Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (11): 125-132.DOI: 10.3778/j.issn.1002-8331.2108-0314

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Improved FA Optimizing LSTM Time Series Prediction Model

ZHANG Zhonglin, ZHANG Yan   

  1. School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
  • Online:2022-06-01 Published:2022-06-01

改进FA优化LSTM的时序预测模型

张忠林,张艳   

  1. 兰州交通大学 电子与信息工程学院,兰州 730070

Abstract: In order to improve the accuracy of time series prediction, a time series prediction model(GAFA-LSTM), which is improved by firefly algorithm(FA) and optimized LSTM, is proposed. Aiming at the problem that FA falls into a local optimum due to the dispersion of population diversity, which affects the optimization effect, a mechanism for increasing population diversity is proposed. First, FA adds the calculation of population diversity after initialization; secondly, it introduces an adaptive diversity increase mechanism under the condition of satisfying the diversity increase mechanism to effectively balance the demand for population diversity in the evolutionary process; finally, in the later iteration, swimming parameters are added to avoid local oscillations. The improved FA is used to optimize the input parameters of the LSTM model to improve the accuracy of the input parameters of the LSTM model. In the experimental part, the improvement effect of the improved FA is tested, and the GAFA-LSTM model is verified. The results show that the improved FA has a better optimization effect, and the GAFA-LSTM prediction model has improved the degree of fit and prediction accuracy to varying degrees compared with other prediction models.

Key words: firefly algorithm, time series forecast, population diversity, long-term and short-term memory network(LSTM), mechanism of diversity increase

摘要: 为了提高时序预测精度,提出了一种改进萤火虫算法(firefly algorithm,FA)优化LSTM的时序预测模型(GAFA-LSTM)。针对FA因种群多样性弥散陷入局部最优,影响寻优效果的问题,提出了种群多样性增加机制。FA在完成初始化后加入种群多样性的计算;在满足多样性增加机制的条件下,引入自适应多样性增加机制,有效平衡进化过程中对种群多样性的需求;在迭代后期加入自适应游动参数来避免局部震荡。将改进后的FA用于LSTM模型输入参数的优化,以提高LSTM模型输入参数的准确性。实验部分对改进FA进行了改进效果测试,对GAFA-LSTM模型进行了模型验证。结果表明改进FA具有较好的寻优效果,GAFA-LSTM预测模型较其他预测模型拟合程度与预测精度都有不同程度的提高。

关键词: 萤火虫算法, 时间序列预测, 种群多样性, 长短期记忆神经网络(LSTM), 多样性增加机制