NOVEL DEEP LEARNING APPROACH FOR SCHEDULE ESTIMATE TO COMPLETION IN CONSTRUCTION PROJECT USING NN-LSTM

• Main Author: DORCAS KORIR
• Other Authors: Min-Yuan Cheng
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/25258036442756742684

碩士 === 國立臺灣科技大學 === 營建工程系 === 106 === Estimation of construction project duration is extremely hard during various construction phases due to complexity, uncertainty and limited information available during a project course. Construction managers estimate construction duration according to their previous experience based on budget planning, a method that can be inaccurate and costly. This study developed a novel inference model that accurately estimate project duration factoring in the dependent and independent factors that significantly influence project duration and captures uncertainty involved in construction field. Duration influencing factors were selected based on the previous studies review, and later categorized into time dependent and independent factors. Subsequently, two artificial intelligence approaches were fused, namely neural networks (NN) and the long short-term memory (LSTM) to create a novel Neural Network - Long Short-Term Memory (NN-LSTM) model. The NN-LSTM was applied to estimate schedule to completion (ESTC) for historical cases where NN captured the impact of independent factors in project duration as LSTM captured the long temporal dependency for sequential inputs. After the analysis, 14 influencing factors and 11 historical cases were selected to establish the case database used for learning purposes. 10-cross validation was used to partition the training and testing dataset. The learning results indicated good performance with MAPE of 4% and mean absolute error of 2% proving the model more reliable than the currently prevailing formula. Moreover, upon comparison with other methods, NN-LSTM proved to be superior to SVM, LSSVM, BPNN, ESIM, ELSIM, SOS-LSSVM and the earned value method (EVM). The model qualifies to replace the subjective estimation based on experience alone. It provides project managers with reliable schedule estimates that facilitates proper planning and help in monitoring project’s performance in terms of time, prompting timely actions in case of foreseen delay and making of informed decisions.