Arti cial Bee Colony Optimization Algorithm with

New Full Dimension Updating Strategy and Its

Application
Yuangang LI
Shanghai Business School
Xinrui GAO
Civil Aviation University of China
Yingjie SONG
Shandong Technology and Business University
Wu DENG

Civil Aviation University of China

Research Article

Keywords: Arti cial bee colony, Full dimensional search, New evolutionary phase, Optimization, Train
operation adjustment

Posted Date: March 7th, 2024

DOI: https://doi.org/10.21203/rs.3.rs-4013944/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License

Additional Declarations: No competing interests reported.

 Artificial Bee Colony Optimization Algorithm with
New Full Dimension Updating Strategy and Its
Application
Yuangang LI 1, Xinrui GAO2, Yingjie SONG 3,*, Wu DENG2,*
1
Faculty of Business Information, Shanghai Business School, Shanghai 200235, China
2
College of Electronic Information and Automation, Civil Aviation University of China, Tianjin
300300, China
3
School of Computer Science and Technology, Shandong Technology and Business University, Yantai
264005, China
corresponding. [email protected] (Yingjie Song) ; [email protected](Wu Deng)

1 Abstract—For the low accuracy and slow convergence of lot of scholars have presented new evolutionary operators,
artificial bee colony (ABC) algorithm in solving complex parameter control strategy, multi-population to improve the
optimization problems, a new full dimensional updating ABC. Sharma et al. [20] improved initialization operation of
ABC/best/1 evolutionary strategy is designed to propose an ABC by Halton points as the solution set of initial
improved ABC based on the new full dimensional updating
distribution. Coelho et al. [21] presented an improved ABC
strategy(FNABC) in this paper. Because of the low efficiency of
one-dimensional search, the full dimensional update search
based on Gaussian distribution. Gao et al. [22] presented an
strategy and ABC/best /1 evolutionary strategy are used to improved ABC in the previous cycle. An initial population
design a new full dimensional update ABC/best/1 evolutionary is generated and the scout bee is sent to a new food source
strategy, which expands the search space, improves the mining location [23]. Zhang et al. [24] presented a new evolutionary
ability and search efficiency. And a new evolutionary phase of formula in ABC. Karaboga et al. [25] presented an improved
full dimensional update strategy is designed to balance the ABC with neighborhood search. Wang et al. [26] presented
global search ability and mining ability. Finally, the FNABC is an ensemble ABC variant with multi-strategy. Gao et al.
compared with eight state-of-the-art ABC variants in solving [27] presented an ILABC using information learning to
12 functions. The experiment results indicate that the FNABC
dynamically adjust the size of subpopulation. Kiran et al.
has better search ability. Additionally, the FNABC is applied to
solve a real-world train operation adjustment problem. The
[28] presented ABCVSS based on the update rules of
results show that it can obtain the ideal results of the train multiple solution and five search strategies. Zhou et al. [29]
operation adjustment problem. designed a search equation to generates a new solution. Cui
et al. [30] presented DFSABC-elite. Zhou et al. [31]
Index Terms—Artificial bee colony, Full dimensional search, presented an improved ABC with global neighborhood
New evolutionary phase, Optimization, Train operation strategy. Lu et al. [32] presented an improved ABC based on
adjustment. improved elitist neighborhood strategy. Yu et al. [33]
presented an AABC. Gao et al. [34] presented a LL-ABC
I. INTRODUCTION with elite learning. Chen et al. [35] presented an ABC-EO
Many problems in engineering can be converted into based on extremum optimization strategy. Aslan et al. [36]
the optimization problems [1–3], which are NP-Hard presented a fast convergent iqABC. Rashed et al. [37]
problems. The traditional methods difficultly solve these presented an improved multi-objective ABC. Dong et al.
problems [4–6]. In recent years, some optimization methods [38] presented a DMSABC using Lévy flight with dual
are presented to solve them, such as ant colony optimization search mechanism of differential self-perturbation. Chandra
(ACO), grey wolf optimizer (GWO), and so on [7–15]. et al. [39] presented an improved ABC using DE evolution
Although these algorithms can get better solutions in solving strategy. Pan et al. [40] presented a KDABC with K-means
these problems than traditional methods, they still have initialization and dynamic neighborhood search mechanism.
lower accuracy and high time complexity. With the increase Ewees et al. [41] presented an ABCSCA by combining ABC
of problem complexity, new optimization method has with SCA. Wang et al. [42] presented a selection strategy,
become hot topics [16–18]. and combined with reverse learning strategy to propose
ABC algorithm is a stochastic evolutionary algorithm NSABC.
by Karagoga [19]. Because its principle is simple, and easy In the process of train operation, train operation will be
to implement, ABC takes on fewer control parameters, delayed due to various reasons. If appropriate adjustment
stronger robustness and search ability. But the ACO is easy measures are not taken, the operation plan of the whole line
to premature and low convergence. In the ABC, both will be affected [43]. The traditional train operation
employed bees and onlooker bees do one dimensional adjustment is mainly operated manually. The existing train
neighborhood search near the food source, which have operation adjustment methods can be divided into two types.
strong global search ability, insufficient local search ability Lee et al. [44] established schedule and route optimization
and low search efficiency. For the shortcomings of ABC, a model with the minimizing weighted track occupancy.
Tornquist et al. [45] established a mixed integer
1
This work was supported by the National Natural Science Foundation programming model based on minimizing train running
of China under Gran U2133205 and 62341605.

