Methods for optimizing logistics chains in construction using artificial intelligence algorithms to minimize downtime and increase labor productivity at civil facilities

Abstract

The study develops and validates a comprehensive method for optimizing construction logistics chains using artificial intelligence algorithms to minimize non-productive downtime and raise labor productivity; its relevance is driven by the high share of time losses due to supply disruptions, limited controllability under reactive approaches, and low predictability of deliveries. The aim is to build a predictive-adaptive system for managing flows of materials, equipment, and crews and to quantify its impact under real site conditions. Materials and methods: the empirical base comprises 12 civil projects(residential and social infrastructure) in the Central Federal District(2020-2023), split into control and experimental groups; data sources include ERP (requests/deliveries), GPS/GLONASS telematics, equipment operation logs, video surveillance, and production department reports; modeling was performed in AnyLogic(agent-based and discrete-event), with statistical verification via ANOVA and Student’s t-test(SPSS). The algorithmic core employs LSTM for forecasting demand and delivery lead times, genetic algorithms for routing and equipment allocation, and reinforcement learning for dynamic resource scheduling; integrated metrics include on-time delivery share, downtime structure, logistics costs, and productivity. Results: total downtime fell from 28,65 to 12,83%(-15.82 percentage points), including waiting for materials from 12,83 to 2,19%; the mean transport arrival deviation decreased from 112,45 to 14,71 minutes, and the share of deliveries within a ±15-minute window rose from 18,36 to 89,52; logistics costs declined by 36,50% owing to reductions in storage costs(-70.96%) and penalties for transport idle time(-93,83%); aggregate productivity increased by 24,88%(concrete works+28,11%, labor input for masonry-22,34%, slab installation-24,35%). A strong association was found between delivery discipline and productivity (R=0.91;+0,35% productivity for each percentage point of on-time delivery). Conclusions: the proposed system shifts logistics into a predictive, controllable mode and delivers a robust economic effect; future work includes integration with BIM/digital twins, expansion of RL policies, and scaling across a portfolio of projects subject to data quality and regulatory constraints.