Anomaly Detection and Analysis Based on Deep Autoencoder for the Storage Tank Liquid Level of Oil Depot

Abstract

In the industrial control system of oil depot, the change of storage tank liquid level is closely related to the transportation process and site management of oil depot. It is of great significance to detect the abnormality of storage tank liquid level data for the safe and stable production of oil depot. A data-driven anomaly detection strategy is proposed by analyzing the non-periodic time series data of the oil storage tank liquid level. Based on the convolutional autoencoder algorithm to learn the features and patterns of a large number of samples, the strategy is carried out by reconstructing the samples and calculating the reconstruction error, which not only does not rely on the labeled samples, but also improves the detection precision. This paper chooses three algorithms of convolutional autoencoder, RNN (Recurrent Neural Network) autoencoder and LSTM (Long Short-Term memory) autoencoder for experimental analysis. Experiments were carried out on the oil tank historical data set and NAB (Numenta Anomaly Benchmark) simulation data set respectively. The results show that the accuracy of convolutional autoencoder is 98% and the F1 score is 82%, which is more practical for the scenes with real-time requirements.