Distributed computing systems underpin many large-scale infrastructures, from cloud platforms and scientific simulations to enterprise applications. Achieving high performance in heterogeneous systems remains difficult because workloads vary, resources contend, and nodes behave dynamically. Traditional load balancing assumes uniform node capacity and a steady task arrival rate, so it cannot adapt to asynchronous resource capacities or to bursty, unpredictable task arrivals. This paper proposes a Hybrid Meta-Heuristic Load Balancing Framework (HMLBF) to address these gaps. HMLBF combines an elitist backward bidirectional migration scheme built on Particle Swarm Optimization (PSO) and an Adaptive Genetic Algorithm (AGA), an Exponential Weighted Moving Average (EWMA) workload prediction module, and an explicit task migration cost model. It brings proactive load forecasting, adaptive optimization, and migration overhead into a single deployable framework, which sets it apart from existing hybrid approaches. The system continuously monitors node status, redistributes tasks when nodes fail, predicts the next set of tasks, and jointly optimizes makespan, resource utilization, and the Load Imbalance Index. Across a broad range of simulations in the CloudSim 4.0 toolkit, HMLBF outperforms five established baselines: Round Robin, Min-Min, Max-Min, standard PSO, and standard GA. It reduces makespan by up to 34.7%, raises resource utilization by 28.3%, and lowers the Load Imbalance Index by up to 41.2%. Ablation studies confirm the contribution of each component. The results demonstrate the practical usability and effectiveness of HMLBF for real-world cloud and edge computing deployments.

Distributed computing, load balancing, heterogeneous systems, particle swarm optimization, genetic algorithm, task scheduling, cloud computing, resource utilization, makespan optimization.