Battery Energy Storage Systems (BESS)

• Battery Energy Storage Systems (BESS) are integrated electrochemical storage solutions that provide grid balancing, renewable integration, and ancillary services.
• They employ advanced control algorithms, scalable model predictive, and decentralized strategies to manage real-time power dispatch and optimize system performance.
• Lifecycle-aware optimization and market integration techniques enhance economic viability and reliability across diverse power network applications.

A Battery Energy Storage System (BESS) comprises electrochemical storage assets, power conversion systems, supervisory controls, and integration interfaces designed for time-shifting, grid balancing, and ancillary services within power networks of all scales. Modern BESS deployments span standalone grid-scale installations, behind-the-meter microgrid support, and embedded assets for frequency regulation, peak shaving, and renewable integration. This article addresses fundamental principles, mathematical formalism, practical architectures, operational paradigms, and key advancements in BESS, with emphasis on recent research in optimal sizing, management, market integration, and reliability.

1. BESS Architectures, Models, and Operational Principles

BESS topologies typically include large arrays of lithium-ion or flow batteries, cell/module-level DC/DC conversion, battery management systems (BMS), and grid-side inverters. At the cell level, open-circuit voltage $v_{\mathrm{oc}}(\mathrm{SOC})$ and internal resistance $R$ define the electrochemical and electrical dynamics (Cassano et al., 25 Mar 2024). BESS operation is governed by constraints:

• Terminal voltage and current bounds map to SOC-dependent power limits:

$$P_{\min}(\mathrm{SOC}) = \max\left\{P_v^{\min}(\mathrm{SOC}),\,P_i^{\min}(\mathrm{SOC})\right\}, \quad P_{\max}(\mathrm{SOC}) = \min\left\{P_v^{\max}(\mathrm{SOC}),\,P_i^{\max}(\mathrm{SOC})\right\},$$

where voltage and current subscripts denote Thevenin-equivalent limits (Cassano et al., 25 Mar 2024).

• State-of-charge (SOC) and thermal dynamics are modeled as:

$$\dot q_j(t) = -\frac{i_j(t)}{Q_j}, \quad C_{\mathrm{th},j}\dot T_j(t) = R_j(q_j(t))i_j^2(t) - \frac{T_j(t)-T_\mathrm{env}}{R_\mathrm{conv,j}},$$

for each cell $j$ (Farakhor et al., 4 Mar 2025).

System aggregation for large-scale BESS is achieved via cell clustering (k-means or hierarchical) on $(q_i, T_i, R_i)$, producing convexified dynamics at the cluster level that enable scalable model predictive control (MPC) (Farakhor et al., 2023).

BESS are dispatched with a variety of topologies (Hakuta et al., 16 Apr 2024):

• Centralized MPC with full-state awareness,
• Decentralized broadcast control: individual batteries compute responses based on global tracking-error signals and local SOC, obviating peer-to-peer communication, and providing plug-and-play scalability for heterogeneous battery fleets,
• Hybrid architectures leveraging cycle-based correction and cluster-wise allocation for peak shaving and efficiency enhancement (Gan et al., 14 Feb 2025).

2. Sizing, Economic Valuation, and Lifecycle-Aware Optimization

BESS sizing targets the trade-off between installation cost, operational revenue, and system reliability. Two main financial objectives appear:

• Levelized Cost of Energy (LCOE):

$$\mathrm{LCOE}(E_\mathrm{BESS},P_\mathrm{BESS}) = \frac{\sum_{n=0}^N \frac{\mathrm{CAPEX}_n + \mathrm{OPEX}_n}{(1+r)^n}} {\sum_{n=0}^N \frac{E_{\mathrm{gen},n}}{(1+r)^n}},$$

for off-grid/hybrid microgrid applications (Richard et al., 2020).

• Net Present Value (NPV):

$$\mathrm{NPV}(E_\mathrm{BESS},P_\mathrm{BESS}) = \sum_{n=0}^N \frac{\mathrm{Income}_n - (\mathrm{CAPEX}_n + \mathrm{OPEX}_n)}{(1+r)^n},$$

with constraints on SoC, power ramp rates, and application-specific constraints.

Simulation-based frameworks (e.g., SPIDER) perform multi-year, time-resolved studies, integrating:

• Degradation (first-order linear or empirical fade models),
• Tabulated efficiency maps $\eta(\mathrm{SOC}, I, T)$,
• Empirically derived battery aging and replacement strategies (Richard et al., 2020).

