to the distance between the clusters is used to determine the number of unknown protocols. Simulation results show that this method can cluster the data frames of unknown bitstream protocol quickly and accurately.

To guarantee the security of carrier-based aircraft during wave-off maneuver process, in this paper, we propose a prediction approach of wave-off risk evaluation based on neural network. By means of the analysis on wave-off maneuver, the control model of carrier-based aircraft is constructed. Comparing the military thrust control system with the military thrust and elevator comprehensive system, we discuss the form of the wave-off safety area. Then, an improved wave-off risk evaluation approach based on neural network is represented. After defining the wave-off risk standard and safe standard, the Wave-Off Safe Area, Wave-Off Risk Area, and Wave-Off Warning Area are indicated to judge the safety of wave-off maneuvering. The experimental results show that both velocity and interference sink rate are influence factors for the signal of Landing Signal Officer, and the wave-off risk evaluation technology refines the area division of landing decision-making time. As such, the obtained results have certain reference values for pilots training and landing safety of carrier-based aircraft specifically.

There exists a complicated and variable feature of fault transient traveling wave in the hybrid transmission line. This paper proposed a ranging method by combining the single-terminal traveling wave method and two-terminal traveling wave method based on the polarity of the traveling wave. First, the two-terminal traveling wave method was used to calculate the approximate fault distance. Based on the approximate fault distance, and combined with the parameters of the line and the law of traveling wave polarity in the fault section, the reflection wave from the fault point and the reflection wave from the junction of the hybrid line could be accurately identified. The precise distance of the fault was finally calculated based on the time of the reflected wave and the initial traveling wave. The method not only evades errors caused by the synchronization system of global position system (GPS) in the two-terminal traveling wave method, but also overcomes the difficulty in recognizing the traveling-wave head in the single-terminal traveling wave method. Simulation results show that the method is suitable for accurate fault location of hybrid lines with different fault distances, fault types, and grounding resistance.

Various abnormal observations often occur in a network system, which result in serious reliability damage of the network system. These abnormal observations are much more complicated than those in a traditional system due to the size and varieties of the network system. In fact, an abnormal observation in the network system could be caused by any program, software or application running in the resource pool of the network system. Therefore, there indeed exist a kaleidoscope of causes of abnormal observations that decrease the reliability of the network system. Effectively guaranteeing the reliability of the network system becomes a critical challenge. In this paper, we present a reliability management system (RMS) by using deep learning (DL) and model driven approaches. This RMS can endow the network system with distinctive capabilities of classifying abnormal observations, as well as quantifying and guaranteeing system reliability. The proposed RMS first uses the DL to derive a classification model to find a suitable repair action for an occurred abnormal observation. Then, it also builds a reliability model to evaluate the reliability metric. Now, the RMS can use the reliability model to update the reliability metric of the system after adopting the repair action. If the repair action derived is not suitable, the corresponding classification may have some errors. Then, it would feedback an error to DL for coordinating the classification model. Therefore, the proposed RMS is capable of AI-based anomaly diagnosis and model-driven reliability guarantee.

This paper proposes a process to deploy a cluster of distributed sandboxes to trace, track and attribute possible threats and attacks. The deployed distributed sandboxes detect suspicious threats collaboratively and publish the results as transactions to a directed acyclic graph (DAG) at the expense of verifying transactions existing in DAG in terms of signatures and relevant threats/attacks. A set of policies are designed to maintain DAG effectiveness, efficiency and fairness. Based on data on the DAG, association analysis is performed to produce threat intelligence for the deployment decision in the next round. To reduce the deployment cost, a stochastic programming is developed to take the uncertainty into consideration. Preliminary experiments are carried out to evaluate the feasibilities of the proposed approach.

Software defect prediction is an important way to find software defects and improve software quality. In recent years, with the rapid development of the software industry and the continuous improvement of data mining technology, a large amount of labeled software data is continuously generated to form a software data stream. However, traditional software defect prediction based on batch learning cannot fully adapt to this form of data flow, so a method of software defects prediction incremental model using ensemble learning (SDPIE) is proposed to learn and process real-time software data streams. In this process, the under-sampling method and selective learning method are used to solve the problem of software data imbalance and the increase of the incremental learning classifier scale. Through a comparison with two incremental learning algorithms and two batch learning algorithms, it is shown that the SDPIE algorithm can effectively control the size of the algorithm and the prediction performance is better than the other four comparison algorithms.

