1. Introduction
According to the report published by the International Energy Agency (IEA) in 2017, CO
emissions of transportation sector from fuel combustion reached 7.74 billion tons in 2015 [
1], of which the rail sector emitted 336 million tons, accounting for 4.34% [
2]. One major source of carbon in railway transportation is the trains using fossil fuel directly and indirectly. To minimize fossil fuel consumption in rail transportation, hydrogen is regarded as a feasible and effective solution. As a kind of clean and renewable energy source, hydrogen can greatly reduce carbon emission and the use of non-renewable energy when applied to transportation. As a typical study, Chang et al. established an energy consumption model based on traffic data from various parts of China, demonstrating that the use of hydrogen fuel cells in transportation represents one of the future development trends, especially in public transportation [
3]. The feasibility of fuel cell hybrid trains (FCHTs) had been verified in many studies [
4,
5]. Although the technical cost is relatively high at the current stage, FCHTs have some advantages over electric trains and diesel trains on railways with lower traffic density [
6]. In this case, optimizing train control and energy management of hybrid power systems is one of the main focuses in the study of cost reduction.
The optimal train speed trajectory with minimum traction energy demand exists when a train is running between two adjacent stations based on Pontryagin’s maximum principle (PMP) [
7,
8]. Based on PMP, the optimal train control modes, i.e., accelerating with maximum acceleration rate, braking with maximum braking rate, cruising, and coasting, and their optimal sequences, can be obtained for one inter-station journey. The PMP-based method obtains the optimal solution via solving the co-state function and it is referred to as the “indirect method” [
9]. Another commonly used method is the direct method, which transforms the optimal control problem into a mathematical programming problem, typically represented by pseudospectral method, mixed integer linear programming (MILP), and other mathematical programming methods. In [
10], both pseudospectral and MILP methods were proposed to optimize the train control strategies under fixed arrival time. The MILP method can easily improve the model to add other engineering constraints [
11]. In addition, dynamic programming (DP) and various heuristic algorithms are also used to locate the optimal train control strategies. DP simplifies a complex problem by decomposing it into simpler sub-problems in a recursive manner. In [
12], two heuristic algorithms, ant colony optimization and genetic algorithm, and DP were applied to search the speed trajectory, and the results show that the optimum train trajectory cannot be solved analytically, and numerical methods must be used.
While the train optimal control of FCHTs shares similar characteristics with other rail vehicles, the optimization of the speed trajectory of FCHTs can generally locate the minimum energy demand from the traction system. However, there are two important traction components of FCHT, i.e., energy storage devices (ESDs) and fuel cell, that deserve further study when the onboard energy management strategies are studied. For FCHTs, the characteristics of ESDs should be considered along with the train optimal control strategies, especially when the ESDs are charging and discharging along with the train control strategies such as braking and motoring. In [
13], Snoussi et al. used sliding mode control for energy management, which well absorbed the peak power in fuel cell trains. In the past few years, many studies have focused on the the ESDs applied to electric trains to increase the recovery of regenerative braking energy and to improve the system performances. In [
14], ESD is equipped on the electric train to increase the use of regenerative energy and stabilize the voltage, and the DP algorithm-based heuristic is designed to find the optimal speed trajectory. In [
15], Wu et al. employed the MILP to locate the optimal train control, and the results indicated that ESD has a significant influence on the speed trajectory of electric train.
