[pt] Os modelos orientados por score, também conhecidos como modelos generalizados de score autorregressivo (GAS), representam uma classe de modelos de séries temporais orientados por observação. Eles possuem propriedades desejáveis para modelagem de séries temporais, como a capacidade de modelar diferentes distribuições condicionais e considerar parâmetros variantes no tempo dentro de uma estrutura flexível. Neste trabalho, apresentamos ScoreDrivenModels.jl, um pacote Julia de código aberto para modelagem, previsão e simulação de séries temporais usando a estrutura de modelos baseados em score. O pacote é flexível no que diz respeito à definição do modelo, permitindo ao usuário especificar a estrutura de atraso e quais parâmetros são variantes no tempo ou constantes. Também é possível considerar várias distribuições, incluindo Beta, Exponencial, Gama, Lognormal, Normal, Poisson, Student s t e Weibull. A interface fornecida é flexível, permitindo aos usuários interessados implementar qualquer distribuição e parametrização desejada. / [en] Score-driven models, also known as generalized autoregressive score (GAS) models, represent a class of observation-driven time series models. They possess desirable properties for time series modeling, such as the ability to model different conditional distributions and to consider time-varying parameters within a flexible framework. In this dissertation, we present ScoreDrivenModels.jl, an open-source Julia package for modeling, forecasting, and simulating time series using the framework of score-driven models. The package is flexible with respect to model definition, allowing the user to specify the lag structure and which parameters are time-varying or constant. It is also possible to consider several distributions, including Beta, Exponential, Gamma, Lognormal, Normal, Poisson, Student s t, and Weibull. The provided interface is flexible, allowing interested users to implement any desired distribution and parametrization.

Air quality assessment plays a crucial role in understanding the impact of air pollution onhuman health and the environment. With the increasing demand for accurate assessment andprediction of air quality, advanced data analytics modelling techniques offer promisingsolutions. This thesis focuses on leveraging advanced data analytics to assess and analyse airpollution concentration levels in Italy over a 4km resolution using the FORAIR_IT datasetsimulated in ENEA on the CRESCO6 infrastructure, aiming to uncover valuable insights andidentifying the most appropriate AI models for predicting air pollution levels. The datacollection, understanding, and pre-processing procedures are discussed, followed by theapplication of big data training and forecasting using Apache Spark MLlib. The research alsoencompasses different phases, including descriptive and inferential analysis to understand theair pollution concentration dataset, hypothesis testing to examine the relationship betweenvarious pollutants, machine learning prediction using several regression models and anensemble machine learning approach and time series analysis on the entire dataset as well asthree major regions in Italy (Northern Italy – Lombardy, Central Italy – Lazio and SouthernItaly – Campania). The computation time for these regression models are also evaluated and acomparative analysis is done on the results obtained. The evaluation process and theexperimental setup involve the usage of the ENEAGRID/CRESCO6 HPC Infrastructure andApache Spark. This research has provided valuable insights into understanding air pollutionpatterns and improving prediction accuracy. The findings of this study have the potential todrive positive change in environmental management and decision-making processes, ultimatelyleading to healthier and more sustainable communities. As we continue to explore the vastpossibilities offered by advanced data analytics, this research serves as a foundation for futureadvancements in air quality assessment in Italy and the models are transferable to other regionsand provinces in Italy, paving the way for a cleaner and greener future.

本研究中，我們計算OGA演化投資策略在五類時間數列模型上之表現，這五類模型分別是線性模型、雙線性模型、自迴歸條件異質變異數模型、門檻模型以及混沌模型。我們選擇獲勝機率、累積報酬率、夏普比例以及幸運係數做為評斷表現之準則，並分別推導出其漸近統計檢定。有別於一般計算智慧在財務工程上之應用，利用蒙地卡羅模擬法，研究中將對各表現準則提出嚴格之統計檢定結果。同時在實証研究中，我們考慮歐元兌美元及美元兌日圓的tick-by-tick匯率資料。故本研究主要的重點之一，乃是對於OGA演化投資策略，於這些模擬及實証資料上的有效性應用，作了深入且廣泛的探討。 / In this study, the performance of ordinary GA-based trading strategies are evaluated under five classes of time series model, namely, linear ARMA model, bilinear model, ARCH model, threshold model and chaotic model. The performance criteria employed are the winning probability, accumulated returns, Sharpe ratio and luck coefficient. We then provide the asymptotic statistical tests for these criteria. Unlike many existing applications of computational intelligence in financial engineering, for each performance criterion, we provide a rigorous statistical results based on Monte Carlo simulation. In the empirical study, two tick-by-tick foreign exchange rates are also considered, namely, EUR/USD and USD/JPY. As a result, this study provides us with a thorough understanding about the effectiveness of ordinary GA for evolving trading strategies under these artificial and natural time series data.

