84 research outputs found
Physics-Based Propagation Models Enabled by Machine Learning
Get PDFAs a growing number of wireless services with high performance demands is offered, their intelligentplanning and efficient management becomes more urgent. To that end, channel propagation
models are indispensable. Such models can be used to optimize the position of wireless access points,
assess interference from and towards neighboring systems, and facilitate network-level performance
evaluation studies.
Propagation modeling has been widely dependent on time-consuming and arduous measurementsurveys. Alternatively, fast and easy-to-use empirical models have also been utilized. Yet, these
models lack predictive accuracy. A partial solution to the problem is offered by the use of computational
methods based on electromagnetic theory, such as finite difference time domain, vector
parabolic equation (VPE) and ray tracing (RT) methods. However, the higher accuracy of these
methods is often combined with computationally expensive simulations.
Motivated by the need to create accurate, yet computationally efficient and flexible propagationmodels, this thesis explores a data-driven approach, leveraging recent advances in machine learning
(ML). That learning is achieved over a training phase, by specifying a set of input features and the
expected output the ML-driven propagation model has to produce. ML models are powerful and
efficient tools that can capture highly complex and non-linear functions at their output. After such
models are trained, they can use their learned parameters to generate fast and accurate predictions
for new cases of considerable complexity.
To enhance the accuracy of the proposed ML-based propagation models, we use physics-basedtraining data generated by high-fidelity solvers, such as VPE and RT solvers. The cost for simulating
the training data is significantly outweighed by the ability of the trained ML models to rapidly
simulate new test cases. This is achieved by utilizing physics-informed input features, such as
information regarding the complexity of the modeling environment and the antenna specifications
of the communication system, and physics-driven cost functions. In addition, to further improve
the computational efficiency of our models, we train using small sets of data, and utilize efficient,
yet simple, ML network structures. Thus, our main goal is to design efficient and accurate ML-based
surrogate models for the VPE and RT solvers, for a diverse set of propagation scenarios. We
successfully demonstrate this accuracy and computational efficiency by evaluating our models on a
wide range of test cases, of significant difficulty and diversity.
The proposed models promise to overcome the classical dichotomy between computational efficiencyand accuracy that has dominated the area of propagation modeling. This can have a
profound impact in many large-scale computations usually encountered in wireless propagation scenarios,
such as providing real-time and accurate localization services, and efficiently selecting the
positions of transmitters in electrically large environments.Ph.D
Υδρογόνωση Ανανεώσιμης Φουρφουράλης προς Φουρφουρόλη με Καταλυτικά Σύμπλοκα του Λευκόχρυσου σε Υδατικό Περιβάλλον
Get PDFΣε αυτήν τη διατριβή μελετούμε την υδρογόνωση της ανανεώσιμης φουρφουράλης παρουσία καταλυτικών συστημάτων λευκοχρύσου στο υδατικό περιβάλλον με σκοπό τον εκλεκτικό μετασχηματισμό της σε φουρφουρόλη. Αρχικά, εξετάζεται μία σειρά 21 in situ σχηματισθέντων καταλυτών λευκοχρύσου από το πρόδρομο σύστημα Na2PtCl6•6H2O τροποποιημένο με φωσφίνες, μονοσχιδείς και πολυσχιδείς αζωτούχους όπως και ΡᴖΝ και ΝᴖΟ χηλικούς υποκαταστάτες υπό ουδέτερες συνθήκες (pH 7, ρυθμιστικό διάλυμα NaH2PO4/NaOH) στο νερό. Τα βέλτιστα αποτελέσματα, με κριτήρια την καταλυτική δραστικότητα, το ποσοστό εκλεκτικότητας και