You are on page 1of 5

The Optimal Circuitous Communication Route Search for Pilot

Protection Based on Ant Colony Algorithm


Xin Chen*, Zhiqian Bo, Xingzhou Dong*, Shenxing Shi*, Bin Wang*

*State Key Laboratory of Control and Simulation of Power System and Generation Equipments (Tsinghua University), China,
chenxin_725@163.com, Xuji Group Corporation, China

Ant colony algorithm is used to find the optimal circuitous


Keywords: pilot protection, ant colony algorithm, circuitous route in the communication network of power system in this
route, dead end, penalty function paper. But in an actual communication network of power
system, each node is just connected to very few nodes. This
Abstract structure leads to the search process come to dead ends
frequently. In order to reduce the probability of ants entering
With the development of smart grid, the self-healing ability of dead ends, penalty functions are added to the algorithm.
protection device becomes much more important. Pilot Simulation results show the whole route accelerating penalty
protection is of good selectivity, but it needs a highly reliable function is the best one, which decreases the probability by
communication network. Once the communication line nearly 20%.
between two devices is out of work, pilot protection will be
forced out of operation. Since there are many ring structures 2 Basic principle of ant colony algorithm
in communication network, circuitous routes can be found to
help pilot protection remain operational. In this paper, ant The ant colony algorithm proposed by Italy scholars Marco
colony algorithm (ACA) are used to find the optimal Dorigo, Vittorio Maniezzo and Alberto Colorni, in early 90s
circuitous route, and simulation results show the method is of last century, is a kind of heuristic bionic evolutionary
fast and accurate. To the actual communication network of algorithm. It simulates the process of ants searching for food
power system, each node is just connected to a few other collectively [7-10].
nodes, so the search process often come to dead ends. Use The whole searching process follows 4 rules:
penalty functions can reduce the probability of coming to a a) Ants can only observe the length of paths between adjacent
dead end and improve the algorithm efficiency. The whole nodes and the pheromone on these paths.
route accelerating penalty function is most efficient. b) The probability of selecting a path is proportional to the
concentration of pheromone on the path, and inversely
1 Introduction proportional to the length of the path.
c) Ants will release pheromone on all paths that they come
Pilot protection is widely used in power system, because of through.
the simple principle and good selectivity [1]. However, a d) Pheromone is evaporable, and it will be updated after one
highly reliable communication network is needed to meet the or a group of ants finish searching.
requirements of high-speed data transmission. Since The ants select paths according to probability, so some of
intellectualization is the most important development them will find new ways. If they find a shorter route, it will
direction of power system, the communication self-healing have a higher level of pheromone concentration. More and
will become an effective way to enhance the reliability of more ants will move on this way gradually. After a period of
pilot protection. time, the shortest way may be found, on which most ants are
In a traditional power system, the pilot protection will be moving.
forced out of operation if the communication line between
relay protection devices is out of work [2]. However, the rest 3 The optimal circuitous communication route
of the communication network can be used to find circuitous
search based on ACA
routes, which does not add investment and just needs a short
time to recover. When the communication line between two nodes, A and B,
In Literature [3], the author finds a circuitous route in the is out of work, the remaining communication network can be
communication network of power system, but doesnt prove used to find the optimal circuitous route.
that is the optimal one. An improved Floyd algorithm is used There are two steps. First, use the node adjacency matrix to
to find the shortest circuitous channel in Literature [4], but find out whether node A and node B are connected. Second,
time delay of communication nodes is not considered. The search for the optimal circuitous route with ACA if the two
author finds the optimal circuitous channel using ant colony nodes are connected.
algorithm in Literature [5,6], taking both time delay and
bandwidth into consideration. However, there is a lack of
analysis on the searching efficiency.

1
3.1 Find the connection relationship with breadth first can be proved that these two nodes are connected through the
algorithm breadth first search algorithm. So it is possible to find a
circuitous route.
In an N-node communication network, the node adjacency
matrix (NAM) C can be represented by an n n matrix. The
definition of cij, the element at i-th row j-th column, is shown
in (1).

1, if node i and node j are connected directly


cij (1)
0, otherwise
In matrix C, the connection relationship between node A and
node B can be confirmed, using the breadth first search
algorithm. Taking node A as the start point, if node B can be
reached, there is at least one circuitous route can be found.
Otherwise, node A and node B are at different isolated islands,
which means there is no possibility to find a circuitous route.

