基于CEEMDAN-云模型特征熵和LSSVM的磨机负荷预测研究

针对球磨机磨矿过程中负荷难以检测和不能准确判断负荷状态的问题,提出了一种基于CEEMDAN-云模型特征熵和LSSVM的磨机负荷预测方法,用完整集成经验分解算法(CEEMDAN)对不同负荷的磨机振动信号进行分解,由相关系数法选取敏感模态分量重构信号,利用逆向云发生器计算重构信号的云模型特征熵作为信号的特征参数,运用正向云发生器生成云模型特征向量的云滴图,结果表明,欠负荷、正常负荷、过负荷之间的熵值差异很大,可以较好地区分和识别磨机负荷状态;将云模型特征向量作为最小二乘支持向量机(LSSVM)的输入,料球比、充填率为输出,建立磨机负荷预测模型;通过磨矿实验验证了该方法的有效性,模型能够准确预测磨机负荷状态。

Prediction of ball mill's load based on IEDA-cloud model feature entropy and LSSVM

Aiming at problems of ball mill’s load being difficult to detect and load state being not able to be determined in process of ball mill grinding ore, a prediction model for mill load based on IEDA-cloud model feature entropy and LSSVM was proposed. The integrated empirical decomposition algorithm(IEDA) was used to decompose vibration signals of a ball mill under different loads. The sensitive modal components were chosen using the correlation coefficient method to reconstruct a signal. The inverse cloud generator was used to calculate the cloud model’s feature entropy of the reconstructed signal as its feature parameter. The forward cloud generator was used to generate the cloud drop diagram of the cloud model feature vector. The results showed that differences among entropy values of under-load state, normal-load one and over-load one are large, and they can be used to better distinguish and identify ball mill’s load states. The feature vector of the cloud model was taken as the input of a least squares support vector machine(LSSVM), ratio of material to ball and filling rate were taken as the output to establish a prediction model for ball mill’s load, the effectiveness of the proposed model was verified by tests of grinding ore, it was shown that the proposed model can be used to correctly predict load state of ball mills.