1
delay time. In order to avoid train delay, Wen Chao et al. vi , j = xi , j + rand (−1,1)( xi , j − xk , j )
[46] used workflow network and triangular fuzzy number to (2)
predict train running conflict caused by interference. Andrea where, is new food source. is a
et al. [47] presented a method to detect and reschedule randomly selected food source. is a random
potential conflict when train running late, so as to reduce
dimension.
continuous train delay.Zhuang et al. [48] used Markov
decision process model and strategy optimization method to C. Onlooker bee phase
improve the effectiveness of high-speed rail operation Onlooker bees will search new food sources according
adjustment. Another one is the heuristic evolutionary to Eq. (3).
algorithm. Yue et al. [49] used the heuristic algorithm to fit ( xi )
pi =
solve the mathematical model of optimizing the train SN

timetable of high-speed railway system. Min et al. [50] used  fit ( x )

i =1
i

GA to solve the train operation conflict problem, but genetic (3)

algorithm is prone to premature, which is not conducive to where, is selection probability, and is corresponding
global optimization. Lei et al. [51] presented using co- fitness value according to Eq.(4).
evolutionary GA. The above algorithms have their own
advantages, which play a positive role in promoting the
research. But in the process of solving the problem, they
(4)
also have some shortcomings of difficulty, high complexity,
where, is function value.
slow convergence and so on.
Although these improved ABC have obtained good D. Scout bee phase
results, there are still some problems. To solve these defects, The associated counter will be checked. If the highest
the full dimensional update evolution strategy is used in the value of the counter is larger than the predefined parameter
phase of employed bees and onlooker bees, and a new limit, then the corresponding food source will exhaust.
evolutionary phase is introduced, and an improved ABC,
namely FNABC using the new full dimensional update III. AN ENHANCED FNABC ALGORITHM
strategy and new evolutionary phase is presented, so as to
improve the convergence accuracy, balance the search A. Motivations
ability and improve convergence and efficiency. A train In solving unimodal functions, ABC has the low
operation adjustment method using FNABC is presented to convergence accuracy, slow convergence speed. In solving
realize the adjustment of delayed trains. multimodal functions, ABC has the slow convergence. At
The main innovations and contributions are listed as the same time, ABC only updates one dimension of
follows: individual vector in each iteration, which limits the search
⚫ A new full dimensional update strategy of efficiency. To solve these problems, the advantages of full
ABC/best/1 is designed to expand the search dimensional update search strategy and ABC/best/1
space and improve the mining ability and search evolutionary strategy are used to design a new full
efficiency. dimensional update ABC/best/1 evolutionary strategy,
⚫ A new FNABC based on the new strategy of all- which can expand the search space of employed bees and
dimensional update and new evolution is onlooker bees, improve the search ability and efficiency. A
presented to improve the optimization new evolutionary phase with full dimensional update
performance. strategy is developed. Finally, a new FNABC based on the
⚫ The minimum deviation between the adjusted train new strategy of full dimensional update and new evolution
diagram and the original train diagram is link is implemented to improve the overall optimization
considered as the optimization objective to performance.
establish the optimization model of train
operation adjustment. B. A new ABC/best/1 evolution strategy
⚫ A train operation adjustment optimization method In formula (2), except for the jth dimension, and
using FNABC is presented. are the same as each other. This one-dimensional updating
method is inefficient, which limits the search ability. To
II. ABC improve the search range and the convergence accuracy, and
accelerate the convergence speed, a full dimensional update
A. Initialization search strategy is designed. At the same time, to improve the
The population of SN food sources will be randomly search ability and speed up convergence, the ABC/best/1
generated by Eq. (1). evolution strategy is developed in the stage of employing
xi , j = ximin
, j + rand (0,1)( xi , j − xi , j )
max min
bees, and the full dimensional update search strategy is use
(1) to realize the full dimensional update ABC/best/1 evolution
where, and . and strategy.
Vi = X best + rand (−1,1)( X1 + X 2 ) (5)
are lower and upper bounds.
where, the test solution is generated by . is the
B. Employed bee phase global optimal solution of whole population, and are
After the initialization, employed bees will search new two random solutions, which must be different from .
food sources according to Eq. (2).

2
C. New evolution phase Functions Range
f1 =  i =i x [-100,100] 0
D 2
To balance search ability and solve the defects of poor i

search ability of ABC, it is necessary to design additional f 2 =  i =1 | xi | +  i =1 | xi |

D D [-10,10] 0
search strategies and add new evolutionary phase as
supplementary operators to solve the search equation, so as f3 ( x) = max | xi |,1  i  n [-100,100] 0

to generate new offspring. Therefore, we introduce a new f 4 ( x) =  i =1 ixi2 [-10,10] 0

D

evolution formula in the new phase Eq. (6).

f 5 ( x) =  i =1 100( xi2 − xi +1 ) 2 + (1 − xi ) 2
D −1 [-10,10] 0
TX i = r1 X i + r2 X best + r3 ( X1 + X 2 ) (6)
f 6 ( x) =  i =1 (  xi + 0.5 )
D 2 [-100,100] 0
where, is the trial solution. is global best solution,
and are two random solutions. r1, r2 and r3 are three random
f 7 ( x) =  i =1 ix 4 + random[0,1) [-1.28,1.28] 0
D
non-negative real number on [0,1], and meet a constraint
condition: r1 + r2 + r3 = 1. f8 ( x) =  i =1 ( xi2 − 10 * cos(2 xi ) + 10)
D [-15,15] 0

