Constrained multi-objective land use allocation optimization under uncertainty

This dataset describes the software and data to quantify uncertainty in Pareto fronts arised from spatial data. This dataset contains the Python code for a multi-objective land use allocation optimization under uncertainty. The program is an extension to CoMOLA from Strauch et al. 2019 (https://doi.org/10.1016/j.envsoft.2019.05.003). For a detailed description of CoMOLA, we refer to the article. Before executing CoMOLA under uncertainty, extreme lower and upper bound samples need to be generated from the land use and soil fertility map with quantified uncertainty. That preprocessing is described reproducable with the following Mendeley Dataset: Hildemann, Moritz Jan; Verstegen, Judith (2021), “Sampling procedure of land use and soil fertility map under uncertainty”, Mendeley Data, V1, doi: 10.17632/6x6cccfc4x.1. For every produced extreme sample, CoMOLA needs to be performed with the corresponding land use and soil fertility map. As the computational effort and computation time are high (15-20 hours) and ten optimizations were performed for every extreme sample, the runs were performed in parallel on a high-performance Linux cluster (MEGWARE cluster with 15.120 cores, 412 nodes and Intel Xeon Gold 6140 18C 2.30GHz processors). The program is executable for Python 3.7 and 3.8 in a Linux environment. The changes compared to CoMOLA include: an update to Python 3.8, removal of R components in objecting the objective values, and the implementation of a seeding procedure to inject single-objective optima into the first generation of the Genetic Algorithm. The seeding procedure allowed faster and better convergence. The generated Pareto fronts can be used postprocessing to quantify the uncertainty in objective and solution space. Pseudo-random states are used to assure reproducibility despite the stochastic processes.

本数据集旨在介绍用于量化空间数据衍生帕累托前沿（Pareto fronts）不确定性的配套软件与数据集。本数据集包含面向不确定性下多目标土地利用分配优化的Python代码。本程序是对Strauch等人2019年提出的CoMOLA模型的扩展（https://doi.org/10.1016/j.envsoft.2019.05.003）。有关CoMOLA的详细说明，请参阅其原文文献。在执行不确定性场景下的CoMOLA前，需从带有量化不确定性的土地利用与土壤肥力地图中生成极端下界与上界样本。该预处理流程可通过以下Mendeley数据集复现：Hildemann, Moritz Jan; Verstegen, Judith (2021), 《不确定性下土地利用与土壤肥力地图采样流程》，Mendeley Data，V1，doi: 10.17632/6x6cccfc4x.1。针对每一组生成的极端样本，均需结合对应的土地利用与土壤肥力地图运行CoMOLA。由于单组极端样本对应的优化计算量与耗时较高（单次需15至20小时），且每组极端样本需执行10次优化，因此所有运行均在高性能Linux集群（MEGWARE集群，配备15120核心、412个节点与Intel Xeon Gold 6140 18C 2.30GHz处理器）上并行完成。本程序可在Linux环境下的Python 3.7与3.8版本中运行。相较于原始CoMOLA模型，本次更新的内容包括：升级至Python 3.8版本、移除目标值计算环节中的R组件，以及实现种子初始化流程，即将单目标最优解注入遗传算法（Genetic Algorithm）的初代种群。该种子初始化流程可加快收敛速度并提升收敛质量。所生成的帕累托前沿可用于后处理环节，以量化目标空间与解空间内的不确定性。尽管流程中存在随机过程，但通过伪随机状态设置可确保实验可复现。