《多传感器数据融合技术在煤矿水害预防中的应用》由会员分享，可在线阅读，更多相关《多传感器数据融合技术在煤矿水害预防中的应用（77页珍藏版）》请在金锄头文库上搜索。
1、太原理工大学 硕士学位论文 多传感器数据融合技术在煤矿水害预防中的应用 姓名：程珍珍 申请学位级别：硕士 专业：检测技术与自动化装置 指导教师：张英梅 20090401 太原理工大学硕士研究生学位论文 I 多传感器数据融合技术在煤矿水害预防中的应用 摘摘 要要 我国是一个煤炭大国，大中小型煤矿星罗棋布，为国民经济的发展提 供了有力的能源支持。但是长期以来，煤炭生产一直被安全问题所困扰。 在煤矿安全灾害中，煤矿水害与瓦斯并列其首，目前随着井下通风设施的 完善，煤矿发生瓦斯事故的概率已经大大降低了，但是煤矿水害事故还是 频频发生，给矿工的生命安全和煤矿的经济效益带来极大威胁。造成煤矿 水害事故发生
2、的原因是多样的，然而现有的技术往往只能检测到某一个信 息，不能将所有的信息在同一时刻进行技术处理，也就不能及时采取措施 或者及时报警，从而造成了大量的人员和财产损失。 针对煤矿井下突水预测技术还不完善的情况，本文提出了一种将多传 感器数据融合技术应用于煤矿井下水害预测的设计方案，主要论述了多传 感器数据融合的一般原理、发展、应用领域，介绍了D-S证据理论和BP神经 网络融合算法，并分别将它们应用于煤矿水害预测中，最后为了改善D-S证 据理论和BP神经网络， 提高融合结果的准确性， 引入了改进型的D-S证据理 论和BP神经网络算法， 并最终提出一种将D-S证据理论和神经网络相结合的 两级融合算法
3、。 通过对比分析，最初考虑使用多传感器信息融合技术的决策级融合方 法来进行对煤矿井下安全状态的预测，过程是：首先，先确定具体的融合 算法，通过对决策级融合几种方法的对比，最终选择使用D-S证据理论作为 主要融合算法，对其融合算法及融合具体过程进行了详细介绍，并进行了 太原理工大学硕士研究生学位论文 II 具体数据的计算。 由于D-S要求各个证据源相互独立，而且其基本概率分配函数难以获 得，所以考虑使用神经网络信息融合算法进行融合，神经网络包括我们所 熟悉的反向传播网络、自组织特征映射Kohonen网络、Hopfield网络等，通 过对比最终选择了目前使用最为广泛、最为熟悉的BP神经网络，对其学
4、习 过程及各参数的选择方法进行了介绍，并使用MATLAB软件对具体数据进 行了训练。 最后根据以上两种算法在融合过程中所出现的不足引入改进型的D-S 证据理论及BP神经网络算法， 并提出了一种将它们相结合的两级融合算法， 即通过不同传感器信息的有效组合，对传感器信号采用不同的子神经网络 进行训练，把每一个子神经网络的输出结果进行归一化处理，将它们作为 证据理论各状态的基本概率分配函数，再用D-S证据理论融合各个证据信 息，得出最后结论。 通过对以上三种方法的融合结果进行分析对比，得出结论：单独使用 D-S证据理论以及单独使用神经网络技术能够实现对井下安全状态的判断， 但是它们准确性较低、不确定
5、性稍高，而两级融合方法所计算出的结果准 确性高、不确定性低。 信息融合作为近年来兴起的一门新学科，在许多领域得到了广泛的研 究和应用，在煤矿水害预测中的应用研究尚处于起步阶段。本文将信息融 合技术引入煤矿水害预测中，借鉴了信息融合的思想和理论，建立了煤矿 水害预测的一般框架，相信这一技术会在煤矿领域得到进一步推广。 关键词：煤矿突水，数据融合，D-S证据理论，神经网络，两级融合 太原理工大学硕士研究生学位论文 III APPLICATION OF MULTI-SENSOR DATA FUSION IN COAL MINE FLOOD ABSTRACT China is a great nati
6、on with rich coal resource. So many coal minds are scattered in the country and provide powerful energy support for the national economy development. But for a long time, the countrys coal production has been puzzled by safety problems. Among the coal mine safety problems, gas explosion and coal min
7、e flood are the most serious. The coal mine flood occurs frequently, brings enormous threat to the workers safety and the coal mines economic efficiency. The main reason is water bursting of the coal mine has various reasons, but the existing technology is often only detected a certain information,
8、can not treat all the information at the same time technically, can not take timely measures or timely alarm , Causing a large number of personnel and property losses. For the prediction technology of coal mine water bursting is not perfect, this paper raised a design that is apply the multi-sensor
9、data fusion technology in coal mine flood forecasting, mainly discussed the general principles, development, Application areas of multi-sensor data fusion, mainly introduced the integration algorithms of D-S theory of evidence(Dempster-Shafer) and BP(Back propagation) neural network, and they were u
10、sed in coal mine flood 太原理工大学硕士研究生学位论文 IV forecast, compared the advantages and disadvantages of D-S theory of evidence and BP neural network, eventually put forward a two fusion algorithm which D-S theory of evidence attached with Neural Network. Initially consider use the decision-making-level of
11、multi-sensor data fusion to integration the various coal mine signals. Firstly, identify the specific integration algorithm, compared the integration advantages and disadvantages of several methods of decision-making-level fusion, ultimately choose to use D-S theory of evidence as a major fusion alg
12、orithm, introduced the fusion algorithm and its concrete process of integration detaily, and carried out the calculation of specific data. As D-S require independent source of evidence, consider use neural network information integration algorithm to fusion all information, neural networks including
13、 what we are familiar with the reverse communication networks, self-organizing feature map Kohonen network, Hopfield network, by comparing, the final choice is BP neural network which the most widely used, most people are familiar with, introduced their learning process and the method of parameters
14、selection, use MATLAB software to training the specific data. Finally raised two-level fusion algorithm that is neural network attached with the D-S theory of evidence, that is, through effective Portfolio of various sensor information, from different sides reflects the coal mines security status, N
15、ormalization to the results of each sub-neural network output, as the basic probability distribution function of evidence theory, and then use D-S theory of 太原理工大学硕士研究生学位论文 V evidence to fusion every evidence information, Draw final conclusions. By comparing the simulation results of the three metho
16、ds and make a conclutiom: D-S evidence theory is used alone as a separate use of neural network technology with lower accuracy, but also to achieve a state of mine safety assessment. The combine of two methods can get higher accuracy results. Information fusion is a new subject rise in recent years, in many fields has been research and application extensively; the research in coal mine flood prediction is still at the initial stage. This article put information fusion technology into co
