Quality monitoring of real-time PPP service using isolation forest-based residual anomaly detection

In order to meet the time-critical and high-precision positioning demands of massive market applications, real-time Precise Point Positioning (PPP) technology has been continuously developed and can achieve a high level of accuracy now. Beyond precision, it is also crucial to ensure the security and stability of real-time high-precision positioning services. Nevertheless, there are few studies that addressed the fault detection and exclusion (FDE) of real-time PPP correction service to protect users from potential faults in the GNSS satellites. This paper introduces a new real-time products quality monitoring method that could effectively identify the potential fault of GNSS satellite corrections using an unsupervised learning algorithm and provide means to alert the real-time PPP users. The ambiguity-fixed Un-differenced Carrier-phase Residual Statistics (UCRS) of large-scale regional stations are first constructed to reflect the status of satellite corrections accurately. A machine learning technique, known as Isolation Forest, is employed to identify outliers in the UCRS to detect situations of potential satellite faults. Then, the UCRS alarm factors are transmitted to users for PPP processing with a modified weighting scheme based on alert information. Experimental validation utilizing 30 monitoring stations in China demonstrates a detection success rate exceeding 95% for orbit faults larger than 5 cm and clock faults larger than 0.2 ns. It is also proved that this method can effectively identify orbit and clock jump in real-time GNSS products that cause additional positioning errors. With the alert information broadcasted by the server, the PPP after FDE (PPP-FDE) presents a significant accuracy improvement of 27–71% compared with traditional PPP processing.