A Data-Driven Method for Diagnosing ATS Architecture by Anomaly Detection

Autonomous Transport System (ATS) architectures enable a wide range of new applications and bring significant benefits to transport systems. However, during the design stage, errors of the architecture can have an impact on the smooth implementation of the ATS, which will endanger the normal operation of the transport systems. To ensure a high autonomy of the ATS architecture, i.e., “functionally evolvable, logically reconfigurable and physically configurable”, the detection of ATS architecture design errors is essential. This paper aims to fill the research gap in the existing research on diagnosing or evaluating ATS architectures. Inspired by word embedding models in natural language processing communities, we propose a data-driven approach to diagnose ATS architectures without prior knowledge or rules. We use an architecture embedding model to generate vector representations of ATS architectures, then train the model through negative sampling of the training dataset to identify the features of abnormal ATS architecture. Finally, we employ the trained model to classify structural errors of the test dataset generated from the ATS architecture. The experimental results show that the proposed method gains a relatively good effect of classifying with an average accuracy of 79.3%, demonstrating the effectiveness of the method.