SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia A model for cascading failures in power system based on complex network Ronak Mirzadaei Mohammad khansari Faculty of New Sciences and Technologies University of Tehran Tehran, Iran E-mail: r.mirzadaei@ut.ac.ir Faculty of New Sciences and Technologies University of Tehran Tehran, Iran E-mail: m.khansari@ut.ac.ir Abstract— Cascading failure is a well know phenomenon in real systems that could cause to blackout and collapse in many infrastructure networks like power system. Although most failures may occur locally but sometimes they can propagate and can largely affect the entire network and lead to global collapse and blackout. So according to the importance of network stability, in such condition quick analysis in order to forecast and prevent cascading failure is so important. In conventional models, by applying differential equations, power system dynamic has been studied. In this paper a novel model based on “Centrality measures” and “Complex network” for analyzing cascading failure in power system has been presented. The main advantage of this model is removing complicated and timeconsuming differential equation calculation and therefore decreasing time of calculations compared to conventional models. For validation, the proposed model for IEEE 30 bus has been implemented and result comparison with power system simulator software shows that the proposed model corresponds to conventional models based on components’ removal order and largest connected component. component, it leads to component outage and this process would continue. Keywords- power grid; cascading failures; complex network; centrality measures I. INTRODUCTION For many realistic infrastructure networks, e.g. the power grid and Internet, the occurrence of some disastrous failures may be due to the faults at only one or a few network components. Because of the coupling property of components in the network, these faults could be extended to a large scale. In this case, many components lose their functionalities consequently and may lead to entire collapse of the network. Such phenomenon is known as the network cascading failure or ‘‘network avalanche’’ [1]. On the other hand, the occurrence of faults leads to removal or exit of a component from the system and thus the load of that component should be handled by other components. If extra load exceeds capacity of a Because of important role of cascading failures in infrastructure networks, a lot of studies have been done for prediction of this process [1]. Thus, much research efforts on the investigation of cascading failures in complex networks have been carried out from different aspects. Most of previous researches have followed two general approaches: in the first approach, cascading failures are merely studied based on the relation between system components, but due the improvement and development of networks, a general and integrated view towards system is needed. The second approach, using graph concepts and definitions, is merely focused on network structural characteristics and doesn’t consider dynamics between components. Recently some models have been developed based on these two approaches by combining dynamics of infrastructure network, like load flow, trip of components, and graph theory. In fact, combination of these two approaches contributes to coming up with a more realistic model of cascading failure [1-8]. In general, four parameters are defined by most of models for cascading failure: loading, capacities of components, probability of trip and load redistribution. DC or AC load flow methods have been used for loading by models which act based on circuit rules and focus on interactions of components. These models use real capacities of components and load flow for load redistribution. In other cascading failure models, instead of real loading and capacities, degree and betweenness centrality measures are used for loading. In all models there are two definite or probable removal of components for probability of trip parameter [1-8]. In this paper, in addition to physical quantities for parameters like loading and capacity, network centrality measures are used for load redistribution in order to model cascading failure process. As a result, the proposed method eliminates complex and time-consuming SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia calculations of electrical DC or AC load flow. In addition, weight, capacity and load for both nodes and edges have been considered for investigating cascading failure process at both node and edge levels of network. The paper is organized as follows. In part II necessary concepts are defined and in the next part, the proposed model is provided. Results and comparison are done based on components’ removal order and largest connected component in part IV. Finally conclusion and future works are presented in part V. II. DIFINITIONS A. Power grid Power grid is one of the most important infrastructure networks and it has three types of substations: 1) Generator substations or plants: which are manufacturer and source of power. 2) Transmission substations: which receive power and transmit it to other substations. 