Enabling high photovoltaic penetration levels in low voltage distribution networks using smart inverter control techniques
Rooftop photovoltaic (PV) installations are rapidly increasing worldwide. Almost onequarter of Australian homes now have rooftop PV systems. However, there is an emerging trend to limit PV connection and generation, and enforce export limits due to actual and perceived network constraints. Rooftop PV systems must have the necessary functions to support the network’s capacity and reliability. Accordingly, in this thesis, the aim is to enable the understanding of the true extent of local voltage excursions to allow more targeted investment, improve the network’s reliability, enhance solar performance distribution, and increase PV penetration levels in low-voltage distribution networks (LVDNs).
Modeling the uncertainty of the PV module is very important to reflect the actual behaviour of these components in real applications in LVDNs. Therefore, the first contribution of this thesis is to improve the accuracy of the physical models of PV modules using artificial intelligence (AI)-based extraction algorithms and machine learning methods. These models are proposed and evaluated based on actual data and applied to three PV technologies: mono-crystalline, poly-crystalline, and thin-film technologies. The results obtained based on the novel proposed AI-based extraction algorithms show its superiority over the existing extraction algorithms. Furthermore, the random forests-based model secured a higher forecasting accuracy than state-of-the-art machine-learning predictors.
In addition, time-varying load data leads to an increase in the complexity and uncertainty of LVDNs. Therefore, the second contribution is to develop new deep learning neural network-based models, with and without optimizing their hyper-parameters, using an adaptive wind driven optimization (AWDO) algorithm to forecast load data at the distribution zone-substation and households levels. The long short-term memory (LSTM)-based forecast algorithm achieved a higher forecasting accuracy than state-ofthe-art predictors at the substation level, while the bi-directional LSTM-based AWDO model was superior to the state-of-the-art deep-learning forecasting models.
Unbalanced three-phase LVDNs modeling, optimisation, and control with accurate mathematical representations are needed for enabling high PV penetration levels in LVDNs. Accordingly, the third contribution is to develop a case study and a benchmark model that shows the effects of smart inverter control techniques on PV penetration levels in LVDNs. This was carried out using the OpenDSS-Julia interface and utilizing the IEEE European Low Voltage Test Feeder adapted with actual smart meter load data. The developed case study’s aim is to replicate the results, so as to review the performance of the proposed smart inverter techniques and improve the reproducibility of results.
The final contribution is to develop an effective inverter control strategy to support the voltage level in three-phase four-wire LVDNs using single-phase inverters connected in arbitrary locations in the network based on centralized coordinated smart inverter control strategies to increase PV penetration levels using Julia platform. The distributed smart inverter control strategies are also considered a backup to the centralized control strategies if any communication failure happens. The new decision support framework shows better results than the existing smart inverter control solutions and overcomes their limitations in enabling high PV penetration levels in LVDNs, without breaching the network constraints and without any extra investments.
