Abstract

The increasing appliance of systems imitating human cognition requires reasonable, in particular transparent and resource-efficient, reference models.

This work demonstrates a course of action in the field of short-term greenhouse gas emission intensity forecasting with statistical Time Series Analysis to plan HPC-system operations in advance. For that, SARIMAX modelling in combination with methods from econometrics and computer science has shown to be qualified for benchmarking by its step-by-step traceability. It also provided high-accuracy rolling day-ahead prognoses on the hourly emission intensity of the German electricity production over a span of half a year. The NMAE of 10.9% and MASE of 73.9 % when predicting their absolute values, and NMAE of 3.4% and MASE of 61.7% for the corresponding hourly changes, prove the performance.

Such an elaborated model, and the applied measures, can be used as a reasonable reference base to benchmark various methods in the area of time series forecasting, whether statistical ones or also those imitating human cognition.

DOI: 10.15439/2026f4766