基于LSSVM的电梯交通模式的模糊识别

Elevator traffic pattern fuzzy recognition based on least squares support vector machine

提出可以对电梯交通模式进行模糊识别的方法.采用最小二乘支持向量机(LSSVM)的回归算法来学习2种交通模式的相对隶属度,通过相对比较法得到当前时刻所有交通模式的隶属度.介绍了LSSVM二值分类算法及传统的多值分类算法,分析LSSVM多值分类与函数回归的关系.分析结果表明,采用函数回归算法可以进行多值分类.若以交通模式的隶属度作为类标,则可采用LSSVM的回归算法来进行2种交通模式的模糊分类.为了提高LSSVM的线性度,分3步逐步细分电梯客流的交通模式.实验结果表明,采用该方法得到的各交通模式隶属度随时间的变化曲线与依据群控专家经验得到的曲线非常相似,识别结果的平均误差小于应用神经网络识别的平均误差,可将识别结果作为电梯群控系统的输入参数.

A method for elevator traffic pattern fuzzy recognition was proposed. The relative membership of two traffic patterns was learned with least squares support vector machine （LSSVM） regression algorithm. Then the memberships of all traffic patterns were obtained by comparing with each other. The LSSVM algorithms of binary classification and traditional multi-class classification were introduced, and the rela- tionship of multi-class classification and function regression was analyzed. Results show that LSSVM func- tion regression algorithm can be used in multi-class classification. If the membership is regarded as class label, then two traffic patterns can be fuzzy recognized by LSSVM regression algorithm. Elevator traffic pattern was recognized in three steps in order to improve the linearity of LSSVM. Experimental results showed that the membership curves of traffic patterns versus time obtained by the method were similar to the curves given by elevator group control experts. The average recognition error using the method was less than that using a neural network, and the fuzzy recognition results can be used as input parameters of elevator group control system.