Predictive Mine Gas Wireless Sensor Network Based on Support Vector Machine

The introduction of ZigBee wireless network protocol into the most important gas monitoring network in coal mine production can not only expand the monitoring range of gas concentration but also make some improvements to the monitoring network based on the delay of gas concentration monitoring. The introduction of support vector machine can get a monitoring network with gas concentration prediction function. Introduction In today's coal production process, the safety of mine gas has been the focus of attention of all sectors of society. Excessive gas concentration is the primary issue that threatens coal mine safety. When the concentration of oxygen in the air reaches about 10% and the gas concentration of 5-16%, the mixture will explode. Therefore monitoring gas concentration in the mine has become an important link in the current coal mine production. The reliability and timeliness of the monitoring system are of great significance to work safety. Due to the special underground environment and complex situations, the traditional manual measurement and the extended performance of wired sensor networks have various unsafe and unstable factors. The new wireless sensor network with its efficient and reliable features has been gradually replacing the traditional monitoring technology [1] . In this paper we use the more novel wireless sensor networks based on ZigBee protocol. In the mine set up with a small-scale gas measurement monitoring system and the host computer to collect real-time statistical data acquisition and forecast gas concentration and get better experimental results. ZigBee Wireless Sensor Network Introduction ZigBee technology is a short-range, low-power, low-rate wireless communication technology. It is based on the IEEE802.15.4 specification, in the data link layer, network layer and application layer to develop a corresponding interface specification. ZigBee nodes include full-function devices and simplified function devices, and these two functional devices form a star, tree and mesh-based topology. This topology generally accommodates 255 child nodes in a network group and can reach a maximum of 65536 child nodes by address expansion [2] . ZigBee wireless sensor network technology is quite prominent: the transmission distance can be 10 meters to 80 meters; its power consumption is very low, the power supply is simple; available working frequency bands are generally divided into 868MHz, 902 ~ 928MHz and 2.4 ~ 2.4835GHz. A working band the transmission rate in the above three bands are not the same respectively at 20kb/s, 40kb/s and 250kb/s or so. Based on these characteristics of ZigBee, there are many advantages for the application of this technology in the monitoring of coal mine gas to the establishment of a wireless monitoring network. The utility model has the advantages of low cost and simple network expansion and is suitable for application in a large-scale monitoring environment. The utility model has the advantages of more reliable reliability, higher gas concentration and wired sensor safety than normal manhole strong anti-interference ability and high fault tolerance. In the rapid development of spread spectrum communication technology downhole wireless sensor networks have begun to gradually apply. From the time and attendance system downhole personnel positioning system are in the underground coal mine wall to solve the problem of wireless signal absorption under the premise of using the current spread-spectrum communication technology to achieve underground wireless network communications. Mine gas concentration monitoring network to ZigBee technology represented by the development of wireless network monitoring and maturing is an inevitable trend. Wireless Monitoring Network Construction Since mine gas is ejected from coal and rock fractures during coal mining it is common to use hose extraction to the total gas extraction pipework at the upper corner of the previously measured gas concentration accumulation. Therefore, the design and construction of wireless sensor networks should give full consideration to the origin of gas. In view of the various mechanisms and factors of gas generation, we choose the network structure to configure each network node because this network structure can well correspond to the ductility of the coal mine face and roadway can solve the sensor fault tolerance in the node configuration sex. A network node is configured at intervals on the seam wall of the roadway and the working face, and by wearing the corresponding sensor to the workers in the first-line mining face of the well, we can get a good dynamic extensibility, and can actively avoid the defects of the gas concentration in the working face, and get more accurate monitoring value. The corresponding algorithm can be programmed and implemented on the corresponding industrial computer and the weighted processing is performed on the active nodes of the working face to obtain better monitoring results. CC2510 built-in positioning system to accurately find the gas concentration exceeded or may exceed the specific location and combined with the latest “coal and gas co-mining technology” we can make a better way of handling. In the choice of sensor nodes taking into account the measurement range and positioning of the measuring point, based on the choice of sensor nodes taking into account the measuring range and measuring point positioning based on the choice of CHIPCON in SmartRF04 technology on the production of CC2510 wireless sensor chip. The hardware architecture of the wireless sensor network node device based on the chip is shown in Figure 1. It is a highly integrated industrial RF transceiver that enables the development of MAC and PHY layer communication protocol standards by ZigBee and operates in the 2.4 GHz band and builds a wide range of monitoring networks through communication protocols and node extensions. Figure 1. Wireless sensor network node device hardware structure. Taking into account the chip has a good data processing module and positioning system module, we can according to the specific situation to improve the traditional monitoring system mode. In view of the special situation that the gas concentration change is hard to grasp the support vector machine prediction model can be introduced into the wireless sensor network. Early warning in the case of gas concentrations may exceed the standard, thereby improving the safety factor to prevent gas accumulation and gas outburst accidents. Support Vector Machine Prediction Model SVM is a kind of statistical learning theory put forward formally by Vapnik in 1995. It takes the structural risk minimization empirical risk minimization and VC dimension as the background. In fact SVM is to find an optimal function so that each subset of a function set can be obtained within the appropriate minimum confidence interval the smallest empirical risk [3,4,5] . We conducted local experiments on the gas concentration data monitored by two sensor nodes set up at the coal mining face. Figure 2. Schematic diagram of sensor position. Figure 3. System flow chart. As shown in Figure 2, we select the coal mining face near the coal wall set up T1 and T2 two sensor network nodes. After ZigBee network communications sent to the host computer and get the appropriate two sets of data values. Then through the pre-processing of the data to get the corresponding weight training samples. Standardized processing of data first, the formula is ' i i i x x x σ − = (1) where, i x is the average value of each component of i x and σ is the standard deviation of each component of i x . The processed data samples according to the flow shown in Figure 3. training and then amended to get a good SVM forecasting model. In the choice of confidence interval we generally need to refer to a series of conditions such as uniform convergence and convergence speed of learning machine. Following the consistency of the learning process finding a balance between real risk and experiential risk avoids the risk of over-learning and under-learning conditions so that the confidence range can be truly adapted to a range of conditions of gas concentration monitoring [6,7] . We know that support vector machines are assumed to be a set of processed data sets, {( , )} i i i A x y } 1 , 1 { , − + ∈ ∈ y R x n (2) that can be separated by a hyperplane ( ) 0( , ) n w x b w R b R • + = ∈ ∈ . (3) where w is the weight vector andb is the offset of the optimal plane shift [8] . We need to normalize this hyperplane. That is (( ) ) 1, 1,..., i i y w x b i l • + ≥ = . (4) In this way we can represent the optimal classification surface as a problem that maximizes the classification distance based on the hyperplane. That is to find 2 1 2 ( ) w w Φ = (5) to minimize the problem. In this way 1 ( ) 1 ( , ( , )) n emp w i i n i R L y f x w