Freeway travel time estimation and prediction using dynamic neural networks

Providing transportation system operators and travelers with accurate travel time information allows them to make more informed decisions, yielding benefits for individual travelers and for the entire transportation system. Most existing advanced traveler information systems (ATIS) and advanced traffic management systems (ATMS) use instantaneous travel time values estimated based on the current measurements, assuming that traffic conditions remain constant in the near future. For more effective applications, it has been proposed that ATIS and ATMS should use travel times predicted for short-term future conditions rather than instantaneous travel times measured or estimated for current conditions. ^ This dissertation research investigates short-term freeway travel time prediction using Dynamic Neural Networks (DNN) based on traffic detector data collected by radar traffic detectors installed along a freeway corridor. DNN comprises a class of neural networks that are particularly suitable for predicting variables like travel time, but has not been adequately investigated for this purpose. Before this investigation, it was necessary to identifying methods for data imputation to account for missing data usually encountered when collecting data using traffic detectors. It was also necessary to identify a method to estimate the travel time on the freeway corridor based on data collected using point traffic detectors. A new travel time estimation method referred to as the Piecewise Constant Acceleration Based (PCAB) method was developed and compared with other methods reported in the literatures. The results show that one of the simple travel time estimation methods (the average speed method) can work as well as the PCAB method, and both of them out-perform other methods. This study also compared the travel time prediction performance of three different DNN topologies with different memory setups. The results show that one DNN topology (the time-delay neural networks) out-performs the other two DNN topologies for the investigated prediction problem. This topology also performs slightly better than the simple multilayer perceptron (MLP) neural network topology that has been used in a number of previous studies for travel time prediction.^

Freeway Travel Time Estimation and Prediction Using Dynamic Neural Networks

Providing transportation system operators and travelers with accurate travel time information allows them to make more informed decisions, yielding benefits for individual travelers and for the entire transportation system. Most existing advanced traveler information systems (ATIS) and advanced traffic management systems (ATMS) use instantaneous travel time values estimated based on the current measurements, assuming that traffic conditions remain constant in the near future. For more effective applications, it has been proposed that ATIS and ATMS should use travel times predicted for short-term future conditions rather than instantaneous travel times measured or estimated for current conditions. This dissertation research investigates short-term freeway travel time prediction using Dynamic Neural Networks (DNN) based on traffic detector data collected by radar traffic detectors installed along a freeway corridor. DNN comprises a class of neural networks that are particularly suitable for predicting variables like travel time, but has not been adequately investigated for this purpose. Before this investigation, it was necessary to identifying methods for data imputation to account for missing data usually encountered when collecting data using traffic detectors. It was also necessary to identify a method to estimate the travel time on the freeway corridor based on data collected using point traffic detectors. A new travel time estimation method referred to as the Piecewise Constant Acceleration Based (PCAB) method was developed and compared with other methods reported in the literatures. The results show that one of the simple travel time estimation methods (the average speed method) can work as well as the PCAB method, and both of them out-perform other methods. This study also compared the travel time prediction performance of three different DNN topologies with different memory setups. The results show that one DNN topology (the time-delay neural networks) out-performs the other two DNN topologies for the investigated prediction problem. This topology also performs slightly better than the simple multilayer perceptron (MLP) neural network topology that has been used in a number of previous studies for travel time prediction.

Estimation and prediction of parameters and breeding values in soybean using REML/BLUP and Least Squares

The aim of this study was to compare REML/BLUP and Least Square procedures in the prediction and estimation of genetic parameters and breeding values in soybean progenies. F(2:3) and F(4:5) progenies were evaluated in the 2005/06 growing season and the F(2:4) and F(4:6) generations derived thereof were evaluated in 2006/07. These progenies were originated from two semi-early, experimental lines that differ in grain yield. The experiments were conducted in a lattice design and plots consisted of a 2 m row, spaced 0.5 m apart. The trait grain yield per plot was evaluated. It was observed that early selection is more efficient for the discrimination of the best lines from the F(4) generation onwards. No practical differences were observed between the least square and REML/BLUP procedures in the case of the models and simplifications for REML/BLUP used here.