The past 30 years have witnessed major developments in vibration based damage detection and identification, also collectively referred to as damage diagnosis. Moreover, the past 10 years have seen a rapid increase in the amount of research related to Structural Health Monitoring (SHM) as quantified by the significant escalation in papers published on this subject. Thus, the increased interest in this engineering field and its associated potential constitute the main motive for this thesis. The goal of the thesis is the development and introduction of novel advanced functional and sequential statistical time series methods for vibration based damage diagnosis and SHM. After the introduction of the first chapter, Chapter II provides an experimental assessment and comparison of vibration based statistical time series methods for Structural Health Monitoring (SHM) via their application on a lightweight aluminum truss structure and a laboratory scale aircraft skeleton structure. A concise overview of the main non-parametric and parametric methods is presented, including response-only and excitation-response schemes. Damage detection and identification are based on univariate (scalar) versions of the methods, while both scalar (univariate) and vector (multivariate) schemes are considered. The methods' effectiveness for both damage detection and identification is assessed via various test cases corresponding to different damage scenarios, multiple experiments and various sensor locations on the considered structures. The results of the chapter confirm the high potential and effectiveness of vibration based statistical time series methods for SHM. Chapter III investigates the identification of stochastic systems under multiple operating conditions via Vector-dependent Functionally Pooled (VFP) models. In many applications a system operates under a variety of operating conditions (for instance operating temperature, humidity, damage location, damage magnitude and so on) which affect its dynamics, with each condition kept constant for a single commission cycle. Typical examples include mechanical structures operating under different environmental conditions, aircrafts under different flight conditions (altitude, velocity etc.), structures under different structural health states (various damage locations and magnitudes). In this way, damage location and magnitude may be considered as parameters that affect the operating conditions and as a result the structural dynamics. This chapter's work is based on the novel Functional Pooling (FP) framework, which has been recently introduced by the Stochastic Mechanical Systems \& Automation (SMSA) group of the Mechanical Engineering and Aeronautics Department of University of Patras. The main characteristic of Functionally Pooled (FP) models is that their model parameters and innovations sequence depend functionally on the operating parameters, and are projected on appropriate functional subspaces spanned by mutually independent basis functions. Thus, the fourth chapter of the thesis addresses the problem of identifying a globally valid and parsimonious stochastic system model based on input-output data records obtained under a sample of operating conditions characterized by more than one parameters. Hence, models that include a vector characterization of the operating condition are postulated. The problem is tackled within the novel FP framework that postulates proper global discrete-time linear time series models of the ARX and ARMAX types, data pooling techniques, and statistical parameter estimation. Corresponding Vector-dependent Functionally Pooled (VFP) ARX and ARMAX models are postulated, and proper estimators of the Least Squares (LS), Maximum Likelihood (ML), and Prediction Error (PE) types are developed. Model structure estimation is achieved via customary criteria (Bayesian Information Criterion) and a novel Genetic Algorithm (GA) based procedure. The strong consistency of the VFP-ARX least squares and maximum likelihood estimators is established, while the effectiveness of the complete estimation and identification method is demonstrated via two Monte Carlo studies. Based on the postulated VFP parametrization a vibration based statistical time series method that is capable of effective damage detection, precise localization, and magnitude estimation within a unified stochastic framework is introduced in Chapter IV. The method constitutes an important generalization of the recently introduced Functional Model Based Method (FMBM) in that it allows, for the first time in the statistical time series methods context, for complete and precise damage localization on continuous structural topologies. More precisely, the proposed method can