大冶铁矿开采沉陷GPS高程拟合CQPSO-LSSVM模型  

Mining Subsidence GPS Height Fitting CQPSO-LSSVM Model of Daye Iron Mine

作　　者：侯林锋 

机构地区：[1]广东工贸职业技术学院测绘遥感信息工程系,广东广州510510

出　　处：《金属矿山》2017年第9期166-169,共4页Metal Mine

摘　　要：传统最小二乘支持向量机拟合模型(Least squares support vector machine model,LSSVM)在进行矿区地表沉降GPS高程拟合时精度较低,为进一步提升矿区地表沉降监测精度,采用协同量子粒子群算法(Cooperative quantum-behaved particle swarm optimization,CQPSO)对LSSVM模型进行了优化。该算法的协同搜索策略是在解空间中使用多个子群取代整个种群,可有效解决由于单个种群、单个搜索策略导致的迭代后期种群多样性下降的早熟问题。以大冶铁矿为例,采用实地获取的矿区地表GPS监测数据对改进最小二乘支持向量机拟合模型(CQPSO-LSSVM)进行试验,并与BP神经网络拟合模型以及量子粒子群算法(Quantum-behaved particle swarm optimization,QPSO)优化的最小二乘支持向量机拟合模型(QPSO-LSSVM)进行比较,结果表明,CQPSO-LSSVM模型的内、外符合精度分别为±2.5 mm、±3.1 mm,BP神经网络拟合模型的内、外符合精度分别为±2.9 mm、±4.6 mm,QPSO-LSSVM模型的内、外符合精度分别为±2.8 mm、±3.5 mm,可见CQPSO-LSSVM模型的拟合精度稍优于其余两者,采用该模型对矿区地表沉降GPS数据进行拟合处理,可获得较高的监测精度。The precise of mining subsidence GPS height fitting of traditional least squares support vector machine model（ LSSVM） is low,which is unable to meet for the high precise mining subsidence monitoring requirement. In order to improve mining subsidence monitoring precise,the LSSVM model is improved by using cooperative quantum-behaved particle swarm optimization algorithm（ CQPSO）. The cooperative search strategy of the CQPSO algorithm can replace the whole population with multiple subgroups in solution space,which is good to solve the premature problem of population diversity declines in the late iteration produced by single population and single search strategy. Taking Daye Iron Mine as the study example,based on the measured mining subsidence GPS height data in the mining area,the test comparison of CQPSO-LSSVM model,BP neural network fitting model and the LSSVM model optimized by quantum-behaved particle swarm optimization algorithm（ QPSO） is done. The results show that： the internal and external coincidence accuracy of CQPSO-LSSVM model are ± 2. 5 mm,± 3. 1mm; the internal and external coincidence accuracy of BP neural network fitting model are ± 2. 9 mm、± 4. 6 mm; the internal and external coincidence accuracy of QPSO-LSSVM model are ± 2. 8 mm、± 3. 5 mm. Therefore,the fitting accuracy of CQPSO-LSSVM model is superior to the others. The mining monitoring precise can be effectively improved by using CQPSO-LSSVM model to process the GPS height data.

关 键 词：开采沉陷 GPS高程拟合 最小二乘支持向量机 协同量子粒子群算法 BP神经网络模型 

分 类 号：TD325[矿业工程—矿井建设]