基于PCA-Informer+模型的周期性甲烷菌体浓度预测研究  

Prediction of periodic methane concentration based on PCA-Informer+

作　　者：杨帆 毛腾跃[1] 占伟 YANG Fan;MAO Tengyue;ZHAN Wei(South-Central Minzu University,College of Computer Science,Wuhan 430074,China;South-Central Minzu University,College of Resources and Environment,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,武汉430074 [2]中南民族大学资源与环境学院,武汉430074

出　　处：《中南民族大学学报(自然科学版)》2025年第3期393-399,共7页Journal of South-Central Minzu University(Natural Science Edition)

基　　金：国家民委中青年英才培养计划(MZR20007);湖北省技术创新计划重点研发专项(2023BAB087);新疆维吾尔自治区区域协同创新专项(科技援疆计划)(2022E02035)。

摘　　要：厌氧发酵是一种重要的生物技术,然而现有的检测技术无法实时测量发酵周期中甲烷菌体的浓度值,导致难以准确监测发酵状态.针对此问题,提出了基于改进Informer深度学习模型的周期性甲烷菌体浓度预测方法.首先,基于Informer构建基础预测模型;其次,利用PCA主成分分析,将特征变量从8维降低至3维,优化模型的输入,提高预测效率;然后,根据周期中每个时间点的重要性建模设计WeightedMSELoss损失函数,以更好地适应周期性甲烷菌体浓度预测任务;最后,融合特征变量、位置编码和周期编码,提高模型捕获长期依赖的能力.实验结果表明:Informer相较于长短期记忆网络(LSTM)、循环神经网络(RNN)、门控循环单元(GRU)在周期性甲烷菌体浓度预测任务上效果最好,且基于Informer改进的PCA-Informer+模型,平均绝对误差(MAE)、均方根误差(RMSE)相较于原Informer模型分别下降了26%、11%,模型效率提高了18%,实现了较为快速准确的甲烷菌体浓度预测.Anaerobic fermentation is an important biological technology,but the existing detection technology can not measure the concentration of methane bacteria in the fermentation cycle in real time,which makes it difficult to accurately monitor the fermentation state.To solve this problem,a periodic methane concentration prediction method based on improved Informer deep learning model was proposed.Firstly,the basic prediction model is constructed based on Informer.Secondly,based on PCA,the feature variables are reduced from 8 to 3 dimensions to optimize the input of the model and improve the prediction efficiency.Thirdly,according to the importance of different time steps in the cycle,the WeightedMSELoss loss function is designed to better adapt to the periodic methane culture task.Finally,feature variables,location coding and period coding are integrated to improve the ability of the model to capture long-term dependencies.The experimental results show that:Compared with long short-term memory network(LSTM),recurrent neural network(RNN)and gate recurrent unit(GRU),Informer has the best performance in predicting periodic methane concentration.Moreover,The PCA-Informer+model is based on Informer improvements,with MAE and RMSE reduced by 26%and 11%respectively,and model efficiency increased by 18%,achieving relatively fast and accurate prediction of methane concentration.

关 键 词：Informer模型 菌体浓度预测 主成分分析 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]