Optimization and Performance of a Neural Network Model Forecasting Water Levels for the Corpus Christi, Texas, Estuary Philippe Tissot*, Patrick Michaud*, Philippe Tissot*, Patrick Michaud*, Daniel Cox** Daniel Cox** *Texas A&M University-Corpus *Texas A&M University-Corpus Christi, Corpus Christi, Texas Christi, Corpus Christi, Texas **Oregon State University, **Oregon State University, Corvallis, Oregon Corvallis, Oregon
Optimization and Performance of a Neural Network Model Forecasting Water Levels for the Corpus Christi, Texas, Estuary. Philippe Tissot*, Patrick Michaud*, Daniel Cox** *Texas A&M University-Corpus Christi, Corpus Christi, Texas **Oregon State University, Corvallis, Oregon. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Optimization and Performance of a Neural Network Model
Forecasting Water Levels for the Corpus Christi, Texas, Estuary
Philippe Tissot*, Patrick Michaud*, Daniel Cox**Philippe Tissot*, Patrick Michaud*, Daniel Cox**
*Texas A&M University-Corpus Christi, Corpus *Texas A&M University-Corpus Christi, Corpus Christi, TexasChristi, Texas
**Oregon State University, Corvallis, Oregon**Oregon State University, Corvallis, Oregon
Texas A&M University-Corpus Christi Division of Near Shore Research
Tides and Water Levels in the Gulf of MexicoTides and Water Levels in the Gulf of Mexico Artificial Neural Network Forecasting of Water Artificial Neural Network Forecasting of Water
Levels and Application to the Corpus Christi Levels and Application to the Corpus Christi EstuaryEstuary
ANN Performance for Water Level ForecastingANN Performance for Water Level Forecasting ANN performance during a Tropical StormANN performance during a Tropical Storm Conclusions Conclusions
Texas A&M University-Corpus Christi Division of Near Shore Research
Texas Coastal Observation Network (TCOON)
Started 1988Started 1988 Over 50 stationsOver 50 stations Primary SponsorsPrimary Sponsors
General Land OfficeGeneral Land Office Water Devel. BoardWater Devel. Board US Corps of EngUS Corps of Eng Nat'l Ocean ServiceNat'l Ocean Service
Texas A&M University-Corpus Christi Division of Near Shore Research
Typical TCOON station
Wind anemometerWind anemometer Radio AntennaRadio Antenna Satellite TransmitterSatellite Transmitter Solar PanelsSolar Panels Data CollectorData Collector Water Level SensorWater Level Sensor Water Quality SensorWater Quality Sensor Current MeterCurrent Meter
Texas A&M University-Corpus Christi Division of Near Shore Research
TCOON Web Site
Texas A&M University-Corpus Christi Division of Near Shore Research
Tides and Water Levels
Tide: The periodic rise and fall of a body of Tide: The periodic rise and fall of a body of water resulting from gravitational water resulting from gravitational interactions between Sun, Moon, and Earth.interactions between Sun, Moon, and Earth.
