Objective: To ensure the consumption safety of grain resources. Methods: According to the relevant national standards, eight chemical pollutants, such As cadmium （Cd）, arsenic （As）, lead （Pb）, chromium （Cr）, aflatoxin （AFs）, fumonisin （FB）, zearalenone （ZEN） and deoxynivalenone （Don）, were identified as the risk assessment indexes of grain quality and safety, and the entropy weight method was adopted. At the same time, taking the data of evaluation index as the input of risk assessment model, four machine learning algorithms, namely, random forest regression （LR）, support vector machine regression （SVM）, BP neural network regression （BP） and K-nearest neighbor regression （KNN）, are selected to construct and compare the models. Results: The prediction correlation coefficient of the model constructed by AHP-RF based on entropy weight was above 0.99. The risk assessment model was used to predict and analyze the grain detection data in August 2019, and the correlation results were consistent with the reality. Conclusion: The risk assessment model based on AHP-RF method can provide targeted reference suggestions for the safety supervision of grain resources.