Exploration of a digital twin concept for income maximisation of the Waikaremoana power scheme

Abstract

Digital twins are digital emulations of real-world objects or systems which mirrors the asset in terms of both behaviour and likeness. Genesis Energy Ltd, an electricity generation company in New Zealand, is interested in developing a digital twin for one of its hydropower assets, the Waikaremoana Power Scheme (WPS). The scheme is a multi-lake cascading system in the North Island with a generation capacity of 138 MW using seven turbines. The ultimate goal for digital twin development with Genesis is to create a tool capable of providing decision making support for traders managing the utilisation of the WPS in the NZ electricity market to maximise income and efficiency while minimising losses. This project is a tentative exploration into how an early digital twin concept could be built for the WPS with the end objective of maximising utilisation through optimising unit commitment and scheduling. Plant data accuracy and reliability was examined as it is a foundational element to any digital twin. It was found that the WPS possessed accurate instruments for parameters like power output and water levels but relied on correlations for many flow readings around the scheme. Data sampling methods were also examined, and it was found that averaged data was better at short sampling intervals due to reduction of noise. A flow model was built in Microsoft Excel using a first principles-based approach, assembled using mass and energy balances along with characteristic equations for the scheme. The accuracy of the model was tested against net flow values via lake sensor readings. It was found that on average, there was a difference of around 2 m3/s for all three lakes but increased in proportion to the model net flow rate. As part of the flow model, the efficiency characteristic functions were found using a regression refinement process, starting from linear regression refining eventually to a multivariable linear (polynomial) model. The model was validated using test data from each of the generation units with excellent or good fits found for all units. . A profit-based optimisation formulation was developed based on literature and the flow model developed. The formulation was applied to a simplified case comprising of a single unit, lake, and time slice problem and testing the relationship between spot price and water value. The problem was solved using Excel Solver and a GRG nonlinear method. Depending on whether the spot price was greater or smaller than water value, the optimiser chose to generate at maximum efficiency before increasing to maximum output as spot price rose. The optimisation encountered difficulties when extending the problem to include multiple units. The Excel Solver was unable to find the global optimum without increasing compute times to unacceptable levels. To proceed further with this optimisation problem, it would need to be moved to a platform with more complex solvers.