Network traffic assignment/equilibrium models have been widely used for transportation planning. For traffic management, one is interested on how traffic patterns change dynamically since equilibrium cannot be reached instantaneously. This dissertation focuses on modeling short-term traffic patterns in a transportation network, and addresses these topics: (1) comparison of loading route flows versus link flows, (2) development of a mesoscopic model for loading route flows in a network, (3) estimation of route flows based on the mesoscopic model, and (4) optimizing traffic signal timings based on the estimated route flows.With regard to the first topic, many models propagate flows through a network using link flows and independent turning probabilities (ITP) at nodes. Chapter 2 describes the effects of the ITP assumption on the traffic patterns that occur based on route flow loading; this provides the motivation for using route flows in this research.Route trajectories are the spatial-temporal realizations of vehicle route flow demands. This dissertation proposes a mesoscopic simulation platform where route flows are propagated through the network and dynamic trajectories are computed. Under interrupted and uninterrupted flow conditions, route trajectories from the mesoscopic model are compared with ones from a microscopic model, the latter model being used to provide realistic data since real data at this level of detail is not currently available. Results from both models match well; the corresponding traffic patterns are very similar, both graphically and statistically.In chapter 5, a model is presented for estimating temporal route flow demands when real-time data is available. The model is formulated as a bi-level optimization problem where a least-square model is constructed at the upper level and the mesoscopic model is utilized at the lower level to relate flow demands and route trajectories. Based on real-time measurements, this model estimates dynamic route flows that are consistent with observed traffic patterns. Computational results using a micro-simulator for "real" data show that the model estimates well the route flows loaded in the micro-simulation model.Finally, in chapter 6, a traffic signal control optimizer is developed based on the mesoscopic model and a gradient search to optimize any given performance index such as average delay. A numerical example shows that the optimizer significantly improves the performance index and the approach may be used for on-line traffic signal control.

As Next Generation (NextG) networks become more complex, the need to develop a robust, reliable network traffic prediction framework for intelligent network management increases. This study compares the performance of machine learning models in network traffic prediction using a custom Service-Level Agreement (SLA) - based loss function to ensure SLA violation constraints while minimizing overprovisioning. The proposed SLA-based parametric custom loss functions are used to maintain the SLA violation rate percentages the network operators require. Our approach is multivariate, spatiotemporal, and SLA-driven, incorporating 20 Radio Access Network (RAN) features, custom peak traffic time features, and custom mobility-based clustering to leverage spatiotemporal relationships. In this study, five machine learning models are considered: one recurrent neural network (LSTM) model, two encoder-decoder architectures (Transformer and Autoformer), and two gradient-boosted tree models (XGBoost and LightGBM). The prediction performance of the models is evaluated based on different metrics such as SLA violation rate constraints, overprovisioning, and the custom SLA-based loss function parameter. According to our evaluations, Transformer models with custom peak time features achieve the minimum overprovisioning volume at 3% SLA violation constraint. Gradient-boosted tree models have lower overprovisioning volumes at higher SLA violation rates. / Master of Science / As the Next Generation (NextG) networks become more complex, the need to develop a robust, reliable network traffic prediction framework for intelligent network management increases. This study compares the performance of machine learning models in network traffic prediction using a custom loss function to ensure SLA violation constraints. The proposed SLA-based custom loss functions are used to maintain the SLA violation rate percentages required by the network operators while minimizing overprovisioning. Our approach is multivariate, spatiotemporal, and SLA-driven, incorporating 20 Radio Access Network (RAN) features, custom peak traffic time features, and mobility-based clustering to leverage spatiotemporal relationships. We use five machine learning and deep learning models for our comparative study: one recurrent neural network (RNN) model, two encoder-decoder architectures, and two gradient-boosted tree models. The prediction performance of the models was evaluated based on different metrics such as SLA violation rate constraints, overprovisioning, and the custom SLA-based loss function parameter.

Variable bit rate (VBR) compressed video traffic, such as live video news, is expected to account for a large portion of traffic in future integrated networks. This real-time video traffic has strict delay and loss requirements, and exhibits burstiness over multiple time scales, thus imposing a challenge on network resource allocation and management. The renegotiated VBR (R-VBR) scheme, dynamically allocating resources to capture the burstiness of VBR traffic, substantially increases network utilization while satisfying any desired quality of service (QoS) requirements. This thesis focuses on the performance evaluation of R-VBR in the context of different R-VBR approaches. The renegotiated deterministic VBR (RED-VBR) scheme, proposed by Dr. H. Zhang et al., is thoroughly investigated in this research using a variety of real-world videos, with both high quality and low quality. A new Virtual-Queue-Based RED-VBR is then developed to reduce the implementation complexity of RED-VBR. Simulation results show that this approach obtains a comparable network performance as RED-VBR: relatively high network utilization and a very low drop rate. A Prediction-Based R-VBR based on a multiresolution learning neural network traffic predictor, developed by Dr. Y. Liang, is studied and the use of binary exponential backoff (BEB) algorithm is introduced to efficiently decrease the renegotiation frequency. Compared with RED-VBR, Prediction-Based R-VBR obtains significantly improved network utilization at a little expense of the drop rate. This work provides evaluations of the advantages and disadvantages of several R-VBR approaches, and thus provides a clearer big picture on the performance of the studied R-VBR approaches, which can be used as the basis to choose an appropriate R-VBR scheme to optimize network utilization while enabling QoS for the application tasks. / Master of Science