This paper shows that characteristics of competitive oligopoly exist in the search engine market by analyzing market shares. We put forward a novel optimization model, named as an oligopoly two-stage-game (OTSG) model in this paper. The model is based on Stackelberg and Cournot models. The model is proposed to maximize the profit of search engine platforms. Combining game theory and economic analysis, the strategies of optimal pricing and advertising quantity for search engine platforms have been investigated at different cost functions. According to the monopoly characteristics in the market, the proposed model considers followers entering and demand elasticity factors. The equilibrium between price and advertising quantity is deduced. The impact of demand elasticity on price and advertising quantity is analyzed in the OTSG model. Furthermore, the relation between the demanded advertising quantity and the remaining adverting quantity is shown in the second stage market. Finally, the simulation results show that Nash equilibrium exists in the second stage market.

Whilst existing research mainly focus on detecting the saliency of dynamic objects based on spatiotemporal features, it is also meaningful to detect the saliency of static objects and label their salient values on the video saliency map, a useful tool for many high-level applications. In view of these, we propose a novel salient object detection model for video sequences, which combines the dynamic saliency and the static saliency into a co-saliency map. First, the salient degree of the general objects in each frame was estimated by the motion-independent algorithm, and the global static saliency map was generated based on the results. Next, the dynamic regions were detected by an improved motion-based approach, and the dynamic saliency map was computed with a local saliency detection method according to the related dynamic regions and the visual fixation map. Finally, a novel co-saliency algorithm was devised to fuse the static and dynamic maps. The final hierarchical co-saliency map reflects the saliency of both dynamic and static objects, and it satisfies the demand of more advanced tasks. Through the evaluation on two existing datasets, it is proven that the proposed model can achieve state-of-the-art performance.

It is a challenging problem to detect small and dense objects in remote sensing images. To address these problems, this paper presents a novel network based on YOLOv3 to detect objects in a one-stage framework, which contains two improvements. Firstly, K-means++ is utilized to determine the number of initial bounding boxes and aspect ratio dimensions, which is beneficial to adaptively adjust the parameters in YOLOv3 and to boost the small object detection in remote sensing images. Secondly, the Soft Non-Maximum Suppression (Soft-NMS) technique is introduced to reduce the confidence values rather than directly discard the candidate boxes whose Intersection-over-Union (IOU) values are larger than a predefined threshold. The Soft-NMS can increase the object detection accuracy for small and dense objects by suppressing the redundant boxes with lower confidence values. Extensive experimental results on the standard NWPU VHR-10 remote sensing dataset demonstrate that the proposed deep network can accurately and efficiently detect a variety of small objects compared with state-of-the-art object detection methods.

Sequencing constraints are restrictions on the order of operations (statements, methods, or calls) during program execution. They trigger software failures if sequencing constraints are violated during software execution; this is a common defect in software systems. Sequencing constraint defects are difficult to find because they are specification-related. To resolve this issue, we propose a defect detection approach based on static analysis. In our approach, first, we formalize the sequencing constraints as regular expressions extracted from software specification, and define regular expression of defect patterns that violate sequencing constraints. Then, the software is converted into a control flow graph (CFG), and the CFG is labeled by the constraint information to generate a sequencing-constraint control flow graph (SC-CFG) that is converted into a path expression using algebraic representation. Finally, a suspiciousness defect report is outputted by pattern matching, which compares defect patterns with string patterns extracted by the path expression. The experiments are performed to validate the effectiveness of our approach. The evaluation results show that our approach is effective at detecting sequencing constraint defects before software testing.

A braking control system is an essential part of urban rail vehicle braking systems. Due to the uncertainty of operational data, direct analysis can affect the authenticity of system reliability. To solve the deviation of system reliability analysis caused by uncertainty, GO Methodology is combined with fuzzy theory to establish the GO graph model of the system according to the characteristics of the braking control system. The membership function of the fuzzy success rate of the system is obtained by using the modified signal flow calculation method combined with the fuzzy number algebraic calculation method, and the system fuzzy success rate with high reliability is selected. The calculated results can provide guidance and suggestions for system maintenance, and they have strong engineering practicability.

Creativity has become more important in modern life due to the development of quality of life. The paper first discusses the relationships among creativity, building environment, and machine learning to illustrate that an energy conservative building environment can raise occupants' creativity and satisfaction. Then, an energy conservative way aided by machine learning is proposed. The building operational patterns are identified using this machine-learning model. The studied database builds air-conditioning energy consumption of a Singaporean school. The clustering process first removes inconsistent and null data. Then, four features are selected from the original 18 features based on the domain knowledge and the characteristics of the dataset. The k-means algorithm is used to discover the hidden building operational patterns in the building energy consumption information. The results show that the three derived clusters can be well interpreted by domain knowledge. An energy conservation measure is carried out based on further investigation on the clusters. The data mining process proposed in this article is a creative way for intelligent building. Future research could also be conducted based on the creative method in this article; possible future directions are highlighted in the last section.