Generation Capacity Expansion Planning (GCEP) requires high temporal resolution to account for the volatility of renewable energy supply. Because the GCEP optimization problem is often computationally intractable, time-series input data are often aggregated to representative periods using clustering. However, clustering removes extreme events, which are important to achieve reliable system designs. We present a method to include extreme periods into time-series aggregation for GCEP that guarantees reliable system designs on the full input data even though only the reduced data set is used for system design. Our method iteratively adds extreme periods to the set of representative periods based on information from the optimization problem itself until the energy system provides power reliably.

We perform a comprehensive analysis on several case studies of both German and Californian energy systems and show that our method leads to meeting electricity demand at all times, whereas when clustering without extreme periods, lost load is between 1.9%–13.5% of total system load. We show that our method outperforms the state-of-the-art method of adding a pre-defined number of extreme periods based on statistical properties of the data itself.

发电容量扩展规划 (GCEP) 需要高时间分辨率来解决可再生能源供应的波动。由于 GCEP 优化问题通常在计算上难以处理，因此通常使用聚类将时间序列输入数据聚合到具有代表性的时期。然而，聚类消除了极端事件，这对于实现可靠的系统设计很重要。我们提出了一种方法，将极端时期包含在 GCEP 的时间序列聚合中，即使只有简化的数据集用于系统设计，它也能保证对完整输入数据进行可靠的系统设计。我们的方法基于来自优化问题本身的信息迭代地将极端周期添加到一组代表性周期中，直到能源系统可靠地提供电力。

我们对德国和加利福尼亚能源系统的几个案例研究进行了综合分析，并表明我们的方法可以始终满足电力需求，而在没有极端时期的聚类时，损失的负载在总系统负载的 1.9%–13.5% 之间. 我们表明，我们的方法优于基于数据本身的统计特性添加预定义数量的极端周期的最新方法。