accurately localize damage anywhere on properly defined continuous topologies on the structure, instead of pre-defined specific locations. Estimator uncertainties are taken into account, and uncertainty ellipsoids are provided for the damage location and magnitude. To achieve its goal, the method is based on the extended class of Vector-dependent Functionally Pooled (VFP) models, which are characterized by parameters that depend on both damage magnitude and location, as well as on proper statistical estimation and decision making schemes. The method is validated and its effectiveness is experimentally assessed via its application to damage detection, precise localization, and magnitude estimation on a prototype GARTEUR-type laboratory scale aircraft skeleton structure. The damage scenarios considered consist of varying size small masses attached to various continuous topologies on the structure. The method is shown to achieve effective damage detection, precise localization, and magnitude estimation based on even a single pair of measured excitation-response signals. Chapter V presents the introduction and experimental assessment of a sequential statistical time series method for vibration based SHM capable of achieving effective, robust and early damage detection, identification and quantification under uncertainties. The method is based on a combination of binary and multihypothesis versions of the statistically optimal Sequential Probability Ratio Test (SPRT), which employs the residual sequences obtained through a stochastic time series model of the healthy structure. In this work the full list of properties and capabilities of the SPRT are for the first time presented and explored in the context of vibration based damage detection, identification and quantification. The method is shown to achieve effective and robust damage detection, identification and quantification based on predetermined statistical hypothesis sampling plans, which are both analytically and experimentally compared and assessed. The method's performance is determined a priori via the use of the analytical expressions of the Operating Characteristic (OC) and Average Sample Number (ASN) functions in combination with baseline data records, while it requires on average a minimum number of samples in order to reach a decision compared to most powerful Fixed Sample Size (FSS) tests. The effectiveness of the proposed method is validated and experimentally assessed via its application on a lightweight aluminum truss structure, while the obtained results for three distinct vibration measurement positions prove the method's ability to operate based even on a single pair of measured excitation-response signals. Finally, Chapter VI contains the concluding remarks and future perspectives of the thesis. / Κατά τη διάρκεια των τελευταίων 30 ετών έχει σημειωθεί σημαντική ανάπτυξη στο πεδίο της ανίχνευσης και αναγνώρισης βλαβών, το οποίο αναφέρεται συνολικά και σαν διάγνωση βλαβών. Επίσης, κατά την τελευταία δεκαετία έχει σημειωθεί σημαντική πρόοδος στον τομέα της παρακολούθησης της υγείας (δομικής ακεραιότητας) κατασκευών. Στόχος αυτής της διατριβής είναι η ανάπτυξη εξελιγμένων συναρτησιακών και επαναληπτικών μεθόδων χρονοσειρών για τη διάγνωση βλαβών και την παρακολούθηση της υγείας κατασκευών υπό ταλάντωση. Αρχικά γίνεται η πειραματική αποτίμηση και κριτική σύγκριση των σημαντικότερων στατιστικών μεθόδων χρονοσειρών επί τη βάσει της εφαρμογής τους σε πρότυπες εργαστηριακές κατασκευές. Εφαρμόζονται μη-παραμετρικές και παραμετρικές μέθοδοι που βασίζονται σε ταλαντωτικά σήματα διέγερσης και απόκρισης των κατασκευών. Στη συνέχεια αναπτύσσονται στοχαστικά συναρτησιακά μοντέλα για την στοχαστική αναγνώριση κατασκευών υπό πολλαπλές συνθήκες λειτουργίας. Τα μοντέλα αυτά χρησιμοποιούνται για την αναπαράσταση κατασκευών σε διάφορες καταστάσεις βλάβης (θέση και μέγεθος βλάβης), ώστε να είναι δυνατή η συνολική μοντελοποίσή τους για όλες τις συνθήκες λειτουργίας. Τα μοντέλα αυτά αποτελούν τη βάση στην οποία αναπτύσσεται μια συναρτησιακή μέθοδος η οποία είναι ικανή να αντιμετωπίσει συνολικά και ενιαία το πρόβλημα της ανίχνευσης, εντοπισμού και εκτίμησης βλαβών σε κατασκευές. Η πειραματική αποτίμηση της μεθόδου γίνεται με πολλαπλά πειράματα σε εργαστηριακό σκελετό αεροσκάφους. Στο τελευταίο κεφάλαιο της διατριβής προτείνεται μια καινοτόμος στατιστική επαναληπτική μέθοδο για την παρακολούθηση της υγείας κατασκευών. Η μέθοδος κρίνεται αποτελεσματική υπό καθεστώς λειτουργικών αβεβαιοτήτων, καθώς χρησιμοποιεί επαναληπτικά και στατιστικά τεστ πολλαπλών υποθέσεων. Η αποτίμηση της μεθόδου γίνεται με πολλαπλά εργαστηριακά πειράματα, ενώ η μέθοδος κρίνεται ικανή να λειτουργήσει με τη χρήση ενός ζεύγους ταλαντωτικών σημάτων διέγερσης-απόκρισης.

