This paper innovatively combines the artificial bee colony (ABC) algorithm and the backpropagation neural network (BPNN) into a precipitation prediction model. The research data were collected by 17 stations in the Wujiang River Basin from 1961 to 2018, and compiled into a time series of precipitation data. Through wavelet analysis on precipitation series, the authors identified the features of precipitation distributions in time and frequency domains at different timescales, and demonstrated the inter-annual trend and abnormalities of precipitation in the basin. Next, the weights and thresholds of the BPNN was optimized by the ABC algorithm, and used to predict the precipitation of the basin in the next two decades. The predicted results were consistent with the periodicity and break points obtained by the wavelet analysis. The Z index was introduced to identify the flood years and drought years in the prediction period. The research results shed new light on climate prediction, flood control and drought resistance.

本文将人工蜂群（ABC）算法和反向传播神经网络（BPNN）创新地组合到降水预测模型中。研究数据收集自1961年至2018年吴江流域的17个站点，并编入降水数据的时间序列。通过对降水序列的小波分析，作者确定了不同时标在时域和频域的降水分布特征，并揭示了该盆地的年际趋势和降水异常。接下来，通过ABC算法优化BPNN的权重和阈值，并将其用于预测未来20年的流域降水。预测结果与小波分析得到的周期性和断点一致。引入Z指数来识别预测期内的洪水年份和干旱年份。研究结果为气候预测，防洪和抗旱提供了新的思路。