Hybrid powertrains—whether combining an internal combustion engine, a hydrogen fuel cell, or multiple battery chemistries—face a fundamental operational challenge: how to split power demands between different energy sources in real time. Every driving decision requires a choice: draw from the primary fuel source, or utilize an energy buffer like a battery or supercapacitor.
If a strategy relies too heavily on the battery, the buffer drains completely, forcing the primary engine or fuel cell into an inefficient operating zone later (De Lorenzo et al. [1]). Conversely, if it underutilizes the buffer, potential energy from regenerative braking is wasted (Yang et al. [2]). As highlighted by Guo et al. [3], the goal of an Energy Management Strategy (EMS) is to dynamically navigate this trade-off to maximize efficiency, minimize emissions, and preserve component health.
The Equivalence Principle and Rule Optimization
To optimize power distribution, a controller must be able to compare apples to oranges—specifically, chemical fuel (or hydrogen) to stored electrical energy. The Equivalent Consumption Minimization Strategy (ECMS) solves this by introducing an "equivalence factor" that translates battery usage into a virtual fuel cost rate (Jung et al. [4]). By minimizing this combined instantaneous cost, the powertrain can make optimal local decisions.
However, tuning these control strategies across changing driving conditions is highly complex. Early approaches used genetic algorithms to fine-tune fuzzy logic rules, keeping internal combustion engines operating close to their optimal efficiency curves (Yang et al. [5]). For extended-range electric vehicles, Wu [6] demonstrated that bio-inspired optimization algorithms can precisely control the on-off thresholds of a range extender, significantly reducing engine operating time and emissions compared to rigid, threshold-based heuristics.
Predictive Optimization Over Horizons
Local optimization is inherently blind to the future. If a vehicle can anticipate an upcoming steep hill, a traffic jam, or a red light, it can pre-deplete or preserve its battery buffer to maximize efficiency. Model Predictive Control (MPC) achieves this by predicting driving conditions over a short future horizon and solving an optimization problem online (Styler et al. [7]).
The primary bottleneck for MPC in production vehicles is computational complexity. To deploy these systems in real-time, Li et al. [8] developed a dual-model predictive control framework. The system uses a Long Short-Term Memory (LSTM) network to recognize driving conditions online and intelligently switches between standard MPC and a faster, explicit MPC (eMPC) variant. Similarly, incorporating temporal classification alongside driving conditions allows controllers to allocate power more accurately without overwhelming onboard processors (Ye et al. [9]).
Learning Transferable Policies
Instead of explicitly predicting the future and solving optimization equations at every time step, Reinforcement Learning (RL) allows an agent to learn an optimal control policy through continuous interaction with a simulated environment (Maino et al. [10]). The agent receives rewards for reducing fuel consumption and penalties for violating battery constraints.
A major hurdle for learning-based EMS is generalization: an RL agent trained on highway driving often performs poorly in urban stop-and-go traffic. To address this, Li et al. [11] investigated noise-injection strategies to improve policy transferability. They discovered that injecting noise directly into the RL agent’s parameter space—rather than its action space—forces the network to learn more robust, underlying physical relationships, leading to faster and more stable adaptation when transferred to entirely new driving cycles.
Hierarchical Decoupling of Motion and Powertrain
A vehicle's energy consumption is fundamentally linked to its velocity profile. True global optimization requires simultaneously planning how the vehicle should move through traffic (eco-driving) and how the powertrain should fulfill that motion. However, solving both problems at once creates an exponential explosion in the control system's state space.
Modern architectures manage this complexity through hierarchical decoupling:
┌────────────────────────────────────────────────────────┐
│ High-Level Motion Planner │
│ Optimizes speed profile based on traffic & signals │
│ (e.g., Approximate Dynamic Programming / DRL) │
└───────────────────────────┬────────────────────────────┘
│
│ Target Velocity / Power Demand
▼
┌────────────────────────────────────────────────────────┐
│ Low-Level Powertrain Controller │
│ Optimizes instantaneous power split (ECMS / DP) │
└────────────────────────────────────────────────────────┘
Jung et al. [4] successfully implemented this by pairing Approximate Dynamic Programming for high-level eco-driving with an ECMS for low-level power split control. In connected urban environments with frequent traffic signals, Dong et al. [12] combined Deep Deterministic Policy Gradient (DDPG) networks for traffic-light-aware speed planning with Dynamic Programming for fuel cell power allocation. This separation allows the vehicle to glide through intersections smoothly while maintaining near-optimal internal energy efficiency.
