基于改进的BiLSTM网络能见度预测研究

Research on visibility prediction based on improved BiLSTM network

能见度对人们日常生活、生产方面的影响越来越大,因此实现对能见度的精确预测就显得尤为重要。近年来很多研究者利用人工智能技术对能见度进行预测,但都聚焦于逐小时数据、精细化程度较低。为了提升能见度预测的精度,本文提出了IM_BiLSTM_Attention网络。一方面,获取大量的逐分钟气象的数据,并计算Spearman相关系数,衡量其与能见度的相关性;另一方面,引入稀疏注意力机制对BiLSTM网络进行改进,进而选择性地关注时间序列中的重要信息以减少注意力分散和噪声数据干扰,提高了能见度预测的精度。通过在数据集上的实验结果表明,IM_BiLSTM_Attention在逐分钟能见度预测问题上效果更优。

Visibility has an increasing impact on people's daily life and production,so it will be especially important to realize the accurate prediction of visibility in a more refined way.In recent years,many researchers have utilized artificial intelligence techniques for visibility prediction,but all of them focus on the hour-by-hour data with a low degree of refinement.In order to improve the accuracy of visibility prediction,this paper proposes IM_BiLSTM_Attention network.On one hand,obtaining a large amount of minute by minute meteorological data and calculating Spearman correlation coefficients to measure their correlation with visibility;on the other hand,the BiLSTM network is improved by introducing the sparse attention mechanism,which selectively focuses on the important information in the time series to reduce distraction and interference from noisy data,and improves the accuracy of visibility prediction.The experimental results on four datasets show that the IM_BiLSTM_Attention is more effective in the minute-by-minute visibility prediction problem.