| Abstract | Utična hibridna električna vozila (PHEV) predstavljaju ključnu prijelaznu tehnologiju prema
punoj elektrifikaciji cestovnog transporta. Zbog kompleksnosti PHEV pogona i izraženih
zahtjeva na kvalitetu upravljanja, tipično se koristi relativno složena, optimalna strategija
upravljanja, čija se sinteza uobičajeno zasniva na kvazistatičkom (tzv. unazadnom) modelu
vozila. U radu se predlaže proširenje unazadnog modela PHEV-a paralelne konfiguracije
podmodelom dinamičkih gubitaka tijekom pokretanja motora s unutarnjim izgaranjem te
promjena stupnja prijenosa automatizirane manualne transmisije. Proširenjem se model po
točnost približava dinamičkom (tzv. unaprijednom) modelu, uz zadržavanje visoke računalne
učinkovitosti svojstvene standardnom unazadnom modelu. Potom se provodi optimiranje
upravljačkih varijabli proširenog unazadnog modela PHEV-a korištenjem algoritma
dinamičkog programiranja (DP). Rezultati optimiranja koriste se za sintezu i provjeru
optimalne strategije upravljanja tokovima snage pogona PHEV-a, koja uzima u obzir dinamičke
gubitke pogona i zasniva se na minimizaciji ekvivalentne potrošnje goriva. Strategija
upravljanja proširuje se algoritmom generiranja optimalne referentne trajektorije stanja
napunjenosti baterije uzimajući u obzir promjenjivi nagib ceste i prisustvo zona s niskim
emisijama. Dodatno su projektirane adaptivna i modelsko prediktivna strategija upravljanja
tokovima snage. Adaptacija parametara upravljačke strategije temelji se na kvadratnom
regresijskom modelu u funkciji značajki voznih ciklusa koje se izračunavaju na pomičnom
horizontu u neposrednoj prošlosti. Regresijski model parametrira se na temelju identifikacije
vrijednosti adaptiranih parametara iz rezultata DP optimiranja provedenog na karakterističnim
voznim mikrociklusima. Modelsko prediktivno upravljanje zasniva se na DP optimiranju
upravljačkih varijabli pogona na pomičnom horizontu u budućnosti, proširenom unazadnom
predikcijskom modelu i regresijskom modelu potrošnje goriva na preostalom dijelu voznog
ciklusa. U takvoj formulaciji, izbjegavaju se težinski faktor ciljne funkcije koji bi bili osjetljivi
na karakteristike voznih ciklusa. Predložene strategije upravljanja sustavno su simulacijski
provjerene i uspoređene s DP-optimalnim referentnim mjerilom kao i međusobno,
kvantificirajući pritom poboljšanja postignuta primjenom adaptivnih i prediktivnih strategija.
Ključne riječi: hibridna utična električna vozila, modeliranje, optimalno upravljanje,
minimizacija ekvivalentne potrošnje goriva, dinamičko programiranje, adaptivna upravljanje,
modelsko prediktivno upravljanje, planiranje reference stanja napunjenosti baterije. |
| Abstract (english) | Plug-In Hybrid Electric Vehicles (PHEV) represent a key transitional technology towards a
fully electrified road transportation system. Due to the complexity of PHEV powertrains and
strict control performance requirements, a relatively complex, optimal control strategy is
typically used, whose design is usually based on quasi-static (so-called backward) vehicle
model. The thesis proposes an extension of a parallel PHEV backward model with sub-models
of dynamic losses occurring during the internal combustion engine start-up and automated
manual transmission gear shifts. By introducing these sub-models, the accuracy of the backward
model is found to approach that of the dynamic (so-called forward) model, while maintaining
the high computational efficiency of the standard backward model. Next, PHEV control
variables optimization is conducted by using a dynamic programming (DP) algorithm and the
extended backward model. The optimization results are used to design and verify an optimal
PHEV power flow control strategy, which takes into account the dynamic powertrain losses
and is based on the equivalent consumption minimisation strategy (ECMS). The control
strategy is extended by an algorithms that generates optimal reference trajectory of battery state
of charge (SoC) for the PHEV blended operating regime, and which takes into account the
effects of varying road slope and low emission zones. Furthermore, adaptive and modelpredictive power flow management strategies are designed. Adaptation of key control strategies
parameters is based on a quadratic regression model inputted by characteristic driving cycle
features, which are calculated on the moving horizon in the immediate past. The regression
model is parameterized based on input/output data extracted from the DP optimization results
obtained for characteristic driving micro-cycles. The model predictive control (MPC) strategy
is based on the DP control variable optimization on a receding prediction horizon, an extended
backward prediction model, and a regression model of the fuel consumption on the remaining
part of driving cycle. Such a formulation does not include cost function weighting factors,
which would otherwise be sensitive to the driving cycle features. The proposed control
strategies are systematically verified by computer simulation, and compared with each other
and the DP- benchmark, while quantifying improvements achieved by applying the adaptive
and predictive control strategies.
