Scholarlink

<p>Nowadays, due to an improvement in technology and increased competition, project managers have the mandatory task of completing the projects in a foresighted time and with the specified resources. In order to finish the projects without delay, the classical <i>PERT-CPM</i> methodology is being extensively used. In this scenario, two critical problems emerge with no analytical solution: <i>Resource Levelling</i> and <i>Resource Allocation</i>. This paper focuses on the first problem from the novel perspective of <i>Artificial Neural Networks</i> within classic <i>Training–Testing</i> protocol. In order to obtain a reliable dataset for the training process two other targets have been the subject of our research: firstly, the goal was to calibrate quantitatively the efficiency of previous heuristic and metaheuristic approaches to, secondly, train our network with the best results. However, to perform that calibration, research was also conducted to find the optimal efficiency descriptor, beyond the classical <i>Sum of Squares.</i> Using the <i>Principal Component Analysis,</i> an optimal mixed descriptor was found, built from the <i>RIC</i> and <i>Sum of Squares</i> performance descriptors. The research obtained an accuracy of 96,2% regarding the predictions made with a trained neural network. Even if the preliminary task for <i>the Training–Testing</i> protocol is demanding, the reward obtained consists of a quick and comfortable resolution for all the subsequent projects.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial Neural Network Approach with Multi-Back Propagation and Training–Testing to Solve the Resource Levelling Problem in PERT-CPM

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月14日

查看全文
  • Inma C Conde,
  • Luis Antonio Palma Martos,
  • María del Rocío Martínez Torres,
  • Francisco Luis Cumbrera Hernández

摘要

Nowadays, due to an improvement in technology and increased competition, project managers have the mandatory task of completing the projects in a foresighted time and with the specified resources. In order to finish the projects without delay, the classical PERT-CPM methodology is being extensively used. In this scenario, two critical problems emerge with no analytical solution: Resource Levelling and Resource Allocation. This paper focuses on the first problem from the novel perspective of Artificial Neural Networks within classic Training–Testing protocol. In order to obtain a reliable dataset for the training process two other targets have been the subject of our research: firstly, the goal was to calibrate quantitatively the efficiency of previous heuristic and metaheuristic approaches to, secondly, train our network with the best results. However, to perform that calibration, research was also conducted to find the optimal efficiency descriptor, beyond the classical Sum of Squares. Using the Principal Component Analysis, an optimal mixed descriptor was found, built from the RIC and Sum of Squares performance descriptors. The research obtained an accuracy of 96,2% regarding the predictions made with a trained neural network. Even if the preliminary task for the Training–Testing protocol is demanding, the reward obtained consists of a quick and comfortable resolution for all the subsequent projects.