MDRP: An Energy-Efficient Multi-Disjoint Routing protocol in WSNs for Smart Grids
Article Category: Research-Article
Published Online: Aug 18, 2020
Page range: 1 - 15
Received: Mar 04, 2020
DOI: https://doi.org/10.21307/ijssis-2020-016
Keywords
© 2020 K. Deepa et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
A smart grid is a network that incorporates information and communication techniques with an electric power grid structure to improve the robustness and performance of the power distribution system (Ullah et al., 2017). The Advanced Metering Infrastructure (AMI) is an essential element of the smart grid, and smart meters are basic components of the AMI networks. In the conventional electric grid, recurrent transmission failures, congestion, and electricity theft are few elements that make the electricity grid ineffective in terms of electricity management. These pitfalls are due to unidirectional communication from the power generation to the consumer, whereas in the smart grid the communication is bidirectional (Ghosaland and Conti, 2019). WSN’s can be employed in the AMI networks of the smart grid to monitor, examine, and manage the various operations of the grid. The sensor nodes are restrained in-terms of processing functionality, energy, and memory (Yu et al., 2012). Data collection and aggregation from nodes are to be done timely as it is required to activate load shutdown under maximum usage, demand side management, emergency alerts, etc. Most of the data collection protocols assure the key application requirements such as delay reduction and trust along with energy efficiency. Packet dropping is one of the most important concerns in WSN’s due to the deployment of nodes in harsh environment. Apart from this poor link quality, non-availability of queue at intermediate nodes, improper selection of forwarder node for next hop may be major reasons for packet dropping (Mothku and Rout, 2019). For effective data communication, it is required to select a forwarder node which has a good link quality, high queue size, and good amount of residual energy (Mahmood et al., 2015; Lai et al., 2017). Due to the random deployment of sensor nodes, the routing mechanism ends in hot spot issues and also leads to exclusive density of nodes inside the monitoring area (Zaheeruddin et al., 2017). To cope with these issues and to obtain better performance of the network, researchers have proposed a work sleep cycle scheme for sensor nodes (Alfayez et al., 2015). The work sleep scheme results in opportunistic routing (OR), which facilitates to improve the overall performance of the network. OR allows the nodes within the network to overhear the transmission of the adjacent nodes for data forwarding. The primary characteristic of OR is its potential to transmit the data packet and to synchronize among the relaying nodes (Boukerche and Darehshoorzadeh, 2015). The performance of OR depends on the OR metric, algorithm involved in selection of nodes, and the coordination method used in coordinating the nodes (Biswas and Morris, 2005). The sensor data produced via deploying WSN in AMI networks may additionally have attributes like delay tolerance and sensitivity issues. In smart grids, the monitoring data (e.g. too many Ev’s charging) are delay sensitive as the data have to be communicated with the meter management system in specified time, while the control data (load shut down) may be considered as delay tolerant with the system requirements of smart grids. Designing an energy proficient routing scheme for the deployed WSN’s in AMI networks faces difficulties like data reliability, definite delay, node fault, packet drop, delay tolerance, sensitivity issues, etc. (Mahmood et al., 2015; Anees et al., 2019). In conventional routing, the time period for hello messages is fixed which leads to high energy consumption when the hello messages are frequently communicated. A probe message forwarding mechanism is also proposed here for communicating the source information. In this work, the data collection can be initiated by the sink whenever required. Inspired by the work of Yang et al. (2018), the probe message transmitting method is adopted here in the routing process. Once the subsequent forwarder node is selected via fuzzy logic and the source information is analyzed from the probe message, the link connectivity between the adjacent nodes is calculated, which results in the identification of opportunistic nodes for connection; hence, a spanning tree can be constructed with the sink node as root and optimal paths can be identified for data transmission. The following are the contributions in this paper: An energy-efficient routing scheme is proposed which can be used in AMI networks of smart grids. Using the probe message and work sleep scheme opportunistic node, connections are obtained. Considering residual energy, link quality, queue size, and closeness of node to sink a routing parameter has been computed. The performance of the proposed scheme is evaluated through simulation and the results are presented.