On the one hand, the characteristics of ESDs play an important role in train control; on the other hand, the nonlinear characteristics of fuel cell efficiency impose further complexity on the onboard energy management and train controls. The energy management strategy determines the power distribution between fuel cell and ESDs, and has a significant impact on the total consumption of hydrogen fuel. In recent years, there have been many methods to optimize energy management strategies, such as state machine strategy, fuzzy logic control, equivalent consumption minimization strategy, etc. [
16,
17]. Current studies are conducted with the fixed power–demand [
18]. In [
19], the convexity of the specific consumption curve is applied to improve the fuel economy. Based on a suggested power–demand curve, a scalable energy management strategy is designed. In [
20], an online extremum-seeking method is used to estimate the maximum efficiency and maximum power points of the fuel cell. In [
21], a fixed speed trajectory in FCHTs is divided into four states: traction, braking, coasting, and station parking, and then the power distributions between fuel cell and the supercapacitor (SC) are distributed through a multi-mode equivalent energy consumption method. The work in [
22] extended to multiple fuel cells, and realized the power splits among multiple fuel cells by means of the equivalent fitting circle method, and optimized the power output of SC by means of equivalent energy consumption method. In [
23], Yan et al. optimized the speed trajectory with the aim of achieving the minimum energy consumption, then determined the hybrid system control strategy based on minimum hydrogen consumption. In [
24], a rule-based energy management strategy was proposed to maximize regenerative braking energy recovery. In [
25], the braking process speed trajectory was obtained based on motor characteristic curve, supercapacitor capacity, maximum acceleration, and other information to ensure that the supercapacitor can obtain more regenerative braking energy, but this method does not consider the efficiency of the fuel cell. In [
26], sequential optimization was applied to improve the potential of fuel efficiency, which develops a speed-smoothing strategy first then optimizes the battery charge based on the smoothed speed profile. However, the train control strategies impose direct impacts on the traction energy demand and are not included in the energy management process in the sequential optimization process. This may undermine the effectiveness of the optimization methods.
In the research area of hydrogen reduction for FCHTs, some recent papers proposed different methods in co-optimizing both train control strategies and onboard energy management concurrently. In [
27,
28], PMP and DP were combined to solve this problem by applying the Hamiltonian as the objective function of DP, but its application was inevitably constrained by ”the curse of dimensionality” and the boundary-value problem. In [
29], Peng et al. proposed dynamic programming for co-optimization of train driving cycles and energy management for fuel cell trains. The parallelization of DP has been proposed to reduce computation time. However, when the dimension of the state variables in the dynamic programming model increases, the calculation time of the model will increase significantly due to the characteristics of the algorithm characteristics. In [
30], Jibrin et al. solved the co-optimization of both energy management and speed trajectory by formulating a convex optimization model using convexity relaxation techniques which greatly improved the calculation efficiency, but convex optimization requires that every constraint in the model is convex, which affects the flexibility of the model. The above-mentioned state-of-the-art studies adopted DP and convex programming in dealing with the co-optimization, and they all promoted the advantages of applying co-optimization for FCHTs.
However, it can be seen from the above research that the existing research on co-optimization mainly adopts dynamic programming [
27,
28,
29] and convex optimization [
30]. Different from the existing methods of co-optimization, a new method is proposed in this paper to solve the co-optimization problem, and the hydrogen-saving mechanism of a fuel cell train is studied by using the co-optimization model. In addition, the energy-saving potential of regenerative braking process and enhancement of fuel cell efficiency during application by taking the advantage of the hybridization of both fuel cell and onboard ESDs needs further investigation. A detailed overview and comparison of the existing research and this paper are given in
Table 1.
This paper aims to develop a time-based MILP model, which has a relatively short computational time, to find the optimal speed trajectory and energy management strategy simultaneously to minimize the net hydrogen consummations. The contributions of this paper are as follows.
- (1)
A time-based co-optimization model based on MILP for FCHT is proposed to tackle the intrinsic nonlinear efficiency characteristics of fuel cell and convert the traditional energy-saving problem into a more realistic hydrogen-saving one. The train control, namely, the train speed trajectory, and energy management between two adjacent stations are optimized simultaneously to guide the autonomous driving.
- (2)
The impact of capacity of ESD on the hydrogen consumption has been studied for FCHTs, and the corresponding hydrogen-saving mechanisms are explored with the quantitative simulation results, giving an insightful analysis and discussion on the mutual influence on optimal train operation.
The remainder of the paper is organized as follows:
Section 2 proposes the mathematical model in detail, including motion model, energy flow model, piecewise linearization, and objective modeling.
Section 3 analyzes the impact of ESD capacity on hydrogen consumption and gives the comparison between co-optimization and sequential optimization under different scenarios.
Section 4 concludes the study and discusses the future research directions.
