Energy-Efficient Fire Monitoring over Cluster-Based Wireless Sensor Networks

Uncontrolled fires occurring in wild areas cause significant damage to natural and human resources. Many countries are looking for ways to fight forest fires at an early stage using sensor networks, by integrating IT technologies. Studies in the fire-related sensor network field are broadly classified into efficient processing of fire data on sensor nodes and energy efficiency during communications among wireless sensor nodes in case of fire. Most studies of sensor network energy efficiency so far mainly focus on extending the connectivity of the entire network and minimizing isolated nodes by applying power evenly to each sensor node through efficient cluster-based routing. This paper proposes an energy-efficient fire monitoring protocol over cluster-based sensor networks. The proposed protocol dynamically creates and reorganizes the sensor network cluster hierarchy according to the direction of fire propagation over the sensor network clusters. This paper also presents experimental results to show that the proposed protocol is more energy efficient than fire monitoring with existing cluster-based sensor network protocols.


Introduction
The world is facing many risks caused by natural disasters such as forest fires, floods, and abnormal climate changes.IT scientists are studying ways of effectively solving such problems by looking at risks at an earlier stage.This study was conducted on a new network for detecting risk factors and rapidly responding to these problems.Wireless sensor networks are regarded as the best systems for applications in those environments.However, current wireless sensor networks have problems in fire monitoring over wide areas because of limited battery capacity and the short life spans of the sensor nodes [1].Unlike general environmental data monitoring such as precipitation or climate monitoring, fire monitoring requires very frequent real-time data transmissions until the fire is put out.Consequently, the sensor nodes in fire monitoring applications consume much more energy in relatively short periods of time than do sensor nodes in general environmental data monitoring applications, and they finally have shorter life spans [2].
For fire monitoring applications, flat-based routing protocols for sensor networks, such as SPIN (Sensor Protocols for Information via Negotiation), are not appropriate since all the sensor nodes that detect fire start sending fire data to the sink node individually and consume energy of the sensor network very quickly.The existing cluster-based routing protocols for sensor networks, such as LEACH (Low-Energy Adaptive Clustering Hierarchy) and TEEN (Thresholdsensitive Energy Efficient sensor Network protocol), are more energy-efficient for fire monitoring applications than the flat-based routing protocols since each cluster head can collect fire data from local sensor nodes, encapsulate the collected data into a single data packet, and transmit the packet to the sink node at a time.However, as the number of clusters that detect fire increases due to the wide propagation of the fire, the energy efficiency of fire monitoring with cluster-based routing protocols will decrease [2][3][4][5][6].
This paper proposes EFMP (Energy-efficient Fire monitoring Protocol), a fire monitoring protocol operating over cluster-based sensor networks.To further increase the energy efficiency of cluster-based sensor networks for fire monitoring, EFMP reduces the number of transmissions of fire data from the cluster heads to the sink node by dynamically creating and reorganizing the sensor network cluster hierarchy International Journal of Distributed Sensor Networks according to the fire propagation over the sensor network clusters.Fire monitoring experiments showed that EFMP consumes 41% and 12% lesser energy on average per node than do LEACH and TEEN, respectively.Furthermore, the number of sensor nodes that survived in the EFMP-based experiment was 12% and 14% more than that in the LEACHand TEEN-based experiments, respectively.
The rest of this paper is organized as follows.Section 2 analyzes the existing energy-efficient sensor network protocols and the problems faced when using these protocols in fire monitoring applications.Section 3 describes the design of the EFMP proposed in this study.Section 4 compares the performance of EFMP in fire monitoring environments with the performances of existing cluster-based sensor network protocols using NS-2.Finally, Section 5 summarizes the paper and future directions as conclusion.

Related Works
This section explains typical energy-efficient routing protocols for sensor networks such as SPIN, LEACH, and TEEN, which can be used in fire monitoring applications.