There exists a wide variety of routing protocols in the literature. Here, we surveyed suitable papers that are related to proposed work. To make the electric grid more efficient, NIST proposed a conceptual framework in Smart Grids. According to the framework both intra-domain and inter-domain communication must take place between all the building blocks in Smart Grids. It also suggests that both information and electrical flow should be supported in Smart Grids between all the building blocks of the Smart Grids. Data collection, data processing, and data aggregation among energy subsystems constitute information flow and electrical flow within the electrical energy distribution, transmission and generation (Fang et al., 2011; NIST, 2013). The purpose of information flow is to regulate the power distribution and the electrical flow takes care of power delivery, demand response, and so on. For effective information exchange within the Smart Grids domains, a highly intellectual communication infra-structure is required. WSN’s can also be used to prevent power theft, handle demand response seamlessly, and make real-time decisions at the prosumer end. Ma et al. (2013) have presented the communication architecture of Smart Grids. Fadel et al. (2015) demonstrated that WSN’s can be utilized effectively in Smart Grids entities. LEACH (Heinzelman et al., 2000) is one of the vital and broadly used protocols for routing in WSN’s. The routing is accomplished through cluster heads that are elected periodically based on a predefined probability value. LEACH gives identical possibility for every sensor node to emerge as cluster head. However, routing through LEACH does not take into account few parameters like energy consumed by each node, geographical positioning of the nodes in the case of asymmetrical clusters. HEED was proposed by authors (Younis and Fahmy, 2004) which rectifies the shortcomings of LEACH in terms of uneven formation of clusters. It also rotates the cluster heads uniformly across the sensor network in multiple rounds of communication based on rotation policy. PEGASIS (Aliouat and Aliouat, 2012) was the upgradation of LEACH. A series of sensor nodes is created for transmitting and receiving the collected data from sensors. But this technique is not appropriate for huge networks because it wastes energy in forming the cluster and electing the cluster heads. In EADEEG (Priya et al., 2017), the cluster heads are elected based on a metric. The metric is the energy of adjacent nodes to the energy of the self-node itself ratio. In EADEEG, nodes close to centroid are summarized and few nodes that are close but not close to centroid are not assigned to any cluster resulting in isolated points. FEAR was proposed by authors (AbdulAlim et al., 2013) which used a ranking scheme to rank the neighboring nodes. A tree is constructed for each transmission based on energy. In this method, however, clusters are not formed; instead, a tree is constructed to enhance the network life time. Although communication via clustering of sensor nodes is highly energy efficient, it yet ends with certain issues in the network. As the cluster heads are chosen primarily based on a preset probability value, real-time load balancing can’t be accomplished. Also, multiple route request packets and control messages for cluster formation cause overhead problem. Most of the clustering protocols don’t consider the location of the base station for routing which is one of the main reasons for hot spot problem in multi-hop communication networks. To overcome this, some unequal clustering strategies are available in the literature like EAFA, EDUC, and UHEED (Yu et al., 2011; Ever et al., 2012; Bagci and Yazici, 2013). Mhemed et al. (2012) proposed a cluster formation protocol to identify the next possible head in the cluster through fuzzy logic by considering distance as a main parameter to enhance the network lifetime. A super-cluster head approach was presented by authors (Selvi et al., 2016), in which the base station receives the sensed data from the super-cluster head rather than from cluster heads. The super-cluster head aggregates the data collected by the normal cluster heads. The routing is decided by fuzzy logic to improve the performance of the network. Most of the fuzzy-based routing protocols discussed in the literature are not able to tune the membership function as input–output pairs change with environment. A genetic-based virtualisation approach was used by authors (Kaiwartya et al., 2017) to handle torrent delay and energy consumption in IoT networks. WCA (Jian-wu et al., 2008) was proposed by authors in which weights