την σταθερότητα του καταλύτη, καταγράφονται από τον καταλύτη Pt τροποποιημένο με το μετά νατρίου άλας της τρισουλφουρωμένης τριφαινυλοφωσφίνης (TPPTS). Στην συνέχεια, διερευνάται η επίδραση των μοριακών λόγων NaH2PO4/Pt και TPPTS/Pt και του χρόνου αντίδρασης. Επιχειρείται η αναγνώριση της φύσεως του καταλυτικού συστήματος Na2PtCl6•6H2O/ TPPTS, προτείνεται μία εξήγηση σχετικά με την επίδραση των υποκαταστατών και συζητούνται πιθανές πορείες για τον σχηματισμό των προϊόντων. Μετά εξετάζεται η επίδραση της φύσεως ενός ευρέος φάσματος πρόδρομων καταλυτών και επιλέγεται το PtCl2 έναντι του Na2PtCl6•6H2O. Παρουσία του καταλυτικού συστήματος PtCl2/TPPTS γίνεται μελέτη των παραμέτρων του μοριακού λόγου TPPTS/Pt και φουρφουράλης/Pt, της θερμοκρασίας, της πίεσης, του όγκου της υδατικής φάσης και της ποσότητας του πρόδρομου καταλύτη. Ο πρόδρομος καταλύτης PtCl2 τροποποιημένος με TPPTS σε ένα μοριακό λόγο TPPTS/Pt = 3, θερμοκρασία 130 οC, 40 bar υδρογόνου, μοριακό λόγο φουρφουράλης/Pt = 2000, παρουσία 20 ml υδατικού διαλύτη, χαμηλή συγκέντρωση λευκοχρύσου 44 ppm, σε pH 7.0 και χρόνο αντίδρασης 30 min παρουσίασε την υψηλότερη δραστικότητα ίση με 20090 TOFs ανά ώρα με υψηλή εκλεκτικότητα σε φουρφουρόλη 98.1 mol% και χαμηλές εκλεκτικότητες σε παραπροϊόντα, τετραϋδροφουρφουρόλη και 1-πεντανόλη, της τάξεως 0.3 mol% και 1.6 mol% αντίστοιχα.In this thesis we study the hydrogenation of renewable furfural selectively into furfuryl alcohol in the presence of platinum catalytic systems in aqueous medium. First, a series of 21 in situ synthesized platinum catalysts from the precursor Na2PtCl6•6H2O modified with phosphines, monodentate and polydentate nitrogen-containing and PᴖN or PᴖO chelating ligands has been investigated under neutral conditions (pH 7, buffer NaH2PO4/NaOH) in water. The best results in terms of catalytic activity, selectivity and catalyst stability have been obtained with the platinum catalyst modified with triphenylphosphinetrisulfonic acid trisodium salt (TPPTS). Next, the effects of molar ratios of NaH2PO4/Pt and TPPTS/Pt and reaction time have been examined. Identification of catalyst’s nature is attempted, rationalization of ligand effect is suggested and possible routes of products formation are discussed. Furthermore, the effect of a broad spectrum of catalyst precursors has been disclosed and PtCl2 is selected instead of Na2PtCl6•6H2O. Moreover, in the presence of PtCl2/TPPTS catalyst the parameters of the molar ratios of TPPTS/Pt and furfural/Pt, temperature, pressure, volume of aqueous solvent and catalyst loading have been determined. The PtCl2 catalyst precursor modified with TPPTS at a molar ratio of TPPTS/Pt = 3, temperature 130 oC, 40 bar hydrogen, molar ratio of furfural/Pt = 2000 with addition of 20 ml aqueous solvent at a low concentration of platinum of 44 ppm, a pH value of 7.0 and reaction duration of 30 minutes exhibited the highest activity of 20090 TOFs per hour with high selectivity towards furfuryl alcohol of 98.1 mol% and low selectivities to the byproducts, tetrahydrofurfuryl alcohol and 1-pentanol in the range of 0.3 mol% and 1.6 mol%, respectively
Empirical evaluation and systematic review of prognostic factors and prognostic models for chronic diseases
No full textPrognosis is defined as the likelihood of displaying a condition at a specific time period, based on specific characteristics, named prognostic factors. A prognostic model is a mathematical function that relates multiple prognostic factors to the probability (risk) of a particular outcome. Prognosis research and prognostic models for chronic diseases, such as cardiovascular disease and cancer, is crucial for patient counseling and clinical decision making. Our first aim was to map the expanded used of the Framingham Risk Score (FRS), one of the most important prognostic models in cardiovascular disease, in