摘要
针对传统火灾探测系统对火灾特征信号响应灵敏度均匀性差,而基于神经网络的智能处理方法又存在泛化能力差和过学习等问题。建立了一种基于支持向量回归机(SVR)模式识别方法与传感器阵列相结合火灾预警模型。SVR方法根据统计学习理论中结构风险最小化原则,将气体传感器、烟雾传感器和温度传感器组成的传感器阵列数据进行融合,将复杂的非线性问题转化成了高维平面内的线性问题,克服了传统方法和神经网络方法的缺陷。实验结果表明,使用支持向量回归机的火灾预警模型的预测精度优于神经网络方法,提高了火灾预警系统的可靠性和准确度。
The traditional fire detection system has poor uniformity of response sensitivity of the fire characteristic signals.And the intelligent method based on neural networks has problems of poor generalization ability and over learning.A new fire alarm model based on support vector regression（SVR） and sensor array is proposed to solve the above problems.SVR uses the structural risk minimization of statistical learning theory to fuse data of sensor array composed by gas sensors,smoke sensors and temperature sensors,the complex nonlinear problem is transformed into linear problem of high dimensional plane.The drawbacks of traditional methods and neural networks mediod are overcome.The experimental results show that prediction accuracy of SVR fire alarm model is better than neural network.The reliability and accuracy of the fire alarm system are increased.
出处
《测控技术》
CSCD
2015年第8期19-22,共4页
Measurement & Control Technology
基金
国家自然科学基金重大国际合作研究项目(60910005)
