CO2 Emissions Forecast in Precast Concrete Production

Abstract

With increasing demands regarding a detailed estimation of environmental impacts of materials in new construction projects, this research was intended to produce a forecasting model of CO₂ emissions in precast concrete production using Artificial Neural Network (ANN). Due to its capability to correlate non-linear and non-unique problem, ANN has received increasing attention for forecasting applications in recent years. Prior to the model development, a set of questionnaire was distributed to several precast concrete plants all around Japan to obtain data regarding indicators which are generally influential to CO₂ emissions in the production. From 107 plants, the result of Principal Component Analysis (PCA) showed that ordinary Portland cement, coarse aggregate, fine aggregate, heavy oil, kerosene, and electricity were considered to be significant indicators and further used as inputs in developing the CO₂ emissions model. A three-layer perceptron with backpropagation neural network approach was proposed to train the network. The different number of hidden neurons, distribution of data sets, learning rate, and momentum were tested to minimize the error between actual and forecasted outputs. The model with 51 hidden neurons using a set of 0.1, 0.9 and 0.3 for learning rate, momentum and initial weight, respectively produced the best result. Indicated by the MAPE value which is less than 10%, this newly developed model shows an excellent accuracy for forecasting the CO₂ emissions in the future. It was also validated by the result of sensitivity analysis that the developed model generated a negligible impact on the CO₂ emissions due to variations of the six significant indicators.