Potential analysis of a predictive energy management strategy designed to increase the efficiency of the powertrain in a hybrid vehicle with use of online available route information

doi:10.21203/rs.3.rs-7793395/v1

Hybrid vehicles, with their combined use of internal combustion engines and electric motors, present a unique opportunity to leverage intelligent control strategies for optimal performance. The method presented in this paper aims to improve the efficiency of a hybrid powertrain through an increased usage of the advantages of the electric components. While conventional hybrid operation strategies must determine the torque-split on either rule-based decision logics or on strategies that are optimized for certain test scenarios a predictive strategy can optimize the torque-split under consideration of the upcoming load requirements. Therefore, this paper explores the development and implementation of real-time predictive driving and operating strategies for hybrid vehicles to enhance fuel efficiency and reduce environmental impact.

The data pertaining to road networks and traffic conditions, currently accessible from numerous map providers, can be effectively utilized to further amplify the benefits offered by hybrid vehicles. These information’s will be used to improve the accuracy of the predicted driving situations, and this subsequently improves the effectiveness of the predictive hybrid strategy by increasing the accuracy of the predicted load requirements. Advancements in prediction model accuracy have been shown to enhance the effectiveness of predictive hybrid control strategies, leading to higher energy efficiency and lower emissions.

The underlying studies will focus on further enhancing the prediction models and undertaking empirical testing to ascertain their efficiency. Future research will aim to refine these models further and conduct real-world testing to validate their effectiveness. The enhanced predictive strategy achieves a consumption reduction of up to 12.18%, approaching 14.4% reduction obtained under perfect prediction, while significantly reducing engine state transitions from 110 to 30, thereby improving both efficiency and driving comfort.

The hybridization of the drivetrain expands the degrees of freedom (DOF) in vehicle operation, enabling the internal combustion engine (ICE) to be operated at more efficient operating points. This is particularly relevant in the context of increasing demands for CO₂ reduction and energy consumption optimization in road transport. Hybrid systems provide the capability to combine electrical and thermal propulsion sources, achieving efficiency gains and meeting environmental targets. Therefore, energy management strategies (EMS) for calculation of the most efficient hybrid operation modes are required. Those calculations include not only discrete operation modes but also the efficient hybrid operation which is a continuous problem of for example the torque-split optimization.

Key functions such as recuperation and load point shifting contribute significantly to energy savings. While recuperation is primarily limited by the available battery capacity, the boundary conditions for load point shifting are more complex. Efficient load point shifting requires not only sufficient free storage capacity in the battery but also a well-informed decision regarding the operation of the ICE. For example, if the battery is fully charged, the energy management strategy will prioritize pure electric operation. However, this approach may introduce inefficiencies if it results in load point shifts occurring later, under less optimal efficiency conditions.

Traditional rule-based strategies, such as Charge-Depleting (CD) and Charge-Sustaining (CS), often fail to account for these efficiency losses adequately. Similar, strategies based on the current power requirement as seen in Figure 1 can miss out on energy saving potentials. Predictive hybrid operating strategies, on the other hand, minimize such inefficiencies by forecasting future operating conditions and incorporating them into decision-making. Achieving this requires precise prediction of load demands along the anticipated driving task.

The method presented in this paper is specifically designed for application in P2 hybrid electric vehicles (HEVs). This drivetrain architecture (Figure 2) positions the electric motor (EM) between the ICE and the transmission, allowing both propulsion sources to operate independently. This is particularly advantageous for functions such as load point shifting and pure electric driving. Due to these characteristics, the proposed strategy is suitable for both plug-in hybrid electric vehicles (PHEVs) and full hybrid electric vehicles (FHEVs). The specific challenges and opportunities of these architectures, such as optimizing power distribution and respecting electrical operating limits, are addressed in the proposed EMS.

This paper includes a method for predictive speed forecasting that uses map data and a driver model to predict future driving behavior. From the predicted altitude and speed profile, future load demands can be estimated and integrated into an optimized EMS. The proposed method aims to further enhance the efficiency of hybrid drivetrains and increase their practical applicability. Overall the proposed strategy encompasses three main components:

An online request from different digital map providers to gather the required data for the following optimization strategies.
A predictive driving strategy that anticipates future driving conditions and computes the vehicle velocity for the prediction horizon; and
An energy management strategy that determines the most efficient way to distribute power between the engine and the electric motor based on the predictive inputs while meeting the driver requirements. Simulation and experimental results suggest that these strategies can significantly improve fuel economy without compromising the driver's comfort and journey time.

