Integration of multi agent reinforcement learning with golden jackal optimization for predicting average localization...

Area of Science:

• Computer Science
• Network Engineering
• Artificial Intelligence

Background:

• Accurate sensor node localization is critical for Wireless Sensor Networks (WSNs) in applications like environmental monitoring and smart cities.
• Dynamic environmental conditions, varying network densities, and parameter interdependencies pose significant challenges to localization accuracy, increasing Average Localization Error (ALE).
• Existing methods often struggle with generalization in dynamic environments due to static parameter configurations or limitations in adapting to real-time network variations.

Purpose of the Study:

• To propose a novel Multi-Agent Reinforcement Learning (MARL) algorithm integrated with Golden Jackal Optimization (GJO) to improve WSN localization accuracy.
• To develop a framework that dynamically learns optimal parameter adjustments, minimizing localization error and its variability in dynamic network conditions.
• To enhance the generalization capabilities of MARL across diverse WSN configurations through GJO hyperparameter tuning.

Main Methods:

• Implementation of a Multi-Agent Reinforcement Learning (MARL) algorithm for dynamic parameter adjustment.
• Integration of Golden Jackal Optimization (GJO) to fine-tune MARL hyperparameters for improved generalization.
• Evaluation using a benchmark dataset and analysis of performance metrics including MSE, MAE, RMSE, R², and MAPE.

Main Results:

• The proposed optimized MARL framework demonstrated significant improvements in WSN localization accuracy.
• Achieved a Mean Squared Error (MSE) of 0.02, Mean Absolute Error (MAE) of 0.11, Root Mean Square Error (RMSE) of 0.14, R-squared (R²) of 0.88, and Mean Absolute Percentage Error (MAPE) of 2.5%.
• Outperformed existing methods like Grid Search RF, Bayesian Optimized RF, Gradient Boosting, and Deep Neural Networks in benchmark evaluations.

Conclusions:

• The novel MARL-GJO framework effectively addresses the challenges of dynamic WSN environments for accurate localization.
• The dynamic learning and optimization approach significantly enhances localization accuracy and adaptability compared to static or heuristic models.
• This research offers a robust solution for improving localization performance in critical WSN applications.