Water level prediction for supporting the right decision making is crucial. The study of daily gage height in Lam Phi-Kun Canal provides essential information to plan for flood prevention and hydroelectric generation of this canal. Daily gage heights in 2011 were used as training instances. We forecasted the gage heights for year 2012 using four different predictors, namely, Linear Regression, Multilayer Perceptron, MLPDQ1 and MLPDQ2. The latter two algorithms are named after our improvements to the existing Multilayer Perceptron by introducing a concept of time-dependent data division and a concept of weight adjustments of the polynomial trend line. Our proposed algorithms are implemented in Java language. The performance evaluation reveals that our two algorithms have distinctly smaller errors than the traditional two predictors. The average errors of our algorithms are less than one meter. We recommend our algorithms for other applications such as rainfall forecasting, sediment forecasting, sales forecasting, and energy consumption trends.
[1]
Ian H. Witten,et al.
The WEKA data mining software: an update
,
2009,
SKDD.
[2]
Apichat Heednacram,et al.
Implementation of Cuckoo Search in RBF Neural Network for Flood Forecasting
,
2012,
2012 Fourth International Conference on Computational Intelligence, Communication Systems and Networks.
[3]
Rajeshri R. Shelke,et al.
Simplified Approach of ANN: Strengths and Weakness
,
2012
.
[4]
Jirapon Sunkpho,et al.
Real-time flood monitoring and warning system
,
2011
.
[5]
Apichat Heednacram.
Suspended Sediment Forecast of Khlong Bang Yai, Phuket
,
2014
.
[6]
อนิรุธ สืบสิงห์,et al.
Data Mining Practical Machine Learning Tools and Techniques
,
2014
.
[7]
Apichat Heednacram,et al.
Improved Cuckoo Search in RBF Neural Network with Gaussian Distribution
,
2013
.