基于PSO-LSSVM球磨机负荷参数预测及监测系统开发

Ball Mill Load Parameter Prediction and Monitoring System Development Based on PSO-LSSVM

针对球磨机筒体振动信号中存在非线性、非平稳性及环境噪声强等问题,提出一种基于粒子群优化最小二乘支持向量机(PSO-LSSVM)球磨机负荷参数(填充率和料球比)预测方法,并开发基于LabVIEW的球磨机负荷参数监测系统。通过粒子群(PSO)优化最小二乘支持向量机(LSSVM)算法中的正则化参数y和核函数宽度,简化求解过程,提高模型训练速度。以球磨机筒体振动信号的Hilbert-边际谱样本熵为输入,以球磨机筒体内部的填充率和料球比为输出,建立基于PSO-LSSVM的磨机负荷参数预测模型。与LSSVM预测结果比较,该模型的预测精度较高,填充率平均绝对误差降低0.05、平均绝对百分误差降低8.09%;料球比平均绝对误差降低0.04、平均绝对百分误差降低2.76%。在线测试结果表明该在线监测系统准确率为64.37%,且系统运行一次的平均时间为45 s,可实现球磨机负荷参数的实时预测。

Aiming at the problems of nonlinear,non-stationary and strong environmental noise in the vibration signals of the ball mill barrel,a prediction method of ball mill load parameters(filling rate and material-to-ball ratio)based on particle swarm optimization and least squares support vector machine(PSO-LSSVM)was proposed,and a monitoring system of ball mill load parameters based on LabVIEW was developed.Particle swarm optimization(PSO)was used to optimize the regularization parameter y and kernel function width of the model in the least square support vector machine(LSSVM)algorithm to simplify the solution process and improve the training speed of the model.Taking Hilbert-marginal spectrum sample entropy of ball mill barrel vibration signal as the input,and filling rate and material-to-ball ratio inside the ball mill barrel as the output,the PSO-LSSVM prediction model was established to predict the mill load parameters.Compared with the prediction results of LSSVM,the prediction accuracy of this model is higher.The average absolute error of filling rate is decreased by 0.05 and the average percentage error is reduced by 8.09%.The average absolute error and the average percentage error of feed ball ratio are reduced by 0.04 and 2.76%respectively.The accuracy of the on-line monitoring system is 64.37%,and the average time of each system running is 45 s,which can realize the real-time prediction of ball mill load parameters.