Estimation and prediction of average vehicle occupancies using traffic accident records

As congestion management strategies begin to put more emphasis on person trips than vehicle trips, the need for vehicle occupancy data has become more critical. The traditional methods of collecting these data include the roadside windshield method and the carousel method. These methods are labor-intensive and expensive. An alternative to these traditional methods is to make use of the vehicle occupancy information in traffic accident records. This method is cost effective and may provide better spatial and temporal coverage than the traditional methods. However, this method is subject to potential biases resulting from under- and over-involvement of certain population sectors and certain types of accidents in traffic accident records. In this dissertation, three such potential biases, i.e., accident severity, driver’s age, and driver’s gender, were investigated and the corresponding bias factors were developed as needed. The results show that although multi-occupant vehicles are involved in higher percentages of severe accidents than are single-occupant vehicles, multi-occupant vehicles in the whole accident vehicle population were not overrepresented in the accident database. On the other hand, a significant difference was found between the distributions of the ages and genders of drivers involved in accidents and those of the general driving population. An information system that incorporates adjustments for the potential biases was developed to estimate the average vehicle occupancies (AVOs) for different types of roadways on the Florida state roadway system. A reasonableness check of the results from the system shows AVO estimates that are highly consistent with expectations. In addition, comparisons of AVOs from accident data with the field estimates show that the two data sources produce relatively consistent results. While accident records can be used to obtain the historical AVO trends and field data can be used to estimate the current AVOs, no known methods have been developed to project future AVOs. Four regression models for the purpose of predicting weekday AVOs on different levels of geographic areas and roadway types were developed as part of this dissertation. The models show that such socioeconomic factors as income, vehicle ownership, and employment have a significant impact on AVOs.

Estimation and Prediction of Average Vehicle Occupancies using Traffic Accident Records

As congestion management strategies begin to put more emphasis on person trips than vehicle trips, the need for vehicle occupancy data has become more critical. The traditional methods of collecting these data include the roadside windshield method and the carousel method. These methods are labor-intensive and expensive. An alternative to these traditional methods is to make use of the vehicle occupancy information in traffic accident records. This method is cost effective and may provide better spatial and temporal coverage than the traditional methods. However, this method is subject to potential biases resulting from under- and over-involvement of certain population sectors and certain types of accidents in traffic accident records. In this dissertation, three such potential biases, i.e., accident severity, driver¡¯s age, and driver¡¯s gender, were investigated and the corresponding bias factors were developed as needed. The results show that although multi-occupant vehicles are involved in higher percentages of severe accidents than are single-occupant vehicles, multi-occupant vehicles in the whole accident vehicle population were not overrepresented in the accident database. On the other hand, a significant difference was found between the distributions of the ages and genders of drivers involved in accidents and those of the general driving population. An information system that incorporates adjustments for the potential biases was developed to estimate the average vehicle occupancies (AVOs) for different types of roadways on the Florida state roadway system. A reasonableness check of the results from the system shows AVO estimates that are highly consistent with expectations. In addition, comparisons of AVOs from accident data with the field estimates show that the two data sources produce relatively consistent results. While accident records can be used to obtain the historical AVO trends and field data can be used to estimate the current AVOs, no known methods have been developed to project future AVOs. Four regression models for the purpose of predicting weekday AVOs on different levels of geographic areas and roadway types were developed as part of this dissertation. The models show that such socioeconomic factors as income, vehicle ownership, and employment have a significant impact on AVOs.

Role of assessment components and recent adverse outcomes in risk estimation and prediction: use of the Short Term Assessment of Risk and Treatability (START) in an adult secure inpatient mental health service

The Short Term Assessment of Risk and Treatability is a structured judgement tool used to inform risk estimation for multiple adverse outcomes. In research, risk estimates outperform the tool's strength and vulnerability scales for violence prediction. Little is known about what its’component parts contribute to the assignment of risk estimates and how those estimates fare in prediction of non-violent adverse outcomes compared with the structured components. START assessment and outcomes data from a secure mental health service (N=84) was collected. Binomial and multinomial regression analyses determined the contribution of selected elements of the START structured domain and recent adverse risk events to risk estimates and outcomes prediction for violence, self-harm/suicidality, victimisation, and self-neglect. START vulnerabilities and lifetime history of violence, predicted the violence risk estimate; self-harm and victimisation estimates were predicted only by corresponding recent adverse events. Recent adverse events uniquely predicted all corresponding outcomes, with the exception of self-neglect which was predicted by the strength scale. Only for victimisation did the risk estimate outperform prediction based on the START components and recent adverse events. In the absence of recent corresponding risk behaviour, restrictions imposed on the basis of START-informed risk estimates could be unwarranted and may be unethical.

