The Fuzzy Logic Congestion Detection (FLCD) algorithm is a recent proposal for congestion detection in IP networks which combines the good characteristics of both traditional Active Queue Management (AQM) algorithms and fuzzy logic based AQM algorithms. The Membership Functions (MFs) of the FLCD algorithm are designed using a Multi-Objective Particle Swarm Optimization (MOPSO) algorithm, in order to achieve optimal performance on all the major performance metrics of IP congestion control. The FLCD algorithm achieves better performance when compared to the basic Fuzzy Logic AQM and Random Explicit Marking (REM) algorithms. Since the optimization process is undertaken offline and is based on a single optimization script, the performance of the FLCD algorithm may not be optimal under different network conditions, due to the fact that the IP environment is characterized by dynamic traffic patterns. This paper proposes two online self-learning and organization structures that enable the FLCD algorithm to learn the system conditions and adjust the fuzzy rule base in accordance with prevailing conditions. The self-organized FLCD algorithm is compared with the unorganized FLCD, the basic Fuzzy Logic AQM and the Adaptive Random Early Detection (RED) algorithms using simulations with dynamic traffic patterns. Performance results show that the self-organized FLCD algorithm is more robust than the other algorithms. Compared to the unorganized FLCD, the new scheme improves the UDP traffic delay for short round trip times and also reduces packet loss rates. In terms of jitter, fairness and link utilization, it exhibits a similar performance to the unorganized FLCD algorithm.