Underactuated ships are facing problems of interference of time-varying multi-source disturbance and sudden changes in control force at the turning points during path following. Aiming at the above two problems, this paper proposed an optimal control method by combining the extended state observer with model predictive control. Firstly, the ESO extended state observer was designed to compensate for time-varying multi-source disturbance and provide low-frequency control input. Secondly, a nonlinear MPC controller was designed to optimize the control force. Finally, the stability of the ESO-MPC cascade controller was proved based on the nonlinear separation principle. The simulation comparison with active disturbance rejection control method and MPC control method verified the effectiveness of the proposed ESO-MPC cascade control method.

Numerical simulation of the resistance and sailing state of high speed amphibious vehicle in still water was conducted with the finite volume method based on Reynolds averaged equation to obtain the resistance characteristics of high speed amphibious vehicle in water. The mesh convergence of mesh was discussed by comparing the resistances of different mesh densities, and the numerical calculation method was validated by comparing the simulation results with model test results. In addition, the different resistance components of the amphibious vehicle were analyzed based on the calculation results of the multiple molding. Then, the changes of resistance and sailing states of the amphibious vehicle model with openings were calculated and analyzed. The results show that, for the amphibious vehicle in transition stage, the proportion of wave-making resistance to the total resistance increases with the increase of velocity (the maximum rate can reach over 60%), the proportion of viscous pressure resistance to total resistance is 30%~40%, and the proportion of friction resistance to total resistance is less than 10%. The resistance increases when the amphibious vehicle added with openings, and the resistance increases by about 11% and the draft increases by about 5% when Fr▽ is 1.297. Compared with the amphibious vehicle model with non-openings,  the amphibious vehicle model with openings shows changes in proportions of different resistance components to the total resistance: the proportion of wave-making resistance decreases and the proportion of viscosity-pressure resistance increases.

To solve the problem of low recognition efficiency and accuracy of Convolutional Neural Network (CNN) in automatic detection of gray image cracks in two-dimensional pavement, this paper proposed a three-stage road crack extraction algorithm based on feature fusion between layers of transposed CNN. The algorithm includes area judgment, image segmentation, multi-layer feature fusion and other modules. Then this study constructed a classification segmentation network and trained several transposed convolution networks of multi-fusion classification network intermediate layer and divided network output layer. Their operation effect was compared with that of CrackNet. The results show that when the minimum recall rate of CNN-Ⅰ is set to 0.95, the accuracy is 0.497, and the threshold value is 0.003152. According to the training results of CNN-Ⅱ, the accuracy of classification segmentation network is 0.78, recall rate is 0.73, F-1 score is 0.75, and the time for calculating a picture is shortened to less than 0.79 ms. The crack information extracted by multi-layer feature fusion method is more accurate because this method retains the continuity of the crack and realizes the optimization of automatic recognition and extraction of pavement cracks based on CNN.

Manual inspection is one of the main tasks in the management and maintenance of bridges. The inspection results of various damages provided by manual visual inspection or measurement is the key basic data for the decision-making of bridge operation and maintenance. However, due to the lack of standardized batch-processing method for manual inspection data, this kind of data is not fully utilized in the past practice. With the traditional methods, the process of those data is laborious, high cost and low timeliness. Therefore, these methods can not meet the requirements of bridge management and research. In consideration of those massive original damage data collected by manual inspection, this study proposed a standardized manual-marked pavement damage data-processing method based on image processing. Firstly, the labeled electronic CAD data was collected by manual inspection, and the damage formed image rasterization batch-processing was carried out to produce the damage distribution bitmap divided by the anchor points of suspenders (or stay cables). Secondly, the area, length and other information of the damage were extracted through the searching and assessment of the image connected components. Finally, the standardized data was processed, analyzed and output by the computer according to the grid labels in no time, and the damage data analysis was carried out, which including damage type analysis, location analysis, horizontal distribution analysis and vertical distribution analysis. The proposed method realizes the rapid and standardized big data processing and collation of manual inspection damage data, and provides an effective basic data processing method for the related researches and targeted guidance of bridge maintenance. In addition, the proposed standardized batch-processing method is quite portable and can provide reference for the processing and analysis of large amount and wide range of other component damage data and other types of data in manual inspection.