D. The detailed steps of the FNABC f9 ( x) =  i=1 ( yi2 − 10 * cos(2 yi ) + 10) [-15,15] 0
D

Step1: Initialize parameters of FNABC. Set dimension  xi , if | xi | 0.5

(D), the maximum iterations (T), the limit value ( ), 
where, yi =  round (2 xi )
 , else | xi | 0.5
the current iteration (t=1), the current limit ( ) and 2
so on. 1 x [-600,600] 0
( i=1 xi2 ) − ( i=1 cos( i ))
D D
f10 ( x) = 1 +
Step2: Initialize the population of the FNABC. 4000 i
According to equation (1) the initial population is produced. f11 ( x) =  i=1| xi |
D i +1 [-1,1] 0
Step3: In employed bee phase, implement greedy
f12 ( x) =  i =1 | xi sin( xi ) + 0.1* xi | [-10,10] 0
D
selection strategy. According to equation (5), the offspring
solution is generalized. If is replaced by , will be
set to 0, else, will be increased by 1. B. Experiment environment and parameter settings
Step4: In onlooker bee, implement roulette selection
In the experiment, some parameters are tested and
strategy. According to equation (3), the solution involved in
modified, and the most reasonable initial values of these
evolution is selected. Implement greedy selection strategy. parameters are obtained. These parameters can get better
According to equation (5), the offspring solution is solution and more reasonable running time, and solve the
generalized. If is replaced by , will be set to 0, problem effectively. The number of honey sources is
else, will be increased by 1. , the maximum iterations are , and the
Step5: In scout bee phase, according to equation (5), if limit value is . The environment includes CPU Intel
, replace . Core i5-6200u, windows win10 and 8GB ram. This
Step6: In the new phase, implement greedy selection algorithm uses python2020 programming.
strategy. According to equation (6), the offspring solution is To prove the effectiveness of FNABC, two ABC variants,
generalized. ABC/best/1 and NABC, which are constructed on the basis
Step7: If the end is met, the result is output. Otherwise, of ABC, are also selected. ABC/best/1 is to replace the
return to Step 3. original solution in basic ABC with the modified
ABC/best/1 solution search equation. NABC is based on the
IV. NUMERICAL EXPERIMENTS AND ANALYSIS basic ABC, adding a new evolutionary phase of full
dimensional update, that is, the improved new phase search
A.Test functions operator is triggered.
To prove the validity and superiority of FNABC, 12 The dimension sizes include and , and
functions are used. The expression, range and minimum the parameters of ABC, ABC/best/1, NABC and FNABC
value of 12 functions are shown in TABLE I. Among the 12 are consistent. Each algorithm runs 30 times independently,
functions, the functions of F1, F4, F6 and F7 are and records the optimal value, the worst value, the average
unimodality-separable functions. F2, F3 and F5 are value, the median and the SD as the evaluation index. The
unimodality-nonseparable functions. F8, F9, F11 and F12 results are shown in Table 2 and Table 3.
are multimodality-separable functions. F10 is the
C. Experiment results and analysis
multimodality-nonseparable function to test the global
search ability and stability. The FNABC is compared with The obtained results for and are given in
other 8 ABC variants [43]. TABLE VI and TABLE II.
TABLE I THE EXPRESSION AND OPTIMAL VALUE OF EACH FUNCTION TABLE II THE OBTAINED EXPERIMENT RESULTS

3
 Fig.1. Convergence curve of for D=30 Fig.2. Convergence curve of for D=30

Fig.3. Convergence curve of for D=30 Fig.4. Convergence curve of for D=30

4
 Fig.5. Convergence curve of for D=30 Fig.6. Convergence curve of for D=30

Fig.7. Convergence curve of for D=30 Fig.8. Convergence curve of for D=30

Fig.9. Convergence curve of for D=30 Fig.10. Convergence curve of for D=30

Fig.11. Convergence curve of for D=30 Fig.12. Convergence curve of for D=30
TABLE III THE OBTAINED EXPERIMENT RESULTS

5
 Fig.13. Convergence curve of for D=50 Fig.14. Convergence curve of for D=50

Fig.15. Convergence curve of for D=50 Fig.16. Convergence curve of for D=50

6
 Fig.17. Convergence curve of for D=50 Fig.18. Convergence curve of for D=50

Fig.19. Convergence curve of for D=50 Fig.20. Convergence curve of for D=50

Fig.21. Convergence curve of for D=50 Fig.22. Convergence curve of for D=50

Fig.23. Convergence curve of for D=50 Fig.24. Convergence curve of for D=50

From TABLE II and TABLE III, it can see that function F5, and NABC performs better on the other 11
ABC/best/1 and NABC can achieve better optimization functions. The reason is that the improved search equation
results than the ABC. For ABC/best/1 and NABC, they have and full dimensional updating method in NABC can provide
different performance in test function. Specifically, a better balance, and improve the convergence accuracy and
ABC/best/1 performs better only on the Rosenbrock speed of the algorithm. After synthesizing these two

7
improvements, FNABC can provide the best results, and can effectively and have better global optimization ability.
converge to the global optimal value in the remaining nine
D. Experimental results compared other 8 improved ABC
functions except Rosenbrock function F5 and quartic with
algorithms
noise function F7. At the same time, it can be seen from Fig.
1 and Fig. 2 that among these 4 algorithms, FNABC also has To further verify the optimization performance of
the fastest convergence. For the 9 functions converging to FNABC, 8 other ABC are compared with it. Through the
solution of 12 functions, the comparison results are shown
the optimal value, FNABC only needs 300 times to
in TABLE IV.
complete the global convergence. This proves that the two
improved strategies of FNABC can work together
TABLE IV THE OBTAINED EXPERIMENT RESULTS

