CAJ | 학술논문

针对氧浓度信号存在非线性、随机性和易受干扰，难以建立准确测量模型的问题，提出一种RBF神经网络软测量技术应用于超声氧浓度计的方法，该装置测量氧气的温度和超声波在定长管道中氧气传播的时间作为RBF神经网络的输入量进行拟合，采用梯度下降法确定RBF基函数的中心及输出层权值，氧气浓度值作为网络输出量。试验结果表明：采用RBF神经网络曲面拟合所测得氧浓度测量值与顺磁式氧浓度分析仪测量结果绝对误差在1．5％以内，具有一定的工程实用性。
침대양농도신호존재비선성、수궤성화역수간우，난이건립준학측량모형적문제，제출일충RBF신경망락연측량기술응용우초성양농도계적방법，해장치측량양기적온도화초성파재정장관도중양기전파적시간작위RBF신경망락적수입량진행의합，채용제도하강법학정RBF기함수적중심급수출층권치，양기농도치작위망락수출량。시험결과표명：채용RBF신경망락곡면의합소측득양농도측량치여순자식양농도분석의측량결과절대오차재1．5％이내，구유일정적공정실용성。
As the oxygen concentration signal is non-linear,random and prone to interference,it is difficult to establish an ex-act measurement model.Based on RBF neural network,a method used in the ultrasonic oxygen concentration meter was proposed. We used the equipment to measure the temperature of oxygen and the transmission time of ultrasonic in fixed length pipe and used the results to get the oxygen concentration by RBF neural network.The temperature of oxygen and the time of ultrasonic wave in the oxygen were regarded as the input for RBF neural network and we took the gradient descent method as the output for the network. Experimental results show that the absolute error can be controlled within 1.5%between ultrasonic oxygen concentration meter and paramagnetic oxygen concentration analyzer when usingRBF neural network,and which has certain practical value in the works.