In response to the increasing challenges of Cyber Electromagnetic Activities (CEMA) in urban settings, characterized by dense electromagnetic (EM) signals and rising data traffic, this research introduces an Agent-Based Model (ABM) aimed at prioritizing critical signals. The primary goal of this research is to deploy a Unmanned Aerial Vehicle (UAV) swarm operating to selectively interfere with communication lines in a CEMA environment. The research goal is segmented into a multi-stage approach, focusing on the following mission tasks for the UAV swarm: i) Transmitter Search, ii) Communication Line Search, iii) Communication Line Mapping, and iv) Interference Point Finding. This research proposes and evaluates various methodologies for these task. A methodological contribution is the development of the Heuristicdriven Utility by Regression-based Metrics Synthesis (HURMS) framework. This framework addresses the subswarming coalition formation problem in Multi Transmitter Search. The HURMS framework utilizes the benchmarking Monte Carlo Tree Search (MCTS), a heuristic search method, to enhance the Maximum Utility Configuration (MUC), a transparent utility-based method, overcoming heuristic search limitations and complexities in creating utility functions. While the HURMS-enhanced MUC method effectively located all transmitters in the task, comparative analysis showed MCTS to be about 45% faster and 56-71% more successful in transmitter detection. This highlights potential areas for enhancing the MUC algorithm further under the guidance of the HURMS framework. Furthermore, the Particle Swarm Optimization (PSO) Search was utilized as a velocity controller in the context of Multi Transmitter Search methods, guiding the speed and direction of UAVs. Regarding Single Transmitter Search, a significant observation was that the PSO Search was approximately 38% faster than the novel Pair-gradient Search method in locating a single transmitter. The study also examines Communication Line Mapping by comparing two frontier-based methods with the Multi-agent Tabular Q-Learning method. The frontier-based methods provided better coverage, while the Tabular Q-Learning excelled in precision and adaptability for multi-agent mapping. Data from these methods were applied to find the interference point. Additionally, a Reinforcement Learning (RL)-trained agent was used for Interference Point Finding, proving to be faster but less accurate.