Design and Application Research of a Smart Saltern System Based on Ultra-Lightweight AI

The salt industry, as a vital sector within mining, faces significant challenges in achieving efficient and automated production due to its harsh operational environment. This paper presents the design and application of a Smart Saltern System grounded in ultra-lightweight Artificial Intelligence (AI) to address these challenges. The system employs a grid-based deployment of 10 automatic weather stations (AWS) for multidimensional data acquisition and integrates three core ultralightweight AI modules: Meteorological AI, Plastic Film Management AI, and Aquaculture AI. These modules, leveraging lightweight Long Short-Term Memory (LSTM), Decision Tree, and Random Forest models deployed on edge devices, enable intelligent decision-making for meteorological prediction, automated plastic film control, and water quality anomaly detection. A central Salt Crystal Intelligent Engine Center ensures lowlatency data transmission and command distribution via RESTful API/MQTT, complying with China's <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3 ~\text{km} \times 3 ~\text{km}$</tex> grid forecast standards (GB/T 33703-2017). Experimental validation through simulations and a controlled pilot demonstrates that the system achieves rapid response times (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;45 ~\mathrm{s}$</tex> for plastic film control, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&lt;105 ~\mathrm{s}$</tex> for aquaculture alerts) and high predictive accuracy (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$&gt;85$</tex> % for rainfall prediction, F1-score of 0.87 for anomaly detection). Results indicate a potential reduction of production losses by 12 % and operational costs by approximately 30 %. This system provides a scalable, efficient, and empirically validated paradigm for intelligent management in the mining sector, particularly in resource extraction environments like salterns.