Energy-Aware ACC for EVs with Regenerative Braking

The problem

Conventional ACC reacts to every change in the lead vehicle's speed with near-immediate acceleration or friction braking, wasting energy. An EV can recover energy by regenerating, but only within motor and battery limits; braking beyond those limits uses friction brakes and wastes energy as heat. Anticipating the lead vehicle and allowing the gap to vary within safe bounds can shift braking into the regenerative range. How much energy this saves without hurting safety or comfort is a practical research question.

Objectives

• Build an EV longitudinal model with a motor efficiency map, regenerative limits and battery.
• Design a conventional ACC baseline with a constant time-gap policy.
• Design an MPC-based ACC with a cost that includes electrical energy use and penalises friction braking.
• Constrain safety gap, speed, acceleration and jerk in the MPC formulation.
• Compare energy use, regenerated energy, gap and comfort on realistic traffic scenarios.

What you need

Key concepts

Regenerative braking limit

The motor can only generate up to a torque that falls with speed, and the battery can only accept limited charge power. Braking beyond these limits must use friction brakes.

Brake blending

Splitting a deceleration demand between regenerative and friction braking. An energy-aware controller tries to keep demands inside the regenerative envelope.

Model predictive control

At each step, MPC solves an optimisation over a short horizon using a model and predicted lead-vehicle behaviour, applies the first input and repeats. It handles constraints directly.

Spacing envelope

Instead of tracking one exact gap, the controller keeps the gap between a safe minimum and a comfortable maximum, giving freedom to smooth speed changes.

Step-by-step guide

1. 1

   Define the problem and metrics

   Write the research question: how much energy can an energy-aware ACC save over a conventional ACC on Indian urban and highway traffic, while keeping safety and comfort within limits? Define metrics: net battery energy per kilometre, regenerated energy, friction braking energy, minimum time gap, time below the safe gap, root-mean-square jerk and maximum deceleration. Define the safe and comfortable gap bounds as constraints with justification from the literature.

   Checkpoint: A metric and constraint table with the research question stated precisely.

2. 2

   Build the EV powertrain model

   Model the vehicle's longitudinal dynamics with mass, rolling resistance, drag and grade. Add a motor with a torque-speed envelope and an efficiency map over torque and speed for motoring and generating, a single-speed gearbox and a battery with internal resistance and charge power limits. Implement brake blending that uses regeneration first up to the available limit and friction brakes for the remainder. Track energy flows in each path.

   Checkpoint: An energy balance that closes within a small error on a test drive cycle.

3. 3

   Generate lead-vehicle scenarios

   Build lead-vehicle speed profiles from Indian drive cycles such as MIDC segments, plus synthetic urban stop-and-go and highway following with gentle and firm braking. Add grade segments to some profiles. Include cut-in events. Use fixed seeds and keep a separate set of profiles for tuning and testing, so the reported energy savings are not tuned to the test data. Record the mean speed, stop count and acceleration statistics of each profile set.

   Checkpoint: Separate tuning and test sets of profiles, plotted and documented.

4. 4

   Build the conventional ACC baseline

   Implement a constant time-gap ACC with a well-tuned feedback controller and the same comfort limits as the MPC will have. Tune it on the tuning set for good gap tracking and comfort. This is the baseline the energy-aware ACC must beat, so make it competent rather than a straw man; a poorly tuned baseline inflates savings and weakens the thesis.

   Checkpoint: Baseline gap tracking and comfort metrics on the tuning set within requirement limits.

5. 5

   Formulate the MPC

   Use a prediction model with states for gap, relative speed and ego speed, and ego acceleration demand as the input, with a first-order actuator lag. Constrain gap above the safe minimum, acceleration and jerk within comfort limits, and speed within the set speed. In the cost, penalise deviation from a reference gap lightly, penalise jerk, and penalise deceleration beyond the speed-dependent regenerative limit, which approximates friction braking loss.

   Checkpoint: A written MPC formulation with every term, weight and constraint stated.

6. 6

   Implement in MPC Designer

   Create the plant model, set the sample time (for example 0.1 s) and the prediction and control horizons, and enter constraints in the MPC Designer app. Because the regenerative limit depends on speed, use adaptive MPC or update the deceleration bound at each step from the current speed. Export the controller to the MPC Controller block and connect it to the EV model and mode logic.

   Checkpoint: The MPC runs in closed loop on a simple braking scenario without constraint violations.

7. 7

   Handle the lead-vehicle prediction

   MPC needs a prediction of the lead vehicle's speed over the horizon. Start with a constant-speed assumption, then try constant acceleration with decay, and optionally a preview from a V2V signal for the extension. Compare their effect on energy and safety on the tuning set. Make clear in the report which prediction assumption gives each result, since it strongly affects savings.

   Checkpoint: Energy and minimum gap results for each prediction assumption on the tuning set.

8. 8

   Tune the weights

   Sweep the weights on gap deviation, jerk and friction-braking penalty, and plot energy saving against comfort and gap variation. Choose a weight set that meets all constraints and offers a clear saving. Check that the solver stays within the sample time on your computer and note the number of iterations, since real ECUs have limited computation and an MPC that is too slow is not usable.

   Checkpoint: A Pareto-style plot of energy against comfort with the chosen weights marked and solver times recorded.

9. 9

   Evaluate on the test set

   Run the baseline and energy-aware ACC on every test profile and compute all metrics. Present results per profile type, because savings in stop-and-go traffic usually differ from highway following. Include cut-in and hard-braking cases to make sure the MPC behaves safely when anticipation is impossible and full deceleration within ACC limits is needed, raising a takeover warning when required. Report means and spreads across seeds for each profile type.

   Checkpoint: A test-set results table per profile type with energy, safety and comfort metrics for both controllers.

10. 10

    Run sensitivity studies

    Vary vehicle mass (empty and fully loaded), battery state of charge (which limits regeneration when the battery is nearly full), motor efficiency map and prediction horizon length. Report how the energy saving changes. Show what happens when regeneration is limited by a full battery, since the MPC should then behave like a conventional ACC without safety loss. Keep the weights fixed across these runs so changes come only from the varied factor.

    Checkpoint: Sensitivity plots of energy saving against mass, state of charge and horizon length.

11. 11

    Write up the research

    Present the EV model, baseline, MPC formulation, tuning, test results and sensitivities. Show where savings come from by comparing friction braking energy and motor operating points, rather than reporting only total energy. Discuss limits such as model accuracy and lead-vehicle prediction. Release models and profiles so results can be reproduced. Include a short section on what would be needed to run the controller on a production ECU, such as solver choice, fixed-point or single-precision arithmetic and timing budgets. Add an appendix listing every model parameter with its source.

    Checkpoint: A thesis chapter with attributable energy savings, safety evidence and reproducible models.

Testing & validation

• Energy balance of the EV model closes on a test cycle.
• Net energy per kilometre, regenerated energy and friction braking energy for both controllers.
• Minimum time gap and time below the safe gap.
• RMS jerk and maximum deceleration.
• Sensitivity to mass, state of charge, efficiency map and horizon length.

Expected results

An energy-aware MPC ACC for an EV with a measured, attributable energy saving over a well-tuned conventional ACC on Indian traffic profiles, with safety and comfort kept within stated limits.

Take it further

• Use V2V preview of the lead vehicle's planned speed in the MPC.
• Add road grade preview from a map to plan regeneration on descents.
• Generate code for the MPC and measure execution time on an embedded target.
• Compare with a learned controller trained by reinforcement learning.

Viva and review questions

Learn it at DIYguru

These DIYguru courses and programmes teach what this project needs.