The aerodynamic interference on the adjacent three separated deck bridges in different altitude difference ratios was investigated by CFD and is verified by the wind tunnel test. The altitude difference ratio (H/D0) is within the range of -2 and +2, where H is the difference of centroid height between the middle box beam and the two sides symmetrical composite girders and D0 is the height of the middle box. The numerical simulation results show that the downstream composite beams have an inhibition effect on the vortex shedding of the middle box beams in a certain range of height difference. At the actual spacing in project, the mean values of the aerodynamics of the adjacent three separated deck bridges are interfered with each other obviously and regularly in the range of H/D0 stu-died in this paper. The aerodynamic interference of middle box girder is the largest, that of the downstream girder takes second place, and that of the upstream girder is the least.

In the seismic control of bridge structures in the transverse bridge direction, viscous dampers and steel dampers are seismic control devices with high reliability, and both have been initially applied in several bridge projects. In this paper, a comparative study of the two types of damping devices was carried out in terms of their ope-rating principles, constructional realization form, mechanical behaviour and seismic damping effects. The mechanical behaviour of the viscous dampers was compared with that of the steel dampers in the literature. As to their seismic damping effects, the difference in response between the two devices and the bridge structure under the excitation of 28 seismic waves was discussed by taking a three-tower cable-stayed bridge as the example. Based on one of the 28 seismic waves, the causes of this phenomenon was analyzed from the perspective of energy. The results show that the different damping devices have different energy dissipation mechanisms, resulting in different dynamic responses, but the seismic response shows no significant change; the viscous dampers are more effective than the steel dampers in controlling the bridge cross-bridge ground vibration at the transition and auxiliary piers under Class I site conditions. 

To effectively define the spatial scope of urban agglomeration considering traffic conditions and urban economic links so as to assist the transportation planning and spatial integrative development of the Guangdong-Hong Kong-Macao Greater Bay Area, firstly, the spatial pattern of raster grid highway traffic accessibility, road network density and economic connection intensity for the Guangdong-Hong Kong-Macao Greater Bay area were calculated and analyzed based on geographic information system (GIS) spatial analysis technology and modified gravity model. Then, different scales of urban hinterland were extracted and divided basing on highway traffic accessibility and urban connection degree. The research shows that the urban traffic accessibility of Guangdong-Hong Kong-Macao Greater Bay Area increases in a circle of “ core-periphery ” with Guangzhou-Foshan, Shenzhen-Guanzhou and other central cities as the center, while the traffic accessibility in  counties presents a distribution pattern of “ high in the middle and low at both ends”. The overall traffic accessibility of Guangdong-Hong Kong-Macao Greater Bay A-rea is at a high level, but all the cities in the Greater Bay Area have not yet achieved 1h access. Under the special development background of “one country, two systems and three customs zones” , the traffic linkages and coordination mechanisms between cities and within cities in the Bay Area are still relatively weak, and the traffic bottleneck problems in the inter-provincial and inter-city administrative boundaries are still prominent. The spatial network trend of economic connection intensity in the research area is significant. The leading cities such as Shenzhen, Hong Kong, Guangzhou, Foshan and Dongguan have obvious radiation effects, while the low value areas are distributed in the marginal areas with poor traffic accessibility. The urban hinterland area considering traffic conditions and economic scale in the Guangdong-Hong Kong-Macao Greater Bay Area does not match the corresponding administrative area. The urban hinterland of the two-wings in the Greater Bay Area is gradually “nibbed” by the central core cities. It is suggested to promote the internal space linkages between the central core cities of the Guangdong-Hong Kong-Macao Greater Bay Area urban and to promote the population flow and traffic flow. Thus can give full play to the intermediary role of the secondary center in cross-regional cooperation, and ultimately achieve the regional traffic planning and management of the super-large area without administrative constraints and realize the sustainable and balanced development of the Greater Bay Area.

