Construct drawing man-hour forecasting based on GA-BP in chemical equipment design process

Abstract

The man-hour costing is the largest cost in the chemical plant design companies because the design processes are undertaken by the staff. The accurate man-hour forecasting can facilitate the design process control and the human resources scheduling optimization so as to cut the costs. This paper presents a framework, combining the Back Propagation (BP) Artificial Neural Network (ANN) and Genetic Algorithm (GA), for the forecasting of Construct Drawing Design Man-hour (CDDM), which is the largest man hour among all work items in the chemical equipment design process. Firstly, the input variables are selected based on the contribution and correlation analysis after describing the optional input variables. Secondly, the forecasting model is presented in details. The experiments are done for the parameters set in the model, such as the hidden neuron number and the training method in the BP ANN, and the GA parameters selection. Thirdly, the simulation results are discussed. The absolute error, the relative error, and the grey relative relational grade are employed to measure the model accuracy and to compare the forecasting accuracy between the GA-BP ANN and pure BP ANN. Finally, the forecasting experiments are also tried to investigate if the variables coming from the earlier design phase can be selected as the input variables. The simulation results show the forecasting model based on GA-BP ANN can be a helpful tool for the CDDM forecasting in chemical equipment design process. The results also show the input variable selection and the experiments for the input variables coming from the earlier design phase are suitable for the CDDM forecasting.