⚡ electrical engineering

ITI-MPC: Intelligent Transportation-Informed Predictive Energy Management for Ammonia-Hydrogen Hybrid Electric Vehicles

This paper proposes an Intelligent Transportation-Informed Model Predictive Control (ITI-MPC) framework that integrates short-horizon demand prediction to optimize energy management in ammonia-hydrogen hybrid electric vehicles, achieving a peak system efficiency of 42.80% while maintaining stable battery state-of-charge.

Original authors: Chenggang Zhang, Zhengxian Chen

Published 2026-09-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chenggang Zhang, Zhengxian Chen

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

The road to a cleaner future for heavy transport is paved with difficult choices. Large trucks and buses need immense amounts of energy to move, yet the batteries that power small cars are often too heavy and slow to charge for these massive jobs. Engineers have looked to hydrogen, a fuel that offers great energy and quick refueling, but storing pure hydrogen requires heavy, high-pressure tanks that are difficult to manage. A promising alternative has emerged in the form of ammonia, a chemical that is easy to store and transport because it is stable and liquid at moderate pressures. However, ammonia cannot burn directly in most engines; it must first be broken down into hydrogen gas through a process called cracking. This creates a complex puzzle for vehicle designers: the system must manage the flow of liquid ammonia, the rate at which it turns into gas, the storage of that gas, and the electricity needed to run the vehicle, all while ensuring the truck never runs out of fuel before reaching its destination.

Researchers at Tsinghua University have developed a new way to solve this puzzle, creating a smart control system specifically for vehicles that run on a mix of ammonia and hydrogen. In their study, they describe a vehicle equipped with three main power sources: an engine that burns a mixture of ammonia and hydrogen to generate electricity, a fuel cell that uses hydrogen to create power, and a battery that stores energy for quick bursts of speed or to capture energy during braking. The challenge lies in deciding how much power each source should provide at any given moment. If the system is too cautious, it might save too much hydrogen for later, leaving the tank full but the engine inefficient. If it is too aggressive, it might use up the hydrogen too quickly, only to find the ammonia cracker cannot produce more gas fast enough to keep up with the engine's needs.

To address this, the team created a strategy called ITI-MPC, which stands for Intelligent Transportation-Informed Predictive Energy Management. Unlike older systems that simply react to the driver's current demand for power, this new system looks ahead. It uses a computer model to predict what the truck's power needs will be over the next few seconds, based on the vehicle's speed and acceleration. By knowing what is coming, the system can make smarter decisions now. For instance, if the prediction shows a steep hill is approaching, the system can start generating extra hydrogen a moment earlier, ensuring the tank is ready when the heavy load arrives. This allows the vehicle to balance the slow, steady production of hydrogen from the ammonia cracker with the immediate needs of the road, avoiding the waste that comes from either over-producing or under-producing fuel.

The researchers tested this new approach using real-world driving data collected from a prototype ammonia-hydrogen truck. They compared their smart system against three other methods: a simple rule-based system that follows fixed instructions, a method that tries to minimize fuel use at every single instant without looking ahead, and a standard predictive system that does not use traffic or road information. The results showed that the new ITI-MPC strategy was the most efficient. It managed to reduce the total amount of ammonia consumed by the vehicle, achieving a system efficiency of 42.80 percent, which was higher than any of the other methods tested. Crucially, the truck did not run out of fuel or leave the battery drained; the system maintained a healthy level of charge in the battery and ensured that the hydrogen tank was used effectively, leaving only a tiny amount of fuel unused at the end of the trip.

The key to this success was not just in having a better engine or a bigger battery, but in how the computer managed the flow of energy over time. The study found that without looking ahead, older systems tended to be overly conservative, hoarding hydrogen in the tank just in case it was needed later, which meant the engine had to work harder and less efficiently to make up for it. The new system, by anticipating the road ahead, could use the hydrogen more freely and confidently, knowing exactly when more would be available from the ammonia cracker. This coordination allowed the vehicle to run closer to its optimal performance limits without risking a fuel shortage. The researchers noted that while the improvement in fuel savings might seem small in percentage terms, it represents a significant gain in efficiency for a system that is already tightly constrained by physics and safety limits.

This work demonstrates that for complex hybrid vehicles, the intelligence of the control software is just as important as the hardware itself. By integrating short-term predictions of the road ahead into the decision-making process, the system can navigate the tricky balance between limited storage and slow fuel production. The study confirms that looking forward, even just a few seconds, allows the vehicle to make better choices today, leading to a more sustainable and efficient way to move heavy loads. As the transportation industry seeks to decarbonize, such intelligent management of multi-source energy systems will likely become a standard requirement for the heavy-duty vehicles of the future.

Technical Summary: ITI-MPC for Ammonia-Hydrogen Hybrid Electric Vehicles

Problem Statement
Decarbonizing heavy-duty transportation requires powertrains that balance high energy density, rapid refueling, and operational flexibility. Ammonia-hydrogen hybrid systems offer a promising pathway, utilizing liquid ammonia for storage and transport, which is then cracked into hydrogen for use in fuel cells and internal combustion engines. However, energy management in these systems presents a complex inter-temporal coordination challenge distinct from conventional hybrid electric vehicles.

