| 张成伟,李慧霞,崔保华,张 焱,王 磊.多模型融合算法在水泥质量数据预测中的应用[J].水泥工程,2023,36(2):5-7 |
| 多模型融合算法在水泥质量数据预测中的应用 |
| Application of multi-model fusion algorithm in quality data prediction for cement |
| |
| DOI： |
| 中文关键词: 多模型融合,模型池,水泥质量,小样本分析,机器学习 |
| 英文关键词: multi-model fusion, model pooling, quality data in cement, small samples analysis, machine learning |
| 基金项目: |
|
| 摘要点击次数: 448 |
| 全文下载次数: 0 |
| 中文摘要: |
| 水泥质量作为水泥生产关键数据，通常存在数据缺失、滞后严重、干扰因素不确定、控制不及时、质量数据不稳定等问题，针对以上问题，一种基于多模型融合的在线预测算法可提前准确地预测水泥质量趋势。该算法首先采集水泥行业质量数据及其相关特征量，然后经过数据预处理（包括数据采样、对齐、异常值处理、平滑等）和特征筛选后得到所需数据。其次，在以上数据中抽取若干段连续的短时间范围数据，分别进行小样本数据分析建模，得到若干个模型存入到模型池，作为预测质量数据的模型来源。最后，根据当前运行工况，融合多个模型进行预测分析，将质量预测结果反馈到控制系统中。在多个水泥厂实施表明，该算法可以显著提高预测精度，尤其可以提前预测质量趋势，减少延时性，通过将质量预测结果引入到控制系统中，可以避免质量调整不及时导致的系统振荡和能源浪费问题，从而提高水泥产量和质量。 |
| 英文摘要: |
| Quality data is the master figures during cement production， but some problems such as key data missed， data acquisition delayed， interference factors uncertainly， controlling untimely and quality data unstable often existed. In response to the above problems， an online prediction algorithm based on multi-model fusion to predict quality data was proposed in this paper. Firstly，The quality data as object and related feature data are collected from cement industry， and the collected data will be preprocessed through data resampling， features alignment， outlier processing， data smoothing， etc. After data preprocessed， feature extracting is needed before model constructed. Next， the model pooling constructed by some unique model is created. The unique model established based on machine learning mainly supported by some short data samples. The model pooling will be used to predict quality data in cement. Finally， quality data will be predicted online according to the real operating conditions by the model pooling algorithm ， and the quality prediction results will be fed back to the control system. By implementation in multiple cement plants， it shows that the prediction accuracy was significantly improved. Especially the quality trend can be predicted in advance， and the time delay can be reduced. By introducing the quality prediction results into control system， system oscillations and energy waste caused by untimely quality adjustments can be avoided. Thereby the production and quality in cement industry can be increased significantly. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|