Wind power is one of the most promising renewable energy resources to replace conventional generation which carries high carbon footprints. Due to the abundance of wind and its relatively cheap installation costs, it is likely that wind power will become the most important energy resource in the near future. The successful development of wind power relies heavily on the ability to integrate wind power effciently into electricity grids. To optimize the value of wind power through careful power dispatches, techniques in forecasting the level of wind power and the associated variability are critical. Ideally, one would like to obtain reliable probability density forecasts for the wind power distributions. As wind is intermittent and wind turbines have non-linear power curves, this is a challenging task and many ongoing studies relate to the topic of wind power forecasting. For this reason, this thesis aims at contributing to the literature on wind power forecasting by constructing and analyzing various time series models and spatiotemporal models for wind power production. By exploring the key features of a portfolio of wind power data from Ireland and Denmark, we investigate different types of appropriate models. For instance, we develop anisotropic spatiotemporal correlation models to account for the propagation of weather fronts. We also develop twostage models to accommodate the probability masses that occur in wind power distributions due to chains of zeros. We apply the models to generate multi-step probability forecasts for both the individual and aggregated wind power using extensive data sets from Ireland and Denmark. From the evaluation of probability forecasts, valuable insights are obtained and deeper understanding of the strengths of various models could be applied to improve wind power forecasts in the future.

The aim of the thesis was to create a tool for time-series prediction based on deep learning. The first part of the work is a brief description of deep learning and its comparison to classical machine learning. In the next section contains brief analysis of some tools, that are already used for time-series forecasting. The last part is focused on the analysis of the problem as well as on the actual creation of the program.

A massive amount has been written about forecasting but few articles are written about the development of time series models of call volumes for emergency services. In this study, we use different techniques for forecasting and make the comparison of the techniques for the call volume of the emergency service Rescue 1122 Lahore, Pakistan. For the purpose of this study data is taken from emergency calls of Rescue 1122 from 1st January 2008 to 31 December 2009 and 731 observations are used. Our goal is to develop a simple model that could be used for forecasting the daily call volume. Two different approaches are used for forecasting the daily call volume Box and Jenkins (ARIMA) methodology and Smoothing methodology. We generate the models for forecasting of call volume and present a comparison of the two different techniques.

Time series forecasting has become a common problem in day-to-day applications and various machine learning algorithms have been developed to tackle this task. Finding the model that performs the best forecasting on a given dataset can be time consuming as multiple algorithms and hyperparameter configurations must be examined to find the best model. This problem can be solved using automated machine learning, an approach that automates all steps required for developing a machine learning algorithm including finding the best algorithm and hyperparameter configuration. This study develops and builds an automated machine learning pipeline focused on finding the best forecasting model for a given dataset. This includes choosing different forecasting algorithms to cover a wide range of tasks and identifying the best method to find the best model in these algorithms. Lastly, the final pipeline will then be tested on a variety of datasets to evaluate the performance on time series data with different characteristics.:Abstract List of Figures List of Tables List of Abbreviations List of Symbols 1. Introduction 2. Theoretical Background 2.1. Machine Learning 2.2. Automated Machine Learning 2.3. Hyperparameter Optimization 2.3.1. Model-Free Methods 2.3.2. Bayesian Optimization 3. Time Series Forecasting Algorithms 3.1. Time Series Data 3.2. Baselines 3.2.1. Naive Forecast 3.2.2. Moving Average 3.3. Linear Regression 3.4. Autoregression 3.5. SARIMAX 3.6. XGBoost 3.7. LSTM Neural Network 4. Automated Machine Learning Pipeline 4.1. Data Preparation 4.2. Model Selection 4.3. Hyperparameter Optimization Method 4.3.1. Sequential Model-Based Algorithm Configuration 4.3.2. Tree-structured Parzen Estimator 4.3.3. Comparison of Bayesian Optimization Hyperparameter Optimization Methods 4.4. Pipeline Structure 5. Testing on external Datasets 5.1. Beijing PM2.5 Pollution 5.2. Perrin Freres Monthly Champagne Sales 6. Testing on internal Datasets 6.1. Deutsche Telekom Call Count 6.1.1. Comparison of Bayesian Optimization and Random Search 6.2. Deutsche Telekom Call Setup Time 7. Conclusion Bibliography A. Details Search Space B. Pipeline Results - Predictions C. Pipeline Results - Configurations D. Pipeline Results - Experiment Details E. Deutsche Telekom Data Usage Permissions

