基于长时序数据多重分解降噪的药品需求预测

药品预测能够确保患者及时获得所需药品，还可有效地控制传染病的蔓延。鉴于药品需求长时序数据具有多周期性和高随机性等特点，文章提出一种基于多重分解降噪的策略并构建综合预测模型。首先，采用多周期时间序列分解（multiple seasonal-trend decomposition using LOESS, MSTL）算法和自适应噪声完全集合经验模态分解对药品需求序列进行二次分解，以充分提取时序特征；其次，针对各分解分量进行小波降噪与组合重构，有效降低数据噪声干扰；在建模阶段，结合卷积神经网络（convolutional neural network, CNN）的特征提取能力和长短期记忆（long short-term memory, LSTM）网络的时序建模优势，构建混合预测模型；为进一步提升性能，采用非洲秃鹫优化算法（African vultures optimization algorithm, AVOA）对模型超参数进行自动优化。实验结果表明，该文所提出的分解降噪策略能够有效提升模型的预测性能，并具有良好的通用性。

Drug prediction can ensure that patients get the drugs they need in time. It can also effectively control the spread of infectious diseases. In view of the characteristics of multi-periodicity and high randomness of long time series data of drug demand, this paper proposes a comprehensive prediction model based on multi-decomposition denoising. Firstly, multiple seasonal-trend decomposition using LOESS(MSTL) algorithm and complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) are used to decompose the drug demand series for the second time to fully extract the time series features. Secondly, wavelet denoising (WD) and combination reconstruction are carried out for each decomposition component to effectively reduce data noise interference. In the modeling stage, a hybrid prediction model is constructed by combining the feature extraction ability of the convolutional neural network(CNN) and the time series modeling advantage of the long short-term memory(LSTM) network. In order to further improve the performance, the African vultures optimization algorithm(AVOA) is used to automatically optimize the hyperparameters of the model. The experimental results show that the decomposition denoising strategy proposed in this paper can effectively improve the prediction performance of the model and has good versatility.