Professional Documents
Culture Documents
Lin 2019
Lin 2019
com
Available online at www.sciencedirect.com
ScienceDirect
ScienceDirect
Energy
Available
Available Procedia
online
online 00 (2018) 000–000
atatwww.sciencedirect.com
www.sciencedirect.com
Energy Procedia 00 (2018) 000–000 www.elsevier.com/locate/procedia
ScienceDirect
ScienceDirect
www.elsevier.com/locate/procedia
Energy
EnergyProcedia 158
Procedia 00(2019)
(2017)2744–2749
000–000
www.elsevier.com/locate/procedia
10th
International Conference on Applied Energy (ICAE2018), 22-25 August 2018, Hong Kong,
th
10 International Conference on Applied Energy
China(ICAE2018), 22-25 August 2018, Hong Kong,
China
Energy Management for a Dual-motor Coupling Propulsion Electric
Energy Management
The 15th Internationalfor a Dual-motor
Symposium on District Coupling Heating Propulsion
and Cooling Electric
Bus based on Model Predictive Control
Bus based on Model Predictive Control
Assessing the feasibility
Cheng Lina,b , Mingjie Zhaoa,b,
of using the heat
*, Hong Pana,b
demand-outdoor
, Shuai Shaoa,b
temperature Chengfunction
Lin , Mingjie
a,b
for a long-term
Zhao *, Hong
a,b,
district
Pan , Shuai
a,b
heat demand
Shao a,b
forecast
National Engineering Laboratory for Electric Vehicles, School of Mechanical Engineering, Beijing Institute of Techonology,
a
1. Introduction
The traditional single motor integrated with automated mechanical transmission (AMT) system has low potential
in energy optimization and dynamic characteristics due to the only one propulsion unit. Therefore, more and more
dual-motor coupling propulsion (DMCP) systems have been researched to improve the comprehensive performance
of whole vehicle [1]. One of the most practical architectures utilizing two motors is coaxial coupling propulsion system
and energy management strategy becomes an essential factor for electric buses equipped with DMCP [2]. As this issue
can be equivalent to an optimization problem with proper constraints, many classical and intelligent optimization
algorithms have been adopted to solve it considering the specific application conditions [3, 4].
With considerable robust property and high calculating efficiency, rule-based control strategy has been widely used
to handle energy management problems. However, it can hardly lead to the optimal operation, then the global
optimization-based energy management strategy has been explored to improve the optimized results [5]. Dynamic
programming (DP) is one of the most effective algorithms which can obtain the global optimal results based on
Bellman principle of optimality. However, it needs the total information of driving cycles in advance that hinder its
online application potential [6]. Another kind of instantaneous optimization method such as model predictive control
(MPC) can find the finite local optimal control sequences with a length of future input [7, 8]. To take advantage of
MPC-based strategy, Markov chain model, which is good at dealing with time series prediction problems, is often
adopted as the prediction tool to estimate the possible required velocities in the next period of time. However, the
Markov chain is a kind of probability-based method, it may head for the wrong trend according to the presetting
statistic data, i.e. even if the vehicle is in an acceleration phase the Markov chain model may also forecast a lower
velocity variation due to the maximum probability. So a relevant velocity trend should be added as supplementary
information to assist the Markov process in leading to a precise result.
In this paper, a practical dual-motor coupling propulsion electric bus (DMCEB) and MPC-based energy
management strategy utilizing a novel piecewise Markov chain method are proposed. In section 2, the basic
information of DMCEB is provided. DP method is implemented to seek the optimal results in section 3. Section 4
introduces a novel piecewise Markov chain model based on acceleration sign prediction. Model predictive control
method is utilized in section 5 to handle the energy management problem in a typical driving cycle. The simulation
results are illustrated, compared and discussed in Section 6 before conclusions drawn in the final section.
The target bus studied in this paper is a dual-motor coaxial coupling propulsion electric bus of which the
architecture of the powertrain is shown in Fig. 1 (a). Here an auxiliary motor (AM) equipped with a two-speed
transmission and a traction motor (TM) connected directly to the main reducer can drive the vehicle in a torque-
coupling way. The efficiency models of main powertrain components are determined and established by experiments
in look-up table form. Rint model is set to describe the battery. The bus is a common 12-meter city bus with a gross
mass of 18,000 kg. The route discussed in the next is the Chinese typical city bus drive cycle (CCBC) and is illustrated
in Fig. 1 (b), with a driving distance of around 5.89 km and a duration of 1314s.