The thesis concerns the topic of forecasting using Neural Networks, particu- larly the return and volatility forecasting in the volatile period of Covid-19. The thesis uses adjusted close daily data from Jan 1, 2000, until Jan 1, 2021, of the S&P index and Prague Exchange Stock index (PX). The comparison was between the vanilla econometrical model, a neural network model, and a hybrid neural network model. Hybrid neural networks were constructed with an additional feature column of the fitted econometrical model. Additionally to comparing the prediction, a risk-return trade-o analysis of the forecasted series was conducted. The test period for all models was from Jan 1, 2020, until Jan 1, 2021, where predictions were made. During the test period, MSE be- tween predicted and true values was extracted and compared. The results are that the hybrid model outperformed both econometrical as well as only neural networks models. Furthermore, the risk-return trade-o forecast provided by the hybrid model fares better than the other ones. JEL Classification C53, C81 Keywords Financial Time Series, Forecasting, Neural Net- works, ARIMA, GARCH Title Forecasting with Neural Network during Covid- 19 Crisis Author's e-mail tiep.luud@gmail.com Supervisor's e-mail barunik@fsv.cuni.cz

Time-series forecasting has long been studied and remains an important research task. In scenarios where multiple time series need to be forecast, approaches that exploit the mutual impact between time series results in more accurate forecasts. This has been demonstrated in various applications, including demand forecasting and traffic forecasting, among others. Hence, this dissertation focuses on graph-based models, which leverage the internode relations to forecast more efficiently and effectively by associating time series with nodes. This dissertation begins by introducing the notion of graph time-series models in a comprehensive survey of related models. The main contributions of this survey are: (1) A novel categorization is proposed to thoroughly analyze over 20 representative graph time-series models from various perspectives, including temporal components, propagation procedures, and graph construction methods, among others. (2) Similarities and differences among models are discussed to provide a fundamental understanding of decisive factors in graph time-series models. Model challenges and future directions are also discussed. Following the survey, this dissertation develops graph time-series models that utilize complex time-series interactions to yield context-aware, real-time, and probabilistic forecasting. The first method, Context Integrated Graph Neural Network (CIGNN), targets resource forecasting with contextual data. Previous solutions either neglect contextual data or only leverage static features, which fail to exploit contextual information. Its main contributions include: (1) Integrating multiple contextual graphs; and (2) Introducing and incorporating temporal, spatial, relational, and contextual dependencies; The second method, Evolving Super Graph Neural Network (ESGNN), targets large-scale time-series datasets through training on super graphs. Most graph time-series models let each node associate with a time series, potentially resulting in a high time cost. Its main contributions include: (1) Generating multiple super graphs to reflect node dynamics at different periods; and (2) Proposing an efficient super graph construction method based on K-Means and LSH; The third method, Probabilistic Hypergraph Recurrent Neural Network (PHRNN), targets datasets under the assumption that nodes interact in a simultaneous broadcasting manner. Previous hypergraph approaches leverage a static weight hypergraph, which fails to capture the interaction dynamics among nodes. Its main contributions include: (1) Learning a probabilistic hypergraph structure from the time series; and (2) Proposing the use of a KNN hypergraph for hypergraph initialization and regularization. The last method, Graph Deep Factors (GraphDF), aims at efficient and effective probabilistic forecasting. Previous probabilistic approaches neglect the interrelations between time series. Its main contributions include: (1) Proposing a framework that consists of a relational global component and a relational local component; (2) Conducting analysis in terms of accuracy, efficiency, scalability, and simulation with opportunistic scheduling. (3) Designing an algorithm for incremental online learning. / Doctor of Philosophy / Time-series forecasting has long been studied due to its usefulness in numerous applications, including demand forecasting, traffic forecasting, and workload forecasting, among others. In scenarios where multiple time series need to be forecast, approaches that exploit the mutual impact between time series results in more accurate forecasts. Hence, this dissertation focuses on a specific area of deep learning: graph time-series models. These models associate time series with a graph structure for more efficient and effective forecasting. This dissertation introduces the notion of graph time series through a comprehensive survey and analyzes representative graph time-series models to help readers gain a fundamental understanding of graph time series. Following the survey, this dissertation develops graph time-series models that utilize complex time-series interactions to yield context-aware, real-time, and probabilistic forecasting. The first method, Context Integrated Graph Neural Network (CIGNN), incorporates multiple contextual graph time series for resource time-series forecasting. The second method, Evolving Super Graph Neural Network (ESGNN), constructs dynamic super graphs for large-scale time-series forecasting. The third method, Probabilistic Hypergraph Recurrent Neural Network (PHRNN), designs a probabilistic hypergraph model that learns the interactions between nodes as distributions in a hypergraph structure. The last method, Graph Deep Factors (GraphDF), targets probabilistic time-series forecasting with a relational global component and a relational local model. These methods collectively covers various data characteristics and model structures, including graphs, super graph, and hypergraphs; a single graph, dual graphs, and multiple graphs; point forecasting and probabilistic forecasting; offline learning and online learning; and both small and large-scale datasets. This dissertation also highlights the similarities and differences between these methods. In the end, future directions in the area of graph time series are also provided.

