Multi-criteria management of energy flows in vehicles
Energy management is the portion of the central control system that divides the driver’s demand between the vehicle’s energy sources, monitors the battery state of charge and the actuation of accessories.
A generally accepted classification is used to divide energy management methods into two main categories, i.e. those that are based on modal logic using expert understanding of the system, and those that use optimisation principles.
Two categories emerge from these methods depending on whether a vehicle’s instantaneous emission factor (speed profile as a function of time) is known.
In either case, the hybridisation degree of freedom is used to optimise a criterion, which is often the reduction in fuel consumption (equivalent in a first approximation to a reduction in CO2 emissions). This criterion can be more complex if pollutant emissions and/or the service life of critical components are taken into account.
Where n is the number of samples across a vehicle-use time frame, mf is the instantaneous fuel flow rate, POL is the instantaneous emission of the pollutant in question, SOH is the battery life criterion, and α, β and γ are the weights assigned to each criterion. A single objective for reducing fuel consumption across the entire cycle in question yields the following criterion:
Two degrees of constraints must be met. The first degree relates to the last objective of the overall control: monitoring the battery state of charge with the preliminary choice of the two possible options (charge sustaining or charge depleting).
The other constraints are equality constraints, which are specific to each hybrid architecture, and inequality constraints, which represent the limitations of the various powertrain subsystems.
The team’s research on offline overall optimisation focuses on the exploration of so-called exact methods that make it possible to find explicit solutions to the problem of optimisation, namely dynamic programming and the calculus of variations.
As regards online optimisation, the team has implemented controls that are based on the calculus of variations or use neural networks paired with machine learning.