Engineers at the University of California, Riverside (UCR) have developed a new way to predict whether an electric vehicle or other battery-powered system can safely complete a specific task. The diagnostic model, called the State of Mission (SOM), merges electrochemical physics with machine learning to forecast battery performance under real-world conditions rather than ideal laboratory settings.
Traditional battery management systems display charge levels, but they can’t tell drivers if a 40 percent reading will get them over a mountain pass on a cold day. SOM addresses that uncertainty by combining internal battery data with environmental inputs such as terrain, traffic, temperature, and driver behavior. “It’s a mission-aware measure that combines data and physics to predict whether the battery can complete a planned task under real-world conditions,” said Mihri Ozkan, a professor of electrical and computer engineering at UCR.
The system relies on neural ordinary differential equations and physics-informed neural networks that learn from large datasets while remaining constrained by the laws of electrochemistry and thermodynamics. This hybrid structure gives SOM adaptability without losing physical credibility, a critical balance that pure data models often lack. Tests using NASA and Oxford University datasets showed the model reduced voltage-prediction errors by 0.018 volts, temperature errors by 1.37 °C, and charge-state errors by 2.42 percent compared with conventional approaches.
Illustrated abstract of State of Mission for Battery Management: UCR
Instead of a simple percentage display, SOM provides actionable outcomes: a vehicle can complete its route, a drone should alter course to conserve power, or a stationary battery should delay discharge to maintain grid stability. “It transforms abstract battery data into actionable decisions, improving safety, reliability, and planning for vehicles, drones, and any application where energy must match a real-world task,” said Ozkan.
The research team sees wide-ranging industry potential. In electric transport, SOM could help manufacturers design smarter range-prediction systems and prevent in-route failures. Drone operators could use it to extend flight times by dynamically adjusting power draw. Utility-scale storage providers could rely on it to forecast capacity under fluctuating demand or weather extremes. Though current computational demands exceed most embedded hardware, the UCR group expects continued optimization and faster processors to make mission-aware diagnostics standard in next-generation energy systems.
