⚡ electrical engineering

Activation-Grade-Scheduled Predictive Control for Battery–Supercapacitor Electric-Vehicle Energy Management

This paper proposes an activation-grade-scheduled predictive controller with a candidate-screening extension (M6) for battery–supercapacitor electric vehicles that successfully eliminates supercapacitor reserve depletion violations but introduces a trade-off by increasing battery current stress and DC-link voltage errors compared to conventional fixed-weight control.

Original authors: Vo Thanh Ha, Nguyen Van Hai, Vo Quang Vinh, Nguyen Duy Trung

Published 2026-10-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Vo Thanh Ha, Nguyen Van Hai, Vo Quang Vinh, Nguyen Duy Trung

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Modern electric vehicles rely on a single, massive battery to power everything from the motor to the air conditioning. While these batteries are excellent at storing large amounts of energy, they struggle when asked to handle sudden, violent bursts of power, such as rapid acceleration or hard braking. These sharp spikes in demand can overheat the battery, wear it out prematurely, and even cause the vehicle's electrical system to become unstable. To solve this, engineers often pair the main battery with a supercapacitor, a device that acts like a high-speed shock absorber for electricity. It can release or absorb energy almost instantly, smoothing out the rough edges of the drive. However, simply adding a second device is not enough; the vehicle needs a smart brain to decide exactly when to use the battery and when to call on the supercapacitor. If the brain makes the wrong choice, the supercapacitor might run out of energy too early, or the battery might still get damaged.

Researchers at the University of Transport and Communications and the Electric Power University in Vietnam have developed a new way to manage this partnership. They created a control system that acts like a dynamic traffic manager, constantly adjusting its priorities based on how the car is driving. Instead of using a fixed set of rules, this system watches the road demand and the state of both energy stores. When the car faces a sudden surge in power needs, the system automatically shifts its focus to protect the battery, allowing the supercapacitor to handle the heavy lifting. When the driving is smooth, it relaxes its guard to ensure the supercapacitor saves enough energy for the next challenge. The team tested this approach in a detailed computer simulation using standard driving cycles that mimic real-world city and highway traffic. They found that by constantly re-tuning its priorities, the new system reduced the average electrical stress on the battery by about 2.35% to 3.40% compared to older, static methods. This reduction in stress is significant because it suggests the battery could last longer and the vehicle could run more efficiently without needing larger, heavier components.

The researchers also explored a more aggressive version of their system designed to strictly prevent the supercapacitor from ever dipping below a safe minimum level of energy. In simulations where the supercapacitor started with a low charge, this stricter version successfully kept the device above its safety threshold for the entire drive, whereas the standard version allowed it to dip below the limit hundreds of times. However, this safety came with a cost. By forcing the system to keep the supercapacitor in reserve, the battery had to work slightly harder, resulting in a small increase in electrical stress and a slightly less stable voltage for the vehicle's electronics. The study highlights a fundamental trade-off in engineering: you can prioritize the safety of one component or the smoothness of the whole system, but achieving both perfectly at the same time is difficult. The new method does not solve the problem entirely, but it provides a flexible tool that allows engineers to choose the right balance for their specific needs.

The work presented in this paper is a simulation study, meaning the results were generated by a computer model rather than a physical car on a road. The researchers used mathematical models to represent the battery, the supercapacitor, and the vehicle's motor, running thousands of calculations to see how different control strategies performed over time. While the results are promising, the authors are careful to note that these findings are not yet proven in the real world. The computer models are simplified versions of reality, and factors like temperature changes, sensor errors, or unexpected road conditions were not fully tested. The team acknowledges that their system relies on a "fallback" mechanism; if the computer cannot find a safe path that keeps the supercapacitor within its limits, it reverts to a less strict mode to keep the car moving. This means the system cannot guarantee that the supercapacitor will never run low in every possible situation.

Despite these limitations, the study offers a clear path forward for improving electric vehicle design. The core idea is that a control system should not be rigid. Just as a driver adjusts their speed and braking based on traffic and road conditions, the vehicle's energy manager should adapt its strategy in real-time. The researchers demonstrated that by making these adjustments automatically, they could reduce the wear and tear on the battery while maintaining the vehicle's performance. The study also emphasized the importance of fairness in energy usage; the system was designed to ensure that the supercapacitor did not get completely drained, leaving the battery to deal with the rest of the journey alone. By keeping a small reserve of energy in the supercapacitor, the system ensures that the vehicle is always ready for the next sudden stop or start.

The practical implications of this work extend beyond just saving a few percent of battery life. As electric vehicles become more common, the demand on their electrical systems will only increase. A system that can intelligently share the load between different energy sources could lead to vehicles that are lighter, cheaper to build, and more reliable over their lifetime. The researchers plan to take their next steps by testing these ideas on actual hardware and with more complex models that account for the unpredictable nature of real-world driving. Until then, their work serves as a strong proof of concept, showing that the key to better electric vehicles may lie not in bigger batteries, but in smarter ways of managing the energy they already have. The study concludes that while no single solution is perfect, the ability to dynamically schedule energy use offers a powerful new tool for engineers striving to make electric transportation more efficient and durable.

Technical Summary: Activation-Grade-Scheduled Predictive Control for Battery–Supercapacitor Electric-Vehicle Energy Management

Problem Statement
Electric vehicles (EVs) utilizing lithium-ion batteries face significant electrical stress due to high-power traction transients, regenerative braking, and rapid load changes. These conditions accelerate battery degradation through ohmic heating and thermal loading. While Battery–Supercapacitor Hybrid Energy Storage Systems (HESS) offer a solution by separating energy (battery) and power (supercapacitor) requirements, existing Energy Management Systems (EMS) often struggle to balance competing objectives. Specifically, conventional Model Predictive Control (MPC) relies on offline-fixed weighting coefficients that cannot adapt to continuously changing traction demands and storage states. Furthermore, existing strategies often fail to explicitly report terminal supercapacitor (SC) energy fairness alongside battery stress metrics, or they lack robustness testing against parameter uncertainty and sensor noise.

