Benchmark Dataset for "Industrial Demand-Side Flexibility: A Benchmark Data Set"

The archive on hand contains a set of benchmark instances for time-flexible industrial processes, with regard to power demand. It also includes some auxiliary data, such as baseline solutions to the instances as well as some intermediate data that the instances were generated from.

Instances

The instances can be found in the 'instances' subfolder. Every JSON file contains one instance. See the 'instance_file_format.{md, html, pdf}' files for a specification of the file format. The file format is suitable to be used with the TCPSPSuite optimization software [2].

In [1], Section 4, we list various parameters that influence the generation of instances. The parameter settings that were active for each instance are stored in the additional field of the instance. For example, the generator__window_mean key in the additional field specifies the window growth mean parameter.

Also, we performed a grouped generation of instances, as explained in Section 4.1 of [1], meaning for every instance with generator__block_count of k, there are other instances with different values for generator__block_count, such that there is a one-to-one correspondence between their jobs. Each such group has a unique group_id in the additional field. To identify which jobs correspond to each other across instances in the same group, the superjob_id key in the additional field of each job can be used. For instances with generator__block_count of more than 1, all blocks (i.e., jobs) of the same job will have the same superjob_id, thus the superjob_id can also be used to identify which jobs are the blocks of the same non-constant job.

Baseline Results

In the 'results' subfolder, we provide the solutions that we were able to compute on all instances. The file 'results/results.csv' lists the numeric results, one line per instance. We report the quality of the best found solution as well as the best lower bound, and computed from that the MIP gap.

In the 'results/solutions' subfolder, the actual schedules computed for each instance are included. The file format is the same as the instance file format, with some fields stripped out, and an additional start_time field for each job that reports the start time for the respective job.

Occurrence Block Decomposition

In 'decompositions/decompositions.csv', you find the decompositions of the motif occurrences we detected. Unfortunately, we are not able to publish the raw time series data itself. However, a subset of the raw input data (without discovered motifs) is available as the HIPE data set [3][4].

The decomposition file format is as follows: Every line corresponds to one occurrence. The first column specifies the motif that the occurrence belongs to. The second column indicates the start time with minute resoulution, i.e., "600" would be a start at 10 a.m.

The remaining columns are titled "Energy A / B" or "Duration A / B". Each such columns specifies the amount of energy resp. the duration of the Ath block in a decomposition into B blocks of the occurrence. E.g., "Energy 3 / 6" would indicate the amount of energy consumed during the third block in a six-block decomposition of the corresponding occurrence.

References

[1] Nicole Ludwig, Lukas Barth, Dorothea Wagner, and Veit Hagenmeyer. 2019. Industrial Demand-Side Flexibility: A Benchmark Data Set. In Proceedings of ACM e-Energy (e-Energy ’19). ACM, New York, NY, USA.
[2] https://github.com/kit-algo/TCPSPSuite
[3] Simon Bischof, Holger Trittenbach, Michael Vollmer, Dominik Werle, Thomas Blank, and Klemens Böhm. 2018. HIPE – an Energy-Status-Data Set from Industrial Production. In Proceedings of ACM e-Energy (e-Energy ’18). ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3208903.3210278
[4] https://www.energystatusdata.kit.edu/hipe.php