AI-Assisted Hybrid PSO–GWO Optimization with Deep Reinforcement Learning Control and LSTM Forecasting for Energy-Aware Load Balancing in Heterogeneous Distributed Systems
ID：125
View protection：Participant Only
Updated time：2026-10-05 20:32:10 Views：38
Online
Abstract
Cloud data centers, edge nodes, and fog infrastructures have made distributed computing environments highly heterogeneous, and static load-balancing policies cannot keep pace with their dynamic workloads. This paper introduces AHMHO, an AI-Assisted Hybrid Meta-Heuristic Optimization framework that combines a hybrid Particle Swarm Optimization–Grey Wolf Optimizer (PSO–GWO) search with a Deep Reinforcement Learning (DRL) controller and an LSTM workload predictor. The DRL controller continuously tunes the search hyperparameters and the objective weights from environmental feedback, and the LSTM forecasts load so that scheduling is proactive. Simulations on CloudSim Plus 8.0 with Google cluster workload traces compared AHMHO with five baselines: First-Come-First-Served (FCFS), Round Robin, standalone PSO, standalone GWO, and Ant Colony Optimization (ACO). Over 30 runs, AHMHO reduced makespan by 9.2% and energy consumption by 13.1% relative to the strongest baseline (GWO), and by 34.7% and 41.1% relative to FCFS, while raising resource utilization from 70.2% to 88.5% and cutting SLA violations from 7.1% to 3.2%. The gains are statistically significant (Wilcoxon signed-rank test, p < 0.01). The per-cycle scheduling time of AHMHO (203.8 ms) is 17% higher than that of GWO but remains compatible with a 500 ms scheduling interval
Keywords
Load balancing, hybrid meta-heuristics, deep reinforcement learning, particle swarm optimization, grey wolf optimizer, LSTM workload forecasting, energy-aware scheduling, heterogeneous distributed computing, cloud computing
Post comments