For the small-scale cascade run-of-river power plant clusters, water abandonment often occurs during periods of high water availability, thus there is an uncertain relationship between generation capacity of run-of-river power stations and meteorological factors such as precipitation. In order to evaluate the amount of abandoned water of runoff power station clusters during flood season, and taking into account the fact that the data recorded by the meteorological stations in the basin are not precise enough while the historical power generation data of runoff power station groups are more complete, this paper first analyzes the hysteresis and cumulative effects of the power generated by runoff power station groups, and then applies a two-step classification method based on the K-means clustering to formulate the sample set used to build the prediction model as well as the sample set of water abandonment scenarios. Then, the multiple linear regression model is developed based on the data of the sample set of non-abandonment scenarios. Furthermore, the above multiple linear regression model is applied to the sample set of scenarios where water abandonment is likely to occur to calculate the ideal power generation of small hydropower in flood season. The amount of water abandoned by small hydropower is derived by subtracting actual power generation of small hydropower from ideal power generation of small hydropower. Finally, the method proposed in this paper is demonstrated by a case study.