3.2 Searching for the optimal circuitous route with ACA


a) Generate the node adjacency distance matrix (NADM) D.
NADM is also an n n matrix. The definition of dij, the
element at i-th row j-th column, is shown in (2).
the length between i-th node and j-th node,

dij if the two nodes are connected directly (2)
1, otherwise

b) Generate the node delay vector (NDV) T, which is an n
dimensional column vector. The i-th element, ti, represents
the time delay when a signal transmitted by the i-th node.
c) The transmission speed of light in an optical fibre
v 2 105 km / s . T 2 vT . T2 represents the node delay
equivalent distance vector (NDEDV). The i-th element, t2i,
represents the length that the i-th nodes time delay is
equivalent to.
d) Generate D2, the node adjacency distance matrix with node
time delay, using both matrix D and vector T2. The definition Figure 1: Ant colony algorithm searching flow chart.
of d2ij, one element in D2, is shown in (3).
1, if dij 1

d 2ij 1 (3)
dij (t 2i t 2 j ), if dij 1
2
e) Using ACA to find the optimal circuitous route between
node A and node B in matrix D2. The flow chart of this
algorithm is shown in Figure 1.
In Figure 1, Nant represents the total number of ants. M
represents the total number of loops. X represents the
pheromone matrix. P represents the shortest path recording
matrix. L represents the shortest path length.

4 Simulation and analysis


In a transmission power system, communication network and
power network have the same structure, because fibres for
communication are in OPGW, which stretches along with
overhead lines [11]. Given that the communication line
between node 23 and node 24 is out of work, it is necessary to
find an optimal circuitous route in order to keep these two
nodes stay in touch. Generate the NAM C of this model. It Figure 2: IEEE 30 nodes model.

2
Set the length between adjacent nodes, ranging from 8km to
50km. Set the time delay of each node, ranging from 15s to Most communication nodes in power system, however, are
25s. And then, generate the NADM D, as well as the NDV T. just directly connected to 2 or 3 nodes, which reduces the
After that, it is easy to generate the matrix D2 through the searching efficiency of ACA.
method mentioned in 3.2. In the experiment, not every ant can successfully get to the
Use ACA to find the optimal circuitous route with different terminal node B. Most ants will enter dead ends and terminate
terminal conditions. If in N consecutive loops (corresponding the searching process, resulting in an invalid search. Specific
to NNant searches) there is no better solution, searching can examples are shown in Figure 4.
be stopped, and output the final results. The optimal
circuitous route is 23-15-12-16-17-10-22-24, which has been
known. Calculate the accuracy of final results of different N
in 100 experiments. Specific statistical results are shown in
table 1.

Average
Terminal Standard deviation of
computation Accuracy
condition - N computation time / s
time / s
3 0.162 0.0476 53%
5 0.242 0.0682 67%
10 0.447 0.113 87%
15 0.648 0.151 96%
20 0.800 0.150 98%
25 0.977 0.188 100% Figure 4: sketch map of coming to a dead end.
30 1.07 0.156 100%
35 1.28 0.175 100% In Figure 4, the start node is node 1 and terminal node is node
9. Suppose the searching route is 1-4-5-6-3-2. When the ant
comes to node 2, all nodes that directly connected to node 2
Table 1: Statistical results of 100 experiments for each have been searched. That is to say, the ant come to a dead end.
terminal condition. Since all nodes in the TSP problem are directly connected,
there is no dead end problem. This is a special problem in the
The searching speed is fast, and it can meet requirements of non-all-adjacent network, especially in the communication
finding the optimal circuitous route online for self-healing. network of power system.
According to the accuracy and computation time, the terminal In the 20 experiments, the average ratio of ants coming to a
condition N is set as 25. dead end is 0.836, and the standard deviation is 0.0159. So
more than 80% searching process is invalid, and this ratio is
4.1 Dead end problem stable.
ACA was first applied to the traveling salesman problem
(TSP problem). For example, there are 5 cities coded from 1 4.2 Ratio of coming to dead ends is not sensitive to degree
to 5. The salesman is at city 1 now. He wants to visits each of adjacency
city exactly once and returns to the origin city. What is the The definition of degree of adjacency (DoA) is shown in (4).
shortest possible route? In this problem, cities are connected Degree of Adjacency
with each other. So no matter which city the salesman is at
the number of nonzero elements in NAM C (4)
now, he can get to at least one city in the next step. That is to
say, there is no dead end problem in TSP problem. the number of elements in NAM C
If it is defined that one node is adjacent to itself, DoA ranges
from 1/N to 1.
In the TSP problem, its DoA is 1. However, the DoA is very
small in communication network of power system, because
most nodes are just directly connected to 2 or 3 ones. For
example, in the IEEE 30 nodes example, its DoA is only
0.1244.
Based on the IEEE 30 nodes model, connect some nodes
which are not connected before and construct 5 networks with
different DoA.
Do 20 experiments with each model, and the statistical results
are shown in Figure 5.
Figure 3: TSP problem.