The most commonly used numerical approach to solving the energy management (EM) problem is the indirect method, based on the Pontryagin Minimum Principle (PMP) or calculus of variations. In this method, the optimal control problem is transformed into a two-point boundary value problem, which can be solved using shooting techniques or collocation methods. This approach is referred to as the Equivalent Consumption Minimization Strategy (ECMS) [3]. The terminology originates from interpreting the adjoint variables derived from the necessary optimality conditions (defined by the Euler-Lagrange equations) as equivalent costs for using battery energy. Various variants of this method exist, such as Instantaneous ECMS, Adaptive ECMS, and Telemetry ECMS [4]; [5]. However, this method suffers from sensitivity to the initialization of solutions and challenges in handling inequality constraints [6].

The predictive Energy Management approach aims to increase the efficiency of the hybrid powertrain by operating the internal combustion engine in such a way that, while adhering to constraints such as SOC limits, operating points with the highest overall vehicle efficiency are selected. Unlike a conventional ECMS, which only considers the current time step in the cost function, the predictive ECMS (P-ECMS) also includes future operating states into the calculation. To manage the increased computational complexity effectively, all necessary maps and simplified models of the drivetrain components are employed. Within each control loop iteration, all feasible operating states over the defined prediction horizon are evaluated. The optimizer then determines the most cost-effective solution, and the corresponding load signals for both the electric and combustion engines are transmitted to the vehicle model. In the subsequent time step, the actual system response is computed and fed back into the operating strategy. This iterative process enables continuous comparison between target and actual values, thereby minimizing deviations from the predicted speed and preventing inaccuracies in subsequent calculations.

For the predictive calculation of possible operating states and the subsequent determination of the most efficient engine operation, it is necessary to estimate the upcoming driving demands and calculate the power required by the drivetrain. Future operating points are therefore dependent on the vehicle, driver, and route. To calculate driving demands using driving resistances, vehicle parameters such as the drag coefficient (c_w-value), vehicle mass, road gradient, and vehicle speed are required. [7]

3.1 Necessary Data set

Vehicle Dependency:

The vehicle parameters can be measured or are determined by the vehicle manufacturer based on the installed hardware components, such as the internal combustion engine and electric motor, and based on their respective operation through the ECU. These parameters are fixed data sets that need to be defined once and stored in the control software of the operating strategy, where they can be retrieved during calculations. The vehicle load plays an important role in computing the power demand and is assumed to computed based on driving conditions.

Route Dependency:

The gradient, which depends on the geographic position of the vehicle, can be calculated using online map data. If information about a planned route is available, an elevation profile can be created for prediction purposes. In navigation systems, route information is already retrieved to calculate travel time and the optimal route. Thus, these data are generally available if a destination is defined at the start of the journey. Using machine learning algorithms, this step can also be automated, for example, if the driver takes the same route to work every morning.

Driver Dependency:

By building a driver model, the speed profile for the predicted route can also be created based on the route data. This driver model requires defining driver characteristics such as desired speed, lateral, and longitudinal acceleration. Measurements from real-world driving are used beforehand to evaluate these parameters. The driver model can then calculate a speed profile for the route, taking physical constraints into account.

The speed and elevation profiles together form the predictive driving demand, which is subsequently used to determine the necessary operating points.

3.2 Vehicle model

Instead of a completely data-driven approach the vehicle model is equation based where parameters are identified using experimental data. Therefore a vehicle model was developed based on measurement data from real-world driving tests and examined on the engine test bench in a hardware-in-the-loop (HiL) simulation. Validation of the simulation model was carried out using various test scenarios, enabling further investigations in the simulation. First, the driving resistance coefficients were measured through coast-down tests, and the drivetrain components were modeled as in the vehicle. The test scenarios included driving the WLTC on a chassis dynamometer and conducting an RDE test drive. For comparison, the energy consumption at the wheels was calculated for each test. To achieve this, the power delivered to the wheels was calculated at each time step from the driving resistances and then integrated to determine the total energy consumption. [8]

3.3 Online data request

In online available Digital Mapping Services such as Google-Maps or HereWeGo, the beforehand described route information is stored and accessible. This includes not only geographic coordinates and elevation data but also speed limits and average vehicle speeds at specific route points. Additionally, information about intersections or traffic lights is available and can be retrieved. These data can also be used to improve the accuracy of speed prediction calculations.