From TABLE IV, it can see that in the case of 30 and 50 line, in normal operation and most emergencies, the trains
dimensions, MABC-NS, MGABC and FNABC can achieve on the up line and the down line operate independently
relatively high convergence accuracy for unimodal functions according to the up and down directions respectively,
with less difficulty. Especially for functions F1, F2, F3, F4 without mutual interference. Only in a few cases, when
and F6, FNABC can converge to the global optimal value. single line blockade occurs and reverse traffic is executed,
When solving these five functions, FNABC has less time the traffic of up line and down line will restrict each other.
complexity and standard deviation. For Rosenbrock function Therefore, the double line can be divided into two
F5, the convergence accuracy of FNABC is much higher independent single lines, this paper considers that trains run
than that of the other 8 algorithms. For the F7 quartic with in the same direction.
In this paper, we consider that there are N stations in a
noise function, the convergence accuracy of FNABC is
dispatching section, and the station set is represented by S =
better than the other 8algorithms. AABC, MEABC, MABC-
{1,2, …, N}. There are M trains running in the dispatching
NS, MGABC and FNABC can converge to the global
section, and the train set is T= {1,2, …, M}. The actual
optimal value in the F8, F9 and F10, which is related to the arrival time and actual departure time of train i at station j
strong exploration ability of ABC. For the multimodal are recorded as a respectively and . Similarly, and
functions F11 and F12, MABCNS, MGABC and FNABC
denotes the punctual arrival time and departure time of
can all obtain higher convergence accuracy, but only
train i at station j. Aiming at the scheduling model, this
FNABC can converge to the global optimal value. To sum
paper improves the line model in algorithm programming,
up, FNABC in the 12 functions of the solution accuracy is
and sets each individual in the population as a
better than the eight comparison algorithms, and in 30 and
matrix, so as to realize the algorithm.
50 dimensions of the solution effect, similarly, has good
scalability and robustness. B. Objective function of train operation adjustment
problem
V.TRAIN OPERATION ADJUSTMENT METHOD The indicators describing the impact of train delay mainly
include the train delay time, the number of trains delayed
A. Construct mathematical modeling of train operation
and the number of stations affected, which can be used to
adjustment
reflect the performance of restoring the normal operation of
The essence of train dispatching is to make the delayed the train after the occurrence of delay. The objective is to
trains and the subsequent affected trains return to punctual minimize the total weighted arrival and departure delay time
operation as soon as possible by adjusting the interval of the delayed trains, that is, the deviation between the
running time, station operation time and train running actual train diagram and the planned train diagram. The
sequence of trains to reduce the impact of train delay and objective function can be expressed as:
improve passenger satisfaction on the premise of ensuring f =  j =1  i=1i [(a ji − a*ji ) + (d ji − d *ji )]
n m
(7)
train safety. Considering that the dedicated line is a double

8
where, is the weight value of train I, Affected by train algorithm. The search method of all-dimensional update is
grade. adopted.
C. Constraints of train operation adjustment problem E. Steps of Train operation adjustment
In the process of train operation, to ensure the operation
Initialize population and set parameters for FNABC
safety, technical means such as block section are adopted,
which are reflected as operation constraints in the
dispatching system. The common constraints are as follows: Determine the population to obtain the initial timetable
1) Minimum interval running time constraint
The operation time of train i between two adjacent Eliminate conflicting timetable according to constrains
stations, j station and j + 1 station shall not be less than the
minimum operation time. Calculate the objective function value and retain the minimum value
a ji − d j +1,i  t run
ji (8)
Update population to obtain a new population
where, t run
ji indicates the minimum running time.
2) Minimum stop time constraint Determine the new population and obtain the new timetable
The stop time of train i at station j is constrained by the
minimum stop time. N
d ji − a ji  e Xji t Tji (9) Meet end condition?

where, e Xji is a 0-1 variable, which indicates whether the Y

Obtain the optimal timetable
down train I stops at the station. When the train stops, it is
T
taken as 1, otherwise it is taken as 0. t ji represents the Fig.25. The flow of the FNABC in train operation adjustment problem
minimum dwell time of train i at station j.
3) Departure time constraint
The actual departure time i at each station shall not be less VI.CASE AND ANALYSIS
than the departure time specified in the figure, that is, it shall A. Experimental data and environment
not be departure in advance.
To verify the effectiveness of FNABC in train operation
d ji  d *ji (10)
adjustment, 6 stations are selected in the down direction of a
4) Departure interval constraint section of Zhengzhou Xi'an railway and 11 trains from 1:00
To meet the safety requirements, the departure time to 7:00 in the afternoon to adjust the delay of three trains
interval between two adjacent trains should not be less than [54]. The train includes three train speed levels. The train
the minimum departure time interval. speed level matrix is L, and the corresponding train weight
d ji − d j ,i −1  I jf (11) matrix is  . The minimum tracking interval of train
where, indicates the departure interval. running in section I = 4 min. The additional time for train
stop and start is 1 min and 2 min. The station track number
5) Arrival interval constraint
matrix is DF. The minimum operation time of each speed
The arrival time interval between two adjacent trains shall
class train in each section is expressed as . The minimum
not be less than the minimum arrival time interval.
standard operation time of each speed class train at each
a ji − a j ,i−1  I dj (12) station is . The planned train timetable is shown in Table
Where, indicates the arrival interval i at station j. 5, and the punctual train diagram is shown in Figure 26.
6) Train Overtaking constraints Parameter setting: FNABC is programmed with Python
In order to train safety, the intersection and overtaking of 2020 to realize the operation adjustment of different train
trains must be carried out in the station, so the train must delays. Set the number of population , the number
follow the principle of first in first out, last in and last out in of iterations , and the limit
the interval. number
L = [1,1,1,1,3,3,3,3, 2, 2, 2]
D. Train operation adjustment method using FNABC  = [3,3,3,3,1,1,1,1, 2, 2, 2]
The train operation adjustment problem belongs to large- DF = [3,3,3, 2, 2,3]
scale NP problem, and the traditional methods are difficult 3 4 3 3 4 3
to get the optimal solution within the specified time. The
 20 36 24  0 0 0  0 0 0 
intelligent optimization algorithm has no prior knowledge,  25   
high precision and fast speed. Therefore, taking full  48 30  0 0 0 0 0 0
QT =  29 40 38  ZT =   ZT =  
advantage of the global optimization performance of  
10 23 21 2 4 3 2 4 3
FNABC, a train operation adjustment solution using 0
 25
 44 
42  0 0 0 0 0
FNABC is presented.    
In this paper, the overall adjustment method is adopted,  2 3 2   2 3 2 
and the ABC is used to search the arrival and departure time TABLE V TIMETABLE OF PUNCTUAL TRAINS
of all trains after the start of late trains and stations until the
STATION
termination conditions are met. The arrival and departure
TRAIN 1 2 3 4 5 6
time of all late trains and their subsequent trains are 1 13:05 13:32 13:58 14:28 14:45 15:11
constructed as a 2N*M matrix as an individual in the swarm 13:11 13:32 13:58 14:34 14:45 15:17