3. Results
The modeling parameters of FCHT in this paper are listed in
Table 3 with references to the design proposed in [
29]. The global optimizer used in this study to solve the MILP model is GUROBI9.0 [
35]. In this section, the optimization results of the time-based MILP model are analyzed. In this study, the supercapacitor is applied as ESD for FCHT. The energy density of the supercapacitor is in the range of 2.5∼15 kWh/t [
36], and it also has a relatively high efficiency, generally around 95% [
37]. In this paper, the energy density is selected to be 10 kWh/t, i.e., 36 MJ/kg. At the same time, the converters (such as GTO and IGBT), motors, and gear system also have energy losses. As mentioned in [
38], the efficiency of them are in the range of 98.5∼99.5%, 90∼94%, and 96∼98%, respectively. In this paper, the efficiency of the converters and gear system is incorporated in the motor efficiency, set as 90%.
Proton exchange membrane fuel cells (PEMFCs) are widely used for their stability and efficiency. A power–efficiency curve of a PEMFC stack is shown in
Figure 5. The power–efficiency model does not consider the characteristics of current and auxiliary system in detail, but only considers the relationship between fuel cell efficiency and output power. When the output power of the fuel cell is low, the efficiency is low due to the factors related to the operation of the air compressor, cooling, and humidifying system. The efficiency decreases due to the increase of current at high load demand. It is noted that since the power–efficiency relationship shows that the lowest efficiency of the fuel cell is 20% with the output power of it being 0, when the efficiency is 20%, the fuel cell can be regarded to be in the non-working status.
3.1. The Impact of ESD Capacity on Hydrogen Consumption
In this section, the effect of ESD on the equivalent hydrogen consumption of a hybrid power system is studied based on the co-optimization model. Two case studies, i.e., Case 1 and Case 2, are proposed to model the FCHT with and without ESDs, respectively. In Case 1, the capacity of ESD is assumed to be 0 MJ to indicate that the FCHT is not equipped with ESD, and ESD capacity in Case 2 is set to be 40 MJ. Comparison of results between the two cases is shown in
Table 4. The initial SOE for ESD in Case 2 is 1.
In
Figure 6a, the optimized speed trajectories are demonstrated. It can be seen that the speed trajectory varies when ESD is introduced. The acceleration of Case 1 without ESD is smaller, so a longer acceleration phase and higher coasting speed are required to meet the time constraint and this leads to more traction energy needed for Case 1.
Power splits between fuel cell and ESD are shown in
Figure 6b, where the solid line represents the optimization results when ESD capacity is 40 MJ, and the dashed line represents the results without ESD. It can be seen from this figure that the fuel cell in Case 2 outputs
for all working hours except for 150 kW for 10 to 20 s. ESD provides extra power so that fuel cell in Case 2 can work with a higher efficiency. In addition, in Case 1, the RBE is consumed directly as heat while the RBE is recovered by 11.51 MJ, which allows 12.83% equivalent hydrogen consumption to be saved in Case 2.
It can be seen that the net hydrogen consumption of the two is significantly different. The net hydrogen consumption of Case 2 can obtain a saving of 29.58% compared with Case 1. This comparison shows that the introduction of ESD can reduce net hydrogen consumption from two aspects. On the one hand, ESD assists fuel cells to avoid high-power but low-efficient working states; on the other hand, RBE can be recovered to improve the energy efficiency.
To show the detailed mechanism on how the fuel cell and ESD influence each other, the relationship between fuel cell and ESD capacity is studied for this hybrid power system. Optimization results with the varied ESD capacities by using the proposed model are shown in
Figure 7a,b. Note that the initial SOE for ESD is 1.
As can be seen from
Figure 7a, as ESD capacity increases, net hydrogen consumption decreases, and this trend is particularly evident in the low capacity, from 0 to 13 MJ. At this stage, the equivalent hydrogen consumption of ESD is 0. It can be seen from
Figure 7b that the reason is that the energy provided by ESD is equal to the energy recovered. This indicates that at low capacity, the increase in recovered regenerative braking energy can significantly reduce net hydrogen consumption.