CAN
bus 70
VCU Electric cable BMS
60
Integrated
TCU Battery 50
MCU
40
Drive axle 30
20
AM TM
10
2-speed Dual-motor coupling
0
transmission
0 200 400 600 800 1000 1200 1400
propulsion system
Time (s)
(a) (b)
Fig. 1. (a) Powertrain architecture of the dual-motor coupling propulsion bus; (b) The Chinese typical city bus drive cycle (CCBC)
2746 Cheng Lin et al. / Energy Procedia 158 (2019) 2744–2749
Author name / Energy Procedia 00 (2018) 000–000 3
Dynamic programming is good at finding the optimal control law at each step to achieve the global energy-saving
results effectively. It firstly calculates the values of objective function based on the aforementioned models and
attaches all inputs to the state and control variables from the end of the sequence. Then the global optimization problem
can be transformed into a series of secondary issues backward from the terminal step and can be solved forward to
determine the optimal control trace. To simplify the control variables as possible, only torque of TM, torque split ratio
(TSR) and transmission shift command are selected. To avoid frequent gear shift phenomenon, a penalty term φ is
added. So the optimization target is to find the control law to minimize the total energy cost considering the frequency
of shift and the expressions are as follows:
xk 1 f ( xk , uk )
x [ SOC , gear ]
k k k
u
k [TSR k , shift k , TTM ]
(1)
N 1 N 1
J L( xk , uk ) ( ECk (uk , k ) shiftk )
k 0 k 0
To solve the above recursive Eq. (1) backward, the state ݔ and control variables ݑ should be discretized into
finite grids (Table 1) The shift command can only be upshift, sustain and downshift and the gear should be chosen
from 1 and 2, so their values are integer and fixed. Interpolation method is also adopted to evaluate the values if ݔାଵ
does not locate on the grids. To accelerate the computational process, an approximately grid resolution is determined.
The idea of Markov chain model is mainly based on a hypothesis indicating that each state in a stochastic time
series process will only related to that of the last n states, which is so called n order Markov process. That means the
following state will be influenced and determined by the past n data. Taking both the driving cycle characteristics and
calculation burden into consideration, first order Markov process is suitable to be employed to forecast the velocity
changes in the short future. To decrease the impact of false velocity trend estimation, the variation sequences of
acceleration signs will be judged in advance. Thus the acceleration and braking processes can be separated and two
independent state transition probability matrices can be determined to predict the next velocity.
There are plenty of methods to deal with the time series prediction problems. The choice criterion mainly depends
on the stationary property of the data sequence. It is obvious that the velocity series is non-stationary time series which
means its unconditional joint probability distribution changes a lot when shifted in time. However, the acceleration
series can be regarded as a trend stationary time series as its mean value is around zero. So auto-regressive moving
average (ARMA) method can be utilized to forecast the acceleration variation. Besides, as we do not care the accurate
value but only focus on the sign of acceleration, a lower order ARMA model can be accepted. Here the order of AR
Cheng Lin et al. / Energy Procedia 158 (2019) 2744–2749 2747
4 Author name / Energy Procedia 00 (2018) 000–000
process and MA process are both set to fifth and the prediction results can be seen in Fig. 2. Though the value of
forecast acceleration is not precise the trend is similar to the real value. Hence the sign prediction is quite suitable and
can be set as supplementary information to assist the velocity prediction in the Markov process.
Fig. 2. (a) The acceleration value prediction in future 5s of each step; (b) the acceleration sign prediction in future 5s of
each step
(a) (b)
Fig. 3. (a) The state transition probability matrix of accelerating process; (b) the state transition probability matrix of braking process
The key factor of Markov chain method is the state transition probability matrix and its judging criteria when
deducing the next velocity is the specific value with the highest probability. With the help of acceleration sign, the
traditional matrix can be divided into two parts to improve the total prediction accuracy as shown in Fig. 3. Since the
Markov here is within first order, the next predictive velocity can be regarded as a new step value and after repeating
the Markov process the required prediction series can be obtained. As illustrated in Fig. 4, the results based on
conventional Markov chain method are not satisfied and its total RMSE is about 2.74km/h, while that based on
piecewise Markov chain method, with a small total RMSE as only 1.4119 km/h, seems quite smooth and fit the real
velocity well. Here the proposed method can improve the accuracy by 48.5%.
