Traffic Load Flow Simulation Theory And Load Modelling Method For Bridges Via Data-Driven Strategies

Bridge vehicle loads are one of the variable loads that deserve significant attention during the life cycle of a bridge,and the accuracy and validity of the related methods and models are of particular concern in practical applications.Current research still utilizes limited traffic and load data to calibrate and evaluate traditional vehicle load models.At the level of analysis methods,the use of data follows the previous basic ideas based on the process of statistics,extrapolation,and calibration.There is still a lack of consideration on how the data can be used and the scenarios where they can be applied.At the same time,it is noted that the increase of data related to bridge vehicle loads has brought about a vast amount of traffic and load information.Still,theories and methods related to the fusion,mining,modeling,and evaluation of these data and information have not been systematically developed yet.There is still a gap between bridge vehicle load models and simulation methods and the current data-driven ideas and technologies,which needs to be addressed and improved.Based on the new perspective of data-driven technologies,this paper starts from the core technology of artificial intelligence and explores the topics related to load response reproduction,load clustering modelling,traffic flow generation,load response prediction and automated driving impact.They correspond to the five perspectives of data: data fusion,data mining,data generation,status prediction,and scene migration.In this way,this paper initially develops the basic framework of flow-density-speed-load,serving the objective of traffic load flow simulation and modelling.For each method,the application of relevant practical cases is elaborated.The main research contents and results are introduced as follows:The current microscopic car-following models are all derived from traffic engineering and only consider the fitting of traffic trajectories.With the objective of accurate reproduction of the bridge load response under traffic load flow,the paper uses a multi-objective optimisation approach to introduce the load information into the microscopic traffic flow model.The advantages and disadvantages of the Intelligent Driver Model and the Gipps’ model in model fitness,model robustness and solution set compactness are compared,and recommended values for the model parameters corresponding to different truck weights and bridge spans are given.The improvement of the microscopic traffic flow model in the basic theoretical framework of flow-density-velocity-load is addressed by supplementing the load parameters.Research on data mining is performed in response to the current situation of missing data for some vehicle load scenarios in bridge loading studies.When the data volume is small,data classification and vehicle weight generation based on macroscopic traffic load parameters is conducted.When the data volume is large,traffic scenario identification based on cross-sectional vehicle time series is conducted.By means of data dimensionality reduction and clustering,the traffic pattern characteristics of the traffic flow data under the influence of load are revealed and then applied to the overall load response analysis of large-span suspension bridges.The construction of the measured database in the basic theoretical framework of flow-density-velocity-load is addressed by efficiently utilising traffic load data.The mainstream load flow simulation methods are mainly based on deterministic traffic evolution implementation,which has inherent limitations in the stochastic nature of short-time traffic evolution.Therefore,it is difficult to readily apply to predicting the short-time most adverse load response.The study introduces vehicle interaction information at different spatial scales by redesigning generators and discriminators in the generative adversarial network.A stochastic traffic flow model is proposed to perform small-scale short-time evolutionary simulations under the premise of satisfying the physical relationship.The analysis of the quality of traffic data,the role of modules and traffic simulation and visualisation are carried out.The microscopic traffic flow model calibrated in Chapter 2 is compared to demonstrate the feasibility and effectiveness of this method,thereby solving the problem of adverse load distribution generation in the basic theoretical framework of flow-density-velocity-load.For bridge safety monitoring and early warning,a significant lag exists in the traffic flow load simulation,which forbids the rapid prediction of the load response.Load response prediction based on historical load response and sectional traffic information are proposed to address this need using the loaded traffic and load response generated in Chapter 3 as the database.Correspondingly,a prediction model based on a spectral-temporal graph neural network and one based on an attentional temporal graph convolutional network are established to this end,respectively.In this process,the weight matrix automatically learned by the first prediction model will be used as an auxiliary input to the second one.The second model incorporates the influence line as an additional output and attempts to construct a neural network with physical information.The two models together solve the problem of simultaneous prediction of multiple load responses in the basic flow-density-velocity-load theoretical framework.There are significant differences in driving patterns and behavioural purposes between human-driven and autonomous vehicles.Significant differences in driving trajectories may lead to changes in load response.The paper draws on the human-driven and autonomous car-following models and incorporates a dynamic formation algorithm for autonomous trucks to propose a mixed traffic flow model with multiple driving modes.It considers the effects of traffic composition,truck formation length,traffic volume and congestion states to achieve large-scale long-time simulations of mixed traffic flows and analyses the response level of vehicle loads under the influence of future traffic.It solves the problem of load responses under mixed traffic flow in the basic theoretical framework of flow-density-velocity-load.