LSTM-Based Anomaly Detection in Manufacturing Environmental Monitoring Data

Abstract

With the proliferation of environmental monitoring data, using machine learning techniques for anomaly detection in environmental time series data has become an active research direction. This study employs Long Short-Term Memory (LSTM) neural network models to detect anomalies in manufacturing emission data. The research first preprocesses the data by handling missing values and conducting stationarity tests. The data will be divided into training and testing sets, with the model trained on normal data and tested for anomalies. Experiments show LSTM outperforms classic methods like Isolation Forest, Matrix Profile, and AutoEncoder in handling enclosed pipeline emission data. This study is primarily due to LSTM's ability to capture long-term dependencies in time series data. Establishing this model facilitates improved environmental protection and safety management, enables automated monitoring and warning, reduces manual intervention, and lowers enterprise environmental compliance risks. This study provides an effective anomaly detection model for monitoring manufacturing emissions, serving as a reliable reference for introducing machine learning into environmental monitoring domains.