基于SVM的矿井提升机载荷辨识研究

发布时间：2018-03-22 19:24

本文选题：矿井提升机　切入点：载荷辨识　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：矿井提升机作为矿山机械的重要组成部分,在矿山生产运行中起着重要的"咽喉"作用。由于矿山生产环境的复杂性,矿井提升机往往也在复杂多变的载荷作用下工作,这些既有负载工作时的正常载荷,也包含故障状态下的危险载荷,这与机组的安全运行有着密切的联系,但是受到工作环境限制,这些载荷往往难以通过直接的方法测量,因此提出了载荷辨识方法,即通过载荷下相对较易测得的动态响应辨识载荷的方法。为了提出有效的载荷辨识方法,需要准确了解矿井提升机的动态响应,在载荷作用下,首先影响矿井提升机的运行特性,然后会影响运行电流,以及固定支座的振动情况,考虑到电流与振动信号的复杂性与处理信号后数据误差的传递性,提出从运行特性辨识矿井提升机载荷的方法。以传递矩阵法假设为依据合理简化矿井提升机物理模型,以能量守恒、功率守恒的原则对转动惯量、刚度、转矩做等效折算,以磁场守恒、功率守恒的原则对异步电机电磁模型进行解耦,进而建立MATLAB/Simulink机电耦合模型与矢量控制仿真模型,从而仿真得到矿井提升机正常载荷与故障载荷(稳态、冲击、瞬态、线性、正弦载荷)作用下的运行特性曲线。发现速度变化特性与载荷特性有着一致的变化关系;数值上,载荷越大,速度降低也越大;而且在正弦载荷作用下,速度变化频率与载荷变化频率一致,频率越大,相同幅值的载荷作用下,速度变化的幅值越小。利用这些特性,可以绕开系统自身特征参数,以运行特性为自变量,载荷特性为因变量,从而实现矿井提升机的载荷辨识。为了能够精确辨识载荷,对比传统回归预测方法与新型回归预测方法,引入SVM辨识载荷的方法。SVM理论完善,具有的良好的泛化能力;需要调节的参数较少,鲁棒性强;只需要简单的优化技术,计算简单;在解决实际问题中总是属于最好的方法之一。因此本文采用SVM模型并与具有代表性的典型的全局寻优算法Grid、启发性直接算法GA与基于群体的启发性算法PSO结合进行载荷辨识,辨识结果表明,Grid与PSO优化SVM往往能得到较好的辨识效果,而Grid优化SVM的辨识效果一般优于PSO优化SVM,这与本文数据分布有关,而且辨识结果最大的绝对误差都发生在最小载荷的预测上,呈现出最大绝对误差到最小的收敛趋势。在数据处理方面,无论是无反馈还是反馈型SVM载荷辨识时,对自变量与因变量较好的处理方式均为自变量不归一化-因变量不归一化和自变量归一化-因变量不归一化;但是自变量归一化与不归一化对最终的辨识效果影响不是很大,可从二者中测试选择最优的,是否PCA降维处理也是如此;而且在反馈控制型SVM载荷辨识时,发现因变量的大小对最终的辨识效果影响比较大,通过适当的比例变化,可以极大的降低最大的绝对误差,而且对整体的辨识效果影响不大,甚至可以优化整体辨识效果。最后,研制试验台,采集加载数据,通过试验验证的方法,证明了 SVM载荷辨识方法在实际应用中的有效性,为矿井提升机的载荷辨识与故障诊断提供了理论依据。
[Abstract]:Mine hoist is an important part of mining machinery, in mine production operation plays an important role in the "throat". Due to the complexity of mine production environment, often work in machine loading under complex mine hoist, the existing load when the normal load, also contains dangerous load fault state next, has a close connection with the safe operation of the unit, but by working environment constraints, these are often difficult to load by direct measurement method, so the proposed load identification method, namely through the dynamic load is relatively easy to measure the response identification method of load. In order to put forward the effective method of load identification, need an accurate understanding of the dynamic response of mine hoisting machine, under load, the influence of operating characteristics of mine hoist, and then affect the operation of the current, and the vibration of the fixed bearing, test Considering the complexity of signal processing and transfer current and the vibration signal data after the error of the proposed method from the operation of mine hoist load characteristics identification. With the transfer matrix method based on simplified assumption of mine hoist physical model based on the energy conservation, power conservation principle of inertia, stiffness, torque and equivalent convert to the magnetic field of asynchronous motor electromagnetic model of decoupling power conservation principle, and then establish the model of MATLAB/Simulink and vector control simulation model of electromechanical coupling, and simulation machine normal load and fault of mine hoist load (steady-state, shock, transient, linear, sinusoidal load) operating characteristic curve under speed variation. And load changes have consistent relationships; numerical, the larger the load is, the greater the speed is reduced; and in sinusoidal load, velocity and load frequency Changes in the same frequency, the higher the frequency, the amplitude of the same load, the amplitude of velocity change is small. Based on these properties, its characteristics can bypass the system parameters, operating characteristics as independent variables and load characteristics as the dependent variable, so as to realize the mine hoist load identification. In order to be able to accurately identify the load forecast method, comparison the traditional regression forecasting method and regression model, improve the introduction of SVM based load identification method of.SVM theory, has the good generalization ability; the need to adjust the parameters of the less robust; only need optimization technology, a simple calculation is simple; always belongs to one of the best methods in solving practical problems. This paper uses the SVM model and the typical global optimization algorithm for load identification Grid, heuristic algorithm GA and PSO combined with direct heuristic algorithm based on population. The identification results show that Grid and PSO Optimization of SVM often can get better identification effect, and Grid identification and optimization effect of SVM is better than PSO SVM the general optimization, and the data distribution, and the identification results of the largest absolute error occurred in the prediction of minimum load, showing a trend of convergence to the maximum absolute error minimum. In the aspect of data processing, either no feedback or feedback type SVM load identification, the independent and dependent variables are way better for variables not normalized variables and independent variables for the normalized normalized variables are not normalized; but the impact of independent variables normalized and non normalized to the identification of the final result is not great, from the two test to choose the best. Whether the PCA dimension is so; but in the feedback type SVM load identification control, found that due to the impact of variable size on the identification of the final result is relatively large, through appropriate The proportion of change, can greatly reduce the maximum absolute error, but has little effect on the identification of the overall effect, even can optimize the overall identification effect. Finally, development of test bench, loading data acquisition method, through experiments, proves the validity of SVM load identification method in practical application, for the machine and load identification fault diagnosis provides a theoretical basis for the mine hoist.

【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TD534

【参考文献】