Automatic Incident Detection Algorithms (AIDA) have been part of freeway management system software from the beginnings of ITS deployment. These algorithms introduce the capability of detecting incidents on freeways using traffic operations data. Over the years, several approaches to incident detection have been studied and tested. However, the size and scope of the urban transportation networks under direct monitoring by transportation management centers are growing at a faster rate than are staffing levels and center resources. This has entailed a renewed emphasis on the need for reliability and accuracy of AIDA functionality. This study investigates a new approach to incident detection that promises a significant improvement in operational performance. This algorithm is formulated on the premise that the current conditions facilitate the prediction of future traffic conditions, and deviations of observations from the predictions beyond a calibrated level of tolerance indicate the occurrence of incidents. This algorithm is specifically designed for easy implementation and calibration at any site. Offline tests with data from the Georgia-Navigator system indicate that this algorithm realizes a substantial improvement over the conventional incident detection algorithms. This algorithm not only achieves a low rate of false alarms but also ensures a high detection rate.

Afin de répondre aux besoins de la 5G, nous évaluons plusieurs visions du cœur de réseau. Nous comparons les performances des visions en mesurant le temps nécessaire pour établir le service pour l’utilisateur. De plus, nous proposons dans cette thèse un algorithme basé sur la théorie du contrôle permettant d’équilibrer la charge entre les instances AMF, et d’augmenter ou de diminuer le nombre d’instances AMF en fonction de la charge du réseau. En outre, nous proposons un nouveau mécanisme pour adapter les ressources du réseau cœur 5G en anticipant les évolutions, de la charge de trafic, grâce à des prédictions via des approches de machine learning. Enfin, nous proposons une solution pour généraliser les réseaux de neurones tout en accélérant le processus. / In order to fulfil the needs of 5G, we evaluate, using a testbed, multiple visions of core networks. We compare the performances of the visions in terms of the time needed to create the user data plane. In addition, we provide an algorithm based on Control Theory allowing to equilibrate the load on the AMF instances, and to scale out or in the AMF instances depending on the network load. Also, we propose a novel mechanism to scale 5G core network resources by anticipating traffic load changes through forecasting via deep learning approaches. Finally, we propose a novel solution to generalize neural networks while accelerating the learning process.

