改进多变量时序模型的露天涌水量预测
Prediction of open air water inflow based on improved multivariate time series model
王孝东 1杨懿杰 1吕玉琪 1刘唱 1陈炫中 1谢博 2杜青文2
作者信息
- 1. 昆明理工大学国土资源工程学院,昆明 650093
- 2. 昆明理工大学公共安全与应急管理学院,昆明 650093
- 折叠
摘要
露天矿坑涌水量变化影响着边坡的稳定性、工程进度和设备使用寿命,在矿山汛期,涌水量的突增给矿山带来巨大的安全隐患.为了做好涌水量突增安全防范,对于汛期涌水量的精准预测成为矿山安全生产的一大难题.针对这一问题,提出了一种基于改进蜣螂优化算法(Sparrow Initialization Dung Beetle Optimizer,SIDBO)优化变分模态分解(Variational Mode Decomposition,VMD)-双向长短周期神经网络(Bi-directional Long Short-Term Memory,BiLSTM)时序模型预测露天矿坑涌水量的方法.对于难以确定VMD参数的问题,利用改进蜣螂优化算法寻找最优VMD核心参数组合.SIDBO算法首先基于t分布的差分策略优化勘探阶段,使用最优解和第二解中位搜寻策略增强全局最优解搜索能力,最后采用麻雀优化算法优化开发阶段.结果表明,与VMD-SIDBO-LSTM 等模型相比较,SIDBO-VMD-SIDBO-BiLSTM模型预测精度更高,均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)、R2分别为5.96、4.96、0.41％、0.98,并将该模型与传统地质方法——水均衡法在实际工程实例中进行对比,该时序模型相对于水均衡法对于矿坑汛期涌水量预测精度提升了 3.8％,为露天矿汛期涌水量预测提供了新的技术方法与思路.
Abstract
Changes in the amount of water surging from open pits affect the stability of slopes,the rate of decline of the project and the service life of the equipment.During the flood season of the mine,the sudden increase in the amount of water surging poses a huge safety hazard to the mine.Therefore,in order to do a good job in the surge of water safety precautions,the accurate prediction of the flood season surge of water has become a major problem of mine safety production.To address this problem,this paper proposes a method based on the Sparrow Initialization Dung Beetle Optimizer(SIDBO)algorithm to optimize the Variational Mode Decomposition(VMD)-Bi-directional Long Short-Term Memory(BiLSTM)time series model to predict the water inflow of open pit.BiLSTM temporal modeling method for predicting water influx in open pits.For the problem of difficult to determine VMD parameters,an improved dung beetle optimization algorithm is used to find the optimal VMD core parameter combinations.The SIDBO algorithm first optimizes its exploration phase based on the t-distributed difference strategy,enhances its global optimal solution searching capability using the optimal solution and the second solution median searching strategy,and finally optimizes its development phase using the sparrow optimization algorithm.The BiLSTM is optimized by SIDBO,and SIDBO optimizes its three parameters:the optimal number of hidden units,the optimal training period,and the optimal initial learning rate.The optimized VMD decomposition components and the mine rainfall data are brought into the super-parameter-optimized BiLSTM to make predictions,and finally the predictions are accumulated and summed up.The results show that the SIDBO-VMD-SIDBO-BiLSTM model has higher prediction accuracy compared with other models such as VMD-SIDBO-LSTM.The four indexes of Root Mean Squared Error(RMSE),Mean Absolute Error(MAE),Mean Absolute Percentage Error(MAPE),and R2 are 5.96,4.96,0.41％,and 0.98,respectively.Comparing this model with the traditional geological method-water equalization method in the actual engineering examples,this time series model can improve the accuracy of water influx prediction for the flood season of the pit by 3.8％,providing a new technical method and idea for the prediction of flood season of the open pit mine and supporting safety production.
关键词
安全工程/算法优化/时序预测/变分模态分解(VMD)/深度学习Key words
safety engineering/algorithm optimization/time series prediction/Variational Modal Decomposition(VMD)/deep learning引用本文复制引用
基金项目
昆明理工大学引进人才科研启动基金项目(KKSY201721032)
出版年
2024
NETL
NSTL
万方数据