Precision scenario-matched modeling and optimized construction of digital-intelligent coal preparation systems
选煤数智化系统精准对口建模与优化构建研究

摘要

针对选煤数智化系统中算法模型与业务场景匹配不足、统一模型泛化套用导致控制效果不稳定的问题，提出“场景任务-数据画像-模型对口-指标验模-部署闭环”的精准对口建模方法。按设备诊断、过程优化、质量预测、故障分析和工艺控制任务，分别匹配神经网络、多元线性回归、支持向量机、聚类分析及专家规则/比例‑积分‑微分（PID）等模型，建立离线验证、工业验证和在线迭代机制。工程应用结果表明，系统故障诊断准确率达93%，精煤回收率提升1.1%，介耗降低16%，日均停机时间减少0.55 h，年直接经济效益341.9万元。研究表明，按场景约束选择模型并持续优化，可提高模型有效性、控制适配性和工程可复制性。

Abstract

To address the mismatch between algorithmic models and business scenarios in digital-intelligent coal preparation systems, a precision scenario-matched modeling method consisting of scenario-task definition, data profiling, model matching, metric-based validation and closed-loop deployment is proposed. For an engineering smart coal-preparation project, neural networks, multiple linear regression, support vector machines, clustering, expert rules and Proportional‑Integral‑Derivative (PID) control are matched respectively to equipment diagnosis, process optimization, quality prediction, fault analysis and process control tasks. Offline validation, industrial verification and online iterative optimization are integrated into the model lifecycle. Engineering application achieved 93% fault-diagnosis accuracy, a 1.1% increase in clean-coal recovery, a 16% reduction in medium consumption, a 0.55 h/day reduction in downtime and annual direct economic benefits of CNY 3.419 million. The results indicate that scenario-constrained model selection and continuous optimization can improve model effectiveness, control adaptability and engineering replicability.