9
 2 13:12 13:39 14:05 14:35 14:52 15:18 5 14:10 14:50 15:38 16:34 17:08 18:02
13:18 13:39 14:05 14:41 14:52 15:26 14:14 14:50 15:38 16:42 17:16 18:10
3 13:19 13:46 14:12 14:42 14:59 15:25 6 14:14 14:56 15:48 16:44 17:18 18:12
13:25 13:46 14:12 14:48 14:59 15:31 14:18 14:56 15:48 16:52 17:26 18:20
4 13:26 13:53 14:19 14:49 15:06 15:32 7 14:20 15:06 15:58 16:54 17:28 18:22
13:32 13:53 14:19 14:55 15:06 15:38 14:28 15:06 15:58 17:01 17:32 18:30
5 14:00 14:46 15:38 16:34 17:08 18:02 8 14:30 15:16 16:07 17:26 18:00 18:54
14:08 14:46 15:38 16:42 17:16 18:10 14:38 15:16 16:31 17:34 18:08 19:02
6 14:10 14:56 15:48 16:44 17:18 18:12 9 15:10 15:43 16:17 17:02 17:36 18:29
14:18 14:56 15:48 16:52 17:26 18:20 15:16 15:43 16:17 17:08 17:36 18:35
7 14:20 15:06 15:58 16:54 17:28 18:22 10 15:20 15:53 16:27 17:10 17:42 18:37
14:28 15:06 15:58 17:01 17:32 18:30 15:26 15:53 16:27 17:16 17:42 18:42
8 14:30 15:16 16:07 17:26 18:00 18:54 11 15:30 16:03 16:37 17:18 17:50 18:45
14:38 15:16 16:31 17:34 18:08 19:02 15:36 16:03 16:37 17:24 17:50 18:50
9 15:10 15:43 16:17 17:02 17:36 18:29
15:16 15:43 16:17 17:08 17:36 18:35
10 15:20 15:53 16:27 17:10 17:42 18:37
15:26 15:53 16:27 17:16 17:42 18:42
11 15:30 16:03 16:37 17:18 17:50 18:45
15:36 16:03 16:37 17:24 17:50 18:50

Fig. 27. Operation adjustment chart for delay situation 1

Fig.26. On time train diagram

B. Assigned results of trains

In the first case, the fifth train arrives 10 minutes late at
the first station, the FNABC adjustment result timetable is
shown in TABLE VI, and the operation adjustment diagram
is shown in Figure 27. The dotted line is the train diagram
after adjustment, and the solid line is the train diagram
before adjustment. The convergence diagram is shown in
Figure 28. The total weighted arrival and departure delay of
the train is 28 minutes. The fifth train resumed punctual
operation at the third station. The sixth train delayed its
arrival at first station for 4 minutes, on the premise of Fig. 28. Convergence curves of the involved four algorithms on delay
meeting the minimum stop time, it started on time. situation 1
Subsequent trains and stations are no longer affected by The second situation is that the ninth train arrives 15
delays. minutes late at the first station. The FNABC adjustment
TABLE VI THE OBTAINED RESULTS FOR LATE SITUATION 1 result timetable is shown in TABLE VIII, and the operation
Methods Best(m) Worst(m) Mean(m) Median(m) Std
ABC 122 1051 306.6552 241 177.2412
adjustment diagram is shown in Figure 29. The dotted line
ABC/best/1 229 1202 658.3 607 223.114 represents the adjusted train diagram. The convergence
NABC 28 28 28 28 0 diagram is shown in Figure 30. The total weighted arrival
FNABC 28 28 28 28 0 and departure delay time of the train is 134 min. The ninth
TABLE VII TIMETABLE OF ADJUSTED TRAINS 1 train resumed on schedule at the fourth station. The tenth
STATION
train resumed on schedule at the third station. Subsequent
TRAIN 1 2 3 4 5 6
1 13:05 13:32 13:58 14:28 14:45 15:11 trains are no longer affected by delays.
13:11 13:32 13:58 14:34 14:45 15:17 TABLE VIII THE OBTAINED RESULTS FOR LATE SITUATION 2
2 13:12 13:39 14:05 14:35 14:52 15:18 Methods Best(m) Worst(m) Mean(m) Median(m) Std
13:18 13:39 14:05 14:41 14:52 15:26 ABC 152 292 206.5714 208 38.76561
3 13:19 13:46 14:12 14:42 14:59 15:25 ABC/best/1 136 342 193.0667 178 45.49114
13:25 13:46 14:12 14:48 14:59 15:31 NABC 134 134 134 134 0
4 13:26 13:53 14:19 14:49 15:06 15:32 FNABC 134 134 134 134 0
13:32 13:53 14:19 14:55 15:06 15:38 TABLE IX TIMETABLE OF ADJUSTED TRAINS 2