SPIN.
SPIN is a method proposed for avoiding repetition of data transmission when the same data are sent to multiple nodes.SPIN sends ADV (Advertisement) messages from sensors collecting data.It asks for readiness to receive data by nearby sensors and the sensor receiving the ADV message sends a REQ (Request) message once it is ready.The sensor receiving the REQ message sends its own sensor data.However, the REQ is not sent if it is not ready.Data are, therefore, sent only to sensors that sent REQs since sensors not sending REQs are not ready.The drawback of SPIN in fire monitoring system environments is that the sensor only sends a data packet once it receives a REQ after sending an ADV packet.If a sensor receives four REQ packets, then the sensor sends four data packets individually.Packets are delivered to nodes that are not relevant to the actual transmission path in this case and this wastes battery power [2,3].The main drawback is high consumption of battery power by numerous transmissions from networks requiring many nodes, as in the case of fire monitoring system environments, since the sensors are connected un hierarchically.In addition, transmission is not completed if there are isolated sensors, and battery power is consumed between nearby nodes and isolated nodes since ADV packets continue to be sent for connection.These problems may occur in the case of a forest fire where a large number of sensors are installed over wide area [7][8][9].

LEACH.
LEACH is a protocol for forming efficient clustering; sensor nodes form local clusters by themselves and distribute energy evenly.One node-forming cluster acts as a cluster head, and the sensor node with the most energy operates as a cluster head; when its energy capacity becomes smaller than those of the other nodes, another node with the most energy then performs the role of cluster head.The battery drains quickly when a sensor node acts as a cluster head; the burden on the node operating as the cluster head is, therefore, reduced by rotating the cluster heads.LEACH is currently on of the most popular method for environmental monitoring system.However, LEACH may waste a lot of energy since each individual cluster head operates independently in a fire monitoring system with a large number of sensor nodes.The number of related clusters increases as the fire spreads widely.Increases in the number of cluster heads sending information and the number of transmissions in the sensor network will result in a reduction in energy efficiency [2,3,10,11].

TEEN.
TEEN is a cluster-based and reactive routing protocol that works with two threshold values: a hard threshold and a soft threshold.A hard threshold is a threshold value for the sensed attribute.Nodes sensing this value only turn their transceivers on if the sensed value is above the defined threshold.A soft threshold is a small change in the value of the sensed value; it triggers the node to switch on its transceiver to become active and transmit.The sensed value is stored as an internal variable in the node.Every time a new cluster head is selected, the threshold values can change.When the cluster node's value exceeds the soft threshold, then it starts sensing again.The advantage of this protocol is the reduced number of deliveries, making it highly efficient in terms of energy consumption and response time [3,4,12,13].Nevertheless, TEEN may also suffer from the drawback of typical cluster-based routing protocols in fire monitoring environments.

Protocol Stack for EFMP.
The method of transmission from the sensor to the cluster head and the cluster head to the sink is based on the existing cluster structure.That is, the methods for partitioning the sensor network into clusters and election of cluster heads are done by the underlying cluster-based routing protocol.
Figure 1 shows the EFMP protocol stack designed in this study.The stack is broadly composed of the system layer and the protocol layer.The system layer includes the sensor node OS and hardware, and the protocol layer includes the MAC (Medium Access Control) protocol responsible for communication by the sensors.The protocol layer has cluster-based routing protocol, such as LEACH and TEEN, over the MAC protocol.The EFMP proposed in this paper is on top of the protocol stack and can improve energy efficiency by reducing the number of transmissions by dynamic hierarchical clustering, while maintaining the existing protocols; this is the main function of the EFMP since it controls cluster hierarchy by operating on existing cluster-based routing protocols.

