AI-Assisted Optimization of Software Energy Consumption

Abstract

The escalating energy footprint of software systems demands urgent intervention, particularly in widely deployed platforms like WordPress, which powers 43.5% of global websites. This thesis develops an AI-assisted methodology to optimize energy consumption in WordPress’s PHP backend, emphasizing rigorous empirical validation. A controlled lab environment (Odroid H3+ rig with PowerGoblin instrumentation) enabled 100 Hz sample-rate energy profiling of computational inefficiencies, while a Random Forest classifier prioritized optimization targets based on execution time, energy, and power metrics. Heuristic-guided refactoring generated context-specific optimizations, such as replacing nested loops with hash-map logic and implementing transient caching. Experimental results demonstrate 99.44% energy reduction for algorithmic bottlenecks (e.g. loop patterns) and statistically significant improvements in memory-intensive operations. The methodology validates AI-driven prioritization as a robust framework for identifying and mitigating energy hotspots in PHP-based systems, advancing sustainable software engineering practices.