The Research Of Filter Model Against Large-Number In The Coal Mine Safety Monitoring System

With economic rapid development in China, the demands for energies are growing. As China is the country which is rich in coal mines, poor in gas and short of petroleum, the exploitation of coal mines plays an extremely important role in our country's economic development. Coal mines are the enterprises, in which accidents are multiple. In recent years, these serious and fearful coal mine accidents occur frequently, the number of the death reaches 6000 every year, not only to the country and people's lives and properties causing serious losses, but also to the Chinese government's image at home and abroad generating a very bad influence. Judging from China's recent coal mine accidents, gas accidents can be considered as significant proportion in these accidents. The gas production is one of the important factors which are harmful to the mine safety.Detection of the concentration of gas is the main link of coal mine production safety, and the accurate measurement and prediction of gas (methane is the main component) concentration directly related to the safety of the mine personnel and equipment. At present, gas detection and alarm systems are installed in some coal mines of China. Now, gas detectors transmit the analog signals through a cable to one central point, then amplify and transmit them to the control house. As the conditions in the mine well are poor, there are all kinds of interference, the weak output signal of the gas sensor is vulnerable to pollution, and causes some pulse signal interferences, also known as "risk of large numbers", which is often caused error alarm. Once the alarm happen, the underground power is automatically cut off, and the production is forced to stop. The technology of the alarm system is laggard, the rate of the false alarm reaches to 80% or 90%. Because of frequent error alarm normal production is greatly affected and caused great losses to the mine enterprise.Although the traditional hardware and software filtering methods can filter pulse signal and eliminate the error alarm, the missed alarm phenomenon will occur when the gas concentration exceeds the limit. The wireless network and voice communication technology based on wireless technologies can be introduced. Although very low leakage alarm rate can be guaranteed, the whole system is very expensive. A new method is in need against the leakage alarm and the error alarm. This paper presents the filtering method based on the neural network. Through the training of numerous samples to neural network, the neural networks can learn and grasp the law of the gas pulse signal interference and the gas outburst from numerous input-output gas data. Because the appearance of pulse signal or "large numbers" is random, people do not know what the mechanism is, and can not establish a precise mathematical model. However, the neural network is composed of a large number of nonlinear processing units, and is the abstract and simulation of the human brain structure and function. Neural networks is possess of the ability of learning, memorizing and concluding as well as the ability of a high degree of non-linear mapping capability. As long as the provision of adequate number of sample models is for learning and training, it will be completed the nonlinear mappings from the n-dimensional input space to m-dimensional output space. The achievement of neural network model against interference needs three steps: (1) the selection of neural networks, using which kind of neural network model is the first step of establishing neural network; (2) the preparation of training sample, the selection of training sample is closely related to the performance of the network. The design of good training samples needs to pay attention to the sample's size and quality, the preparation work is the basis of network design, The scientific and rational choice of samples and data is extremely important for the design of the network; (3) the detailed design of the network, including the selection of the training function and learning function, the design of the network structure, the selection of performance indicators and so on.Through a great deal of experiments and simulations, ultimately determine to adopt back propagation neural network model. Finally, the test samples are carried out performance testing of the model. The test results show that the model can eliminate effectively the error alarm phenomenon caused by the pulse interference and prevent the leakage alarm when the concentration of gas overrun. In the coal mine Safety monitoring and controlling system, there is a good prospect for developing the application of neural networks for eliminating error alarm phenomenon.