Efeitos genéticos aditivos diretos e maternos e heterozigóticos sobre os desempenhos pré e pós-desmama em uma população multirracial Aberdeen Angus × Nelore

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)

Relationship between estimation and real motor performance in school-age children

The relationship between estimated and real motor competences was analyzed for several tasks. Participants were 303 children (160 boys and 143 girls), which had between 6 and 10 years of age (M=8.63, SD=1.16). None of the children presented developmental difficulties or learning disabilities, and all attended age-appropriate classes. Children were divided into three groups according to their age: group 1 (N= 102; age range: 6.48-8.01 years); group 2 (N= 101; age range: 8.02-9.22 years); and group 3 (N=100; age range: 9.24-10.93 years). Children were asked to predict their maximum distance for a locomotor, a manipulative, and a balance task, prior to performing those tasks. Children’s estimations were compared with their real performance to determine their accuracy. Children had, in general, a tendency to overestimate their performance (standing long jump: 56.11%, kicking: 63.37%, throwing: 73.60%, and Walking Backwards (WB) on a balance beam: 45.21%), and older children tended to be more accurate, except for the manipulative tasks. Furthermore, the relationship between estimation and real performance in children with different levels of motor coordination (Köperkoordinationstest für Kinder, KTK) was analyzed. The 75 children with the highest score comprised the Highest Motor Coordination (HMC) group, and the 78 children with the lowest score were placed in the Lowest Motor Coordination (LMC) group. There was a tendency for LMC and HMC children to overestimate their skills at all tasks, except for the HMC group at the WB task. Children with the HMC level tended to be more accurate when predicting their motor performance; however, differences in absolute percent error were only significant for the throwing and WB tasks. In conclusion, children display a tendency to overestimate their performance independently of their motor coordination level and task. This fact may be determinant to the development of their motor competences, since they are more likely to engage and persist in motor tasks, but it might also increase the occurrence of unintended injuries.

Kernel-Based Ranking. Methods for Learning and Performance Estimation

Machine learning provides tools for automated construction of predictive models in data intensive areas of engineering and science. The family of regularized kernel methods have in the recent years become one of the mainstream approaches to machine learning, due to a number of advantages the methods share. The approach provides theoretically well-founded solutions to the problems of under- and overfitting, allows learning from structured data, and has been empirically demonstrated to yield high predictive performance on a wide range of application domains. Historically, the problems of classification and regression have gained the majority of attention in the field. In this thesis we focus on another type of learning problem, that of learning to rank. In learning to rank, the aim is from a set of past observations to learn a ranking function that can order new objects according to how well they match some underlying criterion of goodness. As an important special case of the setting, we can recover the bipartite ranking problem, corresponding to maximizing the area under the ROC curve (AUC) in binary classification. Ranking applications appear in a large variety of settings, examples encountered in this thesis include document retrieval in web search, recommender systems, information extraction and automated parsing of natural language. We consider the pairwise approach to learning to rank, where ranking models are learned by minimizing the expected probability of ranking any two randomly drawn test examples incorrectly. The development of computationally efficient kernel methods, based on this approach, has in the past proven to be challenging. Moreover, it is not clear what techniques for estimating the predictive performance of learned models are the most reliable in the ranking setting, and how the techniques can be implemented efficiently. The contributions of this thesis are as follows. First, we develop RankRLS, a computationally efficient kernel method for learning to rank, that is based on minimizing a regularized pairwise least-squares loss. In addition to training methods, we introduce a variety of algorithms for tasks such as model selection, multi-output learning, and cross-validation, based on computational shortcuts from matrix algebra. Second, we improve the fastest known training method for the linear version of the RankSVM algorithm, which is one of the most well established methods for learning to rank. Third, we study the combination of the empirical kernel map and reduced set approximation, which allows the large-scale training of kernel machines using linear solvers, and propose computationally efficient solutions to cross-validation when using the approach. Next, we explore the problem of reliable cross-validation when using AUC as a performance criterion, through an extensive simulation study. We demonstrate that the proposed leave-pair-out cross-validation approach leads to more reliable performance estimation than commonly used alternative approaches. Finally, we present a case study on applying machine learning to information extraction from biomedical literature, which combines several of the approaches considered in the thesis. The thesis is divided into two parts. Part I provides the background for the research work and summarizes the most central results, Part II consists of the five original research articles that are the main contribution of this thesis.

