Representative period selection for power system planning using autoencoder-based dimensionality reduction
This work addresses a specific bottleneck in power system planning by improving the accuracy of capacity expansion models, though it is incremental in nature.
The paper tackles the problem of selecting representative periods for power system capacity expansion models by proposing an autoencoder-based dimensionality reduction method before clustering, which reduces errors in model outcomes compared to full-space models.
Power sector capacity expansion models (CEMs) that are used for studying future low-carbon grid scenarios must incorporate detailed representation of grid operations. Often CEMs are formulated to model grid operations over representative periods that are sampled from the original input data using clustering algorithms. However, such representative period selection (RPS) methods are limited by the declining efficacy of the clustering algorithm with increasing dimensionality of the input data and do not consider the relative importance of input data variations on CEM outcomes. Here, we propose a RPS method that addresses these limitations by incorporating dimensionality reduction, accomplished via neural network based autoencoders, prior to clustering. Such dimensionality reduction not only improves the performance of the clustering algorithm, but also facilitates using additional features, such as estimated outputs produced from parallel solutions of simplified versions of the CEM for each disjoint period in the input data (e.g. 1 week). The impact of incorporating dimensionality reduction as part of RPS methods is quantified through the error in outcomes of the corresponding reduced-space CEM vs. the full space CEM. Extensive numerical experimentation across various networks and range of technology and policy scenarios establish the superiority of the dimensionality-reduction based RPS methods.