Health-Aware and Multi-Objective Management
Maximizing immediate efficiency can sometimes accelerate the degradation of expensive powertrain components. For instance, aggressive battery cycling or frequent fuel cell start-stop cycles drastically reduce component lifespan. Modern EMS frameworks treat energy management as a multi-objective optimization problem that balances fuel economy against component health.
To protect battery longevity in range-extended vehicles, Liu et al. [13] introduced an adaptive strategy based on bargaining game theory. By modeling the auxiliary power unit and the battery as players in a cooperative game, the system dynamically shifts its optimization priorities based on the battery's state of charge and degradation state.
For complex hydrogen systems featuring multiple fuel cell stacks, Shi et al. [14] utilized Mixed-Integer Quadratic Programming (MIQP) to manage each stack independently. Instead of treating the fuel cell system as a single collective unit, the MIQP controller schedules individual stacks to prevent unnecessary start-stop cycles, distributing the load evenly to minimize both hydrogen consumption and long-term stack degradation.
Similarly, holistic system-level design is required when integrating diverse storage media—such as combining lead-acid batteries, lithium-ion batteries, and supercapacitors into a single hybrid energy storage system (Kim et al. [15]). By matching optimized control strategies with precise transmission parameter designs (Zhou et al. [16]), modern hybrid vehicles can mitigate individual component weaknesses and capitalize on their collective strengths.
References
[[1]] Modelling and Performance Analysis of an Autonomous Marine Vehicle Powered by a Fuel Cell Hybrid Powertrain
[[2]] Parameter Matching of Energy Regeneration System for Parallel Hydraulic Hybrid Loader
[[3]] The Energy Management Strategies for Fuel Cell Electric Vehicles: An Overview and Future Directions
[[4]] Integrated Approximate Dynamic Programming and Equivalent Consumption Minimization Strategy for Eco-Driving in a Connected and Automated Vehicle
[[5]] Research on Genetic-fuzzy Control Strategy for Parallel Hybrid Electric Vehicle
[[6]] On-Off Control of Range Extender in Extended-Range Electric Vehicle using Bird Swarm Intelligence
[[7]] Real-Time Predictive Optimization for Energy Management in a Hybrid Electric Vehicle
[[8]] Energy Management Strategy for Fuel Cell Vehicles Based on Online Driving Condition Recognition Using Dual-Model Predictive Control
[[9]] Research on Energy Management Strategies for Fuel Cell Hybrid Vehicles Based on Time Classification
[[10]] Project and Development of a Reinforcement Learning Based Control Algorithm for Hybrid Electric Vehicles
[[11]] A Comparative Study of Deep Reinforcement Learning-based Transferable Energy Management Strategies for Hybrid Electric Vehicles
[[12]] Hierarchical Optimization Based on Deep Reinforcement Learning for Connected Fuel Cell Hybrid Vehicles through Signalized Intersections
[[13]] Research on Energy Management Strategies Based on Bargaining Game for Range-Extended Electric Vehicle Considering Battery Life
[[14]] Health-aware energy management for multiple stack hydrogen fuel cell and battery hybrid systems
[[15]] A scalable and flexible hybrid energy storage system design and implementation
[[16]] Transmission Parameter Design and Characteristic Analysis of Three-Row Parallel Planetary Gear HMCVT
Key Foundational Work
Charge allocation for hybrid electrical energy storage systems (2011) — contributed 2 concept elements to this direction.
Research on Genetic-fuzzy Control Strategy for Parallel Hybrid Electric Vehicle (2010) — contributed 2 concept elements to this direction.