Using probabilistic forecasts to optimise the revenue of energy production at the Xayaburi Hydroelectric Power Project

The Xayaburi Hydroelectric Power Project is the first run-of-river plant to be developed in the lower Mekong basin. During the operational phase, which is expected to commence in 2019, the electricity will be sold to the Electricity Generating Authority of Thailand (EGAT) as well as to the Electricite Du Laos (EDL), according to Power Purchase Agreements (PPA). The PPA with EGAT requires that the amount of energy that will be declared by the power producer on a weekly basis. Under the agreement, this should be declared well before the start of each week, effectively meaning up to 11 days in advance. EGAT guarantees to buy the energy that is declared. However, if the power producer fails to produce the energy as declared; a penalty due to the shortfall will be incurred. To maximise profit from this run-of-river plant, a reliable discharge and power production forecasting model, is therefore a key ingredient in the optimisation of the energy production. This research aims to develop such a discharge and production forecasting model, in order to forecast and optimise the decision on the energy generation of the Xayaburi Hydroelectric Power project. To develop this forecast model, historical data were first analysed in order to explain how the hydrology and the operation of dams in the upper mainstream of the Mekong River affect the flow regime at the project. Then, the existing forecasting models of the Mekong Basin, developed by the Mekong River Commission in collaboration with international consultants, were applied to forecast flows at the required lead times. These models were adapted to consider only the area relevant to the study site, and extended with a multiple linear regression model to forecast flows upstream of Chaing Saen on the China/Laos border. After that, the predictive uncertainty of the forecasts was explored, and corrected using quantile regression. Lastly, a numerical optimisation algorithm was developed based on an objective function that explicitly considers the declaration of power. The objective of the optimisation is to maximise profit from energy production sale and minimise the penalties. Results of the data analysis part show that the flow regime upstream of the project has changed due to the operation of the dams in China, with a decrease in the flow contribution in the wet season, making the forecasts for the main tributaries in Thailand and Laos more important. The operation of the dams, does, however, adversely influence the predictability of the upstream basin. Results from the flow forecasting part show that the forecasts at Chaing Saen station are quite good, while at Luang Prabang, the station upstream of the project, there is room for improvement. This could be achieved through improved hydrological modelling of the main tributaries and through incorporating real-time flow gauging stations in these tributaries. Post-processing the forecasts using quantile regression shows a clear benefit to the optimisation of energy production, with the relative economic value using the resulting probabilistic forecasts being the highest when compared to the results from other forecasting methods. In summary, it can be concluded that using probabilistic forecasts to optimise the revenue of energy production at Xayaburi Hydroelectric Power Project has significant potential in terms of higher overall revenue, and it is recommended to be developed further and applied in the operation of the power plant