Construction waste recycling filler is obtained from construction waste by crushing． Due to its diversified composition，It can be used for subgrade filling only after sorting． At present，manual screening method is timeconsuming． In this paper，convolution neural network was used for image analysis and the composition of regenerated packing can be obtained automatically． Firstly，a labeled dataset composed of 36000 granular images is created to train different CNN models． Among them，the user-defined resnet34 model with 40% Dropout rate performs the best，and its verification accuracy can reach 97% ． Secondly，the mass of particles was estimated based on the particle type and shape． Finally，the method proposed in this paper was compared with the manual screening method． For most recycled fillers，the quality difference is less than 2% ． This paper aims to improve the utilization of construction waste and it is of great significance to the popularization and application of construction waste backfill subgrade engineering．

In view of the existing problems such as poor energy absorption and buffering capacity of concrete rigid guardrail and the danger to driving safety，a new type of prefabricated anti-collision guardrail was proposed． In order to verify its feasibility，the LS-DYNA finite element software was used to establish a guardrail-bridge deck cooperative force model to simulate the collision process of a small car and a 40 t large truck with the guardrail at different speeds． Numerical results show that the vehicle doesn＇t climb over the guardrail in collision process，the climb height and exit angle are small，and the guardrail improves guidance effects of large trucks better than that of small cars． The peak value of lateral acceleration at the driver＇s position is significantly reduced up to 60. 8% ． The peak values of the Longitudinal and lateral collision force received by the new guardrail are reduced and the longitudinal force component is reduced by up to 49. 2% ． So the new guardrail has great guiding functions and better energy absorption effects，which can effectively improve driving safety． The guardrail design can provide reference for the actual car crash test．

In view of the phenomenon that more and more companies use electric vehicles ( EVs) to distribute cargo in urban areas，this paper presented an optimized EV distribution method with the time window by considering the intelligent charging strategy in the smart grid． The EVs can either charge or discharge when connecting to the smart grid through the vehicle to grid ( V2G) system． The V2G mode provides a more flexible way for EVs to operate． It can help EVs to increase efficiency and lower the cost at the same time． Based on the recharge and vehicle routing problem，this paper proposed a nonlinear integer programming model，considered the charging and discharging decisions of EVs，and proposed an improved genetic algorithm． Finally，25 cases were designed to verify the feasibility of the algorithm． The simulation result shows that the iterative efficiency of the improved genetic algorithm ( GA) is higher than other algorithms and the quality of the optimal solution is improved by 47% ．

The visual environment has long been regarded as an important factor affecting traffic safety． However， limited by image analysis methods，existing studies on the relationship between environmental visual factors and traffic safety are mainly qualitative and it is difficult to conduct large-scale quantitative analysis of the visual environment． This study used rich，easy-to-extract and growing streetscape images as the data source of environmental factors，and extracts the image feature information，location information and sensory information of streetscape through methods such as expanded residual network ( DＲN) ． And Pearson correlation coefficients and ridge regression were used to screen quantitative indicators to construct a quantitative analysis framework for the association between visual environment and traffic safety based on deep learning，providing a new quantitative means to study urban landscapes． In addition，statistical analysis methods were used to identify the influencing factors that lead to changes in road traffic safety conditions． On the one hand，the variability in the contribution of influencing factors associated with the urban landscape to accident rates in different urban areas was explored． For example，an increase in the proportion of ″vegetation″ in commercial and older urban areas has a positive impact on traffic safety， but the opposite is true in suburban areas． On the other hand，patterns that may advance urban planning theory are also discovered． For example，the closer the road unit is to the city center，the safer the traffic conditions． This paper offered a new way of thinking about the link between urban landscape and traffic safety in quantitative terms， and offered the possibility for assessing urban traffic safety conditions efficiently and on a large scale． 

A foreground detection method based on semantic segmentation and background reference was proposed to solve the problem of foreign objects detection in the space between platform screen doors and train doors in metro stations． This method used a depth neural network—DifferentNet to detect foreign objects region in images． Firstly， the background image and the image to be detected were obtained during a metro stop at the platform． The feature pyramid was obtained by extracting the feature information of the images through the encoding part of the network， and feature maps of the two images were merged by concatenation． Then the foreground heat map of the image to be detected was obtained by calculating the feature difference in the decoding part． Finally，the heat map was thresholded and filtered to get detection results． The results show that this method can achieve a high foreground IoU of 81. 2% and F1-score of 89. 5% ． Furthermore，it reached 30 fps in speed． The proposed method performed better than traditional methods and other image segmentation networks without background reference．

