6. GIS Research Center, Feng Chia University,Taiwan
Game or real project/system ?
Winner needs to control
construction project's
cumulative progress
from start to finish
Game by Microsoft - Age of Empires II
YES
We are playing real game.
Real Construction Project Management System
2. Find good partners
1. construction projects
need to be done on time
How to win?
7. GIS Research Center, Feng Chia University,Taiwan
• Soil and Water Conservation Bureau takes care
about construction projects :
– design and execute the conservation engineering of
watershed.
– design and execute the tasks of hillsides conservation,
green engineering, safe drainage and regional soil and
water conservation.
Improving Response Measure for Sediment Disaster in Watershed
8. GIS Research Center, Feng Chia University,Taiwan
Question from big data in existed system?
• Not big, but about 32,000 historical data.
• How to know the reasonable curve
forecast in each project?
• How to find good partners to finish
construction projects on time?
9. GIS Research Center, Feng Chia University,Taiwan
Question from big data in existed system?
• How to know the reasonable curve
forecast in each project?
• How to find good partners to finish
construction projects on time?
13. GIS Research Center, Feng Chia University,Taiwan
13
「工程進度預警」模型-視覺化展示(4/4)
進度報表 將大數據推估出的進度,以報表呈現,用於進度檢討會議
於進度檢討會議報表
增加當月份的推估進度與進度差異
提供匯出
年底前,計畫內進度差異達到
-20%的工程清單
14. GIS Research Center, Feng Chia University,Taiwan
Apply back
propagation neural
network
to build model
Input new data into
the trained model
Find the forecasted
S and P
Check definition table
Classify
four categories
How to build
prediction model?
How to predict the
progress of new
coming
engineering case?
How to classify
Engineering Case?
How to analyze every
project’s S-curve
characteristics?
Apply cubic
polynomial function ,
Curve Fitting to find
the parameters
A
Q
The process of predication
Write R code to run
S Curve Fitting
Find the parameters of a and b
Find S and P
Agricultural Road improvement and maintenance
Waterfront project
Non-waterfront project
Dredging project
Write R code to run
Neural net module
Train model
Compute the parameters of a, b
S-curve can be obtained
New Input:
Budget, duration, type, location, rainfall
15. GIS Research Center, Feng Chia University,Taiwan
The introduction of S-curve
• The S-curve is a graphical representation of a construction
project’s cumulative progress from start to finish and
commonly used as a tool for project progress control, but
the traditional forecasting method by calculating the
cumulative progress at each time point from a schedule is
time-consuming and it would be beneficial for project
progress management if the forecast can be assisted by
faster methods.
Project’s cumulative progress
Time
16. GIS Research Center, Feng Chia University,Taiwan
S Curve formula
x is the time of contract、
y is the cumulative progress
a,b are parameters
x=0, y=0; x=1, y=1
17. GIS Research Center, Feng Chia University,Taiwan
Raw data
normalization
Curve Fitting by R
Find parameters of
a, b
Forecast New Project
Different S-curve obtained
by different projects
New Input of X
(Future time)
The Y value is
calculated by S-curve
Obtain S-curve
18. GIS Research Center, Feng Chia University,Taiwan
Input X: 0.0, 0.1, 0.2, 0.3, 0.4, 0.5
Input Y: 0.0, 0.1, 0.2, 0.4, 0.5, 0.6
Model:
Nonlinear regression model
model: y ~ I(a * x^3 + b * x^2 + (1 – a – b) * x)
data: parent.frame()
a b
-0.9116 0.8963
residual sum-of-squares: 0.003034
Number of iterations to convergence: 1
Achieved convergence tolerance: 9.16e-08
New input of X : 0.6, 0.7, 0.8, 0.9, 1.0
Forecast Cumulative progress : 0.73, 0.84, 0.92, 0.98, 1.00
Cumulative monthly actual progress
Forecast Cumulative progress
Forecast Cumulative Progress Monthly Before the end of Selected Project
19. GIS Research Center, Feng Chia University,Taiwan
Apply the neural network in R
Import Data
Forecast
P and S values
Model Training completed
New
Project’s
Condition
Compute a, b
Observe the difference between the
actual curve prediction curve
Budget, duration, type, location, rainfall
Budget, duration,
type, location, rainfall
20. GIS Research Center, Feng Chia University,Taiwan
# --------- actual curve -------------
---
P = 0.5957
S = 1.0000
A = computeA(P,S)
B = computeB(P,S)
curve(SCurve, 0, 1)
Observe the difference between
the actual curve prediction curve
# -------- prediction curve ---------
--
PNN = 0.3812277520
SNN = 1.0433264985
A = computeA(PNN,SNN)
B = computeB(PNN,SNN)
curve(SCurve, 0, 1, add=T, col=2)
Prediction curve
Actual curve
21. GIS Research Center, Feng Chia University,Taiwan
The number of hidden layers RMSE for P RMSE for S
hidden=1 0.1753558174 0.5418783552
hidden=3 0.1757130704 0.5303154711
hidden=5 0.1755577719 0.5280936548
Training
22. GIS Research Center, Feng Chia University,Taiwan
Question from big data in existed system?
• How to know the reasonable curve
forecast in each project?
• How to find good partners to finish
construction projects on time?
23. GIS Research Center, Feng Chia University,Taiwan
Data of Sampling
and Testing project
checkDataList<-read.csv(“L:0305/DT-EngQuality_Check.csv”)
gradeDataList<-read.csv("L:0305/DT-EngQuality_Grade.csv")
install.packages("rpart")
library(rpart)
check.tree=rpart(result~constructor+inspector, data= checkDataList)
grade.tree=rpart(result~constructor+inspector, data= gradeDataList)
Data of
Supervising project
Project’s quality correlation analysis
1. What combination (Constructor +
Inspector) is good from historical
data?
2. For reference to find good partners.
Constructor
Inspector
24. GIS Research Center, Feng Chia University,Taiwan
Each trained data by decision tree
Import Data
Result: Pass/Failed
Ranking: A/B/C
Trained model
New
combination
Verify the accuracy of model
by testing data
Data of Sampling and Testing project
Data of Supervising project
Sampling and Testing project
Supervise project
Constructor
Inspector