3
Table 2: Comparison of simulation results of different penalty
functions.

If only the pheromone concentration of last section is


modified, the efficiency of ACA will be significantly
improved. But if the pheromone concentration of the whole
route minus dp, the efficiency will only slightly improved,
compared with the original algorithm. It means there will be
some negative effects if the pheromone concentration of first
few section is punished too heavy.
According to the simulation results in table 2, the whole route
accelerating penalty function (dp / k) is the best one, which
decreases the ratio of coming to a dead end by nearly 20%.
Figure 5: relationship between degree of adjacency and ratio
of coming to a dead end.
6 Conclusion
According to results, the ratio of coming to a dead end has a Ant colony algorithm, which is fast and accurate enough, can
rapid decrease if DoA0.17. However, DoA = 0.17 is too be used to find the optimal circuitous communication route
high in a transmission network, which is almost impossible. for pilot protection. And pilot protection will not be forced
So, in real situations, ratio of coming to dead ends is not out of operation when there are only some problem with the
sensitive to DoA. communication line between two protection devices. In order
to further improve the algorithm efficiency, adding a penalty
5 Improve the algorithm efficiency with penalty function to ACA is a good choice. Among several common
functions penalty functions, the whole route accelerating penalty
function (dp / k) is the best one, which increases the
There are several ways to improve the efficiency of ACA. efficiency by nearly 20%.
Adding a penalty function is a simple and less memory space
occupied way. Acknowledgements
Every time an ant comes into a dead end, the pheromone
matrix X will be modified according to the following steps. This paper is sponsored by State Grid Science Project -
a) Get the route this ant has searched. Research and development of novel framework based
b) Calculate the length of the route, which is represented by protection adaptive to the evolution of substation and power
L_deadend. The basic modification of pheromone matrix, grid.
which is represented by dp, equals 1/ L_deadend. The longer
an ant has searched, the smaller the basic modification will be. References
c) Use a penalty function to modify the pheromone matrix
X. [1] George E E, Brownlee W R. Pilot Wire Relay
There are 4 common penalty functions used in this paper. Protection, Transactions of the American Institute of
f1: the pheromone concentration of last section of the route Electrical Engineers, vol. 54, pp. 1262-1269, (1935).
minus dp. [2] Tonghua Wu, Yuping Zheng, Zhe Zhang and Chengguo
f2: the pheromone concentration of the whole route minus dp. Dou. Seamless Handover between Longitudinal
f3: from the last section to the first one, the pheromone Differential Protection and Pilot Protection in
concentration minus dp / k respectively (k=1,2,3). Circumstance of Single Channel, Automation of
f4: from the last section to the first one, the pheromone Electric Power System, vol. 35, no. 17, pp. 72-75,
concentration minus dp / 2k respectively (k=1,2,3). (2011).
The simulation results based on the original IEEE 30 nodes [3] Shiwen Wang, Dongjiao Xu, and Jizhao Lu.
model are shown in Table 2. Constructing Henan electric power communication
network and implementing alternate route for state
Average number of the Standard deviation of communication network, Telecommunications for
Penalty function ratio of coming to a dead the ratio of coming to a Electric Power System, vol. 30, pp. 26-31+36, (2009).
end dead end [4] Qingzhu Wang, Xiuzhe Zhuo, and Fengqing Liu. Path
Original ACA 0.83587695 0.015908212 optimization of electric power optical fiber
f1 0.73400995 0.032043916 communication networks, Telecommunications for
Electric Power System, vol. 33, pp. 18-22, (2012).
f2 0.80447745 0.044552105
[5] Bao Liu, Xianggen Yin, Zhenxing Li, Yingtong Liu, and
f3 0.6798617 0.033851195 Xin Zhang. Study on circuitous channel reconstruction
f4 0.726528 0.030270381 technique of wide area protection based on ant colony
optimization, Proceedings of 2012 China Smart Grid
Seminar, pp. 98-103, (2012).

4
[6] Bao Liu. Research on wide area protection and channel
reconstruction technique coping with power grid
catastrophe, Huazhong University of Science and
Technology, Wuhan, Hubei, China, (2013).
[7] Marco Dorigo, Vittorio Maniezzo, and Alberto Colorni,
Ant system: optimization by a colony of cooperating
agents, IEEE Transactions on Systems Man &
Cybernetics Part B Cybernetics, vol. 26, pp. 29-41,
(1996).
[8] Haibin Duan, Daobo Wang, Jiaqiang Zhu, and Xianghua
Huang. Development on ant colony algorithm theory
and its application, Control and Decision, vol. 19, pp.
1321-1326+1340, (2004).
[9] Bin Wu and Zhongzhi Shi. An ant colony algorithm
based partition algorithm for TSP, Chinese Journal of
Computers, vol. 24, pp. 1328-1333, (2001).
[10] Dorigo. ACO Algorithms for the Traveling Salesman
Problem. Evolutionary Algorithms in Marco
Engineering and Computer Science: Recent Advances in
Genetic Algorithms, Evolution Strategies, Evolutionary
Programming, Genetic Programming and Industrial
Application. John Wiley & Sons, (1999).
[11] Lei Guan, Qingliang You, Liangming Chen, Heping Fan,
Jie Yu, Ming Yuan, et al., Application and
improvement of ADSS/OPGW electric optical cable in
State Grid informatization, Journal of Jianghan
University (Natural Science Edition), vol. 42, pp. 23-27,
(2014).

You might also like