of nodes are calculated before creating the node information table. Due to excessive computation, high energy is consumed by the nodes in WCA, making the network unstable. FLEOR was proposed by authors (Julie and Selvi, 2016) which calculates the shortest path for each round of communication. In this method, the network life time is improved; however, it doesn’t take care into account the packet loss rate and also it is not fault tolerant. Path cost was computed using the link quality by authors in (Anees et al., 2019). The fundamental blocks for OR were presented in by the authors (Zeng et al., 2013). ExOR was proposed by authors (Biswas and Morris, 2005) that is considered as the basic protocol implemented using opportunistic routing which sends packets in batches and the nodes overhear their adjacent nodes to participate in data forwarding. A reinforcement learning-based adaptive OR was suggested by authors (Zhang and Huang, 2006) for ad hoc networks to estimate the optimal hop count. With the aid of work sleep cycle scheme, the various nodes deployed in WSN’s can implement the OR logic for data transmission. Authors (Guntupalli et al., 2018) with the help of work sleep cycle achieved life-time improvement. Authors (Ng et al., 2017) proposed energy efficient routing algorithm based on synchronisation of wake up, traffic and sleep cycle. In smart grids two types of traffic data are routed (delay sensitive and delay tolerant) so it is required to minimize the energy consumption to enhance the network life span. DCBONC was proposed by authors (Yang et al., 2018) for data collection in sensor networks using opportunistic node connections. Here, a random graph is constructed considering the sink as a root node, and then, optimal path is calculated for data communication within the radio range. Certain ideas and concepts from the literature are utilized to propose a solution to enhance the network lifetime using the MDRP routing protocol.
The network and energy model used for routing is discussed in this section. It is assumed the sensor nodes in the AMI network are deployed in random fashion. The nodes are independent, and the AMI gateway is the sink for data collection and monitoring. All the sensor nodes utilizes the work sleep approach to sense and communicate with sink. The nodes can communicate with each other throughout the working mode. Multiple sink strategy is also supported by the network for data collection. A specific id is allocated to each sink node (
Figure 1:
Asynchronous work-sleep cycle approach for nodes in the network.
To receive ‘b’ bits by node
In this section, the energy proficient routing approach is presented. In the proposed data collection scheme, the sink node is stationary and can initiate data collection whenever required. The data collection process includes the following five phases: Initialization phase. Probe message transmission phase. Subsequent node selection phase. Path estimation phase. Routing phase.
Data collection process can be initiated by the sink node randomly whenever required by transmitting a tag message. The tag message includes data collection period and the specific
For the network graph creation, it is considered that sink node remains in a working mode all the time and any sensor node in the network falling within the short radio range can establish communication with the sink at any time and the nodes have the ability to collect the information about their adjacent nodes.
When the tag message is received by the sensor nodes from the sink node, the sensor nodes transmit their information to their sink node (i.e. its working time, status transition, source ID, work sleep scheme and their adjacent node id). The format of probe message which is transmitted to sink is shown in Table 1.
The design of the probe message.
| Field | Sid | Work sleep cycle | STF | Adjacent node id | Sink id | Fid | EOH |
Figure 2:
Illustration of probe message transmission.
Figure 3:
Probe message transmission.
Fuzzy-based subsequent node selection for routing in AMI networks is discussed in this section. Upon receiving the probe message, the sink node initiates data collection process by constructing a random graph with sink as its root. In this work, the subsequent node for data transmission is chosen by fuzzy logic, to choose the subsequent node for transmission Node’s residual Energy (NRE), Link quality (LQ), Queue size (QS) and Status transition frequency (STF) are considered. Every node in the network must find a fuzzy output value to its adjacent node based on the input parameters. Upon calculating the fuzzy output value, the subsequent node for hopping is chosen.