3) Distribution substations: which distribute power in local distribution networks [9]. Power grid is shown as a single line diagram which consists of buses and transmission lines. Buses are substations which have acted as nodes in graph, and transmission lines that connect different substations at different voltage levels (high voltage, middle voltage and low voltage) corresponding to edges [9]. In general, distance is considered as the cost of physical quantities’ transmission between two nodes. Electric distance between nodes i and j is equal to total impedance of transmission lines between these two nodes and is defined as follows. Zi = j Uij Iij network is being laid or when a power network is undergoing expansion. C. Complex Network Most of systems and surrounding phenomena have complex structures and relations, and for understanding, analysis and forecasting future behavior, recognition of their corresponding network is needed. Graph theory is used for network mathematical display in such a way that nodes, components of network and edges show relations between components. Thus each network corresponds with a graph and graph features show network characteristics. If there is a dynamic in network, it would become a weighted network [11]. D. Shortest path in weighted graph Several paths may exist between each of two desired nodes in graph. These paths consist of consecutive edges. The shortest path is the one that passes through edges with the least weight [11]. E. Centrality Centrality is a measure which determines the importance of edge or node in graph. Centrality has different definition based on network type and function. One type of centrality is betweenness centrality which is defined in next part [11]. F. Betweenness Centrality in weighted graph Betweenness centrality of a node (edge) equals the number of shortest paths through which node (edge) passes. Node (edge) betweenness centrality is calculated by the following equation. n = Bi st (1) j In (1), U i is the potential difference between two j nodes, and I i is the flowed current between nodes i and j. Inversion of impedance is called admittance or conductivity between two nodes [10]. Power consists of active and reactive power. Active power is generated in plant and consumed in loads. This kind of power is generated and consumed completely. Reactive power is generated by plant as well but is saved as electric and magnetic fields in inductor and capacitor of network [9]. B. Load flow The starting point of any analysis of power system will be the computation of complex voltages at all the busses. Once the complex voltages have been computed, the power coming out of a bus and the power flowing in all the transmission lines can be calculated. Load flow analysis is a computational tool for this purpose. Load flow is normally used in planning studies when a power 1 = nsti , nsti ∑ 0 In (2), if node (edge) i is in the shortest path from node i s to node t, nst would have the value of one, otherwise it would be zero. This measure is provided for weighted and un-weighted networks. In such a way that the number of possible shortest paths would be obtained from that node or edge in weighted network [11]. G. Largest connected component(LCC) This measure illustrates the number of nodes in the largest connected component of the total number of nodes in the network. In fact the less this amount, the less the efficiency and robustness of network [11]. III. THE PROPOSED MODEL FOR CASCADING FAILUERS On one hand, in the proposed model physical quantities of network like load of components, their capacities and impedance have been considered. On the other hand, centrality measures for load redistribution have been applied for facilitating calculations. In this model, in contrast to previous ones, cascading failure (2) SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia process has been considered at both node and edge levels. In this way, the proposed model is made of two parts (cascading failure on nodes and cascading failure on edges). Due to the nature of power grid, network dynamics can contribute to the two-way transmission between nodes. Therefore, the network is considered as undirected. nodes i and j. According to (3), the weight of edges forming the second shortest path becomes more than the weight of edges forming the first shortest path and therefore the former has less important role in load redistribution. This process applies to the next shortest paths, and as the same way edges in the next paths would have more weight. Most of cascading failures happen in power grid because of the following reasons: Short circuit of transmission lines, fault at power stations and damage to the transmission lines or substations, overloading of electricity mains, bad function of protective device and bad climate that could be modeled as the form of edge removal [12]. After updating weights of all edges in the diffusion network, edge betweenness centrality (weighted) is calculated for each edge of this network. This centrality shows the number of shortest paths between each pair of nodes (based on weight). As a result, there will be multiple paths in the diffusion network in the way that in every path, each edge has its own betweenness centrality. Edge betweenness centrality is an estimation of the load that each edge can pass through. Because each path is composed of some edges in series, and in serial path, current