Search

With ever-increasing disruptions in supply chains throughout the world, like pandemics, political conflicts, and trade wars, power generation is becoming one of the major concerns of global and local economies. The recent increasing energy cost in Europe and North America is one of the main consequences and, at the same time, contributes to these disruptions. Therefore, electricity demand forecasting is crucial in power markets to increase cooperation and integration between players in a power grid. This study aims at reviewing the managerial implications of demand forecasting in the electricity supply chain. Also, some recent statistical and machine learning techniques for electricity demand forecasting used in the literature are analyzed and applied to Ontario's historical dataset. A descriptive analysis of electricity demand characteristics in Ontario, post, and pre-pandemic, is conducted. Furthermore, the forecasting performance of methods like dynamic regression, neural network autoregression, and prophet model are discussed and compared. Another contribution of this study is to include fuzzy hourly demand forecasts for a Canadian dataset.

With ever-increasing disruptions in supply chains throughout the world, like pandemics, political conflicts, and trade wars, power generation is becoming one of the major concerns of global and local economies. The recent increasing energy cost in Europe and North America is one of the main consequences and, at the same time, contributes to these disruptions. Therefore, electricity demand forecasting is crucial in power markets to increase cooperation and integration between players in a power grid. This study aims at reviewing the managerial implications of demand forecasting in the electricity supply chain. Also, some recent statistical and machine learning techniques for electricity demand forecasting used in the literature are analyzed and applied to Ontario's historical dataset. A descriptive analysis of electricity demand characteristics in Ontario, post, and pre-pandemic, is conducted. Furthermore, the forecasting performance of methods like dynamic regression, neural network autoregression, and prophet model are discussed and compared. Another contribution of this study is to include fuzzy hourly demand forecasts for a Canadian dataset.

Electricity price forecasting plays an important role in decision-making on bidding strategies of selling and buying electricity. This thesis computes one-day-ahead quantile forecasts of electricity prices in a highly volatile market by applying regression models to a pool of point forecasts. Three data-driven forecasting methods are implemented to generate day-ahead point forecasts of the Ontario market’s electricity prices. In order to generate the three sets of point forecasts, I use: i) the Triple Exponential Smoothing (TES) method, ii) a Neural Network (NN) that combines layers of Convolutional neurons and Gradient Recurrent Units (GRU), iii) an eXtreme Gradient Boosting (XGB) non-linear regression approach. The performance of the three models is compared against a benchmark that considers the forecast of electricity prices as the average price of the same hour and day during the last four weeks. The TES method decreases the Mean Absolute Error (MAE) of the benchmark model from 10.29 to 9.42. The Convolutional GRU (ConvGRU) model and XGB regression also reduce the MAE to 8.20 and 7.06, respectively. Finally, Quantile Regression Averaging (QRA) is applied to the pool of point forecasts obtained by TES, ConvGRU, and XGB methods to compute day-ahead quantile forecasts of electricity prices. Moreover, the QRA method is further developed in this thesis by employing Gradient Boosting Regression (GBR). It follows from my real data analysis that the GBR method provides more reliable quantiles and tighter prediction intervals with smaller forecasting errors than QRA. The obtained probabilistic forecasts are used to find the optimal energy procurement plan for a large consumer and the linear programming method is applied to solve the problem. The simulation results indicate that using probabilistic forecasts of electricity prices leads to a more flexible and efficient bidding strategy than using point forecasts. Moreover, the regularized probabilistic forecast of day-ahe

Electricity price forecasting plays an important role in decision-making on bidding strategies of selling and buying electricity. This thesis computes one-day-ahead quantile forecasts of electricity prices in a highly volatile market by applying regression models to a pool of point forecasts. Three data-driven forecasting methods are implemented to generate day-ahead point forecasts of the Ontario market’s electricity prices. In order to generate the three sets of point forecasts, I use: i) the Triple Exponential Smoothing (TES) method, ii) a Neural Network (NN) that combines layers of Convolutional neurons and Gradient Recurrent Units (GRU), iii) an eXtreme Gradient Boosting (XGB) non-linear regression approach. The performance of the three models is compared against a benchmark that considers the forecast of electricity prices as the average price of the same hour and day during the last four weeks. The TES method decreases the Mean Absolute Error (MAE) of the benchmark model from 10.29 to 9.42. The Convolutional GRU (ConvGRU) model and XGB regression also reduce the MAE to 8.20 and 7.06, respectively. Finally, Quantile Regression Averaging (QRA) is applied to the pool of point forecasts obtained by TES, ConvGRU, and XGB methods to compute day-ahead quantile forecasts of electricity prices. Moreover, the QRA method is further developed in this thesis by employing Gradient Boosting Regression (GBR). It follows from my real data analysis that the GBR method provides more reliable quantiles and tighter prediction intervals with smaller forecasting errors than QRA. The obtained probabilistic forecasts are used to find the optimal energy procurement plan for a large consumer and the linear programming method is applied to solve the problem. The simulation results indicate that using probabilistic forecasts of electricity prices leads to a more flexible and efficient bidding strategy than using point forecasts. Moreover, the regularized probabilistic forecast of day-ahe