Abstract

A new neural network model aimed at solving the resource leveling (RL) problem in construction is developed. The model is derived by mapping a formulation of the RL problem as a quadratic augmented Lagrangian multiplier (QALM) optimization, onto an artificial neural network (ANN) architecture employing a Hopfield network. Specifically, it is shown that the augmented Lagrangian associated to the RL problem can be interpreted as the energy function of the Hopfield neural network (NN). The ANN architecture consists of two main blocks. The first is the Hopfield NN block, and the second is a control block for the adjustment of Lagrange multipliers in the QALM optimization. The latter is also used for the computation of the new set of weights of the Hopfield block. A new methodology for the derivation of the weight-matrix of the Hopfield-based NN architecture for RL is also proposed. An in-depth study of the matrices generated from the formulation of the RL problem as a QALM optimization, has revealed some very useful structural properties. It is shown that due to these properties, it is possible to develop a flexible and computationally efficient procedure for the formulation and updating of the weight-matrix of the ANN configuration. This procedure allows for the on-line computation of the new set of weights of the Hopfield NN block with the adjustment of Lagrange parameters. The procedure also solves the key issue of mapping the QALM optimization formulation of the RL problem, onto an ANN architecture. Finally, an experimental validation of the proposed model is presented along with its results.