(北京科技大学 土木与资源工程学院，北京 100083)
摘 要: 矿山充填系统在试运行期间的充填料浆配合比变化较大，如何快速准确地获得井下各采场的充填体强度对相邻采场的安全开采来说至关重要。本文首先以料浆浓度(质量分数)、水泥掺量、人工砂尾砂比和养护时间作为输入因子，以室内实验充填体单轴抗压强度SL作为输出因子，建立了一种ANN-PSO预测模型。然后定义了充填体强度预测折减系数k的概念，并通过对比大量相同配合比下的室内实验充填体强度值SL和实际生产充填体强度值SE，计算获得了两者之间的k值。该模型对室内实验充填体强度值SL的预测性能较好，在预测时其平均相对误差为2.41%，可决系数R2为0.992。采用所建模型并联合强度折减系数k，成功预测并分析了某矿山井下263条进路内充填体的实际生产测定值SE，为开采充填采场相邻矿体的支护工作提供了及时有效的指导。
关键字: 充填体强度；智能预测；人工神经网络；粒子群算法; 强度折减
（School of Civil and Resource Engineering, University of Science and Technology Beijing, Beijing 100083, China）
Abstract:Backfill mix proportion changes greatly during the test running of backfill system, thus, obtaining the backfill strength of particular stopes accurately and quickly plays an important role in the safety of mining in adjacent stopes. This paper firstly established an ANN-PSO intelligent prediction model by taking slurry density, cement dosage, ratio of artificial aggregate and tailings and curing time as input factors, and uniaxial compressive strength of laboratory backfill as output factor. Subsequently, the concept of predicted strength reduction coefficient of backfill was defined, and the strength reduction coefficient k was obtained by comparing the backfill strength of laboratory experiments and backfill strength of actual production under the same mix proportion. The results show that the model reveals a good prediction performance for the backfill strength of laboratory experiments, with a mean relative error (EMR) of 2.41% and a determination coefficient (R2) of 0.992. Based on the ANN-PSO model and strength reduction coefficient k, the backfill strengths of actual production of 263 access during the running period are predicted and analyzed, which provide timely and effective guidance for the support works of mining in adjacent stopes.
Key words: strength of backfilling body; intelligent prediction; artificial neural network; particle swarm optimization; strength reduction