Foreign exchange market is the largest financial market in the world and thus prediction of foreign exchange rate values is of interest to millions of people. In this research, I evaluated the performance of Long Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Autoregressive Integrated Moving Average (ARIMA) and Moving Average (MA) on the USD/CAD and USD/AUD exchange pairs for 1-day, 1-week and 2-weeks predictions. For LSTM and GRU, twelve macroeconomic indicators along with past exchange rate values were used as features using data from January 2001 to December 2019. Predictions from each model were then integrated with uncertainty modeling to find out the chance of a model’s prediction being greater than or less than a user-defined target value using the error distribution from the test dataset, Monte-Carlo simulation trials and ChancCalc excel add-in. Results showed that ARIMA performs slightly better than LSTM and GRU for 1-day predictions for both USD/CAD and USD/AUD exchange pairs. However, when the period is increased to 1-week and 2-weeks, LSTM and GRU outperform both ARIMA and moving average for both USD/CAD and USD/AUD exchange pair.

Time series forecasting and modeling are challenging problems during the past decades, because of its plenty of properties and underlying correlated relationships. As a result, researchers proposed a lot of models to deal with the time series. However, the proposed models such as Autoregressive integrated moving average (ARIMA) and artificial neural networks (ANNs) only describe part of the properties of time series. In this thesis, we introduce a new hybrid model integrated filter structure to improve the prediction accuracy. Case studies with real data from University of Kentucky HealthCare are carried out to examine the superiority of our model. Also, we applied our model to operating room (OR) to reduce the inefficiency cost. The experiment results indicate that our model always outperforms compared with other models in different conditions.

n the electrical power grid, the power load is not constant but continuouslychanging. This depends on many different factors, among which the habits of theconsumers, the yearly seasons and the hour of the day. The continuous change inenergy consumption requires the power grid to be flexible. If the energy provided bygenerators is lower than the demand, this is usually compensated by using renewablepower sources or stored energy until the power generators have adapted to the newdemand. However, if buffers are depleted the output may not meet the demandedpower and could cause power outages. The currently adopted practice in the indus-try is based on configuring the grid depending on some expected power draw. Thisanalysis is usually performed at a high level and provide only some basic load aggre-gate as an output. In this thesis, we aim at investigating techniques that are able topredict the behaviour of loads with fine-grained precision. These techniques couldbe used as predictors to dynamically adapt the grid at run time. We have investigatedthe field of time series forecasting and evaluated and compared different techniquesusing a real data set of the load of the Swedish power grid recorded hourly throughyears. In particular, we have compared the traditional ARIMA models to a neuralnetwork and a long short-term memory (LSTM) model to see which of these tech-niques had the lowest forecasting error in our scenario. Our results show that theLSTM model outperformed the other tested models with an average error of 6,1%.

Este trabalho apresenta um estudo sobre o paradigma de Algoritmos Genéticos aplicados a área de Predições de Séries Temporais. O resultado deste trabalho é apresentado na forma de comparação dos resultados obtidos entre o Modelo Clássico de Predição (UCM), Redes Neurais Artificiais (RNAs) e o modelo de Algoritmos Genéticos desenvolvido neste trabalho. Este estudo foi realizado trabalhando-se basicamente com o Índice Mensal de Produção Industrial do Estado do Rio Grande do Sul fornecido pelo IBGE (Instituto Brasileiro de Geografia e Estatística). Os resultados obtidos mostram que os Algoritmos Genéticos podem atingir níveis satisfatórios de precisão em relação aos valores preditos quando comparados com os valores reais. A validação é feita com predições de um passo à frente e de sete passos à frente. Estas predições são em relação aos sete meses iniciais do ano de 1993. / This work presents a study of Genetic Algorithms paradigm applied to Forecasting Time Series. The results are compared with the obtained with the Classic Model of Prediction (UCM), Artificial Neural Networks (RNAs). This study was accomplished using with the Monthly Index of Industrial Production of the State of Rio Grande do Sul, supplied by the IBGE(Instituto Brasileiro de Geografia e Estatística). The results show that the Genetic Algorithms can accomplish a satisfactory precision when compared with the real values. The validation is made with predictions, one and seven steps ahead. These predictions are equivalent to the seven initial months of 1993.