Anomaly Detection (AD) can play a key role in condition monitoring of wind turbines, aiming to detect faults as early as possible, improving their reliability and the efficiency of power generation, while reducing downtime and maintenance costs. Supervisory Control and Data Acquisition (SCADA) systems are widely used, capturing large amounts of status data. Effectively extracting useful information from these massive data is critical. At present, AD methods are mainly divided into shallow kernel and deep-based methods. Deep AD methods show more promising results when applied on large datasets. Typically, AD methods are based on a large number of unlabeled data. However, sometimes in addition to the large number of unlabeled samples, a small number of labeled samples is also available, which can be used to build an improved model. In this paper a semi-supervised AD method based on Convolutional Neural Networks (CNN) and Support Vector Data Description (SVDD) is proposed to automatically monitor the status of wind turbines. The proposed method is validated on a wind turbines’ SCADA dataset, including ice blade events. Extensive experimental results demonstrate that the proposed method achieve better AD performance compared to a state-of-the-art unsupervised Deep SVDD AD method.