CNN-BiLSTM based GAN for Anamoly Detection from Multivariate Time Series Data

Continuous recording of sensor data for monitoring applications does require detection of data patterns which deviate from normal condition. Detection of such events is necessary for implementing early preventive methods to improve overall system performance and potentially identify sensor failure causes. Recently, deep learning techniques based on generative models such as generative adversarial network (GAN) are proposed for anomaly detection from multiple time-series data. In this research, a variant deep learning method-based GAN is proposed for anomaly detection from multivariate time series data. Based on our proposed approach, the generator block consists of both CNN and BiLSTM blocks whilst the discriminator uses BiLSTM. To evaluate the performance of our new approach, multiple recordings from soil moisture measurement system are used to compare our proposed framework to the state-of-the-art techniques. Our proposed CNN-BiLSTM based GAN model presents an improved performance for the soil moisture recordings.