基于优化LSTM网络的多区域协同流感预测方法

Multi⁃Regional Collaborative Influenza Prediction Method Based on Optimized LSTM Network

流感通常表现出季节性、急性起病和传播速度快的特点,因此对流感的准确预测至关重要。针对流感预测精度不佳和长短期记忆网络参数寻优困难导致耗时耗力的问题,提出了一种基于皮尔逊相关系数和采用蜣螂优化算法(DBO)优化长短期记忆网络(LSTM)的多区域协同流感预测方法(MRC-DBO-LSTM)。该模型不仅学习本地的历史数据,还学习与其强相关的区域的历史数据。首先,通过皮尔逊相关系数筛选与预测地强相关的区域,以得到更高维度的输入特征;其次,通过LSTM的门机制衡量这些区域数据的权重来进行特征融合;最后,引入蜣螂优化算法对长短期记忆网络的超参数(如隐藏层数、隐藏层节点数和迭代次数等)寻优,进而生成预测结果。对山西省流感发病率预测的实验结果表明,学习多区域历史数据的DBOLSTM模型的均方误差(MSE)仅为0.0038,与差分整合移动平均自回归(ARIMA)模型相比,MSE降低了99.6%;与季节性差分自回归滑动平均(SARIMA)模型相比,MSE降低了98.7%;与LSTM模型相比,MSE降低了71.0%,与仅使用本地历史数据的DBO-LSTM模型相比,MSE降低了48.6%。结果证明所提模型能够有效提高流感的预测精度。

Influenza usually shows the characteristics of seasonal,acute onset and rapid transmission,so the accurate prediction of influenza is very important.Aiming at the problems of poor accuracy of influenza prediction and the difficulty of optimizing parameters of long short-term memory(LSTM),a multiregion collaborative influenza prediction method(MRC-DBO-LSTM)based on Pearson correlation coefficient and dung beetle optimization algorithm(DBO)was proposed.The model learns not only the historical data of the local area,but also the historical data of the region with which it is strongly related.Firstly,Pearson correlation coefficient was used to select the regions strongly correlated with the prediction place,so as to obtain the input features of higher dimensions.Secondly,the LSTM gate mechanism was used to measure the weight of these regional data for feature fusion.Finally,dung beetle optimization algorithm was introduced to optimize the super parameters(such as the number of hidden layers,the number of hidden layer nodes and the number of iterations,etc.)of the LSTM,so as to generate prediction results.The experimental results of predicting influenza incidence in Shanxi Province show that the RSquared of the MRC-DBO-LSTM model based on multi-regional historical data is 0.988,and the mean square erro(r MSE)is only 0.0038.Compared with the differential integrated moving average autoregression(ARIMA)model,MSE is decreased by 99.6%,MSE is decreased by 98.7%compared to the seasonal differential autoregressive moving average(SARIMA)model,MSE is decreased by 71.0%compared to the LSTM model,and MSE is decreased by 48.6%compared to the DBO-LSTM model using only local historical data.It is proved that the proposed model can effectively improve the prediction accuracy of influenza.