Traffic data prediction is a critical aspect of Advanced Traffic Management System (ATMS). The utility of the traffic data is in providing information on the evolution of traffic process that can be passed on to the various users (commuters, Regional Traffic Management Centers (RTMCs), Department of Transportation (DoT), ... etc) for user-specific objectives. This information can be extracted from the data collected by various traffic sensors. Loop detectors collect traffic data in the form of flow, occupancy, and speed throughout the nation. Freeway traffic data from I-4 loop detectors has been collected and stored in a data warehouse called the Central Florida Data Warehouse (CFDW[trademark symbol]) by the University of Central Florida for the periods between 1993-1994 and 2000 - 2003. This data is raw, in the form of time stamped 30-second aggregated data collected from about 69 stations over a 36 mile stretch on I-4 from Lake Mary in the east to Disney-World in the west. This data has to be processed to extract information that can be disseminated to various users. Usually, most statistical procedures assume that each individual data point in the sample is independent of other data points. This is not true to traffic data as they are correlated across space and time. Therefore, the concept of time sequence and the layout of data collection devices in space, introduces autocorrelations in a single variable and cross correlations across multiple variables. Significant autocorrelations prove that past values of a variable can be used to predict future values of the same variable. Furthermore, significant cross-correlations between variables prove that past values of one variable can be used to predict future values of another variable. The traditional techniques in traffic prediction use univariate time series models that account for autocorrelations but not cross-correlations. These models have neglected the cross correlations between variables that are present in freeway traffic data, due to the way the data are collected. There is a need for statistical techniques that incorporate the effect of these multivariate cross-correlations to predict future values of traffic data. The emphasis in this dissertation is on the multivariate prediction of traffic variables. Unlike traditional statistical techniques that have relied on univariate models, this dissertation explored the cross-correlation between multivariate traffic variables and variables collected across adjoining spatial locations (such as loop detector stations). The analysis in this dissertation proved that there were significant cross correlations among different traffic variables collected across very close locations at different time scales. The nature of cross-correlations showed that there was feedback among the variables, and therefore past values can be used to predict future values. Multivariate time series analysis is appropriate for modeling the effect of different variables on each other. In the past, upstream data has been accounted for in time series analysis. However, these did not account for feedback effects. Vector Auto Regressive (VAR) models are more appropriate for such data. Although VAR models have been applied to forecast economic time series models, they have not been used to model freeway data. Vector Auto Regressive models were estimated for speeds and volumes at a sample of two locations, using 5-minute data. Different specifications were fit--estimation of speeds from surrounding speeds; estimation of volumes from surrounding volumes; estimation of speeds from volumes and occupancies from the same location; estimation of speeds from volumes from surrounding locations (and vice versa). These specifications were compared to univariate models for the respective variables at three levels of data aggregation (5-minutes, 10 minutes, and 15 minutes) in this dissertation. For data aggregation levels of [less than]15 minutes, the VAR models outperform the univariate models. At data aggregation level of 15 minutes, VAR models did not outperform univariate models. Since VAR models were used for all traffic variables reported by the loop detectors, this made the application of VAR a true multivariate procedure for dynamic prediction of the multivariate traffic variables--flow, speed and occupancy. Also, VAR models are generally deemed more complex than univariate models due to the estimation of multiple covariance matrices. However, a VAR model for k variables must be compared to k univariate models and VAR models compare well with AutoRegressive Integrated Moving Average (ARIMA) models. The added complexity helps model the effect of upstream and downstream variables on the future values of the response variable. This could be useful for ATMS situations, where the effect of traffic redistribution and redirection is not known beforehand with prediction models. The VAR models were tested against more traditional models and their performances were compared against each other under different traffic conditions. These models significantly enhance the understanding of the freeway traffic processes and phenomena as well as identifying potential knowledge relating to traffic prediction. Further refinements in the models can result in better improvements for forecasts under multiple conditions.

<p>The study of Internet traffic behavior in a real IP network is the subject of this thesis. Traffic Matrix of a telecommunication network represents the exchanged traffic volume between the source and destination nodes in the network and is a critical input for network studies. However, in most cases, traffic matrices are not readily available. Existing network management protocols such as the ‘Simple Network Management Protocol’ (SNMP) have been used to gather other observable measures, such as link load observations. The first part of this thesis reviews famous methods and approaches that try to infer and estimate the source-destination traffic matrix from the observable link loads.</p> <p>Another important subject in networks is to predict bandwidth requirements in the future. The second part of this thesis reviews some existing methods and approaches of traffic prediction. Recently a traffic prediction method which uses multiple Time-Series analysis, each operating on a different time-scale, has been proposed. This method uses multiple ‘AutoRegressive Integrated Moving Average’ (ARIMA) filters to predict the future bandwidth requirements. Each ARIMA filter operates on a different time scale, i.e., quarter-hour, hour, day, and week. The proposed method associates a weight with each ARIMA filter, and adjusts the weights according to which filter is currently the most accurate predictor. A review of this newly proposed method is presented. Extensive experimental results have been gathered to test the robustness of the method. The filter coefficients of each ARIMA filter have been varied, and the accuracy of the predicted traffic has been measured. Extensive experimental measurements indicate that the model is very robust, and that large changes to each filter's coefficients have only a small effect on the accuracy. In all cases we evaluated, the method is very robust, predicting short-term future traffic demands with typically ≈95% success rates.</p> / Master of Applied Science (MASc)

