Output prediction of alpha-type Stirling engines using gradient boosted regression trees and corresponding heat recovery system optimization based on improved NSGA-II
Output prediction of alpha-type Stirling engines using gradient boosted regression trees and corresponding heat recovery system optimization based on improved NSGA-II
复制标题
DOI：
10.1016/j.egyr.2022.02.244
复制
发表时间：
2022
期刊：
影响因子：
5.2
通讯作者：
Jiying Chen;Zedong Chu;Rui Zhao;Alexander F. Luo;K. Luo
中科院分区：
文献类型：
--
作者：
Jiying Chen;Zedong Chu;Rui Zhao;Alexander F. Luo;K. Luo
气候变化正成为一个紧迫的全球问题，寻找新能源和能源回收技术正成为全球研究的当务之急。斯特林发动机对各种热源的广泛适应性使其成为工业废热回收和太阳能发电的一项有前途的技术。斯特林发动机的运行是一个多物理耦合的传热和力学过程，以及机械摩擦和气体泄漏造成的非线性损失。因此，通过理论分析准确预测斯特林发动机的输出功率是复杂和昂贵的。像梯度增强回归树(GBRT)这样的新兴机器学习算法可以为解决这一问题提供新的方法。GBRT模型由多个决策树组成，这些决策树通过穷尽研究中的所有特征的阈值来分支，以找到数据回归的最佳分裂结构，而GBRT的原理赋予了它找到广泛的区分特征和组合的天然优势，以及强大的泛化能力。在此基础上，建立了阿尔法型斯特林发动机输出功率的GBRT预测模型。使用来自通用汽车4L23斯特林发动机的测试数据作为训练和测试集。对占总样本25%的随机测试集的结果表明，GBRT模型的预测准确率为96.23%。在此基础上，构建了一个由斯特林发动机、光伏电池板和用于工业废热回收的电池组成的区域微电网，并建立了能源供应性能评估体系。最后，基于所提出的功率输出模型，对改进的NSGA-II算法进行了多目标优化，为斯特林发动机的工业化应用提供了指导。
Climate change is becoming a pressing global concern, and the search for new energy and energy recovery technologies is becoming a worldwide research imperative. The broad adaptability of the Stirling engine to a wide variety of heat sources makes it a promising technology for industrial waste heat recovery and solar thermal generation. The operation of the Stirling engine involves a multi-physical coupled process of heat transfer and mechanics as well as non-linear losses due to mechanical friction and gas charge leaking. Therefore, accurate prediction of Stirling engine power output through theoretical analysis is complex and costly. Emerging machine learning algorithms like Gradient Boosted Regression Trees (GBRT) can offer new approaches to solve this problem. The GBRT model consists of multiple decision trees that branch by exhausting thresholds for all features under study to find the best split structure for data regression, and the principle of GBRT gives it the natural advantage of finding a wide range of distinguishing features and combinations, and a powerful generalization capability. A GBRT forecasting model is thus constructed to model the output power of Alpha-type Stirling engines. Test data from the General Motors 4L23 Stirling Engine are applied as the training and test set. Results from the random test set accounting for 25% of the total samples indicate that the GBRT model has a prediction accuracy of 96.23%. Furthermore, a regional microgrid containing Stirling engines, photovoltaic panels and batteries for industrial waste heat recovery is constructed and an evaluation system for energy supply performance is also established. Finally, based on the proposed power output model, multi-objective optimization based on improved NSGA-II is implemented, providing guidance for industrial application of Stirling engines.