The formation of the network graph depends on the work sleep cycle of the nodes and also on the probe message information obtained during the initialization phase. NRE, LQ, QS, STF are the fuzzy inputs considered here. The I/O variables make the use of the triangular membership function to signify the linguistic terms. The mapping between the input space (universe of discourse) and the membership value is described by the membership function. The nodes energy is computed with the help of Equations (1) and (2). The relationship between the input output variables with the corresponding linguistic terms is shown in Table 2. Fuzzy rules are established using a heuristic approach which is based on the following axiom. The subsequent node for hopping is selected based on high energy, available queue size, high status transition frequency, and good link quality. The default rules and membership function guide to obtain the fuzzy output variable. Figure 4A-D describes the fuzzy inputs and their corresponding input linguistic terms used here and Figure 5 describes the fuzzy output variable.
Fuzzy I/O variables and their linguistic terms.
| I/O variables | Linguistic variables |
|---|---|
| Residual energy of node (NRE) | Low, Medium, High |
| Queue size (QS) | Low, Medium, High |
| Status transition frequency (STF) | Low, Medium, High |
| Link quality (LQ) | Poor, Moderate, Good |
| Possibility of turning into subsequent node for hopping | Low, Weak, Medium, High, Very high |
Figure 4:
Membership functions for input variables.
Figure 5:
Membership function for output variables.
The obtained fuzzy variable as output will be converted to a single crisp value which signifies the chance for becoming next node for hopping. The fuzzy output variable is obtained by the default rules and membership function used. In general, the fuzzy logic system consists of four modules fuzzifer, rule set, inference engine, and defuzzifier. In the fuzzifier, the crisp input parameters
The rule set comprises a set of if-then rules that are generated through the Mamdani approach. The set of rules are presented in Table 3. Inference engine infers the output with the help of if-then rules. Defuzzifier converts the fuzzified value into a crisp output value which signifies the chance of become next node for hopping. The centroid technique is used in the defuzzification process. The default rules and membership function guides to obtain the fuzzy output variable. The process of subsequent node selection is described with help of Figure 6.
Fuzzy decision rules.
| Residual energy | Queue size | STF | Link quality | Chance of becoming subsequent node |
|---|---|---|---|---|
| Low | Low | Low | Poor | Low |
| Low | Low | Low | Moderate | Low |
| Low | Low | Low | Good | Weak |
| Low | Low | Medium | Poor | Low |
| Low | Low | Medium | Moderate | Medium |
| Low | Low | Medium | Good | Medium |
| Low | Low | High | Poor | Weak |
| Low | Low | High | Moderate | Weak |
| Low | Low | High | Good | Medium |
| Low | Medium | Low | Poor | Low |
| Low | Medium | Low | Moderate | Low |
| Low | Medium | Low | Good | Low |
| Low | Medium | Medium | Poor | Low |
| Low | Medium | Medium | Moderate | Weak |
| Low | Medium | Medium | Good | Medium |
| Low | Medium | High | Poor | Low |
| Low | Medium | High | Moderate | Weak |
| Low | Medium | High | Good | Medium |
| Low | High | Low | Poor | Low |
| Low | High | Low | Moderate | Low |
| Low | High | Low | Good | Low |
| Low | High | Medium | Poor | Low |
| Low | High | Medium | Moderate | Weak |
| Low | High | Medium | Good | Medium |
| Low | High | High | Poor | Weak |
| Low | High | High | Moderate | Medium |
| Low | High | High | Good | High |
| Medium | Low | High | Poor | Low |
| Medium | Low | High | Moderate | Low |
| High | Low | Low | Poor | Low |
| High | Low | Low | Moderate | Low |
| High | Low | Low | Good | Weak |
| High | Low | Medium | Poor | Low |
| High | Low | Medium | Moderate | Weak |
| High | Low | Medium | Good | Medium |
| High | Low | High | Poor | Low |
| High | Low | High | Moderate | Weak |
| High | Low | High | Good | Medium |
| High | Medium | Low | Poor | Weak |
| High | Medium | Low | Moderate | Medium |
| High | Medium | Low | Good | High |
| High | Medium | Medium | Poor | Medium |
| High | Medium | Medium | Moderate | High |
| High | Medium | Medium | Good | Very High |
| High | Medium | High | Poor | Medium |
| High | Medium | High | Moderate | High |
| High | Medium | High | Good | Very High |
| High | High | Low | Poor | Weak |
| High | High | Low | Moderate | Medium |
Figure 6:
Subsequent hop node selection.
