A Big Data and AI-Driven Intelligent Analysis and Control System for Gas Station Operations

Abstract

The gas station industry in many emerging markets, such as Uzbekistan, is hampered by data silos and insufficient analytical capabilities, constraining operational efficiency and risk management. To address the growing need for lean management and intelligent decisionmaking, this paper presents a novel intelligent analysis and control system that integrates big data processing and artificial intelligence technologies. The system is architected using a layered, decoupled microservices framework, featuring a batch-stream integrated data processing
pipeline (based on Hadoop/Spark/Flink) and a multimodal AI analytics engine. Its core contribution is the proposal and implementation of a
comprehensive "data perception-intelligent analysis-decision support" closed-loop framework. This framework utilizes an LSTM-Attention
hybrid model for high-precision sales forecasting and a novel two-stage anomaly detection algorithm combining Deep Autoencoders and
Isolation Forest to identify complex fraudulent patterns. Experimental results on a real-world dataset from 77 gas stations show that the sales
forecasting model achieves a Mean Absolute Percentage Error (MAPE) below 5%, and the anomaly detection module attains an F1-Score of
92.5%. Deployed via a visual management cockpit and an intelligent early-warning center, the system enables a paradigm shift from traditional experience-based management to data-driven intelligent decision-making, significantly enhancing operational efficiency and risk prevention
capabilities. This study provides a reusable theoretical framework and practical paradigm for intelligent energy retail in environments with
underdeveloped data infrastructure.