Performance and prediction: Bayesian modelling of fallible choice in chess

Evaluating agents in decision-making applications requires assessing their skill and predicting their behaviour. Both are well developed in Poker-like situations, but less so in more complex game and model domains. This paper addresses both tasks by using Bayesian inference in a benchmark space of reference agents. The concepts are explained and demonstrated using the game of chess but the model applies generically to any domain with quantifiable options and fallible choice. Demonstration applications address questions frequently asked by the chess community regarding the stability of the rating scale, the comparison of players of different eras and/or leagues, and controversial incidents possibly involving fraud. The last include alleged under-performance, fabrication of tournament results, and clandestine use of computer advice during competition. Beyond the model world of games, the aim is to improve fallible human performance in complex, high-value tasks.

Estimation of variance components and prediction of breeding values in rubber tree breeding using the REML/BLUP procedure

The present paper deals with estimation of variance components, prediction of breeding values and selection in a population of rubber tree [Hevea brasiliensis (Willd. ex Adr. de Juss.) Müell.-Arg.] from Rio Branco, State of Acre, Brazil. The REML/BLUP (restricted maximum likelihood/best linear unbiased prediction) procedure was applied. For this purpose, 37 rubber tree families were obtained and assessed in a randomized complete block design, with three unbalanced replications. The field trial was carried out at the Experimental Station of UNESP, located in Selvíria, State of Mato Grosso do Sul, Brazil. The quantitative traits evaluated were: girth (G), bark thickness (BT), number of latex vessel rings (NR), and plant height (PH). Given the unbalanced condition of the progeny test, the REML/BLUP procedure was used for estimation. The narrow-sense individual heritability estimates were 0.43 for G, 0.18 for BT, 0.01 for NR, and 0.51 for PH. Two selection strategies were adopted: one short-term (ST - selection intensity of 8.85%) and the other long-term (LT - selection intensity of 26.56%). For G, the estimated genetic gains in relation to the population average were 26.80% and 17.94%, respectively, according to the ST and LT strategies. The effective population sizes were 22.35 and 46.03, respectively. The LT and ST strategies maintained 45.80% and 28.24%, respectively, of the original genetic diversity represented in the progeny test. So, it can be inferred that this population has potential for both breeding and ex situ genetic conservation as a supplier of genetic material for advanced rubber tree breeding programs. Copyright by the Brazilian Society of Genetics.

Performance estimation for concretes made with recycled aggregates of construction and demolition waste of some Brazilian cities

The aim of this paper is to verify the influence of composition variability of recycled aggregates (RA) of construction and demolition wastes (CDW) on the performance of concretes. Performance was evaluated building mathematical models for compressive strength, modulus of elasticity and drying shrinkage. To obtain such models, an experimental program comprising 50 concrete mixtures was carried out. Specimens were casted, tested and results for compressive strength, modulus of elasticity and drying shrinkage were statistically analyzed. Models inputs are CDW composition observed at seven Brazilian cities. Results confirm that using RA from CDW for concrete building is quite feasible, independently of its composition, once compressive strength and modulus of elasticity still reached considerable values. We concluded the variability presented by recycled aggregates of CDW does not compromise their use for concrete building. However, this information must be used with caution, and experimental tests should always be performed to certify concrete properties.