For data retrieval in the simulation, a Python-based application (API) was developed to fetch the required route data via a JSON interface from the internet. During the initialization of the simulation, a GUI opens where, similar to a navigation system, the route can be defined, and some settings for data retrieval can be configured.

In the fields "Departure" and "Arrival," the start and destination points of the route are entered. These can either be simple city names, as shown in the illustration, or alternatively, full addresses or coordinates can be provided. Under "Routing Options," the traffic influence can be enabled or disabled. When enabled, the retrieved "traffic speed" is adjusted to the selected day and time. Activating the "Loading GPX from File" function opens a separate window for selecting a GPX file, and instead of using the "Departure" and "Arrival" fields, the route is retrieved along the GPS points specified in the GPX file.

The "Submit" button allows modifications to the settings for saving the data, and the retrieval process is then initiated. The data is also saved in an Excel spreadsheet, which, as shown in Figure 5, assigns the corresponding details to each data point. These details consist partly of raw data from the retrieval process and partly of calculated values, as previously mentioned.

3.4 Data-preprocessing

The key data for the driving strategy include latitude and longitude, elevation, curve curvature, speed limit, traffic speed, and the so-called intersection bit. The data points are typically located at relevant points on the route, such as intersections, speed limit change points, or curve apexes. Since these points are solely dependent on the infrastructure, the data points are distributed along the route at non-equidistant intervals. Therefore, a calculated "Travel Distance" variable is stored to provide the total distance in meters. This distance can be calculated from the geocoordinates using the Haversine formula and is directly transmitted by the map provider during data retrieval. This enables distance-based assignment of boundary conditions for the subsequent speed calculation. However, gaps of over 50 meters between consecutive points can occur. A speed profile calculation at this resolution is insufficient for the subsequent calculation of operating points, as the dynamic operation of the engines leads to rapid changes in operating states. Therefore, an interpolation to 5-meter intervals is performed, which also allows for case-dependent differentiation in the calculation of the curve radius. This improves the determination of curvature, which is necessary for speed prediction in curves. The curvature is calculated from the coordinates of three consecutive points using the circle equation. As mentioned, the distances between data points can vary, particularly in urban areas, where data points might only exist at the midpoints of two adjacent intersections. In such cases, a curve radius would be calculated that is significantly too large. With 5-meter intervals between the three points used in the circle equation, practically all tight curve radii in typical road traffic can be calculated. However, this approach presents a challenge for long curves in rural areas. Due to the interpolation, many points lie on straight lines where the curvature is zero, and only the points between these straight sections yield overly tight curve radii. This issue can be resolved through a combination of the curvature calculations. For tight urban curves, the radius calculated after interpolation is used, while for rural areas, the values calculated before interpolation are applied. The case differentiation is based on traffic speed: if the traffic speed exceeds a defined threshold, the pre-interpolation calculation is used. [8]

For elevation data, it is common to encounter a resolution of only 1 meter, which can result in very high gradient values for closely spaced points. This would lead to short, high load demands in subsequent calculations, which do not reflect the real driving profile. The issue of the low-resolution elevation profile can be mitigated using a smoothing function. However, configuring the smoothing factors proves challenging: overly high factors smooth out too many road irregularities, while overly low factors fail to address the existing problem. An alternative solution is to replace the retrieved elevation profile with a more accurate one, such as one recorded during a real drive along the route using an inertial navigation system (INS).

The retrieved data is distance-based from the geodata retrieval. By calculating the speed profile, the profile can then be converted into a time-dependent speed profile from a distance-based one. This is necessary for the calculations in the operating strategy, in which the energy consumption of the powertrain components is calculated as a function of time.

4.1 Driver modelling

Driver modeling is based on historical speed profiles of the respective driver. For example, a vehicle measurement can be used. Various characteristics are determined for the driver model. Firstly, the driver's speed selection in general, and secondly the acceleration and braking behavior as well as the cornering speeds. For the latter, the maximum values for longitudinal and lateral acceleration are defined using the 95% quantile from the an example journey like a RDE measurement. This allows the changes in speed to be mapped sufficiently well. A further function limits the jerk, i.e. the derivative of the speed changes, in order to simulate the speed adjustments at different speed limits as well as possible.