Cluster Hierarchy and Roles of Cluster Heads.
The major difference between the sensor network cluster organization of an EFMP and that of existing cluster-based routing protocols is that the EFMP has a hierarchical cluster structure that can be reorganized dynamically.As illustrated in Figure 2, in an EFMP, the cluster heads are layered and classified into master heads and slave heads.The structural difference between a general cluster structure and the proposed EFMP general cluster structure is the presence of a master head that collects and manages data from cluster heads.The master head collects information from the cluster heads and sends the collected information to the sink node.
Existing clusters only collect and send information from the cluster head.They use up large amounts of energy since each cluster head independently transmits data if there are many cluster heads.EFMP achieves better efficiency by reducing the number of transmissions compared with cluster-based cluster head methods by electing a master head and transmitting information by collecting data, in the case of a fire, from sensor nodes, according to information on the fire, as shown in Figure 2. Sensor nodes are the smallest unit forming a sensor network.All the sensor nodes can be slave heads or master heads, depending on the environment.If a master head is elected among the cluster heads, all the cluster heads except the master head become slave heads.
Slave heads send information to the master head.The master head sends information to the sink node in batches by collecting data received from the slave heads.From among the existing master head candidates, the cluster head with the least number of transmissions to the sink node is selected as the master head.If the number of transmissions is the same, then the one that is closest to the sink node, with the least number of sensors detecting fire, or the cluster head with the most battery power, is selected as the master head.
Figure 3 shows the changes in the roles of the nodes according to changes, in the case of a fire, and all the nodes are initialized while in watch mode.
Watch mode refers to the initial state, in which the sensors did not detect a fire.The cluster head that first detects a fire changes from watch mode to master mode.It transforms the nearby cluster heads in watch mode to slave mode since it is in master mode itself, that is, the EFMP system displays much better energy efficiency than existing cluster networks do since the system makes hierarchical networks dynamic by changing the cluster head into a cluster head in watch mode, slave mode, or master mode, according to the direction of the fire.

Types of Packets.
The features of EFMP are electing a master head that collects and manages information from slave heads managing clusters.The location of the master head changes according to information on the fire (direction of the fire).The passage defines the structural design of the EFMP and the algorithm and packet type for deciding a master head.
(1) SIG FIRE Packet.If a sensor node detects fire or fire data, it immediately sends a SIG FIRE packet to its cluster head.The fire is detected if the following condition is met, where TEMP (t n ) represents the temperature measured by a sensor node at the current time t n , ( tn−1 t=t1 TEMP(t))/(n − 1), represents the average temperature measured from time t 1 to t n−1 (i.e., the average of all temperatures measured before t n ): Thus, if the difference between the currently measured temperature and the average of the previously measured temperatures is greater than a specific limit ΔTEMP MAX , the sensor node transmits a SIG FIRE packet to its cluster head.the unique identifier of a sensor node (e.g., the MAC address or IP address of a Zigbee node), and the Sensor data field includes currently measured temperature data.
(2) SIG DATA Packet.SIG DATA is gathered at each cluster head to be transmitted to the sink node.The sensor monitors the fire and sends information to the slave heads by detecting fire data.The sensor node receiving information relays the fire data (SIG FIRE) collected to a master head, and SIG DATA is used when sending information from slave heads to the master head.Figure 5 shows the structure of the SIG DATA packet.The SIG DATA packet consists of Head ID, Sensor ID N, and Sensor data N fields.The Head ID field is a unique identifier of the cluster ID and the Sensor ID N is the sensor node's ID in the cluster header.
Sensor data N indicates the fire data collected from all the sensor nodes in the cluster head.
(3) SIG INFORM Packet.The SIG INFORM packet is used when a new master head provides its own information to a slave head.It is used when the first slave head within a cluster detects a fire and is changed to the master head and provides master head information to nearby sensors as well as information about the new master head when the master head is replaced.The Head ID field is a unique identifier of the cluster ID and Bat capacity is the remaining battery capacity.
Num sensor is the number of sensor nodes, and Sensor ID N is the sensor node's ID in the cluster head.Sensor data N is the fire data collected from all the sensor nodes in the cluster head.
(4) SIG QUERY and SIG RESP Packets.The slave head within the cluster is transformed to the master head once the sensor detects a fire within its own cluster.The slave head becomes a master head candidate and the previous master head asks nearby master head candidates whether it is the appropriate master head.The packet format sending this information is called SIG QUERY and the response is called SIG RESP.Figures 7 and 8 show the structures of the SIG QUERY packet and the SIG RESP packet.The SIG QUERY packet has a unique identifier, Master ID, and Num candidate is the number of master candidates.SIG RESP has a unique identifier, Slave ID.The Hop to master field is the number of hops between the master heads.Candidate ID N is the master head candidate.
(5) SIG LISTEN Packet.Address information on the new master head should be sent to nearby slave heads once a new master head has been elected.Figure 9 shows SIG LISTEN packet.The SIG LISTEN packet provides the address information of the newly elected master head.
(6) SIG TRANS and SIG RESET Packets. Figure 10 shows structures of the SIG TEANS, SIG RESET packet.SIG TRANS is used for registering information on the new master head by receiving the SIG LISTEN packet from the new master head.It is also used when a slave head registers its own address information.The SIG RESET packet is used when the previous master head again becomes a slave head by handing over its authority as a master head to the newly elected master head, and when the reverted previous master head registers its information with the new master head.

