Predicting pit deformation anomalies based on HTM algorithm analysis

Hierarchical Temporal Memory (HTM), an unsupervised online algorithm, has been validated in multiple fields of machine learning, especially prediction and anomaly detection, which can well reduce the false alarm rate by analyzing from multiple dimensions. For instance, traffic flow prediction, ocean observation data anomaly detection, etc. However, the analysis and prediction based on the real-time monitoring data of technical engineering have not been addressed. In this study, a real-time prediction method of foundation pit deformation based on HTM is proposed to monitor the sensor data of different depths in real time, predict the horizontal displacement and vertical settlement, set the abnormal range according to the requirements of the monitoring accuracy level, and calculate the measured and predicted values. Combining the measured and predicted values by cloud computation, a signal value is returned to determine whether it is anomalous or not. The error reporting rate is set according to the number of anomalies. The results show that the HTM can be used to analyze and predict the pit deformation monitoring data effectively and reliably.