ABNORMALITY DETECTION ALGORITHM OF HORIZONTAL DISPLACEMENT MONITORING DATA DURING FOUNDATION PIT EXCAVATION BASED ON TEMPORAL-SPATIAL CHARACTERISTICS

Abstract

Horizontal displacement of the enclosure structure or soil is one of the key indicators to evaluate the safety of deep foundation pit excavation. However, the monitoring data of horizontal displacement will have short- or long-term abnormal fluctuations due to the interference of various non-structural factors, which would directly affect the assessment and identification of the risk of deep foundation pit excavation. In order to avoid missing and misjudging the true large deformation of the foundation pit and consider the requirements for the accuracy and timeliness of the monitoring data, an abnormal data identification and correction algorithm was proposed in this paper considering the temporal-spatial characteristics of the horizontal displacement monitoring data. The algorithm builds a spatial distribution matrix according to the deployment position relationship of multiple sensors, records the historical measured values of sensors in a period of time, and calculates the estimated values of the measured values of sensors at the next time based on multidimensional Kalman filtering. After obtaining the latest horizontal displacement monitoring value at the current time, calculate the difference between the measured value and the estimated value based on the spatial matrix to correct the data, mark the sensor with the difference greater than the adaptive threshold, and finally update the historical measured value, adaptive threshold, Kalman filter parameters, etc. in the window, waiting for the input of the monitoring data at the next time. The abnormal data identification and correction algorithm described in this paper was successfully applied in a deep foundation pit excavation project in Shanghai. The anomaly identification and correction of single sensor data and the horizontal displacement curve were both given. In general, this method can simultaneously discriminate and correct the data of multiple sensors in real time, effectively eliminate the abnormal fluctuation of monitoring data, and has good engineering application value.