Children's performance estimation in mathematics and science tests over a school year: A pilot study

The metacognitve ability to accurately estimate ones performance in a test, is assumed to be of central importance for initializing task-oriented effort. In addition activating adequate problem-solving strategies, and engaging in efficient error detection and correction. Although school children's' ability to estimate their own performance has been widely investigated, this was mostly done under highly-controlled, experimental set-ups including only one single test occasion. Method: The aim of this study was to investigate this metacognitive ability in the context of real achievement tests in mathematics. Developed and applied by a teacher of a 5th grade class over the course of a school year these tests allowed the exploration of the variability of performance estimation accuracy as a function of test difficulty. Results: Mean performance estimations were generally close to actual performance with somewhat less variability compared to test performance. When grouping the children into three achievement levels, results revealed higher accuracy of performance estimations in the high achievers compared to the low and average achievers. In order to explore the generalization of these findings, analyses were also conducted for the same children's tests in their science classes revealing a very similar pattern of results compared to the domain of mathematics. Discussion and Conclusion: By and large, the present study, in a natural environment, confirmed previous laboratory findings but also offered additional insights into the generalisation and the test dependency of students' performances estimations.

High performance shift invariant motion estimation and compensation in wavelet domain video compression

The contributions of this dissertation are in the development of two new interrelated approaches to video data compression: (1) A level-refined motion estimation and subband compensation method for the effective motion estimation and motion compensation. (2) A shift-invariant sub-decimation decomposition method in order to overcome the deficiency of the decimation process in estimating motion due to its shift-invariant property of wavelet transform. ^ The enormous data generated by digital videos call for an intense need of efficient video compression techniques to conserve storage space and minimize bandwidth utilization. The main idea of video compression is to reduce the interpixel redundancies inside and between the video frames by applying motion estimation and motion compensation (MEMO) in combination with spatial transform coding. To locate the global minimum of the matching criterion function reasonably, hierarchical motion estimation by coarse to fine resolution refinements using discrete wavelet transform is applied due to its intrinsic multiresolution and scalability natures. ^ Due to the fact that most of the energies are concentrated in the low resolution subbands while decreased in the high resolution subbands, a new approach called level-refined motion estimation and subband compensation (LRSC) method is proposed. It realizes the possible intrablocks in the subbands for lower entropy coding while keeping the low computational loads of motion estimation as the level-refined method, thus to achieve both temporal compression quality and computational simplicity. ^ Since circular convolution is applied in wavelet transform to obtain the decomposed subframes without coefficient expansion, symmetric-extended wavelet transform is designed on the finite length frame signals for more accurate motion estimation without discontinuous boundary distortions. ^ Although wavelet transformed coefficients still contain spatial domain information, motion estimation in wavelet domain is not as straightforward as in spatial domain due to the shift variance property of the decimation process of the wavelet transform. A new approach called sub-decimation decomposition method is proposed, which maintains the motion consistency between the original frame and the decomposed subframes, improving as a consequence the wavelet domain video compressions by shift invariant motion estimation and compensation. ^

Monitoring and Prediction of Thermal Emergencies in High Performance Computing Systems

Modern scientific discoveries are driven by an unsatisfiable demand for computational resources. High-Performance Computing (HPC) systems are an aggregation of computing power to deliver considerably higher performance than one typical desktop computer can provide, to solve large problems in science, engineering, or business. An HPC room in the datacenter is a complex controlled environment that hosts thousands of computing nodes that consume electrical power in the range of megawatts, which gets completely transformed into heat. Although a datacenter contains sophisticated cooling systems, our studies indicate quantitative evidence of thermal bottlenecks in real-life production workload, showing the presence of significant spatial and temporal thermal and power heterogeneity. Therefore minor thermal issues/anomalies can potentially start a chain of events that leads to an unbalance between the amount of heat generated by the computing nodes and the heat removed by the cooling system originating thermal hazards. Although thermal anomalies are rare events, anomaly detection/prediction in time is vital to avoid IT and facility equipment damage and outage of the datacenter, with severe societal and business losses. For this reason, automated approaches to detect thermal anomalies in datacenters have considerable potential. This thesis analyzed and characterized the power and thermal characteristics of a Tier0 datacenter (CINECA) during production and under abnormal thermal conditions. Then, a Deep Learning (DL)-powered thermal hazard prediction framework is proposed. The proposed models are validated against real thermal hazard events reported for the studied HPC cluster while in production. This thesis is the first empirical study of thermal anomaly detection and prediction techniques of a real large-scale HPC system to the best of my knowledge. For this thesis, I used a large-scale dataset, monitoring data of tens of thousands of sensors for around 24 months with a data collection rate of around 20 seconds.