4.2 Calculation of the speed profile

The speed profile is calculated in several steps. First, a combined speed limit is calculated. This defines the maximum speed of the vehicle at each point on the route. On the one hand, it is limited by the speed limit retrieved from the online route data and limited by the maximum speed corresponding to the driver's request. On the other hand, it is limited by a calculation of the route-dependent maximum cornering speed. The maximum cornering speed is calculated from the accelerations acting on the driver. Since the maximum longitudinal and lateral acceleration values or their superposition are defined as part of the driver model, the resulting acceleration that the driver would experience can be calculated using the respective curve radius. To do this, however, the curve radius must first be calculated. This is calculated using the circular equation from three consecutive points:

To do this, the geocoordinates from the data retrieval are first converted into Cartesian coordinates. The radius can then be calculated using the converted Cartesian coordinates as shown in Formula 1. This is then used to determine the effective accelerations and the maximum cornering speed by comparing them with the driver data. As described above, a case-dependent differentiation is carried out based on the average traffic speed. For higher velocities the calculation of the radius is performed before the interpolation of the online retrieved route points to better fit the long stretched curves on highways and in rural areas. For low speeds, as for example in urban locations and cities, the calculation of the radius in done with the interpolated route points to better match the sharp curves on intersections and traffic lights.

The following figures show the step-by-step adjustments of the velocity prediction profile. Starting from the base speed limit, shown in Figure 6, the velocity profile gets more and more adjusted to the different boundaries.

In Figure 6, the results of the three different maximum speeds, can be seen. These can be combined to form a combined speed limit (Figure 7).

In the next step, the speed transitions are determined via the driver parameterization. First, the subsequent speed progression is adjusted based on the previous current speed, again using the maximum longitudinal acceleration. The speed in the previous step is then reduced in reverse order, i.e. starting from the following speed based on the maximum longitudinal deceleration. Further conditions can also reduce the jerk, i.e. the second derivative of the speed, and/or represent speed fluctuations due to other traffic influences. This results in the final speed profile shown in Figure 8. In this profile, the speed and acceleration limits defined by the driver and the environment are not exceeded and the average driver behavior is mapped.

Afterwards, the speed profile calculated in the manner described above is compared with the real driving profile from the vehicle measurement. Figure 9 shows the respective speed profiles over the route.

With the exception of a few stretches of road, the speed levels are nearly the same for both profiles. However, there may be deviations due to temporary speed limits, for example due to roadworks, or new adjustments to speed limits, if these are possibly not or not yet stored in the online data. In such cases, the deviations cannot be prevented by the calculation, as the input data is incorrect, but previous investigations have shown the robustness of the strategy to those deviations [9]. Figure 10 shows some route sections in detail. These show the deviations caused by the averaged driver parameters. Since, as described above, the 95% quantiles of the driver's accelerations are used for the calculation, there are deviations between the real and simulated driver in a direct comparison of the acceleration curves.

4.3 Optimization of the speed profile

It can be seen in the previous figures that the calculated final speed profile repeatedly shows completely constant sections. The comparison with the speed measurement shows that this driver behavior is often unrealistic if cruise control is not used. Slight speed fluctuations are not uncommon, even on straight stretches of road, due to varying load requirements or fluctuating accelerator pedal operation. When calculating the speed profile, various adjustments can therefore be made to make the predicted speed profile more similar to the speed profile of a real driver with regard to the differences mentioned. Additional adjustments to the speed prediction are therefore made in the next chapter. In the first step to optimize the speed prediction, the jerky acceleration changes are reduced. As the jerk is the 2nd derivative of the speed or the 1st derivative of the acceleration, it can be calculated and a maximum value can be defined. This is dependent on the speed and overall driver behavior and, like the maximum acceleration values previously, is derived from the vehicle measurement and integrated into the calculation of the speed transitions. These changes are only noticeable in areas where there was previously too much jerk, such as when braking from constant speed. In the remaining areas, the speed curve remains the same. The next adjustment is intended to represent the speed fluctuations during constant driving in order to simulate realistic conditions. For this purpose, two fluctuations are superimposed on the speed profile, each of which is intended to represent different influences on the vehicle speed. In practice, drivers or vehicles can be exposed to different types of disturbances [7], [8]. For example, a high amplitude, low frequency oscillation could represent larger but slower changes, such as adapting to a slope or another vehicle. A low-amplitude, high-frequency oscillation could represent smaller, faster adjustments, such as reacting to small speed variations or driver pedal inputs in urban traffic conditions. Another adjustment is the consideration of intersections. Purely based on the geodata, it cannot be determined if the vehicle has to stop or slow down at an intersection due to other vehicles or the traffic light. Therefore, a randomized speed adjustment is implemented that choses between three cases. In the first case no changes are made and the vehicle approaches the intersection at normal speed. In the other cases it forces the driver to either slow to half the maximum speed or stop entirely. Additionally, a combination of the mentioned adjustments is implemented.

