Abstract: The optimization objects of solvent recovering system are to produce high quality lubricating oil and save energy. Because of interference, fluctuation, strong coupling and large stagnancy, the system's mathematical model can't meet the requirement of on-line energy optimization, and thus a combined optimization strategy is put forward, which combines mathematical modeling&optimization with expert system modeling&optimization. A BP neural network is also adopted to realize soft measurement of the index of lubricating oil and a nonlinear predictive control algorithm is adopted to control the tower's temperature. Actual production result proves the success of the strategy.