Lifecycle optimization explicitly incorporates cycle-related degradation using piecewise-linear representations of empirical life-vs-depth-of-discharge (DoD) relationships. In mixed-integer linear programming (MILP), battery wear $L_\mathrm{loss}(t)$ per interval is:

$$L_\mathrm{loss}(t)=|F(\mathrm{SOC}_t)-F(\mathrm{SOC}_{t-1})|,$$

with $F(\mathrm{SOC})$ being an optimally linearized primitive of the per-DoD life loss coefficient (Gui et al., 2019).

Key practitioner guidelines prioritize control strategy over modeling fidelity, with advanced predictive control, accurate load/renewable forecasts, and explicit battery aging models being dominant sizing factors (Richard et al., 2020).

3. Optimal Power Management and Real-Time Dispatch Techniques

Large-scale and real-time BESS operation introduces high-dimensional nonconvex optimization, directly intractable at the cell level for $n\gg100$ (Farakhor et al., 4 Mar 2025). Key advancements include:

• Clustered and Parameterized Power Allocation: The feasible domain is greatly shrunk by parametrizing per-cell or per-cluster power sharing ratios through a small set of coefficients $\theta_1, \theta_2, \theta_3$:

$$\mu_j(t) = \phi_{q,j}(t)\theta_1 + \phi_{T,j}(t)\theta_2 + \phi_{R,j}(t)\theta_3, \quad \sum_j \mu_j(t) \approx 1,$$

where the $\phi$ functions encode SoC, temperature, and resistance preferences (Farakhor et al., 4 Mar 2025, Farakhor et al., 13 Jul 2025).

• Bayesian Inference via Ensemble Kalman Inversion (EnKI): The NMPC problem is recast as a low-dimensional parameter estimation via MAP estimation, solved over a rolling time window by iteratively updating an ensemble of $\theta$ samples based on virtual loss-and-barrier-augmented “observations” $y[t] = J[t] + \psi$ (Farakhor et al., 4 Mar 2025, Farakhor et al., 13 Jul 2025). This achieves real-time feasibility with $\sim$90–98% reduction in computational time compared to direct nonlinear programming, while enforcing balancing and safety constraints.
• Hybrid/Physics-Informed Deep Learning: In multi-phase distribution networks, heterogeneous graph neural nets (GNNs) incorporating detailed topology and physics-aware loss (SoC, C-rate, voltage window penalties) enable real-time dispatch predictions with perfect constraint compliance and $10^5$–$10^6\times$ violation reduction (Ma et al., 10 Dec 2025).
• Cycle-Based and Cluster-Level Heuristics: Peak-shaving frameworks that iteratively correct charge/discharge reference within each operating cycle, combined with instantaneous cluster-level allocation (e.g., via improved particle swarm optimization), maximize capacity utilization rate (CUR) and operational lifetime (Gan et al., 14 Feb 2025).
• Decentralized Control: Direct local gain adaptation as a function of each module’s SoC equalizes utilization and provides failover resilience; only broadcast grid-level tracking error is needed (Hakuta et al., 16 Apr 2024).
• Dynamic Power Constraints (DPCs): SOC-dependent power constraints derived from first principles (Thevenin models) ensure feasible scheduling under high-power, near-capacity operation, reducing voltage/current violations by up to 93% (Cassano et al., 25 Mar 2024).

4. Market Integration, Revenue Models, and Cross-Market Operations

BESS economic viability is strongly shaped by market structure, asset location, and service stacking.

• Energy Arbitrage and Ancillary Services: Mathematical models track day-ahead/intraday prices, unit throughput, and marginal wear cost $W_a$ (Hu et al., 2021). In major European markets, BESS is generally not profitable for pure energy arbitrage at present $W_a\approx100\,\mathrm{€/MWh}$; only volatile/island markets (GB, IE, Sicily) are exceptions. By contrast, frequency containment reserve (FCR) services often yield nearly 100% profitable utilization rate for $W_a\lesssim50\,\mathrm{€/MWh}$ in key Nordic/CWE states.
• Market Participation Algorithms: Mixed-integer bi-level programming captures the strategic bidding of a price-making BESS in energy, reserve, and pay-as-performance regulation markets, with explicit AGC trajectory modeling at sub-hourly intervals. Key constraints include:

$$0\leq P_\mathrm{BS}, P_\mathrm{BD} \leq P^\mathrm{rate},\quad \mathrm{SOC}_t = \mathrm{SOC}_{t-1} + P^{BD}_t \Delta t - P^{BS}_t \Delta t$$

and service stacking restrictions (Khalilisenobari et al., 2020). Stacked market participation outperforms single-service BESS in both revenue and asset longevity.

