Crate routee_compass_powertrain

Compass Powertrain Crate

This crate provides energy-aware route planning models that integrate RouteE Powertrain into RouteE Compass.

Energy Estimation in RouteE Compass

A RouteE Powertrain is an energy estimation model trained on NREL’s FASTSim model. RouteE Powertrain is an ML model which makes it sufficiently fast to run within the loop of a route planner cost model. RouteE Powertrain models are trained and exported via the RouteE Powertrain utility and then loaded into a runtime at the core of this crate.

Model Runtimes

There are two underlying model runtimes available, smartcore and ort (for ONNX models). By default, this crate is loaded with ONNX deactivated. To activate the ONNX feature, pass the onnx feature flag during compilation. Runtime kernels for 3 common OSs have been provided in the onnx-runtime directory within this crate. For more information on cargo features, see The Cargo Book chapter on Features.

The runtime is loaded via the TraversalModel(s) in this crate and used to estimate costs in RouteE Compass searches.

Usage

The TraversalModel in this crate is integrated into compass-app and can be installed by running compass-app with a TraversalModel that uses energy. An example traversal model configuration that uses this crate may look like this:

[traversal]
type = "energy_model"
speed_table_input_file = "edges-posted-speed-enumerated.txt.gz"
speed_table_speed_unit = "kilometers_per_hour"
output_time_unit = "minutes"
output_distance_unit = "miles"

[[traversal.vehicles]]
name = "2012_Ford_Focus"
type = "single_fuel"
model_input_file = "models/2012_Ford_Focus.bin"
model_type = "smartcore"
speed_unit = "miles_per_hour"
grade_unit = "decimal"
energy_rate_unit = "gallons_gasoline_per_mile"
ideal_energy_rate = 0.02857143
real_world_energy_adjustment = 1.166

This TOML section is deserialized into JSON and passed as arguments to the EnergyModelBuilder which in turn loads the [EnergyModelService] in this crate. This in turn builds the [EnergyTraversalModel].

Search

TraversalModels in this crate will add energy estimation to road network search, and will differ in their dependencies and evaluation procedures.

EnergyTraversalModel

Dependencies

• speeds per graph edge as a lookup table
• grade per graph edge as a lookup table (optional)
• an energy_cost_coefficient value, (optional with default of 1.0)
• a real_world_adjustment_factor value, (optional with default of 1.0)

Evaluation

1. lookup speed for the edge in table
2. compute travel time as distance / speed for this edge
3. lookup grade for this edge in table; if missing, use 0.0
4. perform inference to retrieve energy rate from speed and grade values
5. compute energy as energy_rate * distance * real_world_adjustment_factor for this edge
6. compute link cost as (energy * energy_cost_coefficient) + (time * (1 - energy_cost_coefficient))

Real-World Adjustment Factors

In addition to calculating the energy based on a RouteE Powertrain output, an adjustment factor should be applied to capture real-world effects of running a powertrain in an environment. As a result of NREL research, some recommended values for this adjustment are:

powertrain type	factor
combustion vehicle (CV)	1.1660
hybrid vehicle (HV)	1.1252
electric vehicle (EV)	1.3958

A factor of 1.0 equates to 100% of the original energy value.

Modules

• routee