the medical literature. We identified all citations to the original Wilson et al.1998 paper where FRS was originally described. We identified 375 eligible papers corresponding to 471 analyses using the FRS in cohort, case-control or cross-sectional settings. The studied population was different (from healthy) in 25% and 42% of the cohort and cross-sectional analyses, respectively. The studied outcome was different (from coronary heart disease) in 56% of the cohort analyses and in 63% of the case-control studies. Overall, only 33% of the cohort analyses examined the same outcome and population as FRS was originally developed for. In conclusion, a large number of studies use FRS in different settings than originally developed for, for which its performance is unknown and non-validated. Before using prognostic models in different settings than originally developed recalibration/update and external validation is required. Our second objective was to provide a systematic review of secondary prevention models in patients with already established cardiovascular disease. Using MEDLINE and EMBASE databases between January 2004 and June 2013, we identified 79 eligible articles. Most new models were developed in USA and Europe, on patients with coronary artery disease and a mean age of 65 years. The majority of models used Cox or logistic regression, 5 to 7 predictors, and C-statistic and the Hosmer-Lemeshow test for assessing model performance. Framingham Risk Score (FRS), CHADS2, SYNTAX and GRACE were the models more frequently externally validated. Finally, FRS was the model mostly incremented, and CRP and NT-proBNP were the most commonly examined risk factors. In conclusion, the majority of models used in secondary prevention are newly derived models without proper external validation. Finally, we conducted a meta-analysis to assess the association between hypertension (a prognostic factor) and site-specific cancer (a chronic disease). After searching PubMed until November 2017, we identified a total of 148 eligible publications. Considering only evidence from 85 prospective studies, we conducted 30 meta-analyses of hypertension and 16 different cancers. Positive associations were observed between hypertension and kidney, colorectal and breast cancer. Positive associations between hypertension and risk of oesophageal, liver, and endometrial cancers were also observed, but the majority of studies did not perform comprehensive multivariable adjustments.Ως πρόγνωση ορίζεται η πιθανότητα εμφάνισης μιας κατάστασης σε συγκεκριμένη χρονική περίοδο, με βάση συγκεκριμένα χαρακτηριστικά, τους προγνωστικούς παράγοντες. Ένα προγνωστικό μοντέλο είναι μια μαθηματική συνάρτηση που συσχετίζει πολλαπλούς προγνωστικούς παράγοντες με μια συγκεκριμένη έκβαση. Η έρευνα στην πρόγνωση και τα προγνωστικά μοντέλα χρονίων νόσων, όπως η καρδιαγγειακή νόσος και ο καρκίνος, είναι καίρια για τη συμβουλευτική των ασθενών και τη λήψη κλινικών αποφάσεων. Πρώτος στόχος της παρούσας μελέτης ήταν να εξετάσουμε τη χρήση του Framingham Risk Score (FRS), ενός από τα σημαντικότερα προγνωστικά μοντέλα καρδιαγγειακής νόσου, στην ιατρική βιβλιογραφία. Εντοπίσαμε όλες τις παραπομπές στο άρθρο του Wilson και συν. (1998), όπου το FRS αρχικώς περιγράφτηκε. Εντοπίσαμε 375 δόκιμα άρθρα (συνολικά 471 αναλύσεις) που χρησιμοποίησαν το FRS σε προοπτικές μελέτες, μελέτες ασθενών-μαρτύρων και μελέτες επιπολασμού. Ο πληθυσμός που μελετήθηκε ήταν διαφορετικός (από τον υγιή) στο 25% και 42% των αναλύσεων κοόρτης και επιπολασμού αντίστοιχα. Η έκβαση υπό εξέταση ήταν διαφορετική (από στεφανιαία νόσο) στο 56% των προοπτικών αναλύσεων και στο 63% των αναλύσεων ασθενών-μαρτύρων. Συνολικά, μόνο το 33% των προοπτικών αναλύσεων εξέταζαν την ίδια έκβαση και πληθυσμό με την αρχική μελέτη. Συμπερασματικά, ένας μεγάλος αριθμός μελετών χρησιμοποιούν το FRS σε συνθήκες, όπου