Nichtstationarität ist eines der häufigsten, jedoch nach wie vor ungelösten Probleme in der Zeitreihenanalyse und ein immer wiederkehrendes Phänomen, sowohl in theoretischen als auch in angewandten Arbeiten. Die jüngsten Fortschritte in der ökonometrischen Theorie und in Methoden des maschinellen Lernens haben es Forschern ermöglicht, neue Ansätze für empirische Analysen zu entwickeln, von denen einige in dieser Arbeit erörtert werden sollen. Kapitel 3 befasst sich mit der Vorhersage von Mergers & Acquisitions (M&A). Obwohl es keinen Zweifel daran gibt, dass M&A-Aktivitäten im Unternehmenssektor wellenartigen Mustern folgen, gibt es keine einheitlich akzeptierte Definition einer solchen "Mergerwelle" im Zeitreihenkontext. Zur Messung der Fusions- und Übernahmetätigkeit werden häufig Zeitreihenmodelle mit Zähldaten verwendet und Mergerwellen werden dann als Cluster von Zeiträumen mit einer ungewöhnlich hohen Anzahl von solchen Mergers & Acqusitions im Nachhinein definiert. Die Verteilung der Abschlüsse ist jedoch in der Regel nicht normal (von Gaußscher Natur). In jüngster Zeit wurden verschiedene Ansätze vorgeschlagen, die den zeitlich variablen Charakter der M&A-Aktivitäten berücksichtigen, aber immer noch eine a-priori-Auswahl der Parameter erfordern. Wir schlagen vor, die Kombination aus einem lokalem parametrischem Ansatz und Multiplikator-Bootstrap an einen Zähldatenkontext anzupassen, um lokal homogene Intervalle in den Zeitreihen der M&A-Aktivität zu identifizieren. Dies macht eine manuelle Parameterauswahl überflüssig und ermöglicht die Erstellung genauer Prognosen ohne manuelle Eingaben. Kapitel 4 ist eine empirische Studie über Sprünge in Hochfrequenzmärkten für Kryptowährungen. Während Aufmerksamkeit ein Prädiktor für die Preise von Kryptowährungenn ist und Sprünge in Bitcoin-Preisen bekannt sind, wissen wir wenig über ihre Alternativen. Die Untersuchung von hochfrequenten Krypto-Ticks gibt uns die einzigartige Möglichkeit zu bestätigen, dass marktübergreifende Renditen von Kryptowährungenn durch Sprünge in Hochfrequenzdaten getrieben werden, die sich um Black-Swan-Ereignisse gruppieren und den saisonalen Schwankungen von Volatilität und Handelsvolumen ähneln. Regressionen zeigen, dass Sprünge innerhalb des Tages die Renditen am Ende des Tages in Größe und Richtung erheblich beeinflussen. Dies liefert grundlegende Forschungsergebnisse für Krypto-Optionspreismodelle und eröffnet Möglichkeiten, die ökonometrische Theorie weiterzuentwickeln, um die spezifische Marktmikrostruktur von Kryptowährungen besser zu berücksichtigen. In Kapitel 5 wird die zunehmende Verbreitung von Kryptowährungen (Digital Assets / DAs) wie Bitcoin (BTC) erörtert, die den Bedarf an genauen Optionspreismodellen erhöht. Bestehende Methoden werden jedoch der Volatilität der aufkommenden DAs nicht gerecht. Es wurden viele Modelle vorgeschlagen, um der unorthodoxen Marktdynamik und den häufigen Störungen in der Mikrostruktur zu begegnen, die durch die Nicht-Stationarität und die besonderen Statistiken der DA-Märkte verursacht werden. Sie sind jedoch entweder anfällig für den Fluch der Dimensionalität, da zusätzliche Komplexität erforderlich ist, um traditionelle Theorien anzuwenden, oder sie passen sich zu sehr an historische Muster an, die sich möglicherweise nie wiederholen. Stattdessen nutzen wir die jüngsten Fortschritte beim Clustering von Marktregimen (MR) mit dem Implied Stochastic Volatility Model (ISVM) auf einem sehr aktuellen Datensatz, der BTC-Optionen auf der beliebten Handelsplattform Deribit abdeckt. Time-Regime Clustering ist eine temporale Clustering-Methode, die die historische Entwicklung eines Marktes in verschiedene Volatilitätsperioden unter Berücksichtigung der Nicht-Stationarität gruppiert. ISVM kann die Erwartungen der Anleger in jeder der stimmungsgesteuerten Perioden berücksichtigen, indem es implizite Volatilitätsdaten (IV) verwendet. In diesem Kapitel wenden wir diese integrierte Zeitregime-Clustering- und ISVM-Methode (MR-ISVM) auf Hochfrequenzdaten für BTC-Optionen an. Wir zeigen, dass MR-ISVM dazu beiträgt, die Schwierigkeiten durch die komplexe Anpassung an Sprünge in den Merkmalen höherer Ordnung von Optionspreismodellen zu überwinden. Dies ermöglicht es uns, den Markt auf der Grundlage der Erwartungen seiner Teilnehmer auf adaptive Weise zu bewerten und das Verfahren auf einen neuen Datensatz anzuwenden, der bisher unerforschte DA-Dynamiken umfasst. / Nonstationarity is one of the most prevalent, yet unsolved problems in time series analysis and a reoccuring phenomenon both in theoretical, and applied works. Recent advances in econometric theory and machine learning methods have allowed researchers to adpot