In the proposed algorithm, the subsequent node for routing is chosen using fuzzy logic. Once the subsequent node is known, data collection process is initiated by construction of spanning tree. With help of this, the optimal multi-disjoint paths under traffic considerations toward the sink can be computed. To create the tree structure, it is presumed that the sink node is always active and the sensor nodes located within the radio range
The updated path connectivity values from intermediate and distant nodes is obtained by the sink node. In each round of communication, the sink node compares the new PC values with the earlier PC values to choose the maximum PC value. The process is repeated until the sink receives the information from all nodes:
If PC is updated then:
Once the spanning tree is formed the sink broadcasts to the nodes that lie in the long radio range (
Once the spanning tree is formed, the routing scheme is to be defined. The optimal path on the spanning tree is the appropriate path to route the sensed data to sink node. The conventional opportunistic routing just incorporates the asynchronous work sleep cycle of the sensor nodes leading to link failure if any forwarder node in the network is in the sleep state. Also, it doesn’t support the different data constraints of the smart grids, so an alternate multidis-joint path opportunistic routing is required to adapt the nodes different operating status for a reliable data transmission. If a node
Figure 7:
Multi disjoint path selection for data communication on the spanning tree.
Whenever the node
In general, the AMI networks are deployed as static multi-hop networks. The sensor nodes are assumed to be autonomous and alike with communication range of 20 m. The various simulation attributes are described in Table 4. The performance of the MDRP protocol is evaluated with help of Matlab and compared with few other routing protocols like POFA (Chang et al., 2012), DCBONC (Yang et al., 2018), and EXOR (Biswas and Morris, 2005).
Simulation attributes.
| Parameters | Values |
|---|---|
| Size of the network | (500 × 500) m2 |
| No of mobile sink | 1 |
| No of nodes in the network | 500 |
| Mobility pattern | random |
| Time duration for data collection | 600 s |
| Communication range between sensor nodes | 20 m |
| Node’s initial energy | 2 J |
| Size of the buffer | 1,024 bits |
|
|
50 nJ/bit |
|
|
0.0013 pJ/bit/m4 |
| Size of probe message | 120 bits |
| Size of data packet | 1,024 bits |
It is the total energy consumed by nodes to the total number of sensor nodes deployed in the network ratio. The energy level of the network is proportional to the network life span, so this parameter is important. The network life span prolongs with a lesser energy consumption ratio.
From Figure 8, it is obvious that as the number of nodes increases the energy consumed by the nodes in the network also increases. Figure 9 illustrates with increase in radio range the number of hops also increases. It is obvious the number of hops in POFA is large compared to other protocols. In the proposed MDRP protocol with the expansion of the radio range, the network density remains unchanged whereas if the distance between the source and sink increases it may affect the success rate of data transmission. It can also be noted that MDRP performs better than DCBONC as the subsequent node for data communication is calculated through fuzzy logic which helps the nodes to save their energy as the radio range increases. The nodes consume slightly more energy in the proposed MDRP protocol when compared with EXOR and much lesser when compared to POFA. In POFA with increase in network density, the energy consumed by the nodes also increases. This is due to the fact, as network density increases, it selects multiple data forwarders which, in turn, increases the number of hops resulting in more energy consumption. It consumes lightly higher energy when compared to EXOR because MDRP uses asynchronous work sleep strategy for data communication. When the adjacent node is in the sleep state, the node involves in various process like path calculation and selection, fuzzification, etc., which results in energy consumption. It performs better than DCBONC because in MDRP the process of subsequent node selection is not done separately which helps in Energy saving.
Figure 8:
Energy consumed by nodes in the network.
Figure 9:
Hop count variation with increase in radio range.
