基于相关向量回归的高精度动态光纤光栅温度传感预测方法研究

Safe operation of large-scale projects and major equipments is essential for the protection of national security and promotion of economic development. Optical fiber sensing technology is an important technique in the field of safety monitoring for theoretical research and engineering applications. By studying on the characteristics and modeling of fiber bragg grating temperature sensing, a new approach for seeking high-precision of non-contact fiber bragg grating temperature is provided. Since the fiber bragg grating temperature sensing system is a complicated nonlinear system, different from the traditional linear regression calculation between temperature and drift of bragg wavelength, the fiber bragg grating temperature sensing system based on models has two criticial scientific problems such as high-precision modeling for complicated nonlinear system and on-line temperature prediction for engineering applications. To overcome the mentioned problems, the high-precision modeling and prediction of fiber bragg grating temperature sensing based on relevance vector regression is proposed in this project. With the dynamic response kernel model considering nonlinear features of fiber bragg grating temperature sensing, the biased wavelet kernel is analyzed and its corresponding transformation characteristics are derived. The modeling precision is highly improved using the biased wavelet kernel and parameter optimization with fully Bayesian strategy. The prediction based on the improved nonlinear regression model is realized through two steps: offline modeling stage and online model validation stage. Using fast marginal likelihood maximization for hyperparameters, the sparsity factor is constructed and applied in online incremental learning algorithm, which adaptively selects new samples and deletes old relevance vectors for obtaining realtime temperature prediction model. The proposed fiber bragg grating temperature sensing model is finally implemented with high precision and efficient prediction.

大型工程与重大装备的运行安全对保障国家安全、促进经济发展至关重要。光纤传感技术是安全监测领域理论研究与工程应用的一个重要技术手段。光纤光栅温度传感特性及模型研究，为实现高精度、非接触测温提供了新途径。光纤光栅温度传感系统是一个复杂系统，不同于以往温度与光纤光栅峰值波长偏移量之间线性回归形式，基于模型的光纤光栅温度传感系统面临的关键科学问题是非线性系统的高精度建模与在线预测。针对该难题，本项目提出基于相关向量回归的光纤光栅温度传感预测方法，在建立考虑非线性特征的动态响应核函数模型基础上，研究偏移小波核，推导偏移小波变换特性，提出全贝叶斯核参数优化策略，有效提高模型精度；同时使用离线建模和在线模型验证方式，建立超参数最大边缘似然估计快速离线学习算法，进而构建基于模型稀疏因子的在线增量学习算法，建立能够自适应选择新增样本和淘汰向量的在线温度预测模型，最终实现光纤光栅温度传感高精度描述和有效预测。

大型工程与重大装备的运行安全对保障国家安全、促进经济发展至关重要。光纤传感技术是安全监测领域理论研究与工程应用的一个重要技术手段。光纤光栅温度传感特性及模型研究，为实现高精度、非接触测温提供了新途径。本项目提出基于相关向量回归的光纤光栅温度传感预测方法，研究了光纤光栅的温度传感特性，构建了基于偏移小波核相关向量回归的光纤光栅温度传感模型描述体系，建立了在线增量学习的预测模型，完善了基于模型的光纤光栅温度传感预测方法的理论研究，并通过人工数据集合、UCI公共数据集合、铁路和桥梁监测工程应用等验证分析了算法的有效性。针对传统的相关向量回归算法处理大规模样本时的低效率问题，采用分布式计算平台，结合集成学习的思想，研究了分布式相关向量机算法。通过有效运用分布式集群，将大规模数据集的训练任务分布到各个集群的子节点上，显著提升模型的训练速度。. 本项目已经按照项目计划完成了预定的研究任务，取得了期望的研究成果。针对大规模样本的训练问题，扩充了关于分布式算法的研究内容，进一步完善了对光纤光栅温度传感系统中的模型理论以及预测方法的研究。本项目执行期间发表学术论文10篇，其中期刊论文6篇，国际会议论文4篇，SCIE检索4篇，EI检索4篇；申请发明专利6项；获得软件著作权2项。