EFMP Operating Procedures.
EFMP not only sends information acquired by fire detection but also provides hierarchical clusters by selecting a master head according to the direction of the fire and reduces energy consumption by sensors.EFMP is composed of a FIRE Detection part for detecting a fire and a Startup Monitoring part for sending information on fires.This section describes the detailed operating procedures of EFMP in two parts.
(1) Fire Detection in Sensor Nodes.Fire detection procedures refer to sending information to cluster heads by detecting fire for the first time from the information collected by sensors.A sensor node has an initial value of 0 for TEMP SUM , the current temperature as the initial value of TEMP VEA , and 1 for the initial value of n, as shown in Figure 11.The sensor node sends the SIG FIRE packet to the cluster head if the current temperature value is inserted into TEMP CUR and   of a fire breaking out from its own cluster.Figure 12 shows procedure of fire monitoring startup The cluster head recognizing the fire transforms to master mode from watch mode, and it sends a SIG LISTEN packet containing detection of fire to the other cluster heads and sends the SIG DATA packet to the sink node.
(3) Monitoring Procedure.Figure 13 shows monitoring procedure.The sensor node for detecting fire has the function of detecting fire continuously.Detected information is sent to slave heads and the master head through a SIG FIRE packet, and each slave head collects and relays sensor information.The slave head in a fire zone, on detecting a fire within its own cluster of slave heads, transmits that it is a candidate for master in the case of a fire through the SIG INFORM packet.Conversely, slave heads send the SIG DATA packet to the master head, thinking that they are not in a fire zone when they do not detect a fire.The master head sends all the collected SIG DATA received from the multiple slave heads to the sink node in the batch.
(4) Master Election.The master head of an EFMP is not a fixed type but can change continuously, depending on the fire.In other words, a new cluster will participate in fire monitoring when the fire spreads to a new cluster head.The following procedures are used to elect the new master head.
Suppose that M represents a set of master candidates that includes the current master head and all the other heads newly involved in fire monitoring, and the element of M are represented as m 1 , m 2 , . . ., m L .In addition, existing slave heads are represented by s 1 , s 2 , . . ., s N and the sink node is represented by sink.The master election criteria are represented by the following formula, where DIST (a, b) represents the number of transmissions from node a to node b: DIST tot (m x ) represents the total number of transmissions from each slave head to the sink node via the newly elected International Journal of Distributed Sensor Networks master head m x .Thus, for all m x ∈ M, the m x that has the minimum DIST tot (m x ) value is elected as the new master head.
The current master head asks each master head candidate, using the SIG QUERY packet, for the information needed to evaluate the master election criteria, and each master head candidate replies with the SIG RESP packet, as shown in Figure 14.If a new master head is elected according to the master head criteria, the current master head notifies the new master head with the SIG TRANS packet, and then, the new master head sends the other slave heads information about itself through the SIG LISTEN packet.3.5.Fire Monitoring with EFMP.The cluster head in EFMP operates in watch mode before detection of a fire, monitors fires from its own zone, and collects sensor information.The sensor of a cluster head in watch mode, as shown in Figure 15(a), sends fire information to its own cluster head once a fire is detected within its own cluster.
The cluster head receiving information on a fire from the sensor recognizes a fire within its own cluster for the first time and changes into a master head from watch mode.First, the elected master head sends information on the fire within its own cluster to the sink node (Figure 15(b)).It then sends the information about itself to nearby cluster heads in the slave mode through a SIG LISTEN packet (Figure 15(c)) and the slave heads update their master head information.The cluster head where the fire broke out is elected as the first master head.A new master head, elected according to the progress of the fire, sends collected data to the sink node, as shown in Figure 15(d).It acts as the master head until another appropriate master head candidate is elected.protocols.When using LEACH EFMP, 58 out of 300 nodes were still alive (i.e., approx.19% of the entire nodes) after 180 min.However, only 22 nodes (i.e., 7% of the entire nodes) were alive when using the original LEACH protocol under the same condition.TEEN EFMP also gave better results than did the original TEEN.TEEN EFMP had 103 nodes (i.e., 34%) alive, as opposed to the case of the original TEEN protocol, where only 60 nodes (i.e., 20%) were alive.

