Pressure Equipment Status Data Cleaning Method Based on LSTM-VAE-GAIN

Improving status data quality for pressure equipment is crucial for efficiency and fault prediction. Traditional cleansing methods are ineffective due to reliance on complete datasets and single-pattern techniques. To tackle this, this paper proposes an advanced LSTM-VAE-GAIN method. Initially, the kernel function-based local outlier factor (KLOF) algorithm detects anomalies in incomplete datasets, marking them. Then, Long Short-Term Memory (LSTM) and Generative Adversarial Imputation Networks (GAIN) are combined to train and impute missing data using anomaly labels. Additionally, LSTM and Variational Auto-Encoders (VAE) detect anomalies, correcting them with the imputation model. Experimental results show effectiveness in imputing missing data and identifying anomalies, enhancing pressure equipment status monitoring data quality.