Abstract: Electrical capacitance tomography（ECT） technique is a very complex nonlinear problem.To solve the ill-posed image reconstruction problem，a new method based on multi-dimensional support vector regression（MSVR） is proposed.The MSVR with a hyper-spherical insensitive zone is an import branch of SVM，overcoming the over-fitting of neural networks.Similar Iterative Re-Weight Least Square（IRWLS） algorithm is used to simply the realization of MSVR，a nonlinear map between capacitance measurements and the permittivity distribution in image region is built fastly.Using the data and MSVR method，simulation experiments are carried out for six typical flow regimes.The results prove that MSVR is an effective approach to solve image reconstruction for ECT.

摘要： 根据电容层析成像（ECT）中两相流识别问题的特点，提出基于多维输出支持向量回归机（MSVR）的图像重建算法。采用超球空间不敏感损失函数的MSVR是支持向量机理论的一个重要分支，它有效地克服了神经网络算法中的过学习问题，具有较强的泛化能力。近似迭代变权最小二乘法（IRWLS）在保证MSVR回归精度的基础上，有效地简化了其求解过程，可快速建立电容测量值与成像区域介电常数分布之间的非线性映射关系。对包含6种典型两相流流型的仿真数据进行实验。结果表明，该方法泛化能力强，图像重建精度高。