Adaptive corrected parameters algorithm applied in cooling load prediction based on black-box model: A case study for subway station

•The optimal input for cooling load prediction of case subway station is determined.•An adaptive corrected parameters algorithm applied to black-box model is proposed.•The prediction performance of models based on proposed algorithm is enhanced.•The proposed algorithm is better than the single model...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Energy and buildings Ročník 297; s. 113429
Hlavní autoři: Hu, Yuanyang, Qin, Luwen, Li, Shuhong, Li, Xiaohuan, Zhou, Runfa, Li, Yanjun, Sheng, Wei
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier B.V 15.10.2023
Témata:
Adaptive corrected parameters
Black-box model
Cooling load prediction
Input combination
Prediction accuracy
Prediction accuracy
Input combination
Black-box model
Cooling load prediction
Adaptive corrected parameters
ISSN:0378-7788
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:•The optimal input for cooling load prediction of case subway station is determined.•An adaptive corrected parameters algorithm applied to black-box model is proposed.•The prediction performance of models based on proposed algorithm is enhanced.•The proposed algorithm is better than the single model under different sample size. Black-box model is widely utilized for predicting building loads, serving as the foundation for enhancing the operational efficiency of air conditioning systems. However, the input combination is essential to its prediction accuracy and the optimal combination varies with changes in the building. Moreover, individual black-box models exhibit limited prediction accuracy, and there exist only a few methods to enhance the accuracy of any black-box model. In this study, the measured data including system variables, meteorological conditions, time, indoor parameters, and historical load are collected, fifteen kinds of combinations are compared to approach the optimal input combination for case station. Additionally, an algorithm, namely Adaptive Corrected Parameters Algorithm (ACPA) is proposed to enhance the prediction performance of the basic black-box model. The ACPA has the theoretical potential to be applied to any black-box model, enabling the development of a framework composed of basic models, this framework demonstrates enhanced prediction performance by adaptively determining the optimal correction times and hyperparameters for involved models. Three typical black-box models namely Backpropagation Neural Network (BPNN), Radial Basis Function Neural Network (RBFNN) and Random Forests (RF) are employed as basic models, the prediction performance of the three models is studied based on ACPA, considering different correction references and sample sizes. The results indicate that the optimal input combination for the case station includes time, historical load, and system categories, the proposed ACPA can improve the prediction ability of three basic models which can reduce the RMSE of BPNN, RBFNN and RF by 7.10%, 21.08% and 7.08%, respectively. Besides, the ACPA can narrow the prediction gap among the basic models, which can decrease the deviation of MAE by 50% compared with basic models. The prediction performance of models based on ACPA is better than the model itself under different sample size, as a result, the ACPA is recommended for most black-box models used in building load prediction.
ISSN:0378-7788
DOI:10.1016/j.enbuild.2023.113429