Abstract: 【Objective】The objective of this study is to establish a forecasting model for predicting the shelf life of UHT milk. The model is based on somatic cell count, the number of psychrophilic bacteria, heat-stable protease activity of raw milk, preservation temperature of UHT milk and other indicators. 【Method】The quadratic orthogonal three-factor test and multiple linear stepwise regression method were used to design the experiment and to analyze the data. Boiling test and alcohol test are regarded as standard of shelf-life of UHT milk, and sensory score and determination of the degree of protein hydrolysis are regarded as a supplementary criterion. These methods were used to monitor variation of main quality of product, which were made of different qualities of raw milk by the UHT and deposited at different temperatures. 【Result】 Through the quadratic orthogonal rotatable design and analysis, SCC (X1), proteinase activity (X2) and temperature (X3) were used as independent variables, and the obtained equation is Y=53.94-3.46X1-6.56X2-3.52X3+0.89X12+2.67X22+0.19X32-2.75X1X2+1.50X1X3+2.0X2X3. The effect of three factors on shelf life was significant. Analysis of variance showed that the impact factors followed the order: X2(P=0.0002), X3(P=0.0160), X1(P=0.0173). Through multiple linear regression design, the obtained optimized model equation is Y=103.752+0.0297X1-0.0000597X2- 3.661X3-0.000316X5-0.469X6 (R=0.8870, R2=0.7867). 【Conclusion】 Equation can be used to forecast shelf life as the results of model tested and estimated successfully by t-test, F-test, D.W and MAPE, etc. The assessment standards of applicability of raw milk for UHT process was constructed.

Key words: UHT milk , shelf life , prediction model , test