Methodology
The authors propose a hierarchical Activation-Grade-Scheduled Predictive Controller (M5) designed for a fully active battery–supercapacitor HESS topology. The system architecture consists of two layers:

Supervisory Layer (Activation-Grade Scheduling): Instead of a fuzzy logic system that directly outputs power-split commands, this layer acts as a lightweight, deterministic supervisor that dynamically adjusts the internal objective weights of the MPC. It maps operating conditions—traction power demand ( $P_{dem}$ ), demand rate of change ( $\Delta P_{dem}$ ), battery/SC State of Charge (SOC), and DC-link voltage error—into "activation grades." These grades scale four predictive cost weights:
- $w_I$ : Battery current magnitude.
- $w_{\Delta I}$ : Battery current variation (slew rate).
- $w_{Vdc}$ : DC-link voltage deviation.
- $w_{SC}$ : SC energy deviation from a reference.
  High traction severity or rapid demand changes increase penalties on battery current and variation, while DC-link errors increase voltage regulation penalties. A soft terminal-energy term discourages systematic SC depletion without enforcing strict equality at every horizon step.
Predictive Layer (Finite-Grid Optimization): The controller employs a deterministic, two-stage "coarse-to-fine" scalar search to select a single battery power command over a 10-step prediction horizon ( $N_p=10$ , $\Delta t = 0.1$ s).
- Coarse Search: Evaluates 61 evenly spaced candidates.
- Fine Search: Evaluates 31 candidates in a local interval around the best coarse candidate.
  The objective function includes running costs for current, variation, voltage, and energy drift, plus a terminal fairness term and soft penalties for constraint violations (SOC, current limits).
Extension (M6): A candidate-screening variant (M6) was developed to test near-boundary operating reserves. M6 excludes any candidate trajectory where the predicted SC SOC falls outside a specific internal interval $[0.082, 0.948]$ (slightly inside the physical limits of $[0.08, 0.95]$ ). If no candidates pass the screen, the system falls back to the unscreened minimum-cost candidate.

Key Contributions
The paper makes four primary contributions:

Online Multi-Weight Scheduling: It introduces an activation-grade supervisor that dynamically adjusts MPC cost weights based on real-time operating conditions, moving beyond static, offline-tuned coefficients.
Energy-Aware Comparison: It establishes a comparison framework that evaluates battery electrical stress (RMS current, peak current, stress proxy) simultaneously with terminal SC energy fairness, ensuring that battery protection does not come at the cost of excessive SC reserve depletion.
Near-Boundary Operating Reserve Experiment: It evaluates the M6 screening extension, demonstrating the trade-off between maintaining SC operating reserves and electrical performance.
Traceable Numerical Evaluation: The study provides a rigorous evaluation protocol using matched WLTP and UDDS driving cycles, specific input file checksums (SHA-256), and detailed reporting of computational solve times and parameter mismatch scenarios (e.g., $+20\%$ battery resistance, $-20\%$ SC capacitance, and sensor noise).

Results
The study presents results from simulations using a reduced-order plant model with nominal and perturbed parameters:

M5 vs. M3 (Fixed-Weight MPC): Under nominal conditions ( $V_{sc}(0) = 350$ V), the scheduled M5 controller reduced battery current RMS by 2.35% (WLTP) and 3.40% (UDDS) compared to the fixed-weight M3 baseline. DC-link voltage regulation and terminal SC energy changes remained comparable between the two.
Robustness: In sensitivity tests involving plant parameter mismatches and sensor noise, M5 maintained lower battery current RMS than M3, though both controllers exhibited increased current and voltage errors under noisy conditions.
M5 vs. M6 (Screening Extension): In a specific experiment starting with a lower initial SC voltage ( $V_{sc}(0) = 267$ $V_{sc} (0) = 267$ V, $\approx 0.09$ $\approx 0.09$ SOC), the M6 screening successfully eliminated all post-step samples where the SC SOC dropped below the 0.08 operating threshold (reducing counts from 963 to 0 on WLTP and 49 to 0 on UDDS). However, this improvement came with a measurable electrical cost:
- Battery current RMS increased by 1.03% (WLTP) and 2.06% (UDDS).
- DC-link voltage RMSE increased significantly (e.g., from 0.334 V to 0.461 V on WLTP).
- Terminal SC energy changes remained similar between M5 and M6, indicating the screening did not alter the net energy throughput but shifted the burden of regulation to the battery and DC link.

Significance and Claims
The paper modestly claims that the proposed activation-grade scheduling offers a viable method for balancing battery stress reduction and DC-link regulation in HESS applications. The authors explicitly state that the results identify an operating-reserve versus electrical-performance trade-off rather than an across-the-board improvement.

Crucially, the paper does not claim that the M6 screening guarantees robust constraint satisfaction or recursive feasibility. The authors note that the screening margin is heuristic, the prediction model differs from the plant model, and the fallback mechanism prevents hard-constraint guarantees. Consequently, the observed zero crossings in the nominal simulations do not establish robustness against model uncertainty, repeated noise trials, or hardware-in-the-loop conditions. The study positions itself as a step toward adaptive, energy-aware predictive control, emphasizing the need for future validation using independent cycles, quantified uncertainty margins, and experimental hardware testing before making vehicle-level safety or performance claims.

More like this