The following Figure 11 shows for a short route section the resulting different speed profile. To show the overall influence of the velocity prediction a profile purely based on the speed limits and the acceleration and deceleration behavior of the driver is also shown.

Overall, it can be seen that the various speed profiles each have only slight deviations from the vehicle measurement, as the basic calculation in each of them is based on the same route data and the same driver parameterization. Although the combined prediction is also most similar to the profile of a real driven speed profile in terms of the visual course, the accuracy of the actual driven speed deteriorates due to the adjustments. This can also be seen by looking at the mean and median speed deviations in Table 1.

Table 1: Mean and median difference in speed for different velocity prediction methods [8]

Trial	Mean difference in speed	Median difference in speed
Trial	km/h	km/h
Perfect prediction	0	0
Basic prediction	9,68	7,01
Jerk adjustment	9,73	7,10
Fluctuation adjustment	9,07	6,34
Intersection adjustment	10,78	8,37
Combined adjustment	9,07	6,34
Speed limits	11,84	7,86

Since real driver behavior is characterized by stochastic fluctuations and the adjustments to the speed prediction for its reproduction are also partially implemented stochastically an improvement in prediction accuracy would therefore also depend on coincidence, which is why the opposite tends to be the case, as shown here. Nevertheless, it is an advantage that the generated speed profiles do not appear to be a synthetically generated curve. Whether the increased deviations have a negative effect on the subsequent calculations in the predictive strategy was investigated further in [8] and was found to be negligible. The basic functionality of the predictive strategy is further described in the following chapter.

The aim of the predictive operating strategy is to control the hybrid powertrain with maximum efficiency by using the combustion engine when the operating points with the best efficiency can be selected while complying with the SOC limits. For this purpose, all necessary maps and simplified models of the powertrain components are stored and the possible operating states for the defined prediction horizon are calculated in a loop. As soon as the optimum solution is found, the corresponding load signal for the electric motor and combustion engine is transferred to the vehicle model. In the next time step, the adjusted actual values are transferred back to the operating strategy and the calculation is performed again. In this way, a constant target-actual comparison takes place, which prevents deviations from the predicted speed from causing an incorrect calculation.

5.1 Input variables

The most important input variables for the calculation of the optimum torque distribution are the speed and height profile, as these can be used directly to calculate a total torque requirement for mastering the driving task. The predictive operating strategy receives these two variables from the driving strategy. In addition, the operating strategy also requires all vehicle-specific values, meaning all other information required for vehicle modeling in addition to the aforementioned engine maps. For example, the variables relevant for the driving resistances, such as the vehicle mass and the resistance coefficients, but also the shift logic with the gear ratios and the battery parameters. To reduce the computational effort, many characteristic maps are not stored directly, but are mapped using polynomial functions. These types of simplifications are necessary to ensure the real-time capability of the simulation in the test bench environment.

5.2 Functionality

As described before the predictive operating strategy is based on an ECMS. Thereby, the required electrical energy is transformed via an equivalence factor to determine a fuel-equivalent consumption. The equivalence factor can be easily minimized for a defined driving cycle, such as the WLTC, via dynamic programming by a variation of the equivalence factor. However, the result is optimized for the respective reference cycle and can therefore lead to suboptimal results in a different test scenario. For this reason, the predictive operating strategy, which also calculates the optimum solution using a cost function, uses a loop function to adjust the equivalence factor for the route within the prediction horizon. In this case, we speak of a predictive ECMS (P-ECMS).

