A non-linear case-based reasoning approach for retrieval of similar cases and selection of target credits in LEED projects

作者：Cheng Jack C P; Ma Lucky J

来源：Building and Environment, 2015, 93: 349-361.

DOI：10.1016/j.buildenv.2015.07.019

• Artificial neural network
• Case-based reasoning
• Decision support tool
• LEED-NC v2009
• Non-linear
• Target credit selection

Leadership in Energy and Environmental Design (LEED) is a widely used international green building certification program developed by the U.S. Green Building Council (USGBC). Although the need for LEED certification has grown significantly, LEED managers still face challenges in target credit selection and green building technology design. They frequently meet new types of projects with different project characteristics and requirements. Therefore, it would be helpful if LEED managers could refer to other similar certified green building cases when planning and designing LEED projects. However, this is not supported in current studies and research. This paper proposes a case-based reasoning (CBR) approach to provide case studies of similar certified green building projects and suggestions on target LEED credits. Currently, linear formation of Local-Global method is commonly used in the retrieval step of CBR. This paper presents a non-linear formation of Local-Global retrieval based on Artificial Neural Network (ANN), which can provide a higher accuracy. LEED for New Construction (LEED-NC) is the focus of this paper, and 1000 LEED-NC v2009 certified cases were collected for the case base. Pairwise comparison was conducted to generate the local distance (attribute similarity) and the target for training the ANN model. The proposed non-linear CBR approach was tested and evaluated using 20 recently certified projects, and the results showed a prediction accuracy of 80.75% on the LEED credit selection. The results were also compared with those calculated by commonly used linear CBR approaches: Multiple Regression Analysis, Correlation Analysis, and the k-NN approach. The accuracy achieved by these methods was between 71.23% and 77.54%, which was lower than the proposed approach.

• 出版日期2015-11