Abstract
Deregulation in retail energy markets offers consumers the potential for significant savings through plan switching. However, the complexity of tariffs, uncertainty in future demand, and switching costs often lead to consumer inertia. This paper introduces competitive online algorithms designed to automate energy plan selection. By modeling the problem as a Metrical Task System (MTS) with temporally dependent costs, the authors propose deterministic and randomized algorithms that achieve optimal competitive ratios of 3 and 2, respectively. Empirical evaluations using real-world demand data and energy plans demonstrate that these algorithms significantly outperform naive strategies, maintaining costs within 9-15% of the offline optimal.
Introduction
In many deregulated energy markets, residential consumers have the "retail choice" to switch between various electricity providers. Despite the potential for lower rates, participation remains low. This "paradox of choice" is driven by three primary factors: complex tariff structures (e.g., fixed vs. variable rates, tiered pricing), the inherent uncertainty of future electricity demand, and the financial burden of switching costs, such as contract cancellation fees.
To address these hurdles, this research frames the Energy Plan Selection Problem (EPSP) as an online decision-making task. The goal is to determine whether to "stay or switch" energy plans in real-time without requiring accurate forecasts of future demand or market prices.
Algorithmic Framework
The authors formulate the EPSP as a Metrical Task System (MTS), where states represent different energy plans and costs include both electricity usage charges and switching fees. A key innovation is the handling of "temporally dependent" switching costs, such as cancellation fees that decrease linearly over a contract's duration.
Two primary algorithms are proposed:
- gCHASEs (Deterministic): A generalized "conservative" algorithm that switches plans only when the accumulated cost of remaining in the current plan exceeds the cost of switching plus the cost of the new plan. It is proven to be 3-competitive, the theoretical lower bound for deterministic algorithms in this setting.
- gCHASEr_s (Randomized): This version switches plans with increasing probability as potential savings grow. By smoothing out the impact of timing, it achieves a competitive ratio of 2.
As shown in Figure 3, the algorithms respond to the cumulative cost difference $\Delta(t)$ between plans. While the deterministic gCHASEs waits for a specific threshold, the randomized gCHASEr_s provides a more fluid transition, effectively hedging against future uncertainty.
Empirical Evaluation
The algorithms were validated using the 'EPLUS TMY3 residential' dataset from the US Department of Energy, which includes 936 unique annual consumption profiles. These were tested against real-world tariff data from New York and Australia.
Key Results
The empirical performance of the algorithms far exceeded their worst-case theoretical bounds. In simulations with New York energy plans:
- The deterministic algorithm (gCHASEs) maintained an average cost only 14.6% higher than the omniscient offline optimal strategy.
- The randomized algorithm (gCHASEr_s) further improved performance, with most households experiencing costs only 9% higher than the theoretical minimum.
Figure 5 illustrates the "Cost Ratio" (the ratio of the algorithm's cost to the offline optimal cost) across low, base, and high demand categories. Both algorithms consistently outperform naive "Greedy" strategies, which often incur excessive cancellation fees by switching too frequently.
Conclusion and Future Directions
This work provides a robust mathematical foundation for automated energy plan switching. By achieving optimal competitive ratios, these algorithms offer guaranteed performance without the need for complex predictive models. Future research will explore scaling these methods to manage dozens of concurrent plans and integrating them with home energy management systems that control local assets like solar PVs and batteries. Additionally, the impact of large-scale adoption on market price dynamics remains an open question.
Summary updated