The length of the prediction horizon is a key parameter for calibrating the strategy, as the definition of a target SOC, which must be reached by the end of the horizon, creates an additional boundary condition. A shorter horizon therefore results in a greater restriction of the degrees of freedom, as achieving the target SOC has a higher priority than selecting the most efficient operating points. A longer horizon, on the other hand, is more susceptible to inaccuracies and target/actual deviations, as these accumulate over the entire horizon and therefore lead to a greater deviation with a longer horizon. Nevertheless, it makes sense to define a sufficiently long horizon that makes it possible to replace the operating states that are supposedly beneficial for conventional strategies with more efficient ones in the future. For this reason, the strategy is preceded by a controller that examines the entire horizon ahead for its energy recovery potential and shortens it to a suitable horizon length for the predictive operating strategy. In addition to the load requirement due to the driving task, the operating strategy also takes into account the energy consumption by auxiliary units and the strategy reduces frequent switching on/off of the combustion engine via a state change cost variable. The optimum torque distribution calculated for the considered prediction horizon is transferred to the vehicle model as target values for the current time step and the actual values of it are returned as a new starting value in the next time step.

5.3 Influence of the velocity prediction

The velocity and altitude profiles of the previously via GPS retrieved route is shown in Figure 13. The route covers approx. 64 km and was driven in around 1 ½ hours. Although the route does not meet the requirements for a valid RDE drive, the varying operating conditions due to the frequent alternation between cross-country driving at speeds of up to 100 km/h and inner-city traffic at speeds of max. 50 km/h nevertheless provide good representative conditions for testing the energy-saving potential.

The following results show the influence of the velocity prediction on the shown route. The simulation of the route is done with the basic strategy in comparison to predictive strategy with the perfect and the calculated prediction. Shown in Figure 14 on the top are the ICE states.

The colored areas show the timestamps at which the ICE is turned on, which means every time the color changes, an ICE state change has happened. Since one of the functionalities of the predictive strategy is an increase of the driving comfort the simulations with the predictive function reduce the number of total ICE state changes over the duration of the entire route. There are some differences between the simulation with perfect and calculated prediction, but they both show substantially less ICE state changes then the base strategy due to the prediction.

The remaining figures display the difference between the base and the predictive strategies quite clearly. It can already be seen at the beginning of the route that the predictive strategies reduces the many short engine starts in favor of fewer engine starts with longer runtime, whereby it repeatedly builds up a SOC buffer that is subsequently used at the less favorable operating points. Looking at the activation times in the context of the speed profile, it can be seen that in the tests with the prediction, the combustion engine is rarely switched on during the low vehicle velocities as described above, as the efficiency levels are unfavorable there and the engine can often only be switched on for a short time. This results in a consumption advantage over the basic strategy right from the very beginning. The behavior of the strategies before the long downhill run, which begins at approx. 2700 s, is also decisive. The predictive strategy discharges the battery as far as possible to approx. 45% in the preceding uphill section in order to then recover energy through recuperation. The basic strategy arrives at the top of the hill with a higher state of charge of approx. 57 %, which means that the battery is fully charged to 100 % before the end of the descent and the remaining recuperation potential cannot be used. The energy is then lost through mechanical braking.

Overall, the activation times and the operating point selection in the simulation with calculated prediction behave very similarly to the simulation with perfect prediction, which means that the torque of the combustion engine is selected almost identically in some areas. However, the deviations between prediction and simulation also result in time periods in which the combustion engine is operated with lower efficiency. This results in a reduction in efficiency with only a 12.18% reduction in the consumption, as can be seen in Table 2, compared to the perfect prediction which resulted in a consumption reduction of 14.4%. Also, the overall engine state changes were reduced from 110 to 37 and 30 for the different predictions, which is a very significant improvement of the driving comfort.

Table 2: Simulation results for the three different operating strategies on the investigated route [8]

Trial	Fuel consumption	State changes	Consumption reduction
Trial	L/100 km	#	%
Base strategy	5,221	110	0
Perfect prediction	4,564	37	-14.4
Calculated prediction	4,654	30	-12.18

These results confirm that the reduction in consumption of the predictive strategy varies depending on the quality of the speed prediction, in this case the perfect prediction and the basic prediction. Therefore, the proper prediction of the velocity profile is essential for the successful reduction of the fuel consumption with the predictive operating strategy.

