一种新型激活函数的机床能耗预测神经网络研究  

Research on Neural Network with a New Activation Function for Energy Consumption Prediction of Machine Tools

作　　者：刘晶晶[1,2] 刘业峰[2] LIU Jingjing;LIU Yefeng(Department of Basic Courses,Shenyang Institute of Technology,Shenfu New District 113122,China;Liaoning Key Laboratory of Information Physics Fusion and Intelligent Manufacturing for CNC Machine,Shenyang Institute of Technology,Shenfu New District 113122,China)

机构地区：[1]沈阳工学院a.基础课部,辽宁沈抚示范区113122 [2]沈阳工学院辽宁省数控机床信息物理融合与智能制造重点实验室,辽宁沈抚示范区113122

出　　处：《控制工程》2025年第3期492-499,共8页Control Engineering of China

基　　金：沈阳工学院青年骨干教师科研基金资助项目(QN202210);国家自然科学基金面上项目(62073226);辽宁省科技计划联合计划(应用基础研究项目)(2023JH2/101700066);辽宁省教育厅基本科研立项(JYTMS20231179);东北大学流程工业综合自动化国家重点实验室项目(2023-kfkt-03);辽宁省自然科学基金重点领域联合开放基金资助项目(2020-KF-11-09,2021-KF-11-05);沈抚示范区本级科技计划项目(2020JH13,2021JH07)。

摘　　要：构造一种适用于反向传播(backpropagation,BP)神经网络的新型激活函数Lfun(logarithmic series function),并使用基于该函数的BP神经网络进行机床能耗状态的预测。首先,分析Sigmoid系列和ReLU系列激活函数的特点和缺陷,结合对数函数,构造了一种非线性分段含参数激活函数。该函数可导且光滑、导数形式简单、单调递增、输出均值为零,且通过可变参数使函数形式更灵活;其次,通过数值仿真实验在公共数据集上将Lfun函数与Sigmoid、ReLU、tanh、Leaky_ReLU和ELU函数的性能进行对比;最后,使用基于Lfun函数的BP神经网络进行机床能耗状态的预测。实验结果表明,使用Lfun函数的BP神经网络相较于使用其他几种常用激活函数的网络具有更好的性能。A novel activation function for BP neural network is proposed,called‘Lfun’(Logarithmic series function),which is applied to BP neural network to predict the energy consumption of machine tools in production lines.The characteristics and defects of Sigmoid series and ReLU series activation functions are analyzed,and a nonlinear piecewise activation function with parameter is constructed by introducing logarithm function.The function is differentiable and smooth,monotonically increasing,the derivative form is simple,the mean of the outputs is zero,and the variable parameter makes the function form more flexible.Then,the Lfun function is compared with Sigmoid,ReLU,tanh,Leaky_ReLU and ELU function in public datasets by some numerical simulations.Finally,the BP neural network with Lfun function is used to monitor the energy consumption of machine tools in production line.The experimental results show that the BP neural network with Lfun has better performance than that using other activation functions.

关 键 词：激活函数 BP神经网络 机床能耗预测 SIGMOID函数 ReLU函数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]