When ESD capacity increases from 13 to 40 MJ, the net hydrogen consumption of the train decreases slowly, while the equivalent hydrogen consumption of ESD increases and the hydrogen consumption of the fuel cell decreases. From the perspective of energy flow, the hybrid power system is more inclined to use the energy in ESD due to the higher energy efficiency of ESD. This is reflected in the fact that with the increase of ESD capacity, ESD provides more energy for traction, which allows the fuel cell to work at a higher efficiency and obtain a higher average efficiency. In addition, although the net hydrogen consumption decreases slowly, it can be seen from
Figure 7b that the demanded traction energy of the train increases slowly, while the RBE also slightly decreases. This also reflects the higher energy efficiency of the hybrid power system.
It is noted that when ESD capacity is greater than 40 MJ, the net hydrogen consumption remains unchanged, and it can be seen from
Figure 7a that the fuel cell average efficiency is 60%, which is the maximum value. Since the efficiency of ESD in this model is a constant, and in this model, the maximum efficiency of the fuel cell, which is 60%, is used to equate the energy provided by ESD to hydrogen consumption, the energy conversion rate of the system reaches the highest, and the net hydrogen consumption of the system is the minimum. Therefore, when the objective is to minimize hydrogen consumption, 40 MJ can be considered as the optimal ESD capacity of the system.
3.2. Case Study on Flat Track
Section 3.2 and
Section 3.3 are the comparison between co-optimization and sequential optimization. In this section, a simple case is conducted to explore the difference between the two methods on a flat track to avoid the effect of gradient and speed limit on traction or braking power. The capacity of ESD in this case is selected as 40 MJ and its initial SOE is 1, and the travel distance and time are 10 km and 450 s. The performance comparison between the two is listed in
Table 5, where Case 2 and Case 3 are based on co-optimization and sequential optimization, respectively.
The speed trajectory and motor power are shown in
Figure 8a. The solid line and the dashed line represent the results of co-optimization and sequential optimization, respectively. It can be seen that the speed trajectory of sequential optimization can be divided into four stages, acceleration with maximum acceleration, cruising, coasting, and deceleration with maximum deceleration, which meets the optimal control theory, and it should be noted that the train of Case 3 which is based on sequential optimization accelerates to the cruise speed with the maximum traction power, while Case 2 does not.
Figure 8b represents the power output of the fuel cell and ESD. From this figure, it can be seen that ESD of Case 2 and Case 3 have reached the maximum power output in the acceleration stage, and the power difference at this stage is only reflected in the output power of the fuel cell.
In
Figure 8b, it is shown that in the acceleration stage, the fuel cell efficiency of Case 3 is 55%, but for Case 2, except for the fuel cell output power of 11 s to 22 s, which is 150 kW when the efficiency of fuel cell is 58%, the output power of the other acceleration time is 100 kW when the efficiency of fuel cell is 60%. That is to say, the fuel cell in Case 2 worked at the point of higher efficiency during the whole acceleration process. As shown in
Table 5, even though the fuel cell of Case 3 provides less traction energy, the hydrogen consumption of the fuel cell is still higher, which means that the average efficiency of the fuel cell in Case 3 is lower than the one in Case 2.
It can be seen from
Table 5 that the calculation time of sequential optimization is short and it has an advantage in terms of net energy consumption, i.e., 62.28 MJ. In the meantime, the result by co-optimization consumes 64.27 MJ of net energy. However, since the efficiency of the fuel cell is improved during the operation obtained by co-optimization, the net hydrogen consumption in the co-optimization results is lower, saving 3.1% compared to the one in Case 3. The sequential optimization first optimizes the speed trajectory with the minimum energy consumption, so the NEC is less than co-optimization. In terms of hydrogen consumption, the nonlinear characteristics of fuel cell efficiency are considered during optimization in the co-optimization model proposed in this paper. This method allows energy management to actively change the speed trajectory accordingly, thereby changing the corresponding power–demand curve to further improve the energy conversion efficiency of the fuel cell. As a result, the fuel cell efficiency in the operation obtained by the co-optimization result is higher than the one in the results of sequential optimization.
