【作者】 史云峰；

【导师】 吕秀江；

【作者基本信息】 长春工业大学 ， 控制理论与控制工程， 2011， 硕士

【摘要】 火灾威胁着人们的生命和财产安全,精确报警控制系统的研制变得至关重要。目前,还有很多的报警系统功能不够完善,火灾发生时容易出现漏报、误报的情况,主要原因在于火灾报警系统只由单一传感器提供信息源,造成火灾报警系统接收不到完整的火灾预警信号。对于火灾报警系统存在的众多弊端,本文将多传感器信息的聚类融合控制方法运用到火灾报警系统中,解决单一传感器信息源不足的问题。信息融合是集信息处理、概率统计、人工智能、模式识别、认知科学、计算机科学及信息论等技术于一体的一门新发展起来的交叉学科。本文以多传感器信息融合、神经网络知识为基础。神经网络在学习上的优点是有很强的自适应能力和记忆能力以及自组织能力,它能够进行过全局分析以及全面的综合判断,从而形成一种崭新的融合控制策略。在方案中,对数据进行了两次的聚类融合算法上控制,即BP融合和ART-1融合。对BP和ART-1神经网络分别进行研究,并在MATLAB中对它们进行了样本训练和仿真。由于BP神经网络局部误差的原因,其输出值不够精确不能满足报警系统高精度的要求,利用ART-1神经网络对BP神经网络进行完善,使输出的结果更加准确,从而满足报警系统的需要。接着对BP神经网络和ART-1神经网络算法进行了比较,并编写了MATLAB程序。同时,对火灾报警系统运行的硬件环境进行了设计,以及对系统运行时所需的主要芯片进行了简单的分析。通过对本课题研究和仿真实验结果的对比,证明ART-1神经网络的聚类融合控制方法优于BP控制方法,能够满足系统对精确度的要求,证实了本课题理论研究部分的正确性及整体系统设计的可行性。更多还原

【Abstract】 Fire threat to people’s lives and property, good alarm control system is essential. Currently, there are a lot of alarm system function is not perfect, when the fire broke prone to omissions, false positives, mainly due to the fire alarm system consists of a single sensor to provide information source, resulting in less than a complete fire alarm system to receive early warning of fire Signal.The fire alarm system has many drawbacks, this clustering of multi-sensor information fusion control method applied to the fire alarm system, a single sensor to solve the problem of inadequate sources of information. Information fusion is an information processing, probability and statistics, artificial intelligence, pattern recognition, cognitive science, computer science and information theory and other technology in one of a newly developed interdisciplinary. In this paper, multi-sensor information fusion, knowledge-based neural network and the neural network learning and memory ability and self-organizing and adaptive capacity, through a comprehensive judging, global analysis, the formation of a new control strategy.In the scheme, two neural networks using data fusion control, that BP neural network and the ART-1 neural network. On BP and ART-1 neural networks to study, and in MATLAB, they were trained and simulated samples. Local error BP neural network as the reason, the output value is not precise enough alarm system can not meet the requirements of high accuracy, the use of ART-1 neural network to improve the BP neural network, a more accurate value of output to meet the needs of the alarm system. Then on the BP neural network and ART-1 neural network algorithm are compared, and the preparation of the MATLAB program. Meanwhile, the introduction of the fire alarm system is running the hardware environment, and the system is running a major chip required a simple introduction.Through this research and comparison of simulation results show that ART-1 neural network clustering method is better than BP control fusion control method, and calculate the output value of more accurate, confirmed some of the correctness of theoretical research topics And the feasibility of the overall system design.更多还原