η απόδοσή του είναι άγνωστη και μη επικυρωμένη. Πριν από την εφαρμογή προγνωστικών μοντέλων σε διαφορετικές συνθηκες, χρειάζεται αναπροσαρμογή και εξωτερική επικύρωση του μοντέλου. Δεύτερος στόχος της μελέτης ήταν να διεξάγουμε μια συστηματική ανασκόπηση των μοντέλων δευτερογενούς πρόληψης, σε ασθενείς με ήδη εγκαταστημένη καρδιαγγειακή νόσο. Χρησιμοποιήσαμε τις βάσεις δεδομένων MEDLINE και EMBASE (Ιανουάριος 2004 - Ιούνιος 2013) και εντοπίσαμε 79 δόκιμα άρθρα. Τα περισσότερα νέα μοντέλα αναπτύχθηκαν στις Ηνωμένες Πολιτείες και την Ευρώπη, σε ασθενείς με στεφανιαία νόσο και με μέση ηλικία τα 65 έτη. Η πλειοψηφία των μοντέλων χρησιμοποίησε Cox ή λογιστική παλινδρόμηση, 5 έως 7 προγνωστικούς παράγοντες και το C-statistic με το Hosmer-Lemeshow τεστ για τον έλεγχο της απόδοσης του μοντέλου. Τα FRS, CHADS2, SYNTAX και GRACE ήταν τα μοντέλα που συχνότερα επικυρώθηκαν εξωτερικώς. Τέλος, το FRS ήταν το μοντέλο που πρωτίστως προσαυξήθηκε με νέους παράγοντες κινδύνου και συνηθέστερα με τους βιοχημικούς δείκτες CRP και NT-proBNP. Συμπερασματικά, η πλειοψηφία των μοντέλων που χρησιμοποιούνται στη δευτερογενή πρόληψη είναι νέα μοντέλα που δημιουργούνται χωρίς κατάλληλη εξωτερική επικύρωση. Τέλος, διεξήγαμε μια μετα-ανάλυση για να μελετήσουμε τη σχέση μεταξύ της υπέρτασης (προγνωστικός παράγοντας) και της έκβασης του καρκίνου (χρόνια νόσος). Μετά από αναζήτηση στο PubMed μέχρι και το Νοέμβριο του 2017 εντοπίστηκαν 148 δόκιμες μελέτες. Λαμβάνοντας υπόψη μόνο δεδομένα από τις 85 προοπτικές μελέτες διεξήγαμε 30 μετα-αναλύσεις για 16 διαφορετικά ήδη καρκίνου. Θετικές συσχετίσεις παρατηρήθηκαν μεταξύ υπέρτασης και νεφρικού, ορθοκολικού και καρκίνου του μαστού. Παρατηρήθηκαν επίσης θετικές συσχετίσεις μεταξύ υπέρτασης και κινδύνου για καρκίνο του οισοφάγου, του ήπατος και του ενδομητρίου. Ωστόσο, η πλειοψηφία των μελετών αυτών δεν είχε προβεί σε εκτεταμένες πολυπαραγοντικές προσαρμογές για συνήθεις συγχυτικούς παράγοντες
Towards Physics-Based Generalizable Convolutional Neural Network Models for Indoor Propagation
No full textA fundamental challenge for machine learning models for electromagnetics is their ability to predict output quantities of interest (such as fields and scattering parameters) in geometries that the model has not been trained for. Addressing this challenge is a key to fulfilling one of the most appealing promises of machine learning for computational electromagnetics: the rapid solution of problems of interest just by processing the geometry and the sources involved. The impact of such models that can "generalize" to new geometries is more profound for large-scale computations, such as those encountered in wireless propagation scenarios. We present generalizable models for indoor propagation that can predict received signal strengths within new geometries, beyond those of the training set of the model, for transmitters and receivers of multiple positions, and for new frequencies. We show that a convolutional neural network can "learn" the physics of indoor radiowave propagation from ray-tracing solutions of a small set of training geometries, so that it can eventually deal with substantially different geometries. We emphasize the role of exploiting physical insights in the training of the network, by defining input parameters and cost functions that assist the network to efficiently learn basic and complex propagation mechanisms
An Overview of Machine Learning Techniques for Radiowave Propagation Modeling
No full textWe give an overview of recent developments in the modeling of radiowave
propagation, based on machine learning algorithms. We identify the input and
output specification and the architecture of the model as the main challenges
associated with machine learning-driven propagation models. Relevant papers are
discussed and categorized based on their approach to each of these challenges.
