Partner
Vnomics
Industry
Transportation
Goal
Analyze Vnomics driver data to determine if further fuel efficiency could be realized.
Approach
Machine learning
Cluster Analysis
Non-linear modeling
Natural Splines
Variance-component analysis
XGBoost with SHAP Explainable AI Analysis
The Opportunity
Vnomics provides a service to monitor and coach truck drivers for better fuel performance so that fleets can decrease fuel costs by discouraging inefficient fuel behavior and encouraging best practices among drivers. Sensors installed in trucks provide data on performance factors fleet managers and drivers can address to decrease fuel consumption, a major expense in commercial transportation services. As a leader in its field, Vnomics uniquely offers solutions for coaching drivers in real time and provides insightful analytics that separate the driver impact on fuel economy from other factors. It also supplies reports that pinpoint precisely where fuel is wasted because of driver behavior. Vnomics hired RDSC to determine if a deeper analysis of its driver data could identify different or additional behaviors among its drivers and, if so, whether the findings could be used to further reduce fuel costs for its customers.
“In the process of working with trucking fleets and their drivers to improve fuel efficiency, we’ve collected a lot of data on driver behavior and fuel performance. RDSC was able to analyze that data and determine ways we could help our customers to save on fuel costs even more.” – Lloyd Palum, CTO Vnomics
The Challenge
As part of its analysis, RDSC had to create a model that separated non-driver features such as distance and speed, freight load, terrain and elevation, engine torque and power and model year from features that were the direct result of driver performance. At the same time, they had to take into account the nature of the drivers’ vocation: delivery vehicles that stopped often, for instance, needed to be differentiated from long-haul routes. Data-driven tuning adjustments
had to be made such as how to adjust equitably for the effect of weight on fuel efficiency, where RDSC found the common industry measure of “per thousand pounds” resulted in bias.
The Solution
RDSC created clusters of drivers to account for the effect of vocation on fuel efficiency.
After adjusting for distance and freight load, terrain, engine torque, and other non-driver factors, RDSC isolated the information that could only be attributed to either driver behavior or unaccounted-for variation.
RDSC’s analysis then determined the extent to which fuel efficiency could be attributed to driver behavior in each of the clusters. The findings indicated drivers accounted for 20%–40% of the variability in fuel consumption, even on the most modern of tractors. To translate the findings into potential fuel savings, they assumed that the drivers not among the top 25% in fuel-efficient performance could, with more guidance from Vnomics’ systems, change behavior to become more like those in the top 25%.
Results
RDSC’s analysis confirmed that by identifying the best practices of the top 25% of drivers in a fleet, Vnomics’ clients could further reduce fuel costs, from 2% to 6% per cluster in addition to the 7%+ that Vnomics is currently delivering to fleets with their current product.
Additional insights were obtained by using second-by-second trip data sets that illuminated the aspects of driver behavior (braking and acceleration, for example) that were most important.
