This article demonstrates several approaches to the vehicle energy consumption and tailpipe emission reduction opportunities. The article leverages the vehicle storage dynamics through smart and personalized optimization and control approaches in the context of connected vehicles. Recent advances in vehicle connectivity and automation have brought unprecedented information richness and new degrees of freedom that can be synergized with insightful understanding of vehicle powertrain and aftertreatment physical systems. Vehicle automation also provides new degrees of freedom that can be further leveraged by the vehicle control systems to improve vehicle energy efficiency and reduce tailpipe emissions. While vehicle automation levels probably will keep increasing, humans will still be involved in vehicle operations at various levels for the foreseeable future. The prediction of future vehicle’s power demand based on vehicle connectivity can significantly benefit tailpipe emission reductions and fuel economy.
Recent advances in vehicular communication, onboard computation, and vehicle automation technologies have paved the way for substantially improving vehicle energy efficiency and reducing tailpipe emissions. By creatively and systematically combining such information richness and new degrees of freedom with insightful understanding of vehicle system characteristics, many effective approaches can be generated to reduce vehicle operational energy consumption and tailpipe emissions, particularly for vehicles with storage components such as energy storage devices for propulsion systems and reductant storage catalysts for exhaust gas aftertreatment systems. This article describes several approaches exemplifying the vehicle energy consumption and tailpipe emission reduction opportunities by leveraging the vehicle storage dynamics through smart and personalized optimization and control approaches in the context of connected vehicles.
Enabled by Vehicle Connectivity and Intelligence
There are two recent advances in vehicle technologies that can dramatically affect future vehicle operations: one is vehicle connectivity that enables vehicles to exchange information with surrounding vehicles, infrastructure, and many other systems through wireless communication; and the other is vehicle automation that can perform some or all of the vehicle driving functions. While driving comfort, convenience, and safety are the direct advantages that such vehicle connectivity and automation can bring, the information richness and new degrees of freedom can be further utilized in vehicle powertrain and aftertreatment system (PAS) controls to enable substantial vehicle energy consumption and emission reductions. Figure 1 illustrates a plug-in/hybrid electric vehicle (P/HEV) that is connected with a cloud server where various driving-related data can be exchanged. With vehicle connectivity, it is possible to use future information such as road and traffic conditions as well as historical driving data to predict the future vehicle power demands that will be beneficial for optimizing the PAS operations through prediction-based control. On the other hand, vehicle automation also provides new degrees of freedom that can be further leveraged by the vehicle control systems to improve vehicle energy efficiency and reduce tailpipe emissions. For example, in automated driving mode, human drivers relinquish the control authority over vehicle speed and even routing to the vehicle controller, which can then actively adjust the vehicle speed and select the route that can help improve the vehicle energy efficiency and reduce emissions.
Importance of Human Driving Style Variation and Prediction
While vehicle automation levels probably will keep increasing, humans will still be involved in vehicle operations at various levels for the foreseeable future. In real-world vehicle operations, human driving styles can significantly influence vehicle energy consumption and tailpipe emissions as much as 40-50 percent . A driver's torque and speed demands in real-world driving are affected by several time-varying factors including traffic situation and road condition among which the individual human's driving style plays a decisive role because it determines how other factors are transferred into his/her torque and speed demands to the vehicle. Even for the same vehicle and trip, the torque and speed demands from different drivers can be quite dissimilar. For instance, five different human subjects conducted driving simulator tests for the same vehicle-trip that involved no other vehicles or traffic lights but stop signs, and Figure 2 shows that the vehicle speed profiles (and thus the torque demand profiles) by the five drivers are very different. The sales of a vehicle model-year can reach more than 300,000 units, which means that over 300,000 different human drivers, covering an enormously wide range of driving styles, operate the same vehicle model daily. In comparison to such immense variations on the side of human drivers’ driving styles, an identical and fixed vehicle PAS control system on the vehicle side seems improper and unbalanced in the context of minimizing real-world vehicle energy consumption and tailpipe emissions. Given that a vehicle model will be driven by many different human drivers with a wide variety of driving styles, it is impossible to preset the vehicle PAS control system strategies and parameters for accommodating all the individual human drivers in the respectively-optimal ways. Driving style learning and corresponding driver-specific and optimal PAS control must be conducted with respect to the individual drivers onboard to achieve the utmost vehicle energy consumption and tailpipe emission reductions in real-world driving.
Learning-Based and Personalized P/HEV Energy Management and Control
With the rich data including individual vehicle historical driving data made available by vehicle connectivity, vehicle usage information such as the future route and associated information can be predicted from historical driving data on an individual driver-vehicle basis. Such predicted future route and road information can then significantly benefit the energy efficiency particularly for P/ HEV and electric vehicles because of their onboard energy storage devices, e.g. batteries, through predictive control. For example, future road grade information can be very helpful in improving P/HEV energy efficiency over the trip particularly in hilly areas where long uphill and downhill roads are common . However, as different P/HEVs are operated by different human drivers at various geometric locations, prediction of the future road grade information must be done for individual driver-vehicle pairs. In recent work , a Markov-chain model was developed to predict the probabilities of future routes for individual vehicles based on learning from their respective historical driving data and then such predicted future road grade information was used in stochastic predictive energy management for the respective P/HEVs.