Tide and Current GlossaryTide and Current Glossary, National Ocean Service, 2000, National Ocean Service, 2000
Water Levels: Astronomical + Meteorological Water Levels: Astronomical + Meteorological forcing + Other effectsforcing + Other effects
Texas A&M University-Corpus Christi Division of Near Shore Research
Study Site: CC Estuary
Bob Hall Pier
Packery Channel
Naval Air Station
AquariumIngleside
Port AransasNueces Bay
Corpus Christi Bay Gulf of
Mexico
Oso BayPort of Corpus Christi
Texas A&M University-Corpus Christi Division of Near Shore Research
Comparison of Tides and Water Levels
TCOON MeasurementsTide Tables
Corpus Christi Naval Air Station
Texas A&M University-Corpus Christi Division of Near Shore Research
Comparison of Tides, Water Levels, and Winds (squared)
Wa
ter
Le
ve
l (m
)
0 50 100 150 200 250 300 350 400-0.5
0
0.5
Wa
ter
An
om
aly
(m
)
0 50 100 150 200 250 300 350 400-500
0
500
N-S
Win
d S
qu
are
d
0 50 100 150 200 250 300 350 400-400
-200
0
200
E-W
Wid
Sq
ua
red
Julian Day,1997
0 50 100 150 200 250 300 350 400-0.5
0
0.5
1
Texas A&M University-Corpus Christi Division of Near Shore Research
Challenge
Develop a water level forecasting model Develop a water level forecasting model that captures the non linear relationship that captures the non linear relationship between wind forcing and future water level between wind forcing and future water level changeschanges
Take advantage of the large amount of real-Take advantage of the large amount of real-time data available through TCOONtime data available through TCOON
Ability for dynamic learningAbility for dynamic learning
Requires availability of high density of dataRequires availability of high density of data
Texas A&M University-Corpus Christi Division of Near Shore Research
ANN Model
H (t+i)
Output LayerHidden Layer
Observed Winds
Observed Water Levels
Observed Barometric Pressures
Forecasted Winds
Input Layer
Water Level Forecast
(a1,ixi)
b1
b2
(X1+b1)
b3
(X2+b2)
(X3+b3)
(a2,ixi)
(a3,ixi)
Texas A&M University-Corpus Christi Division of Near Shore Research
ANNs Characterisitics
ANN models developed within the Matlab ANN models developed within the Matlab (R13) and Matlab NN Toolbox environment(R13) and Matlab NN Toolbox environment
Simple ANNs are optimumSimple ANNs are optimum Use of ‘tansig’ and ‘purelin’ functionsUse of ‘tansig’ and ‘purelin’ functions Levenberg-Marquardt training algorithmLevenberg-Marquardt training algorithm ANN Trained over 1 year of hourly data ANN Trained over 1 year of hourly data
(8750 forecasts)(8750 forecasts)
Texas A&M University-Corpus Christi Division of Near Shore Research
CCNAS ANN 24-hour Forecasts
0 50 100 150 200 250 300 350 400-0.5
0
0.5
1
Wat
er L
evel
s (m
)
Julian Day,1997
ANN trained over 2001 Data Set
Texas A&M University-Corpus Christi Division of Near Shore Research
CCNAS ANN 24-hour Forecasts
0 50 100 150 200 250 300 350 400-0.5
0
0.5
1
Wat
er L
evel
s (m
)
Julian Day,1997
ANN trained over 2001 Data Set
Texas A&M University-Corpus Christi Division of Near Shore Research
CCNAS ANN 24-hour Forecasts
75 80 85 90 95 100 105 110 115 120 125
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Wat
er L
evel
s (m
)
Julian Day,1997
ANN trained over 2001 Data Set
Texas A&M University-Corpus Christi Division of Near Shore Research
Model Assessment
Based on five 1-year data sets: ‘97, ‘98, ’99, ’00, ‘01 Based on five 1-year data sets: ‘97, ‘98, ’99, ’00, ‘01 including observed water levels and winds, and tide including observed water levels and winds, and tide forecastsforecasts
Train the NN model using one data set e.g. ‘97 for Train the NN model using one data set e.g. ‘97 for each forecast target, e.g. 12 hourseach forecast target, e.g. 12 hours
Apply the NN model to the other four data sets, Apply the NN model to the other four data sets, Repeat the performance analysis for each training Repeat the performance analysis for each training
year and forecast target and compute the model year and forecast target and compute the model performance and variabilityperformance and variability
Texas A&M University-Corpus Christi Division of Near Shore Research
Texas A&M University-Corpus Christi Division of Near Shore Research
Conclusions
ANN models improve considerably on the tides ANN models improve considerably on the tides for regular conditions and frontal passagesfor regular conditions and frontal passages
Once trained computationally very efficient Once trained computationally very efficient Allow great modeling flexibilityAllow great modeling flexibility Accuracy and location of the Wind forecasts will Accuracy and location of the Wind forecasts will
determine model performance beyond 15 hoursdetermine model performance beyond 15 hours Promising for short term, up to 12 hours, water Promising for short term, up to 12 hours, water
level forecasts during stormslevel forecasts during storms
Texas A&M University-Corpus Christi Division of Near Shore Research