2748 Cheng Lin et al. / Energy Procedia 158 (2019) 2744–2749
Author name / Energy Procedia 00 (2018) 000–000 5
Fig. 4. (a) Velocity prediction based on traditional Markov; (b) prediction RMSE based on traditional Markov; (c) velocity
prediction based on piecewise Markov; (d) prediction RMSE based on piecewise Markov
As one of the most efficient advanced control algorithms, Model predictive Control (MPC) mainly consists of three
parts: prediction model, rolling optimization and feedback recalibration. In a certain step, the current states are
regarded as the initial values in the optimization process. The prediction part can provide the necessary information
in a definite time series and a determined optimization method, like DP, can find the optimal results and control
sequence within the forecasting steps considering the rational constraints. Then only the first control variable will act
on the vehicle and transit the state variables into next step. By repeating these procedures, we can get the local optimal
control law gradually. The basic logic flow of such energy management strategy can be summarized as shown in Fig.
5. Here the constraints and references such as gear shift frequency, SOC change, disturbance from environment etc.
can be considered and adjusted online. Moreover, under the MPC control logic, some other conventional offline global
optimization methods can be used to solve the transformed local optimization problem.
Acceleration
Past Driving Cycle Real Velocity Record & Update
Prediction
Information
based on ARMA
Estimated SOC, Gear, etc.
Optimization of Observer
Y Transition Matrix Future Conditions
Acc>0? based on Dynamic
of Acc Process Controlled
Programming Vehicle System Outputs
N Velocity
Transition Matrix Forecast
Inputs
of Brk Process
Reference Driver
Fig. 5. The basic logic flow of the energy management strategy based on MPC
Fig. 6 shows the SOC trace variation results of the strategies based on DP algorithm and the aforementioned MPC
method in the same cycle. The SOC is initialized as 90.0% and the variations of SOC are both gradually decreasing
with obvious fluctuations. The DMCEB consumed only 80.24kWh electricity in CCBC cycle per 100km under the
DP-based strategy whereas 85.75kWh/100km is necessary that under the MPC-based strategy, namely the MPC
Cheng Lin et al. / Energy Procedia 158 (2019) 2744–2749 2749
6 Author name / Energy Procedia 00 (2018) 000–000
energy management is only 6.8% worse than that based on DP. However, as shown in Table 2, the MPC-based strategy
can significantly improve the energy-saving potential comparing with the preliminary rule-based strategy.
0.89
0.885
0.88
0.875
0.87
0 200 400 600 800 1000 1200 1400
Time (s)
7. Conclusions
A novel piecewise Markov-based velocity prediction method adopting acceleration sign is proposed which can
decrease the forecasting RMSE by 48.5%. Energy management strategies based on DP and MPC are conducted
respectively. The results of the latter is only 6.8% worse than that of the former, and can reduce the energy
consumption by 21.42% comparing with a preliminary rule-based strategy. The proposed strategy is quite suitable for
electric city buses of which the route is steady and easy to be predicted accurately.
Acknowledgements
This work was supported by the National Key Technology Research and Development Program of China
(2017YFB0103801). Any opinions expressed in this paper are solely those of the authors and do not represent those
of the sponsors.
References
[1] Hu M, Chen S, Zeng J. Control Strategy for the Mode Switch of a Novel Dual-Motor Coupling Powertrain[J]. IEEE Transactions on
Vehicular Technology, 2018,67(3):2001-2013.
[2] Zhao M, Shi J, Lin C, et al. Application-Oriented Optimal Shift Schedule Extraction for a Dual-Motor Electric Bus with Automated Manual
Transmission[J]. Energies, 2018,11(2):325.
[3] M. Sabri M F, Danapalasingam K A, Rahmat M F. A review on hybrid electric vehicles architecture and energy management strategies[J].
Renewable and Sustainable Energy Reviews, 2016,53:1433-1442.
[4] Serrao L, Onori S, Rizzoni G. A Comparative Analysis of Energy Management Strategies for Hybrid Electric Vehicles[J]. Journal of
Dynamic Systems, Measurement, and Control, 2011,133(3):31012.
[5] H. B, S. A, Y. C. A rule-based energy management strategy for Plug-in Hybrid Electric Vehicle (PHEV): 2009 American Control Conference,
2009[C].2009 10-12 June 2009.
[6] Torres J L, Gonzalez R, Gimenez A, et al. Energy management strategy for plug-in hybrid electric vehicles. A comparative study[J]. Applied
Energy, 2014,113:816-824.
[7] S. J M, H. K F, D. S C, et al. A Stochastic Optimal Control Approach for Power Management in Plug-In Hybrid Electric Vehicles[J]. IEEE
Transactions on Control Systems Technology, 2011,19(3):545-555.
[8] S. D C, D. B, A. B, et al. Stochastic MPC With Learning for Driver-Predictive Vehicle Control and its Application to HEV Energy
Management[J]. IEEE Transactions on Control Systems Technology, 2014,22(3):1018-1031.