- 97 views
- 322 downloads
Deep Learning-based Forecasting and Energy Management Algorithms for Smart Grid Applications
-
- Author / Creator
- Dolatabadi, Amirhossein
-
With the increasing global problems concerning energy security and climate change, new challenges in social progress and human survival have come to the fore. Requiring no fuel, and being renewable and non-polluting, renewable energy (RE) resources, typically from photovoltaic and wind sources, have attracted extensive attention worldwide. However, due to their stochasticity, uncertainty, and intermittency, RE resources could pose considerable challenges to the optimal operation of the energy systems despite their non-polluting and widely available nature. This changing environment necessitates the accurate and efficient operation of the RE-integrated energy system. On the other hand, the rapidly growing applications of artificial intelligence and machine learning techniques can contribute to reducing energy costs, maintaining the balance between generation and demand, and satisfying consumers’ needs.
The existing literature on the scheduling and operation of active distribution systems falls short in several aspects, such as RE prediction performance, cost-effective operation, and realistic scheduling. This research aims to 1) develop a new deep learning-based model integrating the discrete wavelet packet transform (DWPT) and bidirectional long short-term memory (BLSTM) to capture deep temporal features of wind speed time series precisely, 2) investigate a light detection and ranging (LiDAR)-aided deep learning model to learn the powerful spatial-temporal characteristics from the input wind fields, 3) propose a novel model-free deep reinforcement learning (DRL) approach to optimize the compressed air energy storage (CAES) energy arbitrage in the presence of a solar irradiance forecasting model, 4) utilize a deep deterministic policy gradient (DDPG) framework to develop an intelligent controller to schedule the energy hub optimally, and 5) analyze the optimal operation of the biogas-integrated multi-source multi-product facility in the presence of the supervised federated neural architecture search (SFNAS) technique.
The first study in this thesis applies the DWPT to extract the features of the wind time series. It then uses the BLSTM network as a combination of LSTM networks and bidirectional RNNs to capture deep temporal features with high abstraction. The second study extends the network of the first work by combining it with 2-D convolutional neural networks (CNNs) for capturing high levels of abstractions in the wind fields provided by LiDAR. In the third study, the CNN-BLSTM model presented in the second work is used to train a DRL agent that optimizes the self-scheduling of the CAES-PV system. The fourth study upgrades the DRL framework of the third work by introducing the DDPG for more smooth control actions. Finally, the fifth study investigates the dynamic scheduling framework for an energy hub with a biomass-solar hybrid renewable system. Furthermore, an SFNAS technique has been presented to eliminate the need for manual engineering of deep neural network models and the unnecessary computational burden associated with them. The comparative results based on realistic case studies demonstrate the effectiveness and applicability of the proposed frameworks compared to the state-of-the-art methods in the recent literature.
-
- Subjects / Keywords
-
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.