Knowledge-Based Hydraulic Model Calibration

Sep 3, 2003 - Development of a prototype with an existing engine. Context. Approach. Knowledge modelling. Prototype. Dedicated engine. Perspectives. 2 ...
112KB taille 2 téléchargements 290 vues
Knowledge-Based Hydraulic Model Calibration Jean-Philippe Vidal1

Sabine Moisan2

Jean-Baptiste Faure1

September 3, 2003

1

Cemagref, Hydrology-Hydraulics Research Unit, France 2

INRIA Sophia-Antipolis, Orion Project, France

KES2003, Oxford, U.K.

Numerical Modelling Context

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

1

Numerical Modelling Context

Theory Confirmation

Analysis

Reality

Model Validation

Predictive Simulation

Numerical Model

Conceptual Model

Programming

Model Set-up

Code Verification

Simulation Code

Model Calibration

Elements of a modelling terminology, after [Refsgaard and Henriksen, 2002]

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

1

Approach

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Approach • Objective: Development of a calibration assistant for a 1-D river hydraulic code

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Approach • Objective: Development of a calibration assistant for a 1-D river hydraulic code • Method:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Approach • Objective: Development of a calibration assistant for a 1-D river hydraulic code • Method: 1. Expert user knowledge modelling

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Approach • Objective: Development of a calibration assistant for a 1-D river hydraulic code • Method: 1. Expert user knowledge modelling 2. Development of a prototype with an existing engine

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Approach • Objective: Development of a calibration assistant for a 1-D river hydraulic code • Method: 1. Expert user knowledge modelling 2. Development of a prototype with an existing engine 3. Specifications for an engine dedicated to simulation validation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

2

Knowledge Modelling

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps: – Specifications with UML class and activity diagrams

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps: – Specifications with UML class and activity diagrams – Application: knowledge base with an existing language

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps: – Specifications with UML class and activity diagrams – Application: knowledge base with an existing language

• Identification of two knowledge levels:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps: – Specifications with UML class and activity diagrams – Application: knowledge base with an existing language

• Identification of two knowledge levels: – Calibration generic knowledge

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Knowledge Modelling

• Steps: – Specifications with UML class and activity diagrams – Application: knowledge base with an existing language

• Identification of two knowledge levels: – Calibration generic knowledge – 1-D hydraulic knowledge

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

3

Descriptive Generic Knowledge

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

4

Descriptive Generic Knowledge Distinction between:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

4

Descriptive Generic Knowledge Distinction between: • NumericalModel and SimulationCode

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

4

Descriptive Generic Knowledge Distinction between: • NumericalModel and SimulationCode • StaticData and DynamicData

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

4

Descriptive Generic Knowledge Distinction between: • NumericalModel and SimulationCode • StaticData and DynamicData

Event

InputData

DynamicData

uses System

OutputData

produces NