Mathematics and AI Could Make Trains More Energy-Efficient
Researchers at Omsk State Transport University have developed a hybrid model that combines the laws of physics with machine learning to improve the accuracy of train control.
Researchers at Omsk State Transport University have proposed a new system for managing rail traffic. It combines two approaches: traditional physics-based equations that describe traction, braking and resistance forces, and machine-learning algorithms that identify patterns in real-world data. The model was trained using data from more than 1,500 freight train runs on the Zapadno-Sibirskaya Zheleznaya Doroga (West Siberian Railway). Onboard recorders captured train speed, engine operation, brake loads and distances traveled. The resulting model improved trajectory calculation accuracy by 10% to 15%, while the difference between calculated and actual travel times on the routes studied remained under one minute.
It is important to distinguish improved modeling accuracy from proven energy savings: the study has not shown that electricity consumption falls by the same percentage. Real-world testing of the algorithm is still ahead. Nor does the technology turn a train into a driverless vehicle. In the future, it could be integrated into automatic train operation systems or used to help locomotive engineers make operating decisions.
From the Lab to the Rails
The technology has several potential applications in Russia. The first step is planned real-world testing in cooperation with AVP Technology LLC, a developer of automatic train operation systems, and Russian Railways’ Traction Directorate. If the trials confirm the model’s claimed accuracy, the algorithm could become part of software that calculates optimal driving strategies. The system would determine when to accelerate and brake, as well as the appropriate speed on different sections of track, taking actual operating conditions into account.
Freight operations are the most immediate area of interest because the model was trained on data from freight trains. If successful, the technology could reduce electricity consumption and mechanical stress on rolling stock, potentially lowering maintenance and repair costs. It could later be adapted for other routes, with its application to passenger services, including commuter and high-speed trains, explored at a subsequent stage.
How AI Is Coming to the Railways
In 2024, researchers at Omsk State Transport University examined the prospects for energy-efficient control of 2ES6 electric locomotives. They noted that real-world locomotive data are particularly important for building dynamic models. In 2025, a paper on selecting machine-learning methods for energy-efficient train operation appeared in the journal Proceedings of Petersburg Transport University. That same year, researchers published a study of an automatic train operation algorithm for the Sapsan high-speed train, estimating that it could reduce electricity consumption by more than 11%.
The global industry is moving in the same direction. China developed an AI model for energy-efficient train operation in 2022, while international researchers used neural-network methods in 2023 to cut a freight train’s energy consumption by several percentage points. In Europe, the R2DATO project brings together dozens of partners to advance digital and automated railway operations and improve operating efficiency and safety.
Accuracy Today, Savings Tomorrow
At this stage, the hybrid model’s value lies in more accurate calculations, not in resource savings that have already been demonstrated. The reported 10% to 15% improvement refers to trajectory calculations, not a verified reduction in electricity use. The actual economic benefits can be assessed only after operational trials establish the effects on energy consumption, equipment wear and travel times.
The researchers plan to test the algorithm on real routes and compare its performance with existing control methods over the next three years. If the results are positive, the technology could be integrated into locomotive automatic train operation systems and later extended to other types of rolling stock and routes.
In the longer term, hybrid models of this kind could become part of a broader digital railway infrastructure, alongside locomotive digital twins, equipment condition-monitoring systems and transport planning tools.
