|dc.description.abstract||Recently, time series forecasting has acquired a considerable academic and industrial interest in various areas for different applications. Machine learning (ML) algorithms are known for their ability to capture the chaotic temporal non-linear relations in time series data. This research employs various ML concepts and algorithms into two different case studies of time series forecasting: 1-Regional wind power forecasting and 2-Air quality index (AQI) forecasting.
The first case study is conducted to focus on regional wind power forecasting comprehensively from different perspectives. First, the meteorological and spatial parameters with seasonal and temporal features were filtered and selected by a proposed deep feature selection approach consisting of series of steps. Later, multiple ML algorithms, including artificial neural network (ANN), deep neural network (DNN), long short-term memory (LSTM), bagging tree (BT), and support vector machine/regression (SVM/SVR), were used for training one-step-ahead forecasting models. Lastly, an assessment of the constructed models was conducted based on different error criteria metrics. The final comparative discussion concluded that the SVR-based model provided accurate generalized performance when tested on unseen data and surpassed other models, including LSTM. However, when constructing the multi-step ahead forecasting models, the predictions obtained from the multi-input multi-output (MIMO) LSTM approach were reliable with higher accuracies. Overall, for multi-step forecasting, it was concluded that the performance of the MIMO multi-step strategy was superior to the direct multi-step forecasting method, especially by employing algorithms with recursive properties.
It is also essential to mention that chapter 2 of this thesis is a comprehensive literature review of machine learning and metaheuristics methodologies of renewable power forecasting. This review can guide scientists and engineers in analyzing and selecting the appropriate prediction approaches based on the different circumstances and applications.
The second case proposes a comprehensive method to forecast AQI. The proposed methodology was tested on ambient air quality observations at Al-Jahra, a major city in Kuwait. The hourly levels of the six criteria pollutants (O3, SO2, NO2, CO, PM10, and PM2.5) were predicted using artificial neural networks, which then fed into the process of estimating AQI. The prediction of the AQI does not only require the selection of a robust forecasting model, rather it heavily relies on a sequence of pre-processing steps to select predictors and handle different issues in data. One major problem that commonly appears in ambient air quality datasets is data gaps. The presented method dealt with this
by imputing missing entries using miss-forest; a machine learning-based imputation technique. The effectiveness of this imputation method was examined against the linear imputation method for the six criteria pollutants and the AQI. Results obtained showed that models trained using miss-forest imputed data could generalize AQI forecasting and with a prediction accuracy of 92.41% when tested on new unseen data.||en