Abstract Genetic Algorithms (GAs) have been commonly applied in the last two decades as a substitute for the traditional mathematical programming algorithms in searching for optimal operating rules of reservoir systems. However, only a few GA studies related to reservoir operations addressed the inflow uncertainty. In this study, the reservoir operation rules were developed and evaluated by reflecting the uncertainty of reservoir inflows using the non-dominated sorting GA II algorithm (NSGA-II) through the following procedure: (1) the historical inflow data were classified into four clusters using the self-organizing map (SOM) clustering technique; (2) NSGA-II searched for an optimal release rule in each cluster with all the inflow data of the corresponding cluster; (3) a release response function was then derived for each cluster by regressing the calculated optimal release data against the storage at the beginning of the month and the inflow during the month; and (4) finally, the derived release rules for each cluster were tested with three performance indices, namely, reliability, resilience, and vulnerability. The proposed procedure was applied to the monthly operations of the Lake Tana multi-reservoir system in Ethiopia, which has six upstream irrigation reservoirs and one natural lake that has three release outlets for agriculture, hydropower, and instream requirement. Using the NSGA-II, SOM, and a seasonal ARIMA forecasting model in this study, it is concluded that: 1) the overall performance of the proposed optimization procedure with ARIMA forecasts and cluster approach reaches 84% of the perfect information case across multiple performance measures; 2) the use of the one-month-ahead ARIMA forecasts improves the overall system performance by 8% over simply optimizing to the mean flow; and 3) the use of the four clusters and the consequent RRFs further improves the overall system performance by 14%. Furthermore, this study recommends that, in future, significant efforts should be focused on improving the operation performance of the upstream irrigation reservoirs in coordination with the Lake Tana system.

中文翻译：

---

使用遗传算法表示月度水库作业开发的流入不确定性

摘要 遗传算法（GAs）在过去的二十年中作为传统数学规划算法的替代品在寻找储层系统的最佳运行规则方面得到了普遍应用。然而，只有少数与油藏作业相关的 GA 研究解决了流入的不确定性。在本研究中，使用非支配排序GA II算法（NSGA-II）通过以下程序通过反映水库入流的不确定性来制定和评估水库运行规则：（1）将历史入流数据分为四个集群使用自组织映射（SOM）聚类技术；(2) NSGA-II利用对应集群的所有流入数据，在每个集群中寻找最优的发布规则；(3) 然后通过对计算的最佳释放数据与月初的存储量和月内的流入量进行回归，为每个集群导出释放响应函数；(4) 最后，对每个集群的派生发布规则进行了三个性能指标的测试，即可靠性、弹性和脆弱性。拟议的程序被应用于埃塞俄比亚塔纳湖多水库系统的月度运行，该系统有六个上游灌溉水库和一个天然湖泊，有三个用于农业、水电和内流需求的排放口。在本研究中使用 NSGA-II、SOM 和季节性 ARIMA 预测模型，得出以下结论：1) 使用 ARIMA 预测和聚类方法提出的优化程序的整体性能在多个性能指标上达到了完美信息案例的 84%；2) 使用提前一个月的 ARIMA 预测比简单地优化平均流量提高了 8% 的整体系统性能；3) 四个集群的使用和随之而来的 RRF 进一步将整体系统性能提高了 14%。此外，本研究建议，未来应重点努力与塔纳湖系统协调，提高上游灌溉水库的运行性能。3) 四个集群的使用和随之而来的 RRF 进一步将整体系统性能提高了 14%。此外，本研究建议，未来应重点努力与塔纳湖系统协调，提高上游灌溉水库的运行性能。3) 四个集群的使用和随之而来的 RRF 进一步将整体系统性能提高了 14%。此外，本研究建议，未来应重点努力与塔纳湖系统协调，提高上游灌溉水库的运行性能。