and develop new approaches for empirical analyses, some of which will be discussed in this thesis. Chapter 3 is about predicting merger & acquisition (M&A) events. While there is no doubt that M&A activity in the corporate sector follows wave-like patterns, there is no uniquely accepted definition of such a "merger wave" in a time series context. Count-data time series models are often employed to measure M&A activity and merger waves are then defined as clusters of periods with an unusually high number of M&A deals retrospectively. However, the distribution of deals is usually not normal (Gaussian). More recently, different approaches that take into account the time-varying nature of M&A activity have been proposed, but still require the a-priori selection of parameters. We propose adapating the combination of the Local Parametric Approach and Multiplier Bootstrap to a count data setup in order to identify locally homogeneous intervals in the time series of M&A activity. This eliminates the need for manual parameter selection and allows for the generation of accurate forecasts without any manual input. Chapter 4 is an empirical study on jumps in high frequency digital asset markets. While attention is a predictor for digital asset prices, and jumps in Bitcoin prices are well-known, we know little about its alternatives. Studying high frequency crypto ticks gives us the unique possibility to confirm that cross market digital asset returns are driven by high frequency jumps clustered around black swan events, resembling volatility and trading volume seasonalities. Regressions show that intra-day jumps significantly influence end of day returns in size and direction. This provides fundamental research for crypto option pricing models and opens up possibilities to evolve econometric theory to better address the specific market microstructure of cryptos. Chapter 5 discusses the increasing adoption of Digital Assets (DAs), such as Bitcoin (BTC), which raises the need for accurate option pricing models. Yet, existing methodologies fail to cope with the volatile nature of the emerging DAs. Many models have been proposed to address the unorthodox market dynamics and frequent disruptions in the microstructure caused by the non-stationarity, and peculiar statistics, in DA markets. However, they are either prone to the curse of dimensionality, as additional complexity is required to employ traditional theories, or they overfit historical patterns that may never repeat. Instead, we leverage recent advances in market regime (MR) clustering with the Implied Stochastic Volatility Model (ISVM) on a very recent dataset covering BTC options on the popular trading platform Deribit. Time-regime clustering is a temporal clustering method, that clusters the historic evolution of a market into different volatility periods accounting for non-stationarity. ISVM can incorporate investor expectations in each of the sentiment-driven periods by using implied volatility (IV) data. In this paper, we apply this integrated time-regime clustering and ISVM method (termed MR-ISVM) to high-frequency data on BTC options. We demonstrate that MR-ISVM contributes to overcome the burden of complex adaption to jumps in higher order characteristics of option pricing models. This allows us to price the market based on the expectations of its participants in an adaptive fashion and put the procedure to action on a new dataset covering previously unexplored DA dynamics.

The mitigation of greenhouse gases, like CO2 is a challenging aspect for our society. A strategy to hamper the constant emission of CO2 is utilizing carbon capture and storage technologies. CO2 is sequestrated in subsurface reservoirs. However, these reservoirs harbor the risk of leakage and appropriate geophysical monitoring methods are needed. A crucial aspect of monitoring is the assignment of measured data to certain events occurring. Especially if changes in the measured data are small, suitable statistical methods are needed. In this thesis, a new statistical workflow based on cluster analysis is proposed to detect similar transient electromagnetic signals. The similarity criteria dynamic time warping, the autoregressive distance, and the normalized root-mean-square distance are investigated and evaluated with respect to the classic Euclidean norm. The optimal number of clusters is determined using the gap statistic and visualized with multidimensional scaling. To validate the clustering results, silhouette values are used. The statistical workflow is applied to a synthetic data set, a long-term monitoring data set and a repeat measurement at a pilot CO2-sequestration site in Brooks, Alberta.