Abstract The long-term operation of hydro-thermal power generation systems is modeled by a large-scale stochastic optimization problem that includes nonlinear constraints due to the head computation in hydroelectric plants. We do a detailed development of the problem model and state it by a non-anticipative scenario analysis, leading to a large-scale nonlinear programming problem. This is solved by a filter algorithm with sequential quadratic programming iterations that minimize quadratic Lagrangian approximations using exact hessians in L∞ trust regions. The method is applied to the long-term planning of the Brazilian system, with over 100 hydroelectric and 50 thermoelectric plants, distributed in 5 interconnected subsystems. This problem with 50 synthetically generated inflow scenarios and a horizon of 60 months, amounting to about one million variables and 15000 nonlinear constraints was solved by the filter algorithm in a standard 2016 notebook computer in 10 h of CPU.

中文翻译：

---

非线性水热发电系统的最优非预期情景

摘要 水热发电系统的长期运行由一个大规模随机优化问题建模，该问题包括由于水力发电厂水头计算引起的非线性约束。我们对问题模型进行了详细的开发，并通过非预期的场景分析对其进行了说明，从而导致了大规模的非线性规划问题。这是通过具有连续二次规划迭代的滤波器算法解决的，该算法使用 L∞ 信任区域中的精确 Hessian 最小化二次拉格朗日近似。该方法应用于巴西系统的长期规划，该系统拥有 100 多个水力发电厂和 50 个热电厂，分布在 5 个相互关联的子系统中。这个问题有 50 个综合生成的流入场景和 60 个月的时间范围，