FU Hua，DAI Wei（1.Faculty of Electrical and Control Engineering，Liaoning Technical University，Huludao Liaoning 15105，China；.College of Safety Science and Engineering，Liaoning Technical University，Fuxin Liaoning 13000，China）



Gas Concentration Dynamic Prediction Method of Mixtures Kernels LSSVM Based on ACPSO and PSR*

FU Hua1*，DAI Wei2
（1.Faculty of Electrical and Control Engineering，Liaoning Technical University，Huludao Liaoning 125105，China；2.College of Safety Science and Engineering，Liaoning Technical University，Fuxin Liaoning 123000，China）

In order to predict the gas concentration of coalface accurately，the gas concentration dynamic prediction method of mixtures of kernels least squares support vector Machine theory based on phase space reconstructiontheo⁃ry and adaptive chaos particle swarm optimization was proposed.This paper has the coalface gas concentration ob⁃tained by wireless sensor network monitoring system in underground to be the target，the noise of the gas concentra⁃tion was filtered by translation invariant de-noising method，the MK-LSSVM model was trained with gas concentra⁃tion time series data based on phase space reconstruction，and ACPSO algorithm was used to optimize the parame⁃ters of MK-LSSVM model，the prediction accuracy of the whole system was improvement by error correction method. The simulation result shows that，the dynamic prediction method we proposed was able to make prediction result fit the monitoring data well.And the mean absolute percent error was 0.024 1，the relative root mean square error was 0.209 7，the average relative variance was 0.003 11，the results were reasonable and meet the actual needs of the project，which can provide an effective theoretical basis forprediction and prevention work of mine gas.

gas concentration；dynamic prevention；phase space reconstruction；adaptive chaos particle swarm algo⁃rithm；measures-kernel LSSVM

预测工作面瓦斯浓度是防治瓦斯突出灾害的重要措施。受地质构造、煤层厚度等自然因素和开采技术的影响，采煤工作面瓦斯浓度呈现出显著的不均衡性和复杂性，目前，以瓦斯浓度作为研究对象进行预测、预警的方法较多，譬如灰色系统［3］、分形理论［2］、D-S理论［3］、支持向量机［4］、神经网络［5］等，这些方法都是对瓦斯浓度预测的有益探索。然而，在有限的瓦斯浓度观测数据中，瓦斯浓度的非平稳性和随机性对预测结果存在较大影响［6］，需要提高预测模型的精度和泛化能力。因此，提出用混合核最小二乘支持向量机MK-LSSVM（Mixtures Kernel Least Squares Support Vector Machine）网络拟合工作面瓦斯浓度与其历史瓦斯浓度数据之间的非线性函数关系，弥补单一核函数的不足；提出用自适应混沌粒子群优化算法ACPSO（Adaptive Chaos Particle Swarm Optimi⁃zation）所具有的全局搜索能力去获取MK-LSSVM模型的最优参数，同时构建误差校正模型，以提高预测模型的性能。建立相空间重构PSR（Phase Space Reconstruction）MK-LSSVM瓦斯浓度预测模型，实现工作面瓦斯浓度的快速、有效预测，为煤矿的安全监测监控提供良好的理论支持及技术指导。

1　瓦斯浓度时间序列相空间重构

瓦斯浓度序列可以描述为具有混沌非线性特征的时间序列，根据Takens定理，获取瓦斯序列中所蕴含的可以表征动力系统的初始特征信息，则需要创建多维状态空间，使之成为MK-LSSVM模型可用的输入矢量，即对瓦斯浓度时间序列进行相空间重构［7-8］。

对于瓦斯浓度时间序列｛x（t）｝（t=1，…N），根据相空间微熵率重构参数法［9］，以微熵率最小的方法选取合适的延迟时间τ与嵌入维数m进行相空间重构。……