To address the vulnerability of license plate recognition (LPR) data in intelligent transportation systems (ITS) to implicit errors, such as character misrecognition and missed vehicle captures, and the lack of methods for proactively identifying these errors, this study aims to develop a detection method that can proactively and accurately identify LPR devices exhibiting abnormal behavior, thus improving overall data quality and providing decision support for equipment operation and maintenance. First, topological neighborhood relationships between devices were defined using path data derived from vehicle trip chains connecting consecutive LPR checkpoints in the road network, and multidimensional neighbor entropy feature indicators, including total entropy, upstream and downstream neighbor entropy, and asymmetric entropy, were constructed accordingly. Second, a candidate set of abnormal devices suspected of misrecognition or missed captures was initially identified by detecting distortions in neighborhood distributions and fluctuations in vehicle counts caused by implicit errors. Then, spatiotemporal constraints were introduced for cross-verification. Misrecognition devices were identified using physical speed limits, while missed capture devices were identified by incorporating road network topology and frequency-direction consistency constraints. Finally, anomaly confidence scores were calculated for each LPR device, and the final sets of misrecognition and missed capture devices were obtained. Multi-scenario simulation experiments conducted on an actual urban road network show that, across various complex scenarios with different misrecognition and missed capture probabilities and abnormal device counts, the precision of missed capture device identification exceeds 0.8, the recall of misrecognition device identification reaches 1.0, and stable and satisfactory identification performance is maintained under mixed anomaly conditions where both errors coexist.