flows for all edges are equal and also the betweenness centrality would define the portion of current through which the edge could pass so the minimum betweenness centrality in each path is considered for all of the edges in that path. In power grid nodes and edges intrinsically are weighted components, but in previous models just one of them has considered as weighted. In the proposed model, both nodes and edges have been considered weighted and cascading failure process at both node and edge level of network has been studied. An important advantage of this model is usage of betweenness centrality instead of load flow. A. Edge Level In this part, network edges would be weighted with impedance. Thus basic network would be an undirected weighted network. Initial load of edges has been obtained by calculating load flow in ETAP software (for simulating power grid). Edge capacity illustrates tolerance of edge load that is inversely related to impedance (and directly related to admittance). Above data are used in edge level of proposed model. At first, network works normally. At the end, a weighted network including edges with definite load and capacity is obtained. Failure process starts with removal of an edge. In order to study the failure process, a new network is being created from original network based on shortest paths between two points of removed edge and it would be considered as diffusion network. These shortest paths is being chosen based on the weight of their edges so that the first shortest path has the lowest weight. Electrical current in power grid desires to flow through shortest paths, for this reason diffusion network consists of shortest paths between two points of deleted edge. In fact, calculations are done only on a part of this network which is effective in supplying electrical load of removed edge. (Calculation is done in a smaller graph.) First shortest path is more important in diffusion network due to its less weight. That is why edges’ weight in this path is set to have a greater role in load redistribution. According to the above, the new weight of each edge can be obtained from the following equation. wij = wij e ( k −1)W k (3) In (3), wij is the weight of edge between nodes i and j. The index k represents the k-shortest path between two Next equations is used for calculating new load on edges in the network. I Di = Bi ∑B k I deleted − edge (4) k In (4), Bi , Bk and I deleted − edge are betweenness centrality of edge i, the lowest betweenness centrality in path k, and load of removed edge respectively. I Di is the portion of load of edge i from extra load which is the result of removal. I new= I Di + I old −i −i In (5), I new−i is the value of new load of edge i. With the above operations, loads of edges in all shortest paths between two nodes of deleted edge are updated, in other words, load flow is done in the network. In this stage, load of network edges should not exceed their capacity. If load exceeds the capacity in one or more edges, that edge or the edge with highest load will be selected and removed. After removal of new edge, previous process would be continued and new load for network edges will be calculated. Process continues until load of all edges won’t exceed their capacities. It is essential to note that if a path is not found between two nodes, load of removed edge will have no effect on other edges and process will stop. B. Node Level At this level, type, load and capacity are dedicated to nodes. Nodes are generally divided into two categories: generators and loads. Load of each node is obtained from (5) SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia total active and reactive power substation (generator or load). Generators generate power, thus they have certain capacity and following equation is used to calculate node capacity. C= Ii × α i IEEE system bus 30 and its corresponding graph are illustrated in (a) and (b) of fig.1 respectively. The basic network (graph) is illustrated in fig.1. (6) In (6), Ci , I i and α are capacity of node i, load of node i and tolerance parameter respectively (in this model α is set to 1.1) but capacity is not considered for loads. In this part, new parameter called free capacity of generator substations (nodes) are defined which are obtained from the difference between total capacity of generators and their total load. If an edge is removed from a network and the node on both ends of this edge becomes isolated, more detailed examination and calculation is needed because if this node is a generator, the load of generator node would be deducted from free capacity and the load of isolated node would be removed from network. If the node is of load type, the load would be removed from network and this value would be added to free capacity. In each stage of repetition of failure process (edge removal), value of free capacity should be larger than zero because otherwise generators cannot handle loads and network would stop working. IV. (a) Failure process is started with removal of one edge. After the process is stopped, edges’ removal order and largest connected component is reported according to table I. Network edges are divided into two groups based on vulnerability. If any of the edges in group 1 is removed, failure cascades all over the graph and network become