Unlike standard hybrids, ammonia-hydrogen systems are constrained by:

Rate-Limited Cracking: Hydrogen generation is physically limited by the ammonia cracking reaction rate and thermal conditions.
Finite Storage: Hydrogen storage capacity is bounded by safety and volume constraints.
Coupled Dynamics: The availability of hydrogen is jointly determined by cracking kinetics, storage evolution, fuel cell consumption, and engine demand.

Conventional rule-based strategies and reactive Model Predictive Control (MPC) rely solely on instantaneous demand. This "finite-horizon myopia" often leads to suboptimal resource scheduling, such as hydrogen over-generation, forced battery charging, or terminal hydrogen insufficiency, thereby reducing overall system efficiency and feasibility.

Methodology: The ITI-MPC Framework
To address these challenges, the authors propose an Intelligent Transportation-Informed Predictive Energy Management (ITI-MPC) framework. This strategy integrates short-horizon demand prediction directly into a constrained receding-horizon optimization to enable anticipatory multi-energy management.

The framework consists of two tightly coupled components:

Short-Term Power Demand Prediction:
- An Attention-LSTM (Long Short-Term Memory) network predicts traction power demand over a finite horizon (aligned with the MPC window).
- The model uses a sliding observation window of vehicle features (speed, acceleration, historical power) to generate step-specific forecasts.
- Independent models are trained for different prediction horizons (1–5 seconds) to maximize accuracy, minimizing Huber loss.
Prediction-Embedded MPC Optimization:
- State and Control: The system state includes Battery State-of-Charge (SOC) and hydrogen storage mass. Control inputs are engine-generator power and fuel cell power.
- Constraints: The optimizer enforces operational limits including SOC bounds, hydrogen tank capacity, actuator ramp rates, fuel cell minimum run/hold logic, and the bounded cracking rate.
- Objective Function: The cost function ( $J_k$ ) minimizes ammonia consumption, equivalent battery/fuel cell costs, and SOC deviation penalties. Crucially, it includes a prediction-aligned over-generation penalty ( $J_{pred}$ ). This term penalizes power allocation that exceeds predicted future demand, discouraging the accumulation of hydrogen that cannot be utilized due to cracking limits or storage saturation.
- Terminal Logic: A terminal mode activates near the end of a trip to prioritize fuel cell usage if sufficient hydrogen remains, ensuring resource utilization without compromising SOC safety.
- Solution: The controller performs an exhaustive grid search over discretized engine and fuel cell power points to find the optimal power split that minimizes the cost function while satisfying all constraints.

Key Contributions
The paper outlines three primary contributions:

Architecture: A prediction-aware MPC architecture that embeds short-term demand forecasts directly into the receding-horizon cost function, allowing future demand to guide current power allocation.
Formulation: A constrained optimization formulation that explicitly captures the dynamics of hydrogen storage and the bounded capacity of the ammonia cracking process, addressing the unique inter-temporal coupling of this powertrain.
Evaluation: A comprehensive comparative evaluation demonstrating that the proposed strategy outperforms rule-based, ECMS (Equivalent Consumption Minimization Strategy), and conventional (non-predictive) MPC strategies in terms of fuel efficiency and resource utilization.

Results
The proposed ITI-MPC was evaluated using a real-world driving dataset (176.50 km, mixed urban/highway) on an instrumented ammonia-hydrogen prototype. It was compared against rule-based control, ECMS, and conventional MPC under identical conditions.

Prediction Accuracy: The 1-second ahead predictor achieved an $R^2$ of 0.869, maintaining substantial utility ( $R^2 = 0.570$ ) even at a 5-second horizon.
System Efficiency: The ITI-MPC achieved the highest system efficiency of 42.80%, surpassing conventional MPC (42.18%), ECMS (42.11%), and rule-based control (38.88%).
Fuel Consumption: The total equivalent ammonia consumption was reduced to 86.39 kg, a 1.45% improvement over conventional MPC (87.66 kg).
Resource Utilization: The strategy reduced terminal hydrogen storage to 0.0188 kg (compared to 0.0711 kg for conventional MPC), indicating more complete utilization of hydrogen resources rather than conservative reservation.
Real-Time Performance: The average solve time per step remained below 20 ms, satisfying the 0.5 s sampling period requirement.
SOC Stability: The strategy maintained a stable SOC trajectory comparable to other MPC-based methods, confirming that efficiency gains were not achieved through excessive battery depletion.

Significance and Claims
The authors claim that the performance improvement stems primarily from enhanced inter-temporal coordination rather than aggressive control actions. By anticipating future demand, the ITI-MPC avoids the "myopic" behavior of conventional MPC, which often leads to conservative hydrogen reservation or inefficient over-generation.

The paper emphasizes that these gains are significant within the context of tightly constrained multi-energy systems where objectives and constraint sets are fixed. The approach reshapes the current-stage cost using future information without altering the feasible set or actuator constraints. The authors note that the effectiveness of the method is contingent on prediction reliability and suggest that future work could integrate uncertainty-aware weighting or stochastic MPC formulations to further improve robustness under imperfect prediction conditions.

More like this