Abstract Lost circulation is one of the most common issues affecting drilling safety, known for its sudden occurrence. Traditional expert diagnosis methods heavily rely on expert experience, exhibiting a high degree of subjectivity and lag. Conventional machine learning approaches struggle to fully capture the spatial and temporal variations within the multidimensional data of lost circulation, leading to insufficient accuracy. To address the challenge, this paper proposes an early intelligent monitoring hybrid model for lost circulation by concatenating Convolutional Neural Network (CNN) and Long Short-Term Memory Neural Network (LSTM). Initially, sliding window method is employed to structure historical data of lost circulation into time series samples. Subsequently, these time series samples are input into one-dimensional CNN network to extract spatial feature vectors. Following this, the extracted feature vectors are input into LSTM network to uncover temporal feature information. Finally, the Softmax function is applied at the network’s output layer for classification. The proposed model is tested on real dataset of lost circulation. The results show that the performance of this hybrid model is superior to traditional machine learning models, with accuracy of 94%, missed alarm rate of 4%, and false alarm rate of 8% on the test set. Compared with expert diagnosis, hybrid model can detect lost circulation risk earlier and significantly improve monitoring timeliness. This study is of great significance for ensuring drilling safety and improving drilling efficiency.

Early Lost Circulation Monitoring Using a Hybrid CNN-LSTM Model