工业制SO3气体浓度实时检测及多元非线性回归建模

工业接触法制SO₃气体浓度高、腐蚀性强、交叉影响大，高精度实时检测困难，且原料流量与SO₃浓度的关系尚不明确，参数调节依赖经验。针对上述问题，采用AO2电化学氧传感器设计了一种工业制SO₃气体浓度的实时检测系统，并构建了输入空气、SO₂流量与输出SO₃浓度的多元非线性回归模型。SO₃气体浓度实时检测系统主要包括AO2电化学氧传感器、传感器调理电路、AD620放大电路及ADS1256 A/D转换电路，通过检测SO₃发生装置输入空气与输出混合气体中O₂浓度的毫伏级变化，间接测量SO₃浓度，避免了其他敏感气体的交叉影响，实验结果表明检测系统的检测时间为27s，灵敏度为111mV/%，偏差小于0.18%，稳定性良好，证明了检测系统的有效性；利用响应面法建立了预测SO₃浓度的三次多项式非线性回归模型，对该模型进行方差分析后结果显著，测试后均方误差（MSE）为0.0007174，相关系数（R²）为0.9929，拟合度好、预测精度高。

Real-time detection of industrial SO3 gas concentration and multivariate nonlinear regression modeling

In industry, SO₃ gas prepared by oxidation of SO₂ is of high concentration, strong corrosion and great cross-influence, and difficulty by high precision real-time detection. Moreover, the relationship between input flow and SO₃ concentration is not clear, which leads to that the parameter adjustment depends on experience. Aim at the above problems, a real-time system for detection of industrial SO₃ gas concentration was designed by AO2 electrochemical oxygen sensor, and a multivariate nonlinear regression model with input air, SO₂ flow and output SO₃ concentration was established. The real-time detection system of SO₃ gas concentration mainly included AO2 electrochemical oxygen sensor, sensor conditioning circuit, AD620 amplifying circuit and ADS1256 A/D conversion circuit. It indirectly measured SO₃ concentration by detecting the millivolt level change of O₂ concentration in input air and the output mixture gas from SO₃ generator. In this way, the cross influence of other sensitive gases was avoided. The experimental results showed that the detection time of the system was 27s, the sensitivity was 111mV/%, the deviation was less than 0.18% and the stability was good, which demonstrated the effectiveness of the detection system. A cubic polynomial nonlinear regression model for predicting SO₃ concentration was established by response surface method. The results of variance analysis for this model were very significant with the mean square error (MSE) of the model of 0.0007174 and the correlation coefficient (R²) of 0.9929, which proved that the model had good fitting degree and high prediction precision.