Predictive operating strategies are of high relevance for the automotive industry in order to meet multiple and often conflicting requirements and are gaining in importance due to technological development and changing boundary conditions. In the following the relevance of predictive operating strategies is explained from the perspective of an automotive OEM (Original Equipment Manufacturer) as well as shown by means of a practical example.

The automotive industry faces an increasing complexity regarding technology and boundary conditions leading to multidimensional requirements in particular due to customers, regulations and technology. There has been a technological progress during the last years that exceeds the normal speed of development within the automotive industry. This is especially seen in the research and development of powertrains and software including the areas of connectivity and E/E architecture. Concerning automotive powertrains and operating strategies this development progress not only offers new opportunities, but also increases the complexity. With the diversification of the powertrain portfolio, introducing hybrid powertrains and battery electric vehicles, there are new opportunities to increase the efficiency and to reduce the total cost of ownership as well as the environmental impact. Customer expectations regarding connectivity and the autonomous and intelligent behavior of their cars are increasing. Furthermore, regulations have been introduced setting higher standards and restrictions that influence powertrain development and operating strategies. For example zero and near-zero emission zones have been discussed intensively for cities with high levels of air pollution. The discussion included the question if Plug-in-Hybrid-Vehicles (PHEV) need to be capable to operate in these specific zones producing zero tailpipe emissions. Legal restrictions like this have a huge impact on the operating strategy of a PHEV powertrain and require predictive strategies.

In order to meet these multidimensional and fast developing requirements it is essential to introduce intelligent and predictive solutions that benefit from the fast progress of connectivity, the availability of real-time data and opportunities that are offered by hybridization.

One practical example regarding predictive operating strategies is realized in Plug-in-Hybrid-Vehicles in order to be able to drive with zero tailpipe emissions in cities or specific areas, to increase efficiency and thus reduce the environmental impact and to optimize the charging strategy. At the same time customer experience is improved by reducing the cost of use, enhancing the electrical driving experience and relieving the customer of the choice of operating mode.

The objectives of the predictive operating strategy for Plug-in-Hybrid-Vehicles regarding operating modes are to drive fully electric in cities, to use longer distances outside the city to charge the battery and to reach your destination with an almost empty battery in order to recharge externally. This strategy allows an optimized usage of electrical energy and of powertrain components as the battery. It is realized by the regulation of the drive and recuperation behavior along the entire route planned by the driver. The data required for prediction is derived from navigation data including attributes of the routes and respective roads, live traffic information from online services, from the predictive efficiency assistant and from vehicle sensors. Input data is for example the distance to the target, speed limits, characterization of roads, road gradients, emission zones and live traffic information. An activated route guidance via the navigation system is required for the predictive operation strategy. Using this data a rough planning for the entire route and a detailed planning for the next kilometers ahead is calculated. Based on the planning the driver receives instructions when to take his foot off the accelerator. Furthermore, a predictive recuperation is initiated.

The realization of the predictive operating strategy in the functional architecture indicates the increasing complexity as well as the opportunities through technological progress. The function requires an interaction of the driver information system, the infotainment system, the data management system with internal and external data and the system responsible for driving and energy management. Moreover, it is necessary for the system to evaluate the compliance with market requirements before activating the predictive operating strategy. The functional architecture as seen before in Figure 12 contains the interaction of the driver with the navigation system and the selection of the drive mode as initial status. The predictive operating strategy gets information from various systems and defines the driving mode and the charging and respective energy management based on this information. The information about the predictive strategy is returned to the driver via the display.

This work presented a predictive energy management strategy for P2 hybrid electric vehicles. The method combines map- and driver-informed velocity forecasting with a P-ECMS. It selects operating points while respecting SOC and component constraints. A supervisory layer adapts the horizon length to maintain robustness under uncertainty.

On an approximately 64 km mixed route with a duration of about 1.5 hours, the predictive strategies delivered clear benefits compared to a rule-based baseline. Fuel consumption decreased by 12.18% with the calculated prediction and by 14.4% with perfect prediction. Engine state changes dropped from 110 to 30 and 37. Anticipatory battery use before extended downhill segments enabled increased recuperation and contributed to the observed gains in efficiency and drivability.

The magnitude of the benefit depends on the quality of the velocity prediction. Future work will refine elevation and curvature processing and improve data fusion to reduce artifacts. We will tune horizon length and cost weights to cope with traffic variability. We will also extend validation beyond controlled test cycles and hardware-in-the-loop setups, and include policy-driven constraints such as zero-emission zones in the objective function to strengthen applicability in real-world operation.

