基于LSTM的高层住宅电梯群客流模式识别  

Passenger Flow Pattern Recognition of Elevator Group in High-Rise Residential Buildings Based on LSTM

作　　者：顾玲丽 董佳琦 许洪华[2] 徐啸 GU Ling-li;DONG Jia-qi;XU Hong-hua;XU Xiao(School of Electronics and Information Engineering,Suzhou University of Science and Technology,Suzhou Jiangsu 215009,China;Jiangsu Building Smart Energy-saving Laboratory,Suzhou University of Science and Technology,Suzhou Jiangsu 215009,China)

机构地区：[1]苏州科技大学电子与信息工程学院,江苏苏州215009 [2]苏州科技大学江苏省建筑智慧节能实验室,江苏苏州215009

出　　处：《计算机仿真》2023年第4期449-454,共6页Computer Simulation

基　　金：国家自然科学基金(62072324);国家重点研发计划课题(2020YFC2006602);江苏省重点研发计划项目(BE2020026)。

摘　　要：交通流模式识别是电梯群控系统进行优化调度必不可少的前提,针对高层住宅电梯交通流模式识别问题,经典的模式识别方法如支持向量机、模糊推理方法等应用广泛,但其仍存在长期记忆不足,识别准确度不高等缺点。深度学习中的LSTM适合学习电梯交通流时间序列中的长期依赖关系,因此本文提出一种基于LSTM的高层住宅电梯群交通模式识别方法。方法通过LSTM建立神经网络模型,学习电梯交通流时间序列中的长期依赖关系,用Softmax分类器进行分类,同时通过Adam算法优化网络参数。实验结果表明,上述算法下的交通模式识别较传统的支持向量机(SVM)、多元线性回归等方法识别精度上有明显提高。Traffic flow pattern recognition is an indispensable prerequisite for optimized dispatching of elevator group control systems.For the problem of high-rise residential elevator traffic flow pattern recognition,classic pattern recognition methods such as support vector machines and fuzzy inference methods are widely used,but they still have the shortcomings such as insufficient long-term memory and low recognition accuracy.The LSTM in deep learning is suitable for learning the long-term dependencies in the time series of elevator traffic flow,therefor the paper proposes a traffic pattern recognition method for high-rise residential elevator groups based on LSTM.This method builds a neural network model through LSTM,learns the long-term dependence in the elevator traffic flow time series,uses the Softmax classifier to classify,and optimizes the network parameters through the Adam algorithm.Experimental results show that the recognition accuracy of traffic pattern recognition under this algorithm is significantly improved compared with traditional support vector machine(SVR),multiple linear regression and other methods.

关 键 词：电梯群客流 模式识别 长短期记忆网络 均方差标准化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]