Real-time optimization solutions in V2X-based systems for internet of vehicles applications

In this thesis, real-time optimization solutions are developed in V2X based systems for internet of vehicles (IoV) applications and innovative machine learning methods are combined to optimize speed and acceleration estimation and decision-making processes in autonomous vehicles (AVs). The proposed methods include supervised learning-based Chaotic AdaBoost (CAB) and reinforcement learning-based Lévy Flight-enhanced Proximal Policy Optimization (LFPPO). In the comparison of the algorithms, k-Nearest Neighbors (kNN), Artificial Neural Networks (ANN), AdaBoost (AB), Gradient Boosting (GB), Random Forest (RF) and Proximal Policy Optimization (PPO) algorithms were used. In the experiments conducted in Town 10 and Town 5 environments of the CARLA simulator, CAB achieved 99.3% success in acceleration estimation with 0.018 MSE, 0.020 MAE and 0.993 R²; This showed superiority compared to kNN (23.215 MSE, 0.87 R²), ANN (3.297 MSE, 0.982 R²), RF (4.419 MSE, 0.975 R²), GB (1.701 MSE, 0.991 R²) and AB (0.15 MSE, 0.985 R²). LFPPO offered a safer driving profile with a 99% success rate, 42.135 reward level, and 1% collision rate compared to PPO (81% success, 33.895 rewards, 19% collisions). Apache Kafka and MongoDB infrastructure supported the data processing processes; Kafka enabled low-latency prediction processes by providing real-time data streams under the carla-data title in synchronous and asynchronous modes, while MongoDB provided reliable storage and fast access to data. Synchronous and asynchronous prediction processes were implemented using Python script files. CAB and ANN demonstrated superior performance on complex datasets, achieving high accuracy against sensor uncertainties in asynchronous mode (CAB: MSE=0.018, R²=0.993), while kNN and RF provided fast predictions in synchronous mode. This study provides a theoretical contribution to the integration of chaos theory into machine learning and improves the reliability of AVs in urban transportation and logistics applications.