(119d) Decision Tree-Based Optimisation for Flexible Energy Storage Dispatch

Electrical energy storage (EES) devices have shown a number of advantages in both front-of-meter and behind-the-meter applications, especially with the increased penetration of intermittent renewable energy generators (Ma et al., 2018). Some of these merits are the improvements in power reliability, flexibility, and quality, and reduction in the final electricity bill of end users. However, owing to problems of a prolonged pay-back period in large scale deployment of these devices, focus has been drawn to revenue stacking – where a collection of tasks are performed by the EES device in order to generate more revenue (Fisher et al., 2019; Roberts & Brown, 2020). In behind-the-meter applications, for example, the ability to stack the revenue streams of price arbitrage, surplus PV value uplift and ancillary service provision (via aggregation) has the potential to improve the pay-back period. In such situations, the scheduling of the EES device plays a key role in determining the total revenue generated as well as the lifetime of the device.

A number of mathematical models have been developed to optimise the operation of these devices under differing conditions of energy demand, energy generation sources, electricity prices and/or accessible revenue streams (García Vera et al., 2019), however the final results obtained are often a single set of time-dependent decision variables which must be followed to maximise revenue. In reality, these solutions require strict implementation (mostly with the help of computer systems), and do not always perform well when conditions of demand, price or energy generation deviate from assumptions/predictions, even when uncertainty is taken into consideration. A more flexible approach towards energy dispatch involves identifying key metrics/parameters which characterise the optimal operation of the EES device and from which rules can be generated. This has the advantage of being easily implementable and applicable over a wider range of system variability.

To this end, we propose a flexible approach to energy dispatch for EES devices using a decision tree-based optimisation approach. Using output from a two-stage optimisation-based approach, we train a decision tree to obtain highly optimal and feasible dispatch results with no prior knowledge of future load profile and market data. The first stage of the optimisation approach solves an optimal dispatch optimisation model to obtain a set of distinct optimal solutions for training data, and the second stage is a feature extraction stage which finds the optimal price range where charge and discharge actions are executed in order to minimise the total cost. These results are then mapped to a decision tree for obtaining real-time dispatch actions for the EES device. The approach is applied to a microgrid with an EES asset having access to the UK day-ahead energy market. Given a load demand which it must satisfy, this approach proffers the optimal dispatch actions for the EES device in order to ensure minimal total costs.

References

Fisher, M., Apt, J., & Whitacre, J. F. (2019). Can flow batteries scale in the behind-the-meter commercial and industrial market? A techno-economic comparison of storage technologies in California. Journal of Power Sources, 420, 1–8. https://doi.org/10.1016/j.jpowsour.2019.02.051

García Vera, Y. E., Dufo-López, R., & Bernal-Agustín, J. L. (2019). Energy Management in Microgrids with Renewable Energy Sources: A Literature Review. Applied Sciences, 9(18), 3854. https://doi.org/10.3390/app9183854

Ma, T., Shen, L., & Li, M. (2018). Electrical Energy Storage for Buildings. In Handbook of Energy Systems in Green Buildings (pp. 1079–1107). Springer Berlin Heidelberg. https://doi.org/10.1007/978-3-662-49120-1_44

Roberts, D., & Brown, S. F. (2020). Identifying calendar-correlated day-ahead price profile clusters for enhanced energy storage scheduling. Energy Reports, 6, 35–42. https://doi.org/10.1016/j.egyr.2020.02.025