基于Transformer的粮食安全信息融合算法研究

Research on Food Security Information Fusion Algorithm Based on Transformer

粮情监测和环境控制对保证粮库安全具有重要意义,但是由于粮库环境复杂,令相关工作人员对其的实时监测十分困难。这就出现了对粮库多源数据融合分析的需求,特别是对异质传感器信息的融合,即对多传感器时间序列数据进行有效的异常检测。但是如何提高数据融合结果的准确度和稳定性一直是个问题。为了能够实现准确的多源异质数据融合,提高粮库监控的智能化,以自注意力机制为切入口,提出了一种基于无监督的Transformer深度学习模型来进行粮库安全信息融合,该模型能够广泛适用于不同的粮库安全监测场景。实验结果表明,本文基于自注意力机制的信息融合模型具有更好的粮情监控能力,能够及时有效的发现粮库的存储问题,保证储粮安全。

Grain situation monitoring and environmental control are of great significance to ensure the safety of grain depots.However,due to the complex environment of grain depots,it is very difficult for relevant staff to monitor them in real time.This leads to the need for fusion analysis of multi-source data in grain depots,especially,the fusion of heterogeneous sensor information,that is,effective anomaly detection for multi-sensor time series data.However,how to improve the accuracy and stability of data fusion results has always been a problem.In order to achieve accurate multi-source heterogeneous data fusion and improve the intelligence of grain depot monitoring,in this paper,the self-attention mechanism was taken as the entry point,and an unsupervised Transformer deep learning model was proposed for grain depot security information fusion.The model could be widely applied to different grain depot security monitoring scenarios.The experimental results indicated that the information fusion model based on the self-attention mechanism in this paper had a better monitoring ability of grain situation,which could timely and effectively discover the storage problems of grain depots and ensure the safety of grain storage.