Through nine chapters of this work, which include the introduction and conclusion chapters,
backward and forward PHEV simulation models are described, dynamic programming-based control variables optimization results are presented and analysed, optimal, adaptive, and model
predictive power flow control strategies are proposed, and finally a novel method of synthesis
of battery state-of-charge reference trajectory is presented for the blended operating regime and
the general case of varying road grade and low emission zone presence. A brief overview of
each chapter is given below.
Chapter 1: Introduction. Outlines the motivation for the conducted research and gives a
literature review of the relevant topics of the thesis, which include PHEV modelling and optimal
power flow control, adaptive and model predictive control, and battery state-of-charge
reference trajectory planning.
Chapter 2: Backward-looking model of a Plug-in Hybrid Electric Vehicle. The powertrain of
a PHEV given in P2 parallel configuration is described along with basic technical parameters.
PHEV power flows and operating modes are briefly analysed. A basic backward-looking
(BWD) vehicle model is described, which neglects powertrain dynamics effects including
dynamic power losses. The required wheel torque and speed are determined from the driving
cycle data and the vehicle longitudinal dynamics equation, and are used in a backward manner
to calculate the propulsion machines' and battery variables.
Chapter 3: Forward-looking and extended backward model. A more accurate forward-looking
(FWD) PHEV model is presented, which takes into account the dominant powertrain dynamics
effects, evaluated in the direction from propulsion machines to wheels. The corresponding
simulation model is implemented in Simcenter's Amesim software environment, and it includes
a low-level control strategy with actuator dynamics. The low-level control is aimed at achieving
the operating points set by the high-level control strategy, which is done through appropriate
coordination of the main clutch, propulsion machines, automated manual transmission, and
mechanical brakes. Based on the energy balance analysis derived from the FWD model
simulation results, an extended BWD (EXT-BWD) model is proposed, which includes the
engine start-up and gear shift transient losses. These transient losses include power losses due
to main clutch and synchronizers slippage, engine start-up losses, and electric machine-based
dog clutch synchronization losses. The EXT-BWD model is validated by comparing its fuel
consumption and battery state of charge (SoC) time responses with those obtained by the more
accurate FWD model, where DP optimal control variables are used as inputs of both models.
Chapter 4: Control variable optimisation. A PHEV powertrain control variables optimization
method is proposed, which is based on the dynamic programming (DP) algorithm adjusted for
the additional state variables introduced by the EXT-BWD model. With the aim of improving
computing efficiency, the DP algorithm is implemented in the C++ programming language,
while the obtained results are processed and analysed in the Matlab/Simulink environment. The
DP algorithm cost function reflects the conflicting criteria of reducing fuel consumption
throughout the driving cycle while maintaining the target value of SoC at the end of driving
cycle. In addition to the hard constraints on control variables, soft constraints are implemented
to limit the powertrain variables within defined physical limits. Optimization is carried out for
the cases of charge sustaining (CS) and blended (BLND) operating modes. The DP optimal
results obtained by using the BWD and EXT-BWD models are compared with each other, in
order to further highlight the differences in fuel consumption predictions predicted by the two
models. In addition, the optimal SoC trajectories over travelled distance were analysed for
various driving cycles, including the special cases of varying road grade and low emission zone
presence.