• Cross-Market Arbitrage: Integrated MILP-based scheduling across day-ahead, intraday auction, and continuous markets offers significant uplift from price volatility. Sensitivity analyses indicate that forecast quality is crucial, but strategies are robust to moderate prediction errors (Sandbergen, 12 Sep 2025).
• Node Siting and Mobility: High-granularity revenue analytics reveal concentration of profitable nodes in transmission networks; transportable BESS units can realize up to 18% greater revenue via seasonal/monthly relocation despite modest transfer costs (Zhao et al., 2020). Siting mixed-integer programs leveraging LMP volatility clustering optimize placement for risk and revenue diversification.

5. Reliability, SOC Uncertainty, and Lifetime Management

Reliable BESS operation requires advanced SOC estimation and explicit modeling of aging and uncertainty:

• SOC Estimation Uncertainty: Closed-loop estimation using combined Coulomb counting and rest-period voltage recalibration quantifies real-time SOC error, propagating through Kalman-like formulas (Martin et al., 2022):

$$u_{k+1}^2 = (1-\delta_k)^2 u_k^2 + \left(\frac{\eta \Delta t_k}{Q}\right)^2(\alpha + \beta i_k^2) + \delta_k^2 \left(\lambda_1 \frac{d\hat{SOC}_k}{dOCV}+\lambda_2 \frac{R_1 C_1}{t_R}\right)$$

Short rest periods sharply reduce uncertainty, directly impacting revenue and service deliverability in frequency regulation contexts. Overly conservative dispatch (to cover SOC margin) directly reduces marketable capacity.

• Lifecycle- and Degradation-Aware Scheduling: Integrated power tracking and wear cost optimization in MILP and dynamic programming explicitly trade scheduled tracking penalty against battery life loss to yield economically optimal operation (Gui et al., 2019, Zhuo, 2019).
• Cycle Counting for Lifetime Extension: Strategies that maximize CUR and minimize equivalent full-cycle counts (via dynamic reference correction and cluster-level load allocation) yield longer BESS lifespans for a given service commitment (Gan et al., 14 Feb 2025).

6. Future Directions, Challenges, and Scalability

Large-scale BESS management faces challenges of scalability, heterogeneity, and multi-market co-optimization:

• Scalable MPC and Inference: Ensemble Kalman inversion and clustering approaches demonstrate O(100–1000×) computational reductions versus classical high-dimensional NMPC, with robust balancing and near-optimal loss minimization confirmed in hardware (Farakhor et al., 4 Mar 2025, Farakhor et al., 2023).
• Physics-Aware ML and Hybrid Methods: Heterogeneous GNNs and physics-informed losses enforce critical operational constraints in deep learning-based real-time dispatch, bridging the gap between engineering models and generalizable AI (Ma et al., 10 Dec 2025).
• Decentralized Plug-and-Play Control: Broadcast-only decentralized frameworks enable resilience to partial failure and communication bottlenecks, essential for ultralarge heterogeneous fleets (Hakuta et al., 16 Apr 2024).
• Retrofit and Adaptation: Linearization of dynamic power constraints and modular policy representation supports practical retrofitting of both legacy and future schedulers for power-intensive, volatility-rich environments (Cassano et al., 25 Mar 2024).
• Economic and Policy Factors: Falling battery costs, market harmonization, dissemination of high-fidelity asset utilization data, and multi-service market design are priority enablers for further BESS deployment and research advancement (Hu et al., 2021, Zhao et al., 2020).

---

Battery Energy Storage Systems currently constitute a foundational, mathematically sophisticated solution for grid flexibility, renewable integration, and advanced market participation. Advances in scalable optimal power management, heterogeneous fleet control, lifecycle-aware planning, and cross-market operation continue to define the state of the art for academic and industrial research (Farakhor et al., 4 Mar 2025, Farakhor et al., 2023, Richard et al., 2020, Hu et al., 2021, Gan et al., 14 Feb 2025, Cassano et al., 25 Mar 2024, Sandbergen, 12 Sep 2025, Ma et al., 10 Dec 2025).