| 摘要: |
| 摘要DOE-2模型被广泛应用于冷水机组仿真建模,如何根据有限传感器实测数据对某特定冷水机组DOE-2模型的参数进行可靠地辨识,并补偿模型误差,对于节能运行等场景具有重要意义。在实践中由于传感器不足且数据质量不高等问题, DOE-2模型参数的可靠辨识较为困难。因此,本文提出一种基于外部知识库的遗传算法和一种基于人工神经网络的方法分别对DOE-2模型进行参数辨识和误差补偿。结果表明:基于外部知识库的遗传算法可以有效降低DOE-2模型参数辨识时间,并显著提升DOE-2模型预测精度。误差补偿后的DOE-2模型的预测精度显著高于未作补偿的DOE-2模型,前者在预测冷冻水出口温度时的MAE、RMSE、MAPE和CV-RMSE分别降低36.49% ,46.00%,33.16%和45.73% ,R2提高25.75%。 |
| 关键词: 冷水机组建模 遗传算法 参数辨识 人工神经网络 误差补偿 |
| DOI： |
| 投稿时间：2020-09-27 修订日期：2020-11-23
|
| 基金项目:国家自然科学基金(51706197)资助项目。 |
|
| Genetic-Algorithm-Based Parameter Identification and Artificial-Neural-Network-Based Error Compensation for Chiller Model |
| Zhang lizhu,Zhang Chaobo,Chen Qi,Zhao Yang |
| (Institute of Refrigeration and Cryogenics ，Zhejiang University;Zhejiang Energy Group Co.，Ltd.) |
| Abstract: |
| Accurate chiller models are important for the energy conservation of chillers. The DOE-2 model is the most common chiller model. Parameter identification and error compensation are crucial for the development of an accurate DOE-2 model. However, the parameter identification of a DOE-2 model of an actual chiller is usually challenging because chillers are usually equipped with limited sensors and the quality of actual data is usually low. To address the above issues, a genetic algorithm based on an external knowledge base for parameter identification and artificial-neural-network-based (ANN-based) error compensation method are proposed. The results show that the proposed genetic algorithm can significantly reduce the computation load of parameter identification of the DOE-2 model. It can also significantly improve the accuracy of the DOE-2 model. Moreover, the accuracy of the DOE-2 model with the ANN-based error compensation is significantly higher than that of the DOE-2 model without error compensation. The MAE, RMSE, MAPE, and CV-RMSE of the model with error compensation in predicting the chilled water outlet temperature were reduced by 36.49%, 46.00%, 33.16%, and 45.73%, respectively, while R2 of the model with error compensation was increased by 25.75%. |
| Key words: chiller modeling genetic algorithm parameter identification artificial neural network error compensation |
