During the drilling process, water-based drilling mud can infiltrate the formation through mud cake and contaminate the formation water. In order to obtain pure formation water samples, it is necessary to monitor the pollution level in real time. Before sampling, the fluid is gradually transferred from water-based mud filtrate to pure formation water. The degree of formation water contamination can be calculated in real-time by obtaining the absorbance of filtrate, pure formation water, and mixed fluid. In view of the fact that online monitoring of downhole formation water pollution rate can be regarded as a time series prediction problem, Elman neural network model is adopted to train the absorbance data, so as to predict the absorbance of pure formation water. The validation is performed using offshore well data. Combining the absorbance of the formation water based on the Elman neural network and that of the drilling fluid filtrate collected at the initial pumping stage, the real-time formation water pollution rate can be calculated and compared with the laboratory water analysis results. The results show that they are in good agreement. Compared with the traditional algorithm, the new method is efficient and reliable, and has wide applicability and good application value.