Abstract
The rapid growth of large-scale PV plants is increasing the need for energy storage, being batteries the most promising technology in the short-term. Research focused on improving the feasibility of PV-battery systems could help to encourage their development. This study contributes to this process, studying how optimising the operation could make these systems more cost-effective. The optimal operation can be obtained with an optimisation algorithm, that minimises the daily cost of electricity. Conventional methods use this approach, but need to forecast PV generation, electricity price and demand in real-time, with the problems that it entails. A novel method based on Machine Learning (ML) is proposed in this thesis. Past data is used to train the model, that learns to predict the optimal operation. This method is assessed for the case of Cranfield University, where a new PV-battery plant has been proposed. Conventional and ML methods are compared, analysing their performance and calculating their potential savings. After this analysis, ML outperformed the conventional operation, obtaining 13% more accuracy and 5% more savings. For Cranfield University, the cost-benefit analysis showed that including a PV-battery system would be beneficial, saving more than £300k per year. The benefits from using ML operation are high, particularly when increasing PV or battery capacity. Therefore, this methodology could have a great impact on large-scale PV-battery applications.