unstable. If any of the edges in group 2 is removed, failure process will be stopped and network would work normally. TABLE I. SIMULATION RESULTS The network under simulation is a standard test network IEEE (Institute of Electrical and Electronics Engineers) 30 bus. This network includes 30 nodes and 41 edges [13]. Nodes are generator and distribution substations. All the impedance and transformer lines of network are considered as edges. Each node has the characteristics of type, load and capacity. Each edge has the characteristics of weight, capacity and load. Cascading failure starts with the removal of one edge. To study the vulnerability of network edge, process is modeled by eliminating each edge. Simultaneously with the removal of an edge in the proposed model, the corresponding line of ETAP software is removed and then load flow is carried. ETAP software has been used in order to examine and assess results. ETAP offers a suite of fully integrated electrical engineering software solutions including arc flash, load flow, short circuit, transient stability, relay coordination, cable ampacity, optimal power flow, and more. In this paper, an open source programming language R is used for modeling cascading failure, which includes many packages for calculating centrality. The results of the two models are evaluated with two criteria, edges’ removal order and largest connected component (LCC). In this network, at first the proposed model is being explained and then result validation with of ETAP software will be done. The results show that the proposed model corresponds to conventional models based on two mentioned criteria. (b) Fig. 1. (a) IEEE system bus 30 .(b) its corresponding graph in R (b) CLASIFICATION EDGES IN IEEE 30 BUS Measure G r Deleted edge o u p 1 1 2 1 3 2 4 2 6 3 4 4 6 2 5 5 7 6 7 9 11 12 13 25 26 2 Other edges Edges’ removal order Edge 1-3 Edge 1-2 Isolation node 1 Isolation node 1 Unstable Unstable Edge 1-3 Edge 1-2 Isolation node 1 Unstable Edge 2-5 Edge 2-4 Edge 1-3 Isolation edge 1-2 Edge 1-2 Isolation edge 1-3 Unstable Edge 1-3 Edge 1-2 Edge 2-6 Edge 5-7 Edge 2-5 Edge 2-5 No path No path No path Isolation node 5 Isolation edge 5-7 Isolation node 11 Isolation node 13 Isolation node 26 LCC Isolation node 1 Isolation node 5 Unstable Unstable End of process End of process End of process End of process End of process End of process End of process To understand the above table, work process will be explained with removing some edges in two models (the proposed model and ETAP software). 29/ 30 29/ 30 29/ 30 28/ 30 28/ 30 29/ 30 29/ 30 29/ 30 28/ 30 29/ 30 29/ 30 29/ 30 30/ 30 SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia A. The proposed model If edge 1-3 from group 1 is chosen and removed for the start of cascading failure, diffusion network would consist of shortest paths between nodes 1 and 3. New loads of edges are calculated through edge betweenness centrality in diffusion network. Then, loads of all edges are compared with their capacities. In this stage, edge load 1-2 exceeds its capacity and will be removed. In result, node 1 which is a generator type will be isolated and the value of free capacity would become negative. Therefore, other generators cannot supply the loads and normal operation of network will be stopped. In this case, LCC will be 29/30 and network will be changed according to fig.2. show that substation 1 (main generator) is overloaded and network collapse is occurred according to fig.4. Fig. 4. Network resulted from removing link 1-3 in ETAP In ETAP software, transmission line between substation 9 and 11 has been removed, then load flow is performed. In result, it can be found that none of the substations are overloaded and load of removed line is handled by other lines. So network will operate in stable condition according to fig.5. Fig. 2. Graph resulted from removing edge 1-3 in R If edge 9-11 from group 2 is removed, because there is no path between these two nodes, diffusion network would not be formed and load of removed edge will not be distributed and failure process will be stopped. On the other hand, node 11 will be isolated. This node is a load type and removing it will keep value of free capacity positive. In this case, LCC will be 29/30 and network will be changed according to fig.3. Fig. 5. Network resulted from removing link 9-11 in ETAP The main advantage of the proposed model is reducing time-consuming calculation of load flow since the average time for making diffusion network and its calculation is 70 millisecond but the time of each load flow is more than 500 millisecond. V. Fig. 3. Graph resulted from removing edge 9-11 in R B. Validation with ETAP To evaluate the results in ETAP software, transmission line between substation 1 and 3 has been removed. After removing, load flow is performed. Results of simulation CONCLUSION AND FUTURE WORK In most of previous cascading failure models, just initial loading that causes to cascading failure, based on network centrality measures, has been studied. In these models, cascading failure process starts only with the removal of one node which is less probable [2, 4-6, 8]. Cascading failure usually happens in cases such as unexpected accidents and short circuit that leads to edge removal. In the proposed model, at first power grid is in normal operation