Aging-Aware Battery Control via Convex Optimization

Obidike Nnorom Jr, Giray Ogut, Stephen Boyd, Philip Levis · 2025

Read Aging-Aware Battery Control via Convex Optimization by Obidike Nnorom Jr, Giray Ogut, Stephen Boyd, Philip Levis (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-09030.txt.

Introduction

Obidike Nnorom, Giray Ogut, Stephen Boyd, and Philip Levis tackle battery control in this 2025 paper. Cycling a battery earns money through price arbitrage or smooths power demand. But every cycle also ages the battery. Heavy use thus trades short term gains against a shorter life.

The team folds a data based aging model into a convex optimization problem, a form that can be solved reliably. They run it inside model predictive control, which replans as new data arrives. A convex stand-in for the aging dynamics keeps the trade-off manageable in real time. Simulations then measure the balance between earnings and wear in both money making and load smoothing uses.

Catalog record

CREATORObidike Nnorom Jr, Giray Ogut, Stephen Boyd, Philip Levis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-09030

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Aging-Aware Battery Control via Convex Optimization Obidike Nnorom Jr, Giray Ogut, Stephen Boyd, Philip Levis Abstract We consider the task of controlling a battery while balancing two competing objectives that evolve over different time scales. The primary objective, such as generating revenue by exploiting time varying energy prices or smoothing out the load of a computation center, operates on the scale of hours or days. The long term objective is to maximize the lifetime of the battery, which operates on a time scale of months and years. These objectives conflict; roughly speaking, the primary objective improves with cycling the battery more, which ages the battery faster. Using an existing model for battery aging, we formulate the problem of controlling the battery under these competing objectives as a convex optimization problem. We demonstrate the tradeoff between the primary objective and battery lifetime through numerical simulations. 1 Introduction 1.1 Setting and tasks In the last decade, battery storage, whether retail or grid scale, has become increasingly common. The ability to store energy and discharge it at a later time allows for temporal decoupling of energy production and consumption. Providing this service can be lucrative, as the price of electricity can vary by orders of magnitude depending on the time of day and the season. In other basic applications, a battery can be used to smooth out the power produced by a renewable source, or consumed by a load such as a computation center. Like any physical system, batteries degrade over time. The more a battery is cycled, meaning charged and discharged, the more it ages, and the shorter its lifetime becomes. The trade-off between short term tasks such as arbitrage and renewable smoothing, and a long term task of maximizing battery lifetime, is the focus of this paper. Short term objectives. We consider two applications, as simple examples of short term objectives. 1. Energy arbitrage. The battery is charged when electricity is cheap and discharged when it is expensive. We evaluate the performance of the system by its revenue. 2. Load smoothing. The battery is used to smooth out the electric power demand ( i.e. , load) consumed by a computation center, so the net power varies more smoothly over time. We evaluate the performance of the system using the root mean square difference between the load and its previous value. These are just two simple illustrative examples of short term objectives; many others could be handled by the methods we describe in this paper. Long term objective. The battery has a finite lifetime, which is determined primarily by the number of cycles it undergoes. We want to maximize the lifetime of the battery, defined as the time the battery capacity drops below some fraction of its initial value such as 80% or 90%. Without considering battery aging, the two short term tasks described above involve aggressive cycling of the battery, leading to a shortened lifetime. We show how to operate the battery so as to achieve an optimal trade-off of the short term objective and the long term objective. 1.2 Related work Battery aging models. Battery aging is often split into two main effects: calendar aging, which happens when the battery is resting (no charge or discharge), and cycle aging, which happens during active charge–discharge cycles [ 1 ] . Both processes are sensitive to temperature (especially above 30 ∘ C), high current rates, and usage patterns like depth of discharge or state of charge (SoC) [ 2 ] . When it comes to modeling these effects, three main approaches appear in the literature. Electrochemical models [ 3 , 4 , 5 ] try to describe the internal reactions mathematically (for instance, using the Butler–Volmer equations [ 6 ] ) and can be quite detailed but hard to implement in practice. At the other extreme, empirical models fit observed data to capture aging trends, but these can fail outside their specific test conditions and often need large datasets [ 7 ] . Semi-empirical models [ 8 , 9 , 10 , 11 ] blend theoretical ideas with curve-fitting so that the most critical aging drivers (like temperature, SoC, or C-rate) are handled without the complexity of a full electrochemical model or the narrow scope of an entirely data-driven approach. Calendar aging mainly depends on how batteries are stored (in terms of temperature and SoC), with slow chemical reactions gradually eating away at capacity. Cycle aging, on the other hand, is tied to how often and how aggressively the battery is charged or discharged. In actual operation, both processes happen at the same time, so an aging model typically combines or overlays both [ 12 ] . Semi-empirical models are especially attractive in this setting because they balance realism and simplicity, letting us capture key aging behaviors in a way that can be used in practical control algorithms [ 13 , 14 ] . Optimization in batteries. Current optimization techniques used in battery longevity do not directly focus on working with aging models of the batteries. Liu et al. [ 15 ] focuses on optimizing charging behavior of both CC-CV charging and multi stage CC-CV charging. Nonlinear optimization techniques are used to determine the optimal current and voltage values to balance between aging, efficiency loss, and charge time. Chung et al. [ 16 ] focuses on reducing the calendar aging of a PEV battery through an optimal charging scheme. This paper argues that calendar aging is most important for PEV batteries, and proposes a nonlinear interior point method to determine an optimal overnight charging scheme for the PEV battery. Bashir et al. [ 17 ] tackles lifetime maximization of lead-acid batteries by formulating key aging characteristics of the battery as convex formulas. Bad recharge, the time since last full charge, and the lowest state of charge since last recharge are the three primary factors in their objective function. A multi-objective convex optimization problem is then solved to maximize the lifetime of these batteries. Model predictive control (MPC). Model predictive control goes by several other names, such as rolling-horizon planning, receding-horizon control, dynamic matrix control, and dynamic linear programming. Originally developed in the 1960s, MPC offers a powerful framework for managing constraints on states, inputs, and outputs. It has a long history and large literature, and is widely used. Some early work is [ 18 , 19 ] ; for more recent surveys see the papers [ 20 , 21 , 22 , 23 ] or books [ 24 , 25 , 26 , 27 , 28 ] . Papers describing applications of MPC in specific areas include HEVs [ 29 ] data center cooling [ 30 ] , building HVAC control [ 31 ] , wind power systems [ 32 ] , microgrids [ 33 ] , pandemic management [ 34 , 35 ] , dynamic hedging [ 36 ] , revenue management [ 37 , 38 ] , railway systems [ 39 ] , aerospace systems [ 40 ] , and agriculture [ 41 ] . With appropriate forecasting (which in many applications is typically simple) and choice of cost function, MPC can work well, even though it does not explicitly take into account uncertainty in the dynamics and cost, or

Read the complete text