Current implementation. The current implementation of the Pipeline Service consists of the following implemented features. • A fully defined webservice interface of the Pipeline Service. • Implemented translation component that supports translations for both LONI Pipeline, and translation to JDL for gLite Submission. • Enactor that uses an extended gLite-adaptor for the Glueing Service for submission to the grid. Each implemented feature will be described in detail in the subsequent sections.
Current implementation. In this section, we evaluate the effectiveness of the RL-based approach applied for V2X misbehaviour detection by performing experiments using the VeReMi dataset. The VeReMi dataset includes 19 misbehaviour attack types and models two road traffic densities under each attack scenario: high- density (37.03 vehicles per square km) and low-density (16.36 vehicles per square km). For each attack scenario, a log file per vehicle is generated, which contains information transmitted by neighbouring vehicles over its entire trajectory. Each scenario contains a ground truth file to record the observed behaviour of all participating vehicles. The exchanged messages constitute a three-dimensional vector for position, speed, acceleration and heading angle features. The proportion between misbehaving and genuine vehicles is set to 30% - 70%, respectively, for all the simulations. Table 1 depicts the results of the RL-based detection per attack type. Type Attack Accuracy Precision Recall F1-score 1 Constant Position 0.9868 0.9588 0.9984 0.9782 2 Constant Position Offset 0.9981 0.9629 0.9982 0.9803 3 Random Position 0.9886 0.9642 0.9985 0.9810 4 Random Position Offset 0.9886 0.9632 1 0.9812 5 Constant Speed 0.9988 0.9968 0.9995 0.9982 6 Constant Speed Offset 0.9923 0.9766 0.9978 0.9871 7 Random Speed 0.9985 0.9987 0.9963 0.9975 8 Random Speed Offset 0.9915 0.9774 0.9945 0.9858 9 Sudden Stop 0.9811 0.9274 1 0.9623 10 Disruptive 0.9896 0.9664 1 0.9829 11 Data Replay 0.9894 0.9656 0.9999 0.9825 12 Delayed Messages 0.9666 0.9012 0.9976 0.9470 13 DoS 0.9999 0.9999 1 0.9999 14 DoS Random 0.9997 0.9996 1 0.9998 15 DoS Disruptive 0.9991 0.9984 1 0.9992
Current implementation. In this section, we evaluate the effectiveness of the RL-based approach applied for V2X misbehaviour detection by performing experiments using the VeReMi dataset. The VeReMi dataset includes 19 misbehaviour attack types and models two road traffic densities under each attack scenario: high- density (37.03 vehicles per square km) and low-density (16.36 vehicles per square km). For each attack scenario, a log file per vehicle is generated, which contains information transmitted by neighbouring vehicles over its entire trajectory. Each scenario contains a ground truth file to record the observed behaviour of all participating vehicles. The exchanged messages constitute a three-dimensional vector for position, speed, acceleration and heading angle features. The proportion between misbehaving and genuine vehicles is set to 30% - 70%, respectively, for all the simulations. Table 1 depicts the results of the RL-based detection per attack type.
