BP based soft measurement of flash point of lubrication oil

Energy optimal control of solvents recovery device is to minimize the stream consumption of unit product (lubrication oil), and to assure product quality at the same time. The flash point, which is the quality index of product, must be realized online real-time measurement, to assure real-time modeling and closed-loop optimal control of the ketone-benzol de-waxing device. Based on the special nonlinear mapping function of BP neural networks, a kind of soft measurement technology is created to predict the flash point. Its input are assistant variables, its output is the main variable, and the soft measurement is realized by network learning. It is pointed out that the space distribution of learning samples and learning methods largely affects the characteristics of the ANN. For large-scale complex process, multiple networks and local learning method are applied to avoid too much neural units and too slow learning speed. Actual application result obtained proves its effectiveness