بررسی مقایسه ای شبکه های عصبی مصنوعی و تجزیه و تحلیل رگرسیون چندگانه برای پیش بینی زمان بلند کردن وسایل جرثقیل برج

This paper aims to develop a quantitative model for predicting the hoisting times of tower cranes for public housing construction using artificial neural network and multiple regression analysis. Firstly, based on data collected from crane operators and site managers in seven construction sites, the basic factors affecting the hoisting times for tower cranes are identified. Then, artificial neural networks (ANN) and the multiple regression analysis (MRA) are used to model the hoisting time, and from the results, the neural network model and the multiple regression model of hoisting time are established. The modeling methods and procedures are explained. These two kinds of models are then verified by data obtained from an independent site, and the predictive behaviors of the two kinds of models are compared and analyzed. Furthermore, the predictive behaviors of the neural network model are also investigated by a sensitivity analysis. Finally, the modeling methods, predictive behaviors and the advantages of each model are discussed.

In most modern cities, high-rise building construction prevails. Thus, material handling and movement in construction sites have received considerable attention by site managers to ensure efficient material supply. Material hoists, which can provide fast vertical transportation, are widely used because of low cost and easy operation. However, horizontal distribution of materials to the designated work areas is required. The use of tower cranes can perform vertical and horizontal transportation at the same time and thus becomes dominating in high-rise building construction. On the other hand, the number of tower cranes installed for building sites is restricted to avoid clashing between cranes and by its high installation and running costs. Hence, the allocation of cranes for different trades is one of the critical targets in resource planning. In planning crane operations such as concreting, installing precast concrete units and fixing formwork panels, which when combined together will determine the cycling duration for structural frame construction. At present, the times for hoisting, loading and discharging are usually estimated by experience, which may vary between people and create inaccuracies.

Based on a number of data obtained from the crane operators and site managers in the seven construction sites, this paper identifies the basic factors affecting the hoisting times for tower cranes. MLFF with BP algorithm and GRNN with genetic algorithm, and multiple regression technique are used to model the hoisting times. These models are verified by the data obtained from an independent site, and the predictive behaviors of the established models are compared and analyzed by tests, applications and sensitivity analysis. The following points can be concluded from the results of this research. 1. Seventeen basic factors in relation to the hoisting movements affect the hoisting time of tower cranes in construction sites. Among these factors, “hoisting height” is ranked the highest in the supply hoisting and return hoisting models. The area of load and the simultaneous movements are the second highest contributor in the supply time and return time operations, respectively. 2. Neural networks and multiple regression technique can be used to model hoisting times of tower cranes. Furthermore, the predictive performance of GRNN with genetic algorithm models is better than that of the multiple regression model and the MLFF network with BP algorithm in modeling the hoisting times. 3. The hoisting time models derived in this paper enable planners to predict and compare the hoisting times for different loading points of a designated tower crane. On the other hand, the predicted hoisting times allow planners to foresee possible overtime works as the hoisting height increases. In addition, the estimation of the duration for typical floor construction cycles for lower floors and upper floors can be improved by considering the predicted hoisting times.