Ｒeasonable path planning can shorten the travelling time to ensure that rescue forces can arrive at the scene in time and improve the efficiency of emergency rescue． Based on the urban road traffic characteristics and the dynamic evolution of traffic environments，this paper proposed an extended co-evolutionary algorithm ( ECEA) to calculate the optimal rescue path． The ECEA establishes a co-evolutionary optimization mechanism，which means that the path planning process co-evolves with the evolution of traffic environments． Meanwhile，ECEA can flexibly select the search scope to improve the number and quality of alternative solutions． Experimental results show that ECEA outperforms timing co-evolutionary algorithm ( TCEPO) and Online re-optimization ( OLRO) both in the travelling time and robustness under the condition of limited data，thereby improving emergency rescue efficiency．

In order to analyze the impact of freeway three-dimensional spatial alignment continuity decline on traffic safety, this paper collected and statistically processed the alignment data and accident data of five interstate freeways in Washington State, USA from 2011 to 2018. The freeways were segmented with 100 meters length as the fixed length unit. The average spatial curvature of each unit (ASC), the absolute value of the curvature difference with the previous adjacent freeway unit(AVDP), and the absolute value of the curvature difference with the next adjacent freeway unit(AVDN) were calculated as descriptive variables, and the number of freeway unit accidents was used as the described variable. The traditional negative binomial model, zero-inflated negative binomial model, and zero-truncated negative binomial model were used to perform fitting analysis on all freeway units data and the freeway units data where accidents occurred. The results show that: the negative binomial model fits better than the zero-inflated negative binomial model based on all units data; the zero-truncated negative binomial model fits better than the negative binomial model based on the accident-occurring units data. In the four models, there is a large positive correlation between the absolute value of the curvature difference of the adjacent freeway unit and the safety level of the freeway unit, that is, increasing the continuous performance of the three-dimensional curvature of the freeway can improve the traffic safety level. It indicates that AVDP and AVDN can be used as a freeway alignment level evaluation index to provide support for the optimization of freeway alignment design.

A crash risk prediction model for freeway was developed with crash data and real-time traffic flow data to improve road active traffic management. The experimental sample sets were designed in matched case-control study and then the most significant traffic variables that have a crucial impact on the crash were selected by random forest algorithm. Based on the selected variables, the crash risk prediction model was developed in the support vector machine algorithm, and the performance of SVM models in the different kernel functions was compared. Meanwhile, in order to explore the effect of case-control matching ratios on the model performance, multiple sample sets with the different matching ratios were designed for the experiment. The results show that the model can effectively eva-luate the crash risk model according to the real-time traffic flow data. At the same time, the results show that increasing the case-control matching ratio has a particular effect on improving the models performance, and the ratio could be set explicitly according to traffic management needs.

Single-vehicle crash, as a common crash type of freeway crash, leads to great loss to society every year.  To reduce the negative impact brought by this type of crash, this study took freeway single-vehicle crash severity as dependent variable and constructed a econometric model to investigate the major influence factors. And some engineering and management countermeasures were put forward. The single-vehicle crash data of Guangdong Kaiyang freeway in 2013—2015 was collected. To describe the weather condition at the time when crash occurred more comprehensively, specifically and accurately, real-time weather data was matched to each crash according to the crash time and location. In terms of methodology, considering that the crash severity was ordered and there might exist heterogeneity in crash data, random parameter ordered logit model was established to fit the data. The results show that, the impact of vehicle type and humidity on crash severity has significant heterogeneity. The significant variables also include driver type, emergency medical services (EMS) response time, wind speed and crash time. Compared to non-professional driver, professional driver has a 6.22% higher probability for severe crash. Compared to other vehicles like coach, truck has a 0.59% lower probability for severe crash. Compared to daytime, the probability of severe crash at night decreases 0.31%. When the EMS response time increases 1min, the humidity increases 1%, and the wind speed decreases 1m/s, the probability of severe crash would increase 0.01%, 0.02% and 0.10% respectively. Finally, according to the results, some safety countermeasures were put forward.

In order to overcome the shortcomings of the existing prediction methods, such as short prediction steps and insufficient utilization of spatial-temporal characteristics of road networks, and to predict the freeway travel time accurately, five commonly used prediction models, namely RF(Random Forests),XGBoost(Extreme Gradient Boosting),LSTM(Long Short-Term Memory),KNN(K-Nearest Neighbor), and SVR(Support Vector Regression), were taken to carry out multi-steps prediction of freeway travel time based on the origin and destination data set. A fusion model based on Bayesian linear regression method was proposed. The Long-gang to Bu-long section of Shui-guan Expressway in Guangdong province was taken as a case study. We predicted the travel time of every 15 minutes in the next 2 hours. The results show that the prediction performance of RF model and XGBoost model is stable under multi-steps; the LSTM model has superior prediction performance in the case of short prediction steps; the fusion method integrates the advantages of various prediction methods and has higher accuracy and robustness. The experiments also demonstrate that it has the best prediction performance.