International Journal of Distributed Sensor Networks
Figure 18 shows the average energy consumption (including energy consumed for initial cluster formation of the underlying routing protocol) over all the sensor nodes during the experiments; this was calculated using the following formula, where C(S i , t) represents the energy consumed by a sensor node S i at time t, and N represents the total number of sensor nodes: As shown in Figure 18, the average energy consumptions at the end of the experiments with LEACH and LEACH EFMP were 23.5 J and 13.9 J respectively.The simulation results show that LEACH EFMP used 41% lesser energy per sensor node on average than did LEACH under the same fire monitoring condition.The average energy consumptions of TEEN and TEEN EFMP were 14.6 J and 12.9 J, respectively.The simulation results show that TEEN EFMP used 12% lesser energy than did TEEN.

Conclusion
Sensor networks are the most appropriate systems in fire monitoring environments.However, sensor networks have problems in fire monitoring environments because of their limited battery capacity.The energy efficiency of a typical cluster-based sensor network drops when a large number of sensor clusters simultaneously monitor a fire over a wide area, because of the fire monitoring characteristics.In this study, we proposed the EFMP that reduces overall energy consumption of the sensor network by dynamically forming a multilayer cluster hierarchy based on the propagation of the fire and efficiently transmitting data over the hierarchical cluster-based sensor network appropriate for fire monitoring.
The performance evaluation results showed that EFMPbased fire monitoring consumes 41% lesser energy on average per sensor node than does LEACH-based fire monitoring and 12% lesser energy on average per node than does TEENbased fire monitoring.Furthermore, the number of sensor nodes surviving in the EFMP-based experiments was 12% and 14% more than that in the LEACH-and TEEN-based experiments, respectively.
Finally, future work will focus on estimating the size, speed, and direction of a fire by extending EFMP in general fire monitoring environments.We hope that this study will contribute to protecting more lives and properties from fire disasters.

Figure 6
shows structures of SIG INFORM packet.The SIG INFORM packet consists of Head ID, Bat capacity, Num sensors, Sensor ID N, and Sensor data N fields.
|TEMP CUR − TEMP AVE | is greater than ΔTEMP MAX .(2)Fire Monitoring Startup.The cluster head among the sensor nodes receiving the SIG FIRE packet becomes aware