As Next Generation (NextG) networks become more complex, it has become increasingly necessary to utilize more advanced algorithms to enhance the robustness, autonomy, and reliability of existing wireless infrastructure. One such algorithm is network traffic prediction, playing a crucial role in the efficient operation of real-time and near-real-time network management. The contributions of this thesis are twofold. The first introduces a novel cluster-train-predict framework that leverages domain knowledge to identify unique timeseries sub-behaviors within aggregates of network data. This method produces distributions that are more robust towards changes in the spatio-temporal environment. The ensemble of time-series prediction models trained on these distributions posses a greater affinity towards accurate network prediction, selectively employing learned behaviors to handle sources of time-series data without any prior knowledge of it. This approach tends to improve the ability to accurately forecast network traffic volumes. Secondly, this thesis explains the development and implementation of a modular data pipeline to support the cluster-train-predict framework under a variety of conditions. This setup promotes repeatable and comparable results, facilitating rapid iteration and experimentation on current and future research. The results of this thesis surpass traditional approaches in [1] by up to 60%. Furthermore, the effectiveness of this framework is also validated using two additional time-series datasets [2] and [3], demonstrating the ability of this approach to generalize towards other time-series data and machine learning applications in uncertain environments. / Master of Science / As Next Generation (NextG) networks become more complex, it has become increasingly necessary to utilize more advanced algorithms to enhance the robustness, autonomy, and reliability of in-use wireless infrastructure where network traffic prediction plays a crucial role in the efficient operation of real-time and near real-time network management. The contributions of this thesis are twofold. The first explores a novel cluster-train-predict framework that uses an unsupervised learning approach, specifically time-series K-means clustering, to group similar time-series data. In doing so, this approach identifies unique time-series behaviors within network provider data. Since this approach aims to reduce the variance within each aggregate, models can specialize towards particular network behaviors, becoming better suited for a wider variety of network trends during prediction. Because this framework assigns data to each cluster based on these groupings, the framework can adapt towards changes in network infrastructure or underlying shifts in its environment to forecast with a greater degree of certainty and explainability. This framework can even generalize towards out-of-distribution cases where it has no prior knowledge of a source of time-series data outperforming [1] by up to 60%. This approach tends to improve the ability to accurately forecast network traffic volumes. Secondly, this thesis explains the development and implementation of a modular data pipeline to support the cluster-train-predict framework under a variety of conditions with repeatable and comparable results, facilitating rapid iteration and experimentation on current and future research. The results of the framework are also corroborated on two, additional time-series datasets [2] and [3], demonstrating the ability of this approach to generalize towards time-series data, where this framework can also be applied to other machine learning applications in uncertain environments.

In 6G communication, many state-of-the-art machine learning algorithms are going to be implemented to enhance the performances, including the latency property. In this thesis, we apply Buffer Status Report(BSR) prediction to the uplink scheduling process. The BSR does not include information for data arriving after the transmission of this BSR. Therefore, the base station does not allocate resources for the new arrival data, which increases the latency. To solve this problem, we decide to make BSR predictions at the base station side and allocate more resources than BSRs indicate. It is hard to make an accurate prediction since there are so many features influence the BSRs. Another challenge in this task is that the time intervals are tremendously short (in the order of milliseconds). In other traffic predictions, the traffic data in a long term, such as in a week and month, can be used to predict the periodicity and trend. In addition, many external features, such as the weather, can boost the prediction results. However, when the time is short, it is hard to leverage these features. The datasets provided by Ericsson are collected from real networks. After cleaning the data, we convert the time series forecasting problem into a supervised learning problem. State-of-the-art algorithms such as Random Forest(RF), XGboost, and Long Short Term Memory(LSTM) are leveraged to predict the data arrival rate, and one K-Fold Cross-Validation is followed to validate the models. The results show that even the time intervals are small, the data arrival rate can be predicted and the downlink data, downlink quality indicator and rank indicator can boost the forecasting performance. / I 6G-kommunikation kommer många toppmoderna maskinin lärnings algoritmer att implementeras för att förbättra prestanda, inklusive latensegenskapen. I den här avhandlingen vill vi tillämpa Buffer Status Report (BSR) förutsägelse för schemaläggningsprocessen för upplänkning. BSR innehåller inte information för data som anländer efter överföring av denna BSR. Därför tilldelar basstationen inte resurser för den nya ankomstdatan, vilket ökar latensen. För att lösa detta problem bestämmer vi oss för att göra BSR-förutsägelser på basstationssidan och tilldela fler resurser än vad BSR anger. Det är svårt att göra en exakt förutsägelse eftersom det finns så många funktioner som påverkar BSR. En annan utmaning i denna uppgift är att tidsintervallen är oerhört korta (millisekunder). I andra trafikprognoser kan trafikdata på lång sikt, som under en vecka och månad, användas för att förutsäga periodicitet och trend, och många externa funktioner, såsom väder, kan öka förutsägelseresultaten. Men när tiden är kort är det svårt att utnyttja dessa funktioner. Dataset som tillhandahålls av Ericsson samlas in från riktiga nätverk. Efter rengöring av data konverterar vi tidsserieprognosproblemet till ett övervakat inlärningsproblem. Toppmoderna algoritmer som Random Forest (RF), XGboost och LSTM(Long Short TermMemory) utnyttjas för att förutsäga data ankomst astigheten och en K-Fold Cross-Validation följs för att validera modellerna. Resultaten visar att även tidsintervallen är små, datainkomsthastigheten kan förutsägas och nedlänksdata, kvalitetsindikator för nedlänk och rangindikator kan öka prognosprestandan.

In a lot of big cities, the traffic network is overloaded, with congestion and unnecessary emissions as consequence. Therefore, different traffic control methods are useful, especially in case of an incident. One key problem for traffic control is traffic prediction and the aim of this thesis is to develop, calibrate and evaluate a route flow model using only observed travel times and travel demand as input. The route flow model was used to calculate the metric link flow destination distribution, that presents to which destinations the travelers on a link are going in percentage.