Aiming at the multi-objective optimization ( MOO) problem of the range-extended electric vehicle regenerative braking control strategy，a real-time adaptive regenerative braking control strategy was proposed based on the MOO model and optimal optimization theory． Firstly，the vehicle simulation model was established on AVL / Cruise and Matlab /Simulnk software，and a MOO model was built with the system braking performance ( BP) ， regenerative braking loss efficiency ( RBLE) and battery capacity loss rate ( BCLR) as the objective functions based on NSGA-Ⅱ algorithm． Then Parato optimal solution was obtained through off-line optimization under the comprehensive regenerative braking performance． Combined with the optimization results，a real-time adaptive fuzzy controller was designed． The controller considers the road adhesion and the state of battery，and can adjust the distribution of the regenerative braking work-point online． Simulation results on WLTP driving cyclic conditions show that the adaptive regenerative braking control strategy can effectively balance the relationship among BP，RBLE and BCLR，and it can effectively reduce BP and RBLE while maintaining a small BCLR．

The public transport space is a high-risk environment for the spread of the epidemic because of the dense flow of people and the narrow and closed internal space of the vehicles． The characteristics of customized bus “one person，one seat”and “one stop direct”are just in line with the requirements of epidemic prevention and control． Therefore，considering the advantages of conventional bus and customized bus，this paper proposed a collaborative optimization method of urban conventional bus and customized bus under epidemic prevention and control． It established a total cost minimization model considering risk cost，operation cost and travel cost，and designed an improved column generation algorithm to solve the model． The results show that: the collaborative optimization scheme proposed in this paper is significantly better than the independent conventional bus scheme and customized bus scheme，and ensures the service level of passengers; with the decrease of risk level，the number of conventional bus lines increases，the number of customized bus lines decreases，and the total number of service increases with the decrease of epidemic level． The study confirms the beneficial effect of customized bus on epidemic prevention and control and provides a theoretical guidance for the optimization of public transport system under the situation prevention and control．

A numerical prediction method of rolling damping was established for asymmetric catamaran with computational fluid dynamics ( CFD) method and overset grid technology． Firstly，the study simulated the rolling decay curves with different rolling angles and ship speed under different transverse and longitudinal space，and analyzed the rolling damping characteristics． Then based on 3D potential flow theory and considering the correction of the nonlinear rolling damping and nonlinear restoring，the rolling motion response of asymmetric catamaran in regular waves was calculated，and the characteristics of rolling motion under different wave directions were analyzed． Finally， model tests were carried out to verify the numerical prediction methods of rolling damping and rolling motion． It is showed that the model test results are close to the numerical prediction，and the transverse wave disturbance in the side body leads to greater rolling response than that in the main body of the asymmetric catamaran．

To comprehensively consider the influence of the three-dimensional road alignment on the operating speed，the research on the operating speed prediction in the three-dimensional point cloud environment was carried out based on the driver's dynamic visual field． Firstly，the point cloud data was used to calibrate the road and the terrain on both sides，and the three-dimensional point cloud road environment was established． Secondly，through the transformation and projection of point cloud coordinates，the view plane model with the driver's viewpoint as the center was constructed to obtain the driver's static visual field image． Thirdly，according to the principle of space geometry and the parameters of dynamic visual field，the projection ellipse equation was calculated and the driver's dynamic visual field image was obtained． Finally，the road-to-dynamic-visual-field-area-ratio was used as the research index of operating speed，and the operating speed prediction model was established by the vehicle dynamic performances and threshold control． The prediction model was testified by the example of dong-lu-mountain section of Ningxia-Hangzhou freeway． The results show that the average relative error between the predicted values and the measured values is 2. 726% for passenger cars and 9. 023% for trucks． The prediction result of trucks indicates that this section is a bad alignment section and the result is consistent with the measurement result． Therefore，the proposed prediction model can effectively solve the problem that the existing model cannot accurately predict the operating speed inside the section division unit of freeway