Figure 10 illustrates that energy consumed by the nodes increases with time. Energy consumption is the energy consumed by the nodes during data transmission regardless of the status of data delivery. EXOR consumes 95% more energy, POFA consumes 260% more energy, whereas DCBONC consumes 55% more energy when compared with the proposed MDRP protocol. The maximum energy consumption in MDRP protocol is during the probe message forwarding and data transfer. The EOH metric, subsequent nodes selection and the path cost calculation by the sink helps in decreasing the energy consumption in the network.
Figure 10:
Energy consumed vs network simulation time.
It is the ratio of packets received by the sink node to number of packets transmitted by the sensor nodes. When the data packet is forwarded to the adjacent nodes during working state, it is considered as successful packet transmission and it is unsuccessful (packet drop) when a particular node doesn’t find or have any active adjacent nodes (i.e. the node may be in sleeping mode or at transition mode). PDR is evaluated with respect to the radio range in this case. Figure 11 illustrates that PDR in POFA increases with radio range, whereas in other protocols like DCBONC, EXOR, and MDRP the PDR decreases with radio range. In POFA, as the radio range increases, more nodes are involved in data collection increasing the number of hops for data communication which reduces the efficiency in data delivery, whereas in the proposed MDRP protocol, EXOR and DCBONC PDR decrease with increase in radio range. Figure 11 shows that the proposed MDRP protocol performs better in terms of PDR. The proposed MDRP protocol achieves 135%, 60.26%, and 30% over POFA, EXOR, and DCBONC.
Figure 11:
Packet delivery ratio.
It is one of the most significant criteria to assess the network performance. By measuring the number of dead nodes, network lifetime can be estimated. The network lifetime can be computed from the energy drained by 25% of nodes in the network. From Figure 12, it can be observed that with increase in time the nodes death rate also increases. In this simulation, the data collection period is 600 seconds and the data collection process can be initiated randomly within the network. The nodes die faster in the case of POFA when compared to DCBONC EXOR and MDRP. It can be observed from Figure 12 that MDRP performs better than all schemes. This is due to the selection of proper node as next hop node through fuzzy logic and the calculation of optimal paths for the subsequent forwarders in the network. It can also be observed that MDRP performs better than DCBONC. This is due to the proper selection of next hop node. DCBONC uses time frequency parameter to estimate the optimal paths for communication, whereas MDRP uses buffer size link quality and nodes residual energy to attain better network performance.
Figure 12:
Network lifetime.
One of the important parameters to evaluate the performance of energy efficiency in QoS routing protocols is average end to end delay. In this simulation, the average packet delay is considered. Figure 13 illustrates the average packet delay with respect to the packet receiving rate for various protocols discussed in this paper. Here, the change in delay is measured against the rate of packet arrival. MDRP performs better than other protocols due to the queuing model usage in the AMI environment of smart grids. DCBONC don’t differentiate between the traffic types which results in a slightly lower average packet delay. MDRP performs better as it calculates the link and path connectivity for data transmission.
Figure 13:
Avg end to end delay.
An energy Efficient adaptive fuzzy-based multi-disjoint routing protocol is proposed for WSN’s which can be employed in the AMI networks of SG. In the proposed MDRP protocol, the sink node can initiate a data collection randomly whenever required by broadcasting a tag message. Upon receiving the tag message, the sensor nodes formulate their work sleep schedule and the probe message is transmitted to the sink as an acknowledgment. The probe message has information about the source node, work sleep agenda, status transition frequency of the nodes and subsequent node for data transmission. When the probe message is received, the sink node constructs a spanning tree with sink as root. Then, it calculates the path connectivity cost (PC) to find the optimal path for each sensor node. Based on these factors, a routing protocol is proposed that adapts with the operating status of the node.
The possible future work could be integrating this scheme to cognitive radio sensor networks for handling predictive channel assignments. Also, the scheme can be implemented with model predictive controllers for predicting the changes in the demand side management, real-time traffic issues like faults, blackouts, power surges, etc., in smart grids.