Research on Anomaly Detection of Offshore Buoy Water Temperature Data Based on LSTM-DAE Model

At present, there are many quality problems in the hydrological data of ocean buoys, and it is urgent to carry out scientific and standardized data quality control to ensure the accuracy and rationality of the buoy data. This paper focuses on water temperature data collected by offshore buoys, and an LSTM-DAE abnormal value detection model is proposed. In this model, both the encoder and decoder of the denoising autoencoder adopt a long short-term memory network (LSTM) as the core component of the neural network layer. Experimental results show that compared with the traditional LSTM model for outlier detection, the proposed model can better deal with the long-term series characteristics in the buoy data and reconstruct the input data more accurately, so as to improve the effect of outlier detection.