10
 STATION resumed punctual operation at the fifth station, and the
TRAIN 1 2 3 4 5 6 ninth, tenth and eleventh trains resumed punctual operation
1 13:05 13:32 13:58 14:28 14:45 15:11
13:11 13:32 13:58 14:34 14:45 15:17
at the sixth station. The delay caused a large range of train
2 13:12 13:39 14:05 14:35 14:52 15:18 delays.
13:18 13:39 14:05 14:41 14:52 15:26 TABLE X THE OBTAINED RESULTS FOR LATE SITUATION 3.
3 13:19 13:46 14:12 14:42 14:59 15:25 Methods Best(m) Worst(m) Mean(m) Median(m) Std
13:25 13:46 14:12 14:48 14:59 15:31 ABC 479 588 535 532.5 31.53651
4 13:26 13:53 14:19 14:49 15:06 15:32 ABC/best/1 458 804 522.9333 505.5 78.49486
13:32 13:53 14:19 14:55 15:06 15:38 NABC 458 458 458 458 0
5 14:00 14:46 15:38 16:34 17:08 18:02 FNABC 458 458 458 458 0
14:08 14:46 15:38 16:42 17:16 18:10
6 14:10 14:56 15:48 16:44 17:18 18:12 TABLE XI TIMETABLE OF ADJUSTED TRAINS 3.
14:18 14:56 15:48 16:52 17:26 18:20 STATION
7 14:20 15:06 15:58 16:54 17:28 18:22 TRAIN 1 2 3 4 5 6
14:28 15:06 15:58 17:01 17:32 18:30 1 13:05 13:32 13:58 14:28 14:45 15:11
8 14:30 15:16 16:07 17:26 18:00 18:54 13:11 13:32 13:58 14:34 14:45 15:17
14:38 15:16 16:31 17:34 18:08 19:02 2 13:12 13:39 14:05 14:35 14:52 15:18
9 15:25 15:52 16:22 17:02 17:36 18:29 13:18 13:39 14:05 14:41 14:52 15:26
15:28 15:52 16:22 17:08 17:36 18:35 3 13:19 13:46 14:12 14:42 14:59 15:25
10 15:20 15:56 16:27 17:10 17:42 18:37 13:25 13:46 14:12 14:48 14:59 15:31
15:32 15:56 16:27 17:16 17:42 18:42 4 13:26 13:53 14:19 14:49 15:06 15:32
11 15:30 16:03 16:37 17:18 17:50 18:45 13:32 13:53 14:19 14:55 15:06 15:38
15:36 16:03 16:37 17:24 17:50 18:50 5 14:00 14:46 15:38 16:34 17:08 18:02
14:08 14:46 15:38 16:42 17:16 18:10
6 14:10 14:56 15:48 16:44 17:18 18:12
14:18 14:56 15:48 16:52 17:26 18:20
7 14:20 15:06 15:58 16:54 17:28 18:22
14:28 15:06 15:58 17:01 17:32 18:30
8 14:30 15:16 16:37 17:33 18:00 18:54
14:38 15:16 16:43 17:37 18:08 19:02
9 15:10 15:43 16:41 17:19 17:43 18:29
15:16 15:43 16:41 17:22 17:43 18:35
10 15:20 15:53 16:45 17:23 17:47 18:37
15:26 15:53 16:45 17:26 17:47 18:42
11 15:30 16:03 16:49 17:27 17:51 18:45
15:36 16:03 16:49 17:30 17:51 18:50

Fig. 29. Operation adjustment chart for delay situation 2

Fig. 31. Operation adjustment chart for delay situation 3

Fig. 30. Convergence curves of the involved four algorithms on delay

situation 2
The third situation is that the eighth train arrives 30
minutes late at the third station. The schedule of FNABC
adjustment results is shown in TABLE IX, and the operation
adjustment diagram is shown in Figure 31. The dotted line is
the train diagram after adjustment, and the solid line is the
train diagram before adjustment. The convergence diagram
is shown in Figure 32. The total weighted arrival and Fig. 32. Convergence curves of the involved four algorithms on delay
departure delay time of the train is 458 min. The eighth train situation 3

