发电技术 ›› 2025, Vol. 46 ›› Issue (1): 135-144.DOI: 10.12096/j.2096-4528.pgt.23005

• 发电及环境保护 • 上一篇    

基于K-Means聚类与Ridge回归算法的煤炭发热量预测研究

乔世超, 陈衡, 李博, 潘佩媛, 徐钢   

  1. 热电生产过程污染物监测与控制北京市重点实验室(华北电力大学),北京市 昌平区 102206
  • 收稿日期:2023-11-15 修回日期:2024-03-09 出版日期:2025-02-28 发布日期:2025-02-27
  • 通讯作者: 陈衡
  • 作者简介:乔世超(2000),男,硕士研究生,研究方向为电站大数据优化,Gregory_Qiao@outlook.com
    陈衡(1989),男,博士,副教授,研究方向为热力系统优化,本文通信作者,heng@ncepu.edu.cn
  • 基金资助:
    国家自然科学基金青年项目(52106008);国家自然科学基金创新研究群体项目(51821004)

Research on Coal Calorific Value Prediction Based on K-Means Clustering and Ridge Regression Algorithm

Shichao QIAO, Heng CHEN, Bo LI, Peiyuan PAN, Gang XU   

  1. Beijing Key Laboratory of Emission Surveillance and Control for Thermal Power Generation (North China Electric Power University), Changping District, Beijing 102206, China
  • Received:2023-11-15 Revised:2024-03-09 Published:2025-02-28 Online:2025-02-27
  • Contact: Heng CHEN
  • Supported by:
    National Natural Science Foundation of China(52106008);National Natural Science Foundation of China Innovative Research Group Project(51821004)

摘要:

目的 煤炭发热量是衡量煤质的重要评价标准之一,也是动力用煤计价的主要依据。为了能够在降低预测成本的前提下实现对煤炭发热量的高精度快速预测,提出了一种新的预测方法。 方法 采用K-Means聚类算法对相似煤种进行聚类,样本数据来源于山东某电厂自备煤场近6年的4 269条入场化验信息。在聚类的基础上,分别建立工业分析数据低位发热量的Ridge回归模型,以此作为煤炭发热量的预测模型。 结果 所建立的K-Means聚类与Ridge回归混合模型在预测效果上表现出色。与传统的多元线性回归模型相比,该混合模型可将平均绝对值误差最高减少30.525%,均方根误差最高降低60.054%,相关系数最高提高2.320%。 结论 K-Means聚类与Ridge回归混合模型不仅降低了煤炭发热量的预测成本,还提高了预测的精度和速度,为煤炭发热量的预测提供了一种新思路。

关键词: 燃煤电厂, 煤炭发热量, 工业分析, 预测, K-Means聚类, Ridge回归

Abstract:

Objectives The calorific value of coal is one of the important evaluation criteria for measuring coal quality, and is also the main basis for valuation of power coal. In order to realize high-precision and rapid prediction of coal calorific value while reducing prediction costs, a new prediction method is proposed. Methods The K-Means clustering algorithm is used to cluster similar coal types. The sample data comes from 4 269 entry testing information from a self-contained coal yard of a power plant in Shandong in the past 6 years. On the basis of clustering, Ridge regression models are established from industrial analysis data to received base calorific value, which is used as a prediction model for coal calorific value. Results The established K-Means clustering and Ridge regression hybrid model exhibits excellent prediction performance. Compared with the traditional multiple linear regression model, this hybrid model can reduce the mean absolute error by up to 30.525%, reduce the root mean square error by up to 60.054%, and increase the correlation coefficient by up to 2.320%. Conclusions The mixed model of K-Means clustering and Ridge regression reduces the prediction cost of coal calorific value, and also improves the accuracy and speed of prediction, providing a new idea for predicting coal calorific value.

Key words: coal-fired power plants, calorific value of coal, proximate analysis, prediction, K-Means clustering, Ridge regression

中图分类号: