Abstract: Soft measurement technology is also named soft instrument technology. It uses easy-to-measure process variable (It is also named assistant variable or secondary variable.)， and bases on the relation (soft measurement model) between the easy-to-measure process variable and the need measuring process variable (It is also named main variable) which is difficult to measure directly to measure or estimate the need measuring process by all kinds of mathematic calculation and estimation methods. At present， the soft measurement technology has become one hotspot in the
field of process control.
This article chooses the Diesel cracking unit as the research object， collecting the process conditions and product quality indicator to perfor data， the implementation of regression based on data-driven modeling， analysising the relationships between them， and influencing factors to achieve a quantitative prediction of the quality indicators ， using modern intelligence technology of artificial neural network nonlinear method， and comparing with the traditional multiple linear regression method to explain the advantage and parameters optimization option.
Keywords：Residual decomposition; Soft Sensor; Data modeling; The artificial neural network