11
 From the experimental results of the above three cases, can get the best convergence effect of the four algorithms,
we can see that the convergence results of ABC and but FNABC algorithm has less iterations and faster
ABC/best/1 are poor. ABC/best/1 has a better optimal value convergence speed.
than ABC, but it has a larger variance, and its robustness is
C. Comparison and analysis
weaker than ABC. NABC algorithm and FNABC algorithm
TABLE XII CONTRAST ALGORITHMS.
IACO FSO ABC ABC/best/1 NABC FNABC
Delay situation 1(min) 30 30 122 229 28 28
Delay situation 2(min) 98 98 76 68 67 67
Delay situation 3(min) 332 298 263 255 255 255
The references [52,53] are selected to compare the
shortest total arrival and departure delay time of each ACKNOWLEDGMENTS
algorithm. For these three cases, NABC and FNABC have The authors would like to thank all the reviewers for their
the best convergence effect. For the first case, the
constructive comments.
convergence results of ABC and ABC/best/1 are the worst,
and the total arrival and departure delay time of NABC and
FNABC is 2 minutes less than that of FSO and immune FUNDING
ACO (IACO). The total convergence time of IACO is less This work was supported by the National Natural Science
than that of NABC, NABC and NABC. For the third delay Foundation of China (U2133205).
case, the solution effect of FSO is better than that of IACO,
and the convergence results of ABC, ABC/best/1, NABC DATA AVAILABILITY STATEMENT
and FNABC are better than FSO. The total arrival and
departure delay time of FSO is 35 minutes less than that of Not applicable
IACO, and the total arrival and departure delay time of
ABC/best/1, NABC and FNABC are the same, 53 minutes CONFLICTS OF INTEREST
less than that of FSO. But ABC and ABC/best/1 need 2000 The authors declare no conflict of interest.
iterations, the convergence speed is slow, IACO and NABC
only need 300 iterations, while FSO and FNABC only need REFERENCES
20 iterations. In conclusion, FNABC has the best [1] W. Deng, H.M. Zhao, L. Zou, G.Y. Li, X.H. Yang, D.Q. Wu, A novel
convergence accuracy and the fastest convergence speed. collaborative optimization algorithm in solving complex optimization
problems, Soft Comput. 21 (15) (2017) 4387–4398.
VII. CONCLUSION AND FUTURE WORK [2] M.A. Al-Betar, I. Aljarah, M.A. Awadallah, H. Faris, S. Mirjalili,
Adaptive β-hill climbing for optimization, Soft Comput. 23 (24)
For shortcomings of slow convergence speed and low (2019) 13489–13512.
precision of ABC, a new strategy of full dimensional [3] W. Deng, J. Xu, Y. Song, et al., An effective improved co-evolution
ant colony optimization algorithm with multi-strategies and its
updating ABC/best/1 evolution is designed, and a FNABC application, Int. J. Bio-Inspired Comput. 2020 (2020) 1–10.
based on the new strategy and new evolution phase of full [4] Y. Liu, Y. Mu, K. Chen, et al., Daily activity feature selection in
dimensional updating is presented. The effectiveness of the smart homes based on pearson correlation coefficient, Neural Process.
algorithm is verified by 12 functions and a practical train Lett. 51 (2020) 1771–1787.
[5] H. Chen, Q. Zhang, J. Luo, et al., An enhanced bacterial foraging
operation adjustment problem. The results show that the optimization and its application for training kernel extreme learning
FNABC can converge to the global optimal value quickly. machine, Appl. Soft Comput. 86 (2020) 1–24.
For the F5 Rosenbrock function and F7 quatric with noise [6] H. Sharifipour, M. Shakeri, H. Haghighi, Structural test data
function, FNABC has a higher convergence accuracy than generation using a memetic ant colony optimization based on
evolution strategies, Swarm Evol. Comput. 40 (2018) 76–91.
the other 8 improved ABC algorithms. The FNABC has [7] Y. Xu, H. Chen, A.A. Heidari, J. Luo, Q. Zhang, X. Zhao, C. Li, An
good search performance for multimodal function. For the efficient chaotic mutative moth-flame-inspired optimizer for global
train operation adjustment problem, FNABC has high optimization tasks, Expert Syst. Appl. 129 (2019) 135–155.
convergence results, greatly shortens the delay time of the [8] K.V. Price, R.M. Storn, J.A. Lampinen (Eds.), Differential Evolution:
A Practical Approach to Global Optimization, Springer Natural
train, and makes the deviation between the train operation Computing Series, 2005.Y. Xu, H. Chen, J. Luo, Q. Zhang, S. Jiao, X.
diagram and the punctual train operation diagram less. Zhang, Enhanced Moth-flame optimizer with mutation strategy for
Therefore, FNABC can effectively solve the train operation global optimization, Inform. Sci. 492 (2019) 181–203.
adjustment problem and obtain ideal train operation [9] M.A. Al-Betar, M.A. Awadallah, I.A. Doush, A.I. Hammouri, M.
Mafarja, Z.A.A. Alyasseri, Island flower pollination algorithm for
adjustment results. global optimization, J. Supercomput. 75 (8) (2019) 5280–5323.
First, FNABC has not yet found the optimal solution for [10] T. Li, J. Shi, X. Li, et al., Image encryption based on pixel-level
F5 and F7 functions, which needs further study. Secondly, diffusion with dynamic filtering and DNA-level permutation with 3D
the train scheduling model established in this paper only Latin cubes, Entropy 21 (2019) 1–21.
[11] W. Deng, J. Xu, H. Zhao, An improved ant colony optimization
considers one objective, which is not enough for the actual algorithm based on hybrid strategies for scheduling problem, IEEE
train scheduling situation. In the future, we can consider Access 7 (2019) 20281–20292.
improving the model and algorithm for multiple objectives. [12] R. Chen, S. Guo, X.Z. Wang, et al., Fusion of multi-RSMOTE with
fuzzy integral to classify bug reports with an imbalanced distribution,
IEEE Trans. Fuzzy Syst. 27 (2019) 2406–2420.
ETHICAL APPROVAL [13] M.A. Al-Betar, M.A. Awadallah, Island bat algorithm for
Ethics approval was not required for this research. optimization, Expert Syst. Appl. 107 (2018) 126–145.
[14] T. Li, Z. Qian, T. He, Short-term load forecasting with improved
CEEMDAN and GWO-based multiple kernel ELM, Complexity 2020
(2020) 1–20.

