Speed profile optimisation for a semi-trailer truck based on a hybrid fuel consumption model with automatic calibration based on CAN bus data

The paper proposes a method for optimising the speed profile of a long-haul semi-trailer truck that combines a physical fuel consumption model with a machine-learning correction with automatic calibration using on-board CAN bus data. The physical component is built on the tractive balance equation, which parameters, i.e. the power transmission efficiency, rolling resistance coefficient, brake-specific fuel consumption and lumped mechanical losses, are identified without bench testing by minimising the difference between the computed and the measured instantaneous fuel rate using the Powell's method; calibration is verified against the cumulative fuel counter (SAE J1939). The correction is implemented with the LightGBM algorithm that approximates the rest of the physical model; for the first time the current gear signal extracted directly from the CAN bus is used as an additional feature. The optimal speed profile is obtained by dynamic programming subject to a constraint on the travel-time increase; the speed range is lumped into 13 values within 60–80 km/h. The method was verified using a field trial with a Mercedes-Benz Actros 1845 truck on the M-11 «Neva» motorway (Moscow – Saint Petersburg, 672.5 km; about 14 million CAN records processed in under two minutes). The calibration error was 3.6%; introduction of the gear feature reduced the prediction root-mean-square error from 13.7 to 11.5 L/100 km (by 16%) and the cumulative-consumption deviation down to 1.4%. The computed fuel saving reached 9.3% at the travel time increased by less than 0.6%. Numerical simulation using the Moscow – Kazan route (809.1 km) confirmed the scalability: a 10.4% saving at a 2.8% time increase without model re-calibration.