API	Application Programming Interface
CD	Charge-Depleting strategy
CS	Charge-Sustaining strategy
DOF	Degrees of Freedom
ECMS	Equivalent Consumption Minimization Strategy
ECU	Engine Control Unit
EM	Electric Motor
E/E	Electrical/Electronic architecture
FHEV	Full Hybrid Electric Vehicle
GPX	GPS Exchange Format
GPS	Global Positioning System
GUI	Graphical User Interface
HEV	Hybrid Electric Vehicle
HiL	Hardware-in-the-Loop
ICE	Internal Combustion Engine
INS	Inertial Navigation System
JSON	JavaScript Object Notation
OEM	Original Equipment Manufacturer
P-ECMS	Predictive ECMS
PHEV	Plug-in Hybrid Electric Vehicle
PMP	Pontryagin Minimum Principle
RDE	Real Driving Emissions
SOC	State of Charge
WLTC	Worldwide harmonized Light vehicles Test Cycle

D. Görke, Untersuchungen zur kraftstoffoptimalen Betriebsweise von Parallelhybridfahrzeugen und darauf basierende Auslegung regelbasierter Betriebsstrategien (Wissenschaftliche Reihe Fahrzeugtechnik Universität Stuttgart). Wiesbaden: Springer Vieweg, 2016.
R. S. Vadamalu, D. Buch, H. Xiao, and C. Beidl, "Energy management of hybrid electric powertrain using predictive trajectory planning based on direct optimal control," IFAC-PapersOnLine, vol. 48, no. 25, pp. 236–241, 2015, doi: 10.1016/j.ifacol.2015.11.093.
L. Serrao, S. Onori, and G. Rizzoni, "ECMS as a realization of Pontryagin's minimum principle for HEV control," in 2009 American Control Conference, St. Louis, MO, USA, 2009, pp. 3964–3969, doi: 10.1109/ACC.2009.5160628.
A. Sciarretta and L. Guzzella, "Control of hybrid electric vehicles," IEEE control systems magazine, vol. 27, no. 2, pp. 60–70, 2007.
R. Beck, A. Bollig, and D. Abel, "Comparison of two real-time predictive strategies for the optimal energy management of a hybrid electric vehicle," Oil & Gas Science and Technology-Revue de l’IFP, vol. 62, no. 4, pp. 635–643, 2007.
J. T. Betts, "Survey of numerical methods for trajectory optimization," Journal of guidance, control, and dynamics, vol. 21, no. 2, pp. 193–207, 1998.
C. Riegelbeck, A. Stalp, D. Schade, and C. Beidl, "Development of a Prediction Module for a Hybrid Operating Strategy Using Geo-Data," in 22. Internationales Stuttgarter Symposium (Proceedings), M. Bargende, H.-C. Reuss, and A. Wagner, Eds., Wiesbaden: Springer Fachmedien Wiesbaden, 2022, pp. 114–123.
C. Riegelbeck, Effizienzsteigerung des Hybrid-Antriebsstrangs durch Integration einer prädiktiven online-optimierenden Betriebsstrategie (Schriftenreihe des Instituts für Verbrennungskraftmaschinen und Fahrzeugantriebe 31). Düren: Shaker Verlag, 2025.
R. S. Vadamalu and C. Beidl, "Online Optimization based Predictive Energy Management Functionality of Plug-In Hybrid Powertrain using Trajectory Planning Methods," in SAE Technical Paper Series, 2017, doi: 10.4271/2017-01-1254.

No competing interests reported.

Potential analysis of a predictive energy management strategy designed to increase the efficiency of the powertrain in a hybrid vehicle with use of online available route information

Status:

Version 1

Abstract

Figures

1 Introduction

2 Optimization approach

3 Simulation setup

3.1 Necessary Data set

Vehicle Dependency:

Route Dependency:

Driver Dependency:

3.2 Vehicle model

3.3 Online data request

3.4 Data-preprocessing

4 Driving strategy

4.1 Driver modelling

4.2 Calculation of the speed profile

4.3 Optimization of the speed profile

5 Operating Strategy

5.1 Input variables

5.2 Functionality

5.3 Influence of the velocity prediction

6 Practical application

7 Conclusion

Abbreviations

References

Additional Declarations

Status:

Version 1