The chaotic local-region forecasting model is suitable for nonlinear and non-stationary urban daily water supply forecast, and the neighbourhood selection has a direct impact on the model prediction accuracy. The traditional method usually takes the embedded dimension m as a reference, and selects m+1 nearest neighbours by experience. It usually introduces the pseudo nearest neighbours, which leads to the reduction of the prediction accuracy. Accordingly, the evolutionary tracing method is introduced into the prediction of urban daily water supply. By mining the historical evolution of nearest neighbours, the reference samples are optimized to improve the prediction accuracy. The proposed method is validated by the actual daily water supply data. The results show that the optimal approach is significantly improved by using evolutionary tracing method, and the average absolute error is reduced from 2.501% to 1.683%.