Accuracy of reservoir inflow forecasting is an important issue for the reservoir operation and water resources management. The main aim of the current study is to develop reliable models to forecast monthly inflow data. The present research proposed a robust model called co-active neuro-fuzzy inference system (CANFIS) to improve the forecasting accuracy. The reliability of the CANFIS model was evaluated by comparing with two different AI-based models, ANN and ANFIS model. To obtain the best forecasting result, the proposed models were trained utilizing four different Training Procedures. This study was conducted to forecast the inflow data for Haditha Dam on Euphrates River, Iraq. The comparison of models reveals that the CANFIS model is better than ANN and ANFIS model. The results showed that the second training procedure is more suitable for the forecasting models. The CANFIS model yielded a relative error of less than (15%), a low MAE (69.66 m³/s), a RMSE (78.10 m³/s) and a high correlation between the actual and forecasted data (R² = 0.97).

水库入流预测的准确性是水库运行和水资源管理的重要问题。当前研究的主要目的是开发可靠的模型来预测每月的流入数据。本研究提出了一种称为协同神经模糊推理系统（CANFIS）的稳健模型，以提高预测精度。通过与两种不同的基于人工智能的模型，ANN 和 ANFIS 模型进行比较，评估了 CANFIS 模型的可靠性。为了获得最佳预测结果，所提出的模型使用四种不同的训练程序进行训练。本研究旨在预测伊拉克幼发拉底河上哈迪萨大坝的流入数据。模型对比表明CANFIS模型优于ANN和ANFIS模型。结果表明，第二个训练程序更适合预测模型。CANFIS 模型产生的相对误差小于 (15%)，低 MAE (69.66 m³ /s)、RMSE (78.10 m ³ /s) 以及实际数据和预测数据之间的高度相关性 (R ²  = 0.97)。