As illustrated in Figure 3, a personally-owned vehicle will typically be operated on several frequent routes each with its own road grade profile. The daily driving data of the vehicle can be made available and collected through vehicle connectivity and then a learning mechanism can be applied to such historical driving data to generate a stochastic model. Such a model can predict the probabilities of the different branches that the driver may take at an intersection based on the current time and traveling direction. With such a future route prediction, the probability for future road grade information can be readily extracted from an onboard map and thus can be utilized in a stochastic model predictive control (MPC) based energy management strategy for the P/HEV. As an intuitive example, if the model predicts that the vehicle is very likely to go to destination D before reaching the first intersection as shown in Figure 3, then the P/HEV energy management system will make sure the battery state of charge (SOC) is low enough such that the battery can have enough capacity to store the electricity to be generated by regenerative braking on the downhill road towards D. The study shows that for 24 trips of a vehicle driven among these frequent routes, 7-9% energy saving can be expected in comparison to the equivalent consumption minimization strategy (ECMS) P/HEV energy management method .
Another common vehicle operating scenario is fixed-route driving where a vehicle repeatedly travels on a fixed route. Personal commute vehicles, public transportation vehicles, and utility vehicles are typical applications of fixed-route driving. Fixed-route driving is different from fixed-cycle driving where the vehicle speed is a predefined function of time and is typically utilized for standard vehicle emission tests. Even though the route is fixed, however, because of the randomness and variations of traffic and road conditions, vehicle speed profiles (and thus power demands) will not be the same trip-to-trip on the same route. For instance, the traffic lights at the intersections on the route vary trip-to-trip and so does the vehicle speed nearby the traffic lights. In order to improve the operational energy efficiency of P/HEVs traveling on fixed routes, the key is to accurately predict future vehicle speed (or power demand) given the randomness induced by such trip-to-trip traffic and road condition variations. In recent work , a two-level real-time-implementable P/HEV energy management methodology was developed that can characterize the trip-to-trip randomness on the route by learning from historical driving data of individual vehicles and generating the optimal control parameters for real-time P/HEV energy management.
As graphically illustrated in Figure 4, four major steps are required for such a learning-based stochastic P/HEV energy management strategy for fixed-route driving. First, based on the traffic and road condition features such as locations of traffic lights and stop signs as well as road grade and curvature-changing locations on a route, the route is divided into several shorter road segments, for example 20 road segments for the route given in Figure 4. Second, the historical driving data for a given driver-vehicle pair including fuel and electricity consumption over each of the road segments are collected via vehicle connectivity for a sufficiently-long driving period such as a month. Third, based on such a historical driving data set, a stochastic model is developed with fuel and electricity consumption over each road segment as the random variables. Then an offline optimization process is carried out to compute the optimal P/HEV control parameter, such as the electricity-to-fuel equivalence factor λ in the ECMS control method, for each road segment that can minimize the expected energy consumption. Once all these steps are done offline, the optimal control parameter is stored as a 3-D map with the inputs as the current P/HEV battery SOC and road segment index, and the output as the optimal P/HEV control parameter, such as λ, that goes into the real-time P/HEV power management controller. When additional driving data sets with new features are collected, the process can be repeated and the optimal control parameter map can be updated. Based on a study for 24 trips on a fixed route, although the method cannot guarantee the optimal performance for each single trip because of the stochastic nature of the process, it can reduce the total energy consumption of the trips by 7% in comparison to an existing method.
Tailpipe Emission Reductions Via Human Driving Style Learning
In addition to the energy storage dynamics available on electrified vehicles, for vehicles equipped with exhaust gas aftertreatment systems that have reductant storage dynamics such as selective catalytic reduction (SCR) systems shown in Figure 1, prediction of future vehicle power demand based on vehicle connectivity can significantly benefit tailpipe emission reductions and fuel economy. For a vehicle equipped with SCR, because its ammonia storage dynamics are much slower than the engine dynamics, future information on vehicle power demand and engine-out emissions can considerably help achieve appropriate and prediction-based control of SCR ammonia storage level, which in turn can reduce both tailpipe emissions and engine fuel consumption by leveraging the intertwined engine and aftertreatment system operations, particularly in transient operations. Recent work  demonstrates the potential for utilizing the predicted future vehicle power demand to coordinately control both engine and SCR together via a nonlinear model predictive control (NMPC) approach for reducing fuel consumption and tailpipe emissions in the context of connected vehicles. The overall control structure is shown in Figure 5 where both the engine fueling such as start of injection (SOI) and SCR diesel exhaust fluid (DEF) injection are coordinately controlled using the predicted information. For predictive control of engine-aftertreatment systems, prediction of human driver's driving style is critical because different drivers may issue different power demands even for the same driving condition. A new individualized driving style prediction model that can incorporate future road and traffic information has been established in . Such a model can learn from individual drivers’ daily driving data to parameterize the stochastic model and to better predict the respective drivers’ power demands that can be subsequently utilized in the predictive engine-aftertreatment system control.
In sum, recent advances in vehicle connectivity and automation have brought unprecedented information richness and new degrees of freedom that can be synergized with insightful understanding of vehicle powertrain and aftertreatment physical systems to create new methods and approaches for significantly reducing real-world vehicle energy consumption and tailpipe emissions. By capitalizing on the vehicle storage dynamics and vehicle connectivity, smart and personalized powertrain-aftertreatment system control methods can be quite effective.