3.3. Case Study with Speed Limits and Gradients
The above cases are carried out on the assumption that there is no speed limit and gradient in the journey. However, in the real-world railway operations, there are speed limits and gradients along the journey. In this case, ESD capacity, travel distance, and time were 40 MJ, 10 km, and 450 s, respectively. The initial SOE of ESD is 1. Case 4 and Case 5 indicate co-optimization and sequential optimization, respectively. The geographic information is shown in
Figure 9a; a speed limit section, an uphill area, and a downhill area are added.
With extra constraints of gradients and speed limits introduced, from the speed trajectory profiles presented in
Figure 9b, it can be seen that the train cruises at 72 km/h in 196 s to 450 s in the speed limit area, and the train of sequential optimization is still accelerating with maximum acceleration, which requires higher traction power. There are still some differences in the two acceleration phases; the traction power of the train in Case 5, from 0 to 62 s and 252 s to 285 s, is 600 kW, which means that the train has the maximum acceleration.
As shown in
Figure 9c, the efficiency in Case 4 in general is higher than the one in Case 5 due to a better energy management on the train load demand to enable a higher working efficiency of the fuel cell. As can be seen from
Table 6, the net hydrogen consumption in Case 4, i.e., co-optimization, is reduced by 6.48% compared with that in Case 5, i.e., sequential optimization.
Different from the case study on the flat railway track, the speed trajectories of the two methods are different in the cruising phase. It can be seen from the discharge of ESD of Case 4 in
Figure 9d that ESD discharges all the available energy during the acceleration process of the train, so the train in Case 4 is not running at a constant cruise speed to avoid the high fuel cell power. From the energy-efficiency point of view, the train needs higher power during acceleration, so ESD is more inclined to discharge during the acceleration process to increase the train’s speed, and this will support the fuel cell to work in states with higher efficiency.
It can be seen from the above case studies that the power split strategy of both the fuel cell and ESD is determined by the algorithm itself but not preset by rules. In the model, only the total energy demand or generation of the train is considered, and by conducting the optimization, the power and energy split of the fuel cell and the supercapacitor can be configured by the model automatically to achieve the objective of minimizing the hydrogen consumption.
4. Conclusions and Future Work
In this paper, we establish a time-based MILP co-optimization model to optimize the train control and energy management simultaneously to minimize the net hydrogen consumption. Based on this model, two problems are discussed. The first is the relationship between ESD capacity and net hydrogen consumption in a hybrid power system. The results show that ESD can reduce net hydrogen consumption by up to 29.58% in the proposed case study. It provides extra power to enable the fuel cell to work at the point of higher efficiency, leading to a significant reduction of net hydrogen consumption.
Another problem is the difference between co-optimization and sequential optimization. Two specific scenarios are simulated to compare the net hydrogen consumption, and the results show that co-optimization can save 6.48% net hydrogen consumption in the proposed case study. The advantage of co-optimization is that it tends to avoid the working states of high power and low efficiency for a fuel cell, which mainly exist during the acceleration phase, and the power–demand curve can be optimized simultaneously to find a global optimal solution.
In this paper, the optimal ESD capacity is obtained based on the minimum hydrogen consumption, without considering the device initial cost and other engineering factors, and the results may not fully reflect the economic indicators of the railway line operation. In addition, the whole life-cycle optimization of fuel cell and ESD is not considered in the proposed model, and the multi-station operation, considering optimal running time allocation, can also be explored in the future. Therefore, in future, the cost and detailed characteristics of ESD and the whole life-cycle cost can be further studied in the model to achieve a more realistic optimization result, and flexible running time can be considered. It also should be noted that the linearization technique we used in the paper is to transfer the nonlinear optimization problem into an MILP problem to achieve a more efficient method of finding the global optimal solution. However, the approximation applied in the modeling process brings the errors, and the evaluation of the error and the comparison between the linearized and nonlinear method on the same problem is also the potential future direction from the perspective of methodology. The size of voltage of the energy system is also an important variable in the train control, which is of great significance to implement the energy management strategy of the train and ensure that the train runs at the optimal speed trajectory. Therefore, in the future work, the voltage will be considered in the model.