Synthetic load profiles of non-energy intensive industrial sites: A combined bottom-up and top-down approach

The heterogeneous nature of the industrial sector in terms of process and product variety often hinders assessments of its energy consumption. The generation of synthetic load profiles (LPs) however offers a key instrument to support energy suppliers, grid operators and industries themselves by quickly evaluating the impacts of energy efficiency measures, fuel switch, new technologies etc. Currently, such LP generators are only developed for single real-life industry plants or require comprehensive beforehand data. Within this study, we propose a new model for generating synthetic LPs of industrial process chains without the need for extensive, plant-specific data. This solution includes a series of bottom-up and top-down approaches. We analyse different data sources and develop algorithms based on underlying data science methods and technical and economical modelling paradigms like Markov chains or Economy of Scale. Here, we describe novel findings e.g., a method that proves the prediction of industrial shift models from a top-down perspective or effects on the energy demand of industrial plants for rising production capacities. Furthermore, we prove that only some production processes in industrial facilities are responsible for the main share of their energy demand, as the residues can be modelled by numerical analyses. We validate the developed approach by generating synthetic electricity LPs of different real-life industrial plants and comparing the results to measured LPs. This study contains five case studies, of which we found valid approximations of our synthetic LPs to the measured, real-life plants. However, our model's results differ from measured LPs, especially when depicting lower energy demands (< 200 kW). Furthermore, long-term periodicities e.g., part-time working hours on Saturdays have not been incorporated into our model yet, which leads to certain inaccuracies. We quantified these effects within this study. This, nevertheless, leaves open areas for future research work. (c) 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).