By Using the NCEP / NCAR reanalysis daily 850hPa wind data and daily precipitation data of 25 weather stations over the lower Yangtze region, based on the 20-30d second-order Butterworth filter and with the varying-coefficient time lagged regression models and rainfall key-area auto filter,an extended-range real-time forecast model solution was established for low frequency rainfall of Yangtze River downstream in flood season from May to August,and then it was used in hindcast and forecast experiment in the flood season of 1979 and 2013.The results show that the real-time forecasting solutions based on multivariable lagged regressive model (MLR), can objectively reflect the time-varying propagation characteristics of the intra-seasonal oscillation (ISO) for Eurasian circulation that affect the severe rainfall over lower Yangtze region during summer, and offer a main signal for the extended-range forecast for strong precipitation over the area. The rainfall key areas change with the modeled samples in the proposed real-time prediction scheme, and have reflected the diversity and a certain degree of instability of ISO changes. The prediction model can promptly adapt to the change of relevant time-lag structure caused by the internal dynamics of atmospheric processes and other external forcing, and have improve the ability of extended-range forecast to a certain extent.