基于ILSTM-AMSGD神经网络的时间序列预测方法  

Time Series Prediction Method Based on ILSTM-AMSGD Neural Network

作　　者：杨爽 李文静 乔俊飞[1,2,3,4,5] YANG Shuang;LI Wenjing;QIAO Junfei(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Laboratory of Smart Environmental Protection,Beijing University of Technology,Beijing 100124,China;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing University of Technology,Beijing 100124,China;Engineering Research Center of Intelligent Perception and Autonomous Control,Ministry of Education,Beijing University of Technology,Beijing 100124,China;Beijing Institute of Artificial Intelligence,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]北京工业大学智慧环保北京实验室,北京100124 [3]北京工业大学计算智能与智能系统北京市重点实验室,北京100124 [4]北京工业大学智能感知与自主控制教育部工程研究中心,北京100124 [5]北京工业大学北京人工智能研究院,北京100124

出　　处：《控制工程》2023年第10期1793-1800,共8页Control Engineering of China

基　　金：科技创新2030——“新一代人工智能”重大项目(2021ZD0112301);国家自然科学基金资助项目(62021003,61890930-5,62173008)。

摘　　要：针对标准长短期记忆(long short-term memory,LSTM)神经网络的结构参数众多、训练过程耗时长问题,提出一种基于自适应动量随机梯度下降(adaptive momentum stochastic gradient descent,AMSGD)算法的改进型长短期记忆神经网络(ILSTM-AMSGD神经网络),并将其用于时间序列预测中。首先,通过简化结构方程中的递归项权值,减少网络中所需训练的参数。其次,设计一种AMSGD算法对神经网络结构参数进行学习。最后,通过2个基准数据集和1个实际数据集对ILSTM-AMSGD神经网络模型在时间序列预测中的准确性和运行效率进行实验验证。结果表明,递归项权值简化方法可以提高模型的泛化能力,同时AMSGD算法加快了模型的收敛速度。与其他模型相比,ILSTM-AMSGD神经网络模型实现了对时间序列更加高效、准确的预测。Conventional long short-term memory(LSTM)neural networks have many structural parameters and the training process is time-consuming.To solve this problem,an improved LSTM neural network based on the adaptive momentum stochastic gradient descent(AMSGD)algorithm(ILSTM-AMSGD neural network)is proposed and applied to time series prediction.Firstly,the number of training parameters in the network is reduced by simplifying the weights of the recursive term in the structural equation.Then,the AMSGD algorithm is designed to train the structural parameters of the neural network.Finally,with two benchmark data sets and one practical data set,the accuracy and operational efficiency of the ILSTM-AMSGD neural network model are verified in time series prediction.The results show that the weight simplification method of the recursive term can improve the generalization ability of the model.Meanwhile,the AMSGD algorithm improves the convergence speed of the model.ILSTM-AMSGD neural network model can predict time series more efficiently and accurately than other models.

关 键 词：时间序列预测 改进型长短期记忆神经网络 权重精简 梯度下降算法 自适应 动量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]