mode and network stability will be examined by line outages. Therefore, in proposed model SRPioneers © Science and Research Pioneers Institute 7th International Conference on Electrical, Computer, Mechanical and Mechatronics Engineering (ICE-2017), 7-8 September 2017, Tbilisi, Georgia cascading failure starts with one edge removal and diffusion process is studied at both edge and node levels [12]. Combining study of cascading failure at node and edge levels simultaneously would lead us to gain better results in comparison with previous cascading failure models. In some cascading failure models degree and betweenness centrality measures had been used for initializing capacities and component loading. But these initialized values are differed from real values. In the proposed model physical and real quantities have been used for initial load and capacities and in this way, accuracy of calculations has been increased [1-8]. Since electrical current in power grid flows through shortest paths and considering edge betweenness centrality which calculates shortest paths, in the proposed model edge betweenness centrality measure in diffusion network has been used instead of complicated and time-consuming differential equitation calculations of electrical power flow. For validation results in IEEE 30 bus dataset based on components’ removal order and largest connected component, ETAP software has been used. It can be found that simulation results and the proposed model results are similar. Edges in this network are divided into two groups based on vulnerability. If any of the edges in group 1 is removed, failure cascades all over the graph and network become unstable. If any of the edges in group 2 is removed, failure process will be stopped and network would work normally. Stability of edges in group 1 is less than group 2, so damaging these edges in group 1 will disturb the network and it should be protected against various damages. This model can be applied on other standard test systems and real networks with defined parameters. This model can be used for all the infrastructure networks in which there are dynamic quantities, load and capacity of system components which play important roles in cascading failure. With adjusting these values and capacities, this can contribute to design and development of infrastructure systems and model network dynamics (such as consecutive trip of lines). In power systems, fluctuations from cascading failure cause network components to have different interaction on each other and make changes in load diffusion, frequency, domain, and voltage phase and protection device function, control, operator procedures and alarm and monitoring systems [12]. As a result, other dynamics can be added to edges and nodes characteristics in order to obtain a more exact model of cascading failure process. REFERENCES [1] X. Fang, Q. Yang and W. Yan, “Modeling and analysis of cascading failure in directed complex networks,” Safety Science. 2014 Jun 30;65:1-9. [2] S. Boccaletti, V. Latora, Y. Moreno, M. Chavez and DU. Hwang, “Complex networks: structure and dynamics,” Physics reports. 2006 Feb 28;424(4):175-308. [3] ZJ. Bao, YJ. Cao, GZ. Wang and LJ. Ding, “Analysis of cascading failure in electric grid based on power flow entropy,” Physics Letters A. 2009 Aug 17;373(34):3032-40. [4] JW. Wang and LL. Rong, “A model for cascading failures in scale-free networks with a breakdown probability,” Physica A: Statistical Mechanics and its Applications. 2009 Apr 1;388(7):1289-98. [5] J. Zhang, X. Xu, L. Hong, S. Wang and Q. Fei, “Attack vulnerability of self-organizing networks,” Safety science. 2012 Mar 31;50(3):443-7. [6] Wang and Jian-Wei, “Modeling cascading failures in complex networks based on radiate circle,” Physica A: Statistical Mechanics and its Applications 391.15 (2012): 4004-4011. [7] Y. Koç, M. Warnier, P. Van Mieghem, RE. Kooij and FM. Brazier, “A topological investigation of phase transitions of cascading failures in power grids,” Physica A: Statistical Mechanics and its Applications. 2014 Dec 1;415:273-84. [8] X. Peng, H. Yao, J. Du, Z. Wang and C. Ding, “Invulnerability of scale-free network against critical node failures based on a renewed cascading failure model,” Physica A: Statistical Mechanics and its Applications. 2015 Mar 1;421:69-77. [9] R. Albert, I. Albert and GL. Nakarado, “Structural vulnerability of the North American power grid,” Physical review E. 2004 Feb 26;69(2):025103. [10] E. Bompard, D. Wu and F. Xue, “Structural vulnerability of power systems: A topological approach,” Electric Power Systems Research. 2011 Jul 31;81(7):1334-40. [11] M. Newman, Networks: an introduction. OUP Oxford; 2010 Mar 25. [12] I. Dobson, BA. Carreras, VE. Lynch and DE. Newman, “Complex systems analysis of series of blackouts: Cascading failure, critical points, and self-organization,” Chaos: An Interdisciplinary Journal of Nonlinear Science. 2007 Jun 1;17(2):026103. [13] R. Kavitha, “Transient Stability of IEEE-30 bus system using ETAP Software,” International Journal for scientific and engineering research. 2012 dec 12, 3(12).
0
You can add this document to your study collection(s)
Sign in Available only to authorized usersYou can add this document to your saved list
Sign in Available only to authorized users(For complaints, use another form )