Generalized nonlinear regression with ensemble of neural nets: The great energy predictor shootout
- Mobile Research and Development Corp., Princeton, NJ (United States). Central Research Lab.
A novel approach to data modeling has been developed using an ensemble of artificial neural networks. This development has been driven by the need in the petroleum business to model data that has poorly defined independent ,parameters, poor repeatability and/or reproducibility (more than {+-}10% variation in repeated measurements), and relatively few data sets. Their objective for entering ``The Great Energy Predictor Shootout`` was to evaluate the efficiency of their approach. In particular, the authors were interested in how quickly a model could be constructed and the accuracy of the model`s predictions. This scientific competition was ostensibly organized to compare various modeling methodologies applied to both time-dependent (dynamic) and time-independent problems. Both problems were given to the participants with little descriptive information. The objective of this competition was to train (or fit) a model given a complete set of data (independent and dependent variables) and then blindly predict the dependent variables when given only the independent variables of a different data set. The predictions were then sent to the organizers for evaluation. While their entry ranked third overall, the models were generated with less than one man-day of effort -- far less than the two months required by the first-place winner and the one to one-and-one-half months spent by the other winners.
- OSTI ID:
- 33311
- Report Number(s):
- CONF-9406105-; TRN: IM9518%%132
- Resource Relation:
- Conference: 1994 American Society of Heating, Refrigerating, and Air Conditioning Engineers (ASHRAE) annual meeting, Orlando, FL (United States), 25-29 Jun 1994; Other Information: PBD: 1994; Related Information: Is Part Of ASHRAE transactions 1994: Technical and symposium papers. Volume 100, Part 2; PB: 1372 p.
- Country of Publication:
- United States
- Language:
- English
Similar Records
Predicting hourly building energy use: The great energy predictor shootout -- Overview and discussion of results
Great energy predictor shootout II: Measuring retrofit savings -- overview and discussion of results