In most energy optimization models, a battery is treated like a simple bucket: you pour energy in, you take energy out, and the bucket stays the same. To make the math easy, researchers often assume that every unit of energy cycled through the battery causes a fixed, linear amount of "wear."
However, a lithium-ion battery is not a bucket; it is a complex, sensitive chemical reactor. Its health doesn't just decline based on how much work it does, but how it does that work. Factors like temperature, depth of discharge (DoD), and current state of charge (SoC) create a highly non-linear degradation profile. A growing body of research argues that ignoring these non-linearities leads to "unrealistic performance expectations" Dirks et al. and suboptimal financial decisions in energy markets.
The Linear Fallacy
The appeal of linear models is computational simplicity. When planning large-scale systems—like a city's electric bus fleet or a regional power grid—linear approximations allow for fast optimization using Mixed Integer Programming (MIP).
But as Arnob Ghosh and Vaneet Aggarwal point out, these models fail to capture the physical reality of battery aging. For instance, discharging a battery from 100% to 80% is significantly less stressful than discharging it from 20% to 0%, even though the energy moved is the same. By treating these two events as identical, models may inadvertently encourage behaviors that accelerate battery death, such as deep discharges or high-temperature operations.
The Hidden Costs of Simplification
When we simplify battery wear, we also simplify the economics of energy storage. In energy arbitrage—buying low and selling high—the profit margin is often razor-thin. If a model assumes a constant linear cost for degradation, it might suggest a trade that looks profitable on paper but actually costs more in battery health than it earns in revenue.
James Anderson and Yiqian Wu have shown that when price predictions are imperfect, these simplified wear models lead to "suboptimal cycling strategies." Similarly, Muhammad Tuoqeer Anwar and colleagues found that in frequency regulation markets, where batteries must respond to rapid grid fluctuations, linear approximations fail to account for the intense stress of varying discharge rates and temperature spikes.
Integrating Electrochemical Reality
To bridge the gap between mathematical models and chemical reality, researchers are moving toward high-fidelity electrochemical aging profiles. Instead of just counting kilowatt-hours, these models track physical phenomena like Solid Electrolyte Interface (SEI) layer growth—a primary driver of capacity loss.
Jongchan Baek and Soohee Han demonstrated that by using Deep Reinforcement Learning (specifically Soft Actor-Critic) paired with these high-fidelity models, a system can learn to charge a battery in a way that prioritizes longevity over speed, respecting the non-linear limits of the chemistry.
The Path Forward: Complexity vs. Scalability
The central challenge in this field is the trade-off between accuracy and scale. As Xinyi Zhao and Hongbin Sun note, researchers face a difficult choice: use accurate chemical models that are too slow for large-scale grid planning, or use fast linear models that might be wrong.
The emerging consensus, seen in work by Jiaju Qi and Dusit Niyato, involves hierarchical approaches. These systems decouple high-level fleet management from low-level power control, allowing the model to handle thousands of vehicles while still respecting the non-linear physical constraints of each individual battery. By moving beyond the "bucket" analogy, these researchers are building energy systems that are not just profitable in the short term, but sustainable for the long life of the hardware.