Chapter 5: High-level control strategy. A high-level power flow control strategy of a parallel
PHEV is proposed, which combines a rule-based (RB) controller with an equivalent
consumption minimization strategy (ECMS). The RB controller regulates the battery SoC,
calculates the transmission input power demand, and determines the engine on/off state. The
ECMS performs instantaneous optimization (i.e., optimal allocation) of the control variables,
which include the engine torque and the transmission gear ratio. Two versions of the control
strategy are considered, depending on whether the BWD or EXT-BWD model is used within
the ECMS strategy. With the aim of reducing the number of gear shifts in the case of BWD
model-based strategy, a gear shift delay algorithm is introduced through expansion of the
ECMS cost function. The two versions of control strategy are verified by simulation and
compared by using the EXT-BWD and FWD powertrain models in the CS regime and for
several driving cycles.
Chapter 6: Adaptive control strategy. The insights obtained from the control variable
optimisation results given in the fourth chapter are used to establish an adaptation mechanism
of the control strategy proposed in the fifth chapter. The DP optimal results are collected on
micro-cycles that are obtained by segmenting the certification cycles, in order to reflect the
influence of specific local statistical features of the driving cycle (e.g., city driving, cruising, aggressive driving). The values of three high-level control parameters are identified from the
DP optimal results, and quadratic, linear-in-parameters regression models are formed and fed
by the micro-cycle features calculated on a moving history time horizon. The adaptive
management strategy is compared with the non-adaptive strategy described in the fifth chapter
by means of simulation verification using the EXT-BWD and FWD models.
Chapter 7: Model predictive control. A synthesis of the high-level control strategy based on
the MPC approach is proposed. The cost function of the proposed MPC strategy consists of the
fuel consumption on the prediction horizon and the predicted optimal remaining fuel
consumption from the end of prediction horizon to the end of driving cycle. The remaining fuel
consumption is described by a regression model fed by the remaining trip distance, SoC value
at the end of prediction horizon, and optionally the mean value and standard deviation of the
vehicle speed profile on the remaining section of driving cycle. The regression model is
established and parametrised on the basis of DP-optimal results for several driving cycles. In
this way, MPC does not use a multi-criteria objective function, i.e. the one that would combine
fuel consumption and SoC error penalty at the end of the prediction horizon. This avoids
adjusting the weighting coefficients of the objective function, which would otherwise be
sensitive to the characteristics of driving cycles. At the same time, the SoC at the end of the
driving cycle is implicitly accounted for, instead of explicitly (and thus generally sub-optimally)
via the SoC at the end of the prediction horizon. The control variable optimization on the
moving prediction horizon is performed by the DP algorithm, and the proposed strategy is
implemented and verified through simulation by using the EXT-BWD PHEV model.
Chapter 8: Battery state-of-charge reference planning for blending operating regime. Two
methods of SoC reference profile synthesis with respect to travelled distance are proposed based
on the insights obtained from the DP optimal SoC responses analysis presented in the fourth
chapter. Special emphasis is on the cases of varying road grade and low emission zone presence.
The offline synthesis method sets the SoC reference profile before the trip based on the
knowledge or a sufficiently precise estimation of the trip distance, the average vehicle speed or
the trip duration, and the road grade profile along the trip. Therefore, the proposed method is
suitable for vehicles operating on pre-known and repetitive routes, with the historical vehicle
tracking data available (e.g., city buses and delivery vehicles). The online SoC reference
synthesis method relies on the recorded driving data during the trip to determine the upcoming
SoC reference profile, and it only requires knowledge or a sufficiently precise estimate of the driving cycle total distance. Validation of the proposed synthesis methods is performed for
RB+ECMS and EXT-BWD model, and the results are compared with those obtained by using
the linear SoC reference profile.
Chapter 9: Conclusion. Concluding remarks are given and possible future work directions are
discussed. The main contributions of the doctoral thesis are listed at the end of the chapter, and
they include: (i) computationally efficient backward-looking PHEV model, which accounts for
powertrain transient loss effects described by analytical models; (ii) optimal PHEV power flow
control strategy, which takes into account the powertrain transient losses and, thus further
reducing the fuel consumption, and improving the drivability and drive comfort; (iii) adaptive
PHEV power flow control strategy, which adapts the optimal control strategy parameters with
respect to driving cycle features that are identified on a moving horizon in the immediate past,
(iv) model predictive PHEV power flow control strategy, which realizes the final battery SoC
condition in a robust way involving a regression model of the remaining fuel consumption from
the end of prediction horizon to the end of driving cycle; and (v) method for SoC reference
trajectory synthesis in PHEV blended operating regime based on the available driving cycle
features. |
