Interest in distributed approaches to machine learning has increased significantly in recent years due to continuously increasing data sizes for training machine learning models. In this thesis we describe three popular machine learning algorithms: decision trees, Naive Bayes and support vector machines (SVM) and present existing ways of distributing them. We also perform experiments with decision trees distributed with bagging, boosting and hard data partitioning and evaluate them in terms of performance measures such as accuracy, F1 score and execution time. Our experiments show that the execution time of bagging and boosting increase linearly with the number of workers, and that boosting performs significantly better than bagging and hard data partitioning in terms of F1 score. The hard data partitioning algorithm works well for large datasets where the execution time decrease as the number of workers increase without any significant loss in accuracy or F1 score, while the algorithm performs poorly on small data with an increase in execution time and loss in accuracy and F1 score when the number of workers increase.

The aim of this dissertation is to develop an ensemble algorithm for short-term electricity consumption forecasting and to assess the impact of meteorological factors and other features on the quality of the model. The work addresses issues related to improving the accuracy of electricity consumption forecasting using open meteorological data through ensemble machine learning methods and hyperparameter tuning algorithms. A review and analysis of existing methods for time series forecasting are conducted, taking into account the specific characteristics of electricity consumption time series forecasting. Additionally, a machine learning algorithm is developed using various factors as features for model training. The developed algorithm is tested on electricity consumption data from the Siberian Regional Dispatch Office. / Целью диссертационной работы является разработка ансамблевого алгоритма для краткосрочного прогнозирования электропотребления и оценка влияния на качество модели метеофакторов н других признаков. В работе рассматриваются вопросы повышения точности прогнозирования электропотребления с использованием открытых метеорологических данных с применением ансамблевых методов машинного обучения, и алгоритма подбора гиперпараметров моделей. В работе проведен обзор и анализ существующих методов для прогнозирования временных рядов с учетом особенностей прогнозирования временного ряда электропотребления. Также в работе разработан алгоритм машинного обучения с использованием различных факторов в качестве признаков для обучения моделей. Разработанный алгоритм протестирован на данных электропотребления в зоне ответственности ОДУ Сибири.

This work addresses the issues of improving the accuracy of forecasting the generation of photovoltaic power plants based on open meteorological data using machine learning methods and a preprocessing algorithm for the initial data. / В работе рассматриваются вопросы повышения точности прогнозирования генерации фотоэлектрических станций на открытых метеорологических данных с использованием методов машинного обучения, и алгоритма предварительной обработки исходных данных.

[pt] Boosting é uma técnica de aprendizado de máquina que combina diversos classificadores fracos com o objetivo de melhorar a acurácia geral. Em cada iteração, o algoritmo atualiza os pesos dos exemplos e constrói um classificador adicional. Um esquema simples de votação é utilizado para combinar os classificadores. O algoritmo mais famoso baseado em Boosting é o AdaBoost. Este algoritmo aumenta os pesos dos exemplos em que os classificadores anteriores cometeram erros. Assim, foca o classificador adicional nos exemplos mais difíceis. Inicialmente, uma distribuição uniforme de pesos é atribúda aos exemplos. Entretanto, não existe garantia que essa seja a melhor escolha para a distribuição inicial. Neste trabalho, apresentamos o Boosting at Start (BAS), uma nova abordagem de aprendizado de máquina baseada em Boosting. O BAS generaliza o AdaBoost permitindo a utilização de uma distribuição inicial arbitrária. Também apresentamos esquemas para determinação de tal distribuição. Além disso, mostramos como adaptar o BAS para esquemas de Aprendizado Semi-supervisionado. Adicionalmente, descrevemos a aplicação do BAS em diferentes problemas de classificação de dados e de texto, comparando o seu desempenho com o algoritmo AdaBoost original e alguns algoritmos do estado-da-arte para tais tarefas. Os resultados experimentais indicam que uma modelagem simples usando o algoritmo BAS gera classificadores eficazes. / [en] Boosting is a Machine Learning technique that combines several weak classifers with the goal of improving the overall accuracy. In each iteration, the algorithm updates the example weights and builds an additional classifer. A simple voting scheme is used to combine the classifers. The most famous Boosting-based algorithm is AdaBoost. This algorithm increases the weights of the examples that were misclassifed by the previous classifers. Thus, it focuses the additional classifer on the hardest examples. Initially, an uniform weight distribution is assigned to the examples. However, there is no guarantee that this is the best choice for the initial distribution. In this work, we present Boosting at Start (BAS), a new Machine Learning approach based on Boosting. BAS generalizes AdaBoost by allowing the use of an arbitrary initial distribution. We present schemes for the determination of such distribution. We also show how to adapt BAS to Semi-supervised learning schemes. Additionally, we describe the application of BAS in different problems of data and text classifcation, comparing its performance with the original AdaBoost algorithm and some state-of-the-art algorithms for such tasks. The experimental results indicate that a simple modelling using the BAS algorithm generates effective classifers.

The aim of this work is to develop a method for constructing an optimal dynamic topology of a 6/10 kV distribution network using a network sectioning algorithm (operation of switching devices) based on load forecasting to achieve the best indicators in terms of quality, reliability, and reduction of technological losses. / Целью работы является разработка метода построения оптимальной динамической топологии распределительной сети 6/10 кВ с применением алгоритма секционирования сети (работы коммутационных аппаратов) на основе прогноза нагрузки для достижения наилучших показателей по качеству, надежности, снижению технологических потерь.