Emphasis is given on presenting the prospects and open problems in this
promising and rapidly evolving area.Comment: 15 pages, 9 figures, 2 small tables and 1 1-page tabl
Structural Optimisation of Permanent Magnet Direct Drive Generators for 5MW Wind Turbines
Get PDFThis thesis has been submitted in fulfilment of the requirements for a postgraduate degree (e.g. PhD, MPhil, DClinPsychol) at the University of Edinburgh. Please note the following terms and conditions of use: • This work is protected by copyright and other intellectual property rights, which are retained by the thesis author, unless otherwise stated. • A copy can be downloaded for personal non-commercial research or study, without prior permission or charge. • This thesis cannot be reproduced or quoted extensively from without first obtaining permission in writing from the author. • The content must not be changed in any way or sold commercially in any format or medium without the formal permission of the author. • When referring to this work, full bibliographic details including the author, title, awarding institution and date of the thesis must be given
Generalizable Machine-Learning-Based Modeling of Radiowave Propagation in Stadiums
No full textProviding high throughput and quality of service in modern stadiums necessitates the placement of hundreds of access points (APs). Optimizing the locations of APs in such venues via measurements requires significant resources. Even simulation methods, such as ray-tracing, can be computationally costly. We provide a solution to this problem by building a propagation model based on machine learning (ML) that rapidly predicts received signal strengths in stadiums. We train the model with a small set of simulated data generated by a ray-tracer. We use input features, such as the electrical distance between the transmitter and the receiver, and the antenna gain along the direct path between the two, to generalize to new transmitter locations, antenna patterns and stadium geometries. Geometry and pattern generalization have not been included in existing propagation models for stadiums. Finally, we present a novel sampling approach for the input features in a given stadium, ensuring the computational efficiency and accuracy of the ML model. The results demonstrate the accuracy of our propagation model for new transmitter locations, patterns and stadiums. The trained model is also considerably faster than a ray-tracer, making it an efficient tool for resource planning tasks, such as optimal placement of APs
Fast Selection of Indoor Wireless Transmitter Locations with Generalizable Neural Network Propagation Models
No full textThe continuous emergence of new wireless communication systems increases the need for intelligent planning prior to their deployment. This planning phase includes determining the position and transmit power of wireless access points, to meet quality of service objectives along with electromagnetic compatibility standards. How to effectively place a set of transmitters in an environment is a long-standing problem that has been widely studied over the years. This process has mainly relied on expensive measurement campaigns, low-fidelity empirical models, or high-fidelity but time-consuming simulations. Recent advances in scientific machine learning create new possibilities for overcoming the standard dichotomy between speed and accuracy.
In this paper, we train a deep neural network (U-Net) to predict received signal strength levels for a variety of different geometries and for positions of multiple transmitters. Then, we leverage the computational efficiency of the trained model to determine the position of access point transmitters in new geometries. This approach dramatically accelerates the process of selecting the position of access points, meeting multiple optimization objectivesin an efficient manner. </p
- …
