Short-team planning for mining operations using robust optimization

该研究计划旨在开发和解决一种新的方法，以便在非常短期和短期内对地表和地下采矿作业进行更稳健的规划。作业规划确定了三个月至一年的开采顺序。它确定了在此期间将开采的材料块及其顺序。该调度必须考虑几个操作约束，例如块之间的优先级（其中包括遵守有关最大坡度的安全规则），作业活动之间的优先顺序（钻孔、爆破、运输）、每天可处理的矿石量的限制、设备（如铲子和钻头）的可用性和位置，通过坡道等进入提取区。 传统上，确定性线性规划模型已被用于解决业务规划。这些模型是有效的，但他们缺乏现实主义，因为使用的几个参数不能被认为是确定性的：岩石的机械和化学性质，卡车的行驶时间，装载时间，等待时间在服务点等这些确定性模型假设系统参数的准确值，通过使用标称值，即平均值的基础上地质统计研究和历史数据。在现实世界的应用中，例如用于短期和非常短期规划的应用中，数据的不确定性在两个重要方面威胁到由确定性线性规划模型获得的解的相关性：（i）当决策者试图实施该解决方案时，该解决方案实际上可能并不可行，以及（ii）该解决方案在可行时，可能会导致比真正的最优策略更大的成本（或更小的收入）。本提案中描述的研究计划旨在通过提出一种称为鲁棒优化的建模和解决方案来克服这些缺点。

This research program aims to develop and solve a new approach for more robust planning of surface and underground mining operations in the very short-term and short-term. Operational planning determines the mining sequence for a period of three months to one year. It identifies the blocks of material to be mined out during this period and the order in which they will be. This scheduling must take into account several operational constraints such as the precedence among the blocks (among other things, to comply with safety rules on the maximum degree of slope), the precedence among the operational activities (drilling, blasting, transportation), the limit on the amount of ore that can be treated per day, the availability and the location of equipment such as shovels and drills, the access to extraction areas via ramps, etc. Traditionally, deterministic linear programming models have been used for solving operational planning. These models are effective, but they lack of realism because several of the parameters used can not be considered as deterministic: the mechanical and chemical properties of rock, the travel time of trucks, loading time, waiting time at service points, etc. These deterministic models assume accurate values for the system parameters by using a nominal value, i.e. an average value based on geostatistical studies and historical data. In real-world applications, such as those used for short-term and very short-term planning, a uncertainty in the data threatens the relevance of the solution obtained by the deterministic linear programming model in two important aspects: (i) the solution might not actually be feasible when the decision-maker attempts to implement it, and (ii) the solution, when feasible, might lead to a far greater cost (or smaller revenue) than the truly optimal strategy. The research program described in this proposal aims to overcome these drawbacks by proposing a recent approach to modeling and resolution called robust optimization.