12
[15] M.A. Al-Betar, M.A. Awadallah, H. Faris, I. Aljarah, A.I. Hammouri, Boltzmann Selection probability[J]. Swarm and Evolutionary
Natural selection methods for Grey Wolf Optimizer, Expert Syst. Computation, 2019, 49.
Appl. 113 (2018) 481–498. [36] S. Aslan, H.Badem, D. Karaboga, Improved quick artificial bee
[16] R.H. Huang, T.H. Yu, An effective ant colony optimization algorithm colony (iqABC) algorithm for global optimization[J]. Soft Computing
for multi-objective job-shop scheduling with equal-size lot-splitting, A Fusion of Foundations Methodologies & Applications, 2019.
Appl. Soft Comput. 57 (2017) 642–656. [37] Rashed, Salem, Mustafa, et al. An Artificial Bee Colony Algorithm
[17] Y. Liu, X. Wang, Z. Zhai, et al., Timely daily activity recognition for Data Replication Optimization in Cloud Environments[J]. IEEE
from headmost sensor events, ISA Trans. 94 (2019) 379–390. Access, 2019, 8:51841-51852.
[18] W. Deng, Differential evolution algorithm with wavelet basis function [38] C. Dong, Z.Xiong, X. Liu , et al. Dual-Search Artificial Bee Colony
and optimal mutation strategy for complex optimization problem[J]. Algorithm for Engineering Optimization[J]. IEEE Access, 2019:1-1.
Applied Soft Computing, 2020. [39] M. Chandra, R. Niyogi, Web Service Selection Using modified
[19] D. Karaboga, An idea based on honey bee swarm for numerical Artificial Bee Colony Algorithm[J]. IEEE Access, 2019, PP(99):1-1.
optimization[R]. Kayseri: Erciyes University,2005. [40] X. Pan, Q. Zhang, H. Pan, Improved Artificial Bee Colony Algorithm
[20] T.K. Sharma , M .Pant, Halton Based Initial Distribution in Artificial and Its Application to Fundus Retinal Blood Vessel Image
Bee Colony Algorithm and Its Application in Software Effort Binarization[J]. IEEE Access, 2020, PP(99):1-1.
Estimation[J]. IEEE, 2011. [41] A.A. Ewees, M.A. Elaziz, M. Al-Qaness, et al. Improved Artificial
[21] L.Coelho, P. Alotto, Gaussian Artificial Bee Colony Algorithm Bee Colony Using Sine-Cosine Algorithm for Multi-level
Approach Applied to Loney's Solenoid Benchmark Problem[J]. IEEE Thresholding Image Segmentation[J]. IEEE Access, 2020, PP(99):1-
Transactions on Magnetics, 2011, 47(5):1326-1329. 1.
[22] W. Gao, S. Liu, A modified artificial bee colony algorithm, Comput. [42] H. Wang, W. Wang, S. Xiao, et al. Improving Artificial Bee Colony
Oper. Res. 39 (3) (2012) 687–697. Algorithm Using a New Neighborhood Selection Mechanism[J].
[23] W.F. Gao, S.Y. Liu, L.L. Huang, A novel artificial bee colony Information Sciences, 2020, 527.
algorithm based on modified search equation and orthogonal learning, [43] S.Q. Ni, C.L. Zhao, H. Zhuang, H.X. Lv. High-speed railway signal
IEEE Trans. Cybern. 43 (2013) 1011–1024. system[M]. Beijing: China Railway Press, 2013: 1-137.)
[24] X. Zhang, S.L. Ho, et al. A Modification of Artificial Bee Colony [44] Y. Lee, C. Chen, A. Y, heuristic for the train pathing and timetabling
Algorithm Applied to Loudspeaker Design Problem[J]. IEEE problem[J].Transportation Research Part B ， 2009，43（8/9）：
Transactions on Magnetics, 2014, 50(2):737-740. 837-851.
[25] D. Karaboga, B. Gorkemli, A quick artificial bee colony (qABC) [45] J. Tornquist, J. A.Persson, N-tracked railway traffic rescheduling
algorithm and its performance on optimization problems, Appl. Soft during disturbance[J].Transport Research Part B， 2007，41（3）：
Comput. 23 (2014) 227–238. 342-362.
[26] H.Wang, Z.Wu, S. Rahnamayan, H. Sun, Y. Liu, J.-S. Pan, Multi- [46] C. Wen, Predicting high-speed train operation conflicts using
strategy ensemble artificial bee colony algorithm, Inf. Sci. 279 (2014) workflow nets and triangular fuzzy numbers[J]. Proceedings of the
587–603. Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid
[27] W.F. Gao, L.L. Huang, S.Y. Liu, C. Dai, Artificial bee colony Transit, 2015: 229 (3),268-279.
algorithm based on information learning, IEEE Trans. Cybern. 45 [47] Andrea D’Ariano, Marco Pranzo and Ingo A.Hansen. Conflict
(2015) 2827–2839. Resolution and Train Speed Coordination for Solving Real-Time
[28] M.S. Kiran, H. Hakli, M. Gunduz, H. Uguz, Artificial bee colony Timetable Perturbations[J]. IEEE Transactions on Intelligent
algorithm with variable search strategy for continuous optimization, Transportation systems. 2007: 8(2), 208-222.
Inf. Sci. 300 (2015) 140–157. [48] H. Zhuang, S.W. He, Y.C. Dai, Train Operation Adjustment Model
[29] X. Zhou, Z. Wu, H. Wang, S. Rahnamayan, Gaussian bare-bones and Strategy Optimization Method for High Speed Railway[J]. China
artificial bee colony algorithm, Soft. Comput. 20 (3) (2016) 907–924. Academy of Railway Sciences, 2017: 38(2), 118-126.)
[30] L. Cui, G. Li, Q. Lin, Z. Du, W. Gao, J. Chen, N. Lu, A novel [49] Y.X. Yue, S.F. Wang, L.S. Zhou, Optimizing train stopping patterns
artificial bee colony algorithm with depth-first search framework and and schedules for high-speed passenger rail corridors[J].
elite-guided search equation, Inf. Sci. 367 (2016) 1012–1044. Transportation Research Part C ,2016:63,126–146.
[31] X. Zhou, H. Wang, M. Wang, J. Wan, Enhancing the modified [50] Y.H. Min, M.J. Park, S.P. Hong, An appraisal of a column generation-
artificial bee colony algorithm with neighborhood search, Soft. based algorithm for centralized train conflict resolution on a
Comput. 21 (10) (2017) 2733–2743. metropolitan railway network[J].Transportation Research Part B ，
[32] J. Lu, X. Zhou, Y. Ma, M. Wang, An elite group guided artificial bee
2011，45（2）：409-429
colony algorithm with a modified neighborhood search, in, in: The
15th Pacific Rim International Conference on Artificial Intelligence, [51] M. Lei, X.L. Meng, Research on adjustment of high-speed railway
Springer, 2018, pp. 387–394. operation based on coordinated evolutionary genetic
[33] W.-J. Yu, Z.-H. Zhan, J. Zhang, Artificial bee colony algorithm with algorithm[J].Journal of Railway Science and Engineering,2017：14
an adaptive greedy position update strategy, Soft. Comput. 22 (2) （6）,1137-1145.)
(2018) 437–451. [52] W.W. Dou, Optimization Research of Train Operation Adjustment
[34] W. Gao, H. Sheng, J. Wang, S. Wang, Artificial bee colony algorithm Model Based on Immune Ant Colony Algorithm, Lanzhou Jiaotong
based on novel mechanism for fuzzy portfolio selection, IEEE Trans. University, 2012.
Fuzzy Syst. 27 (5) (2018) 966–978. [53] J.C. New, Research on Methodology of Train Operation Adjustment
[35] M.R. Chen, J.H. Chen, G.Q. Zeng , et al. An improved artificial bee Based on Artificial Fish Swarm Algorithm, Lanzhou Jiaotong
colony algorithm combined with extremal optimization and University, 2018.

13