Real-time anomaly detection system for grain storage

Abstract

The invention discloses a real-time anomaly detection system for grain storage, and the system comprises the steps: firstly collecting historical data collected by each sensor in a grain storage process, and carrying out the preprocessing; then a neural network model used by a real-time anomaly detection system is built, and auto-encoder training and adversarial training are carried out on the model; inputting the preprocessed to-be-detected data into the neural network model, outputting a reconstructed time sequence, and performing interval adjustment on the abnormal score according to the reconstructed time sequence; and finally, further carrying out point adjustment on the obtained abnormal score, marking an abnormal point, and giving an alarm in real time. According to the method, an auto-encoder system structure formed in a two-stage adversarial training framework is adopted, on one hand, whether input data contains abnormal data or not can be recognized through training, the inherent limitation of an auto-encoder is overcome, and therefore good reconstruction is executed; and on the other hand, the AE architecture system has obtaining stability during adversarial training, and the problems of collapse and non-convergence modes encountered by the GAN model are solved.

Description

technical field [0001] The invention relates to the technical field of real-time abnormality detection, and mainly relates to a real-time abnormality detection system for grain storage. Background technique [0002] Given the size and complexity of food storage, the number of sensors used to infer normal and abnormal behavior has increased dramatically over time, making traditional expert-based methods of supervision slow or error-prone. [0003] The autoencoder architecture enables it to learn in an unsupervised manner. The use of adversarial training and its architecture allow it to isolate anomalies while providing fast training. Therefore, the present invention designs a real-time anomaly detection system based on grain storage on the basis of the above technology. Better early warning of possible anomalies in the grain storage environment. SUMMARY OF THE INVENTION [0004] Purpose of the invention: In view of the problems existing in the above background technology...

AI Technical Summary

Problems solved by technology

[0002] Given the size and complexity of grain storage, the number of sensors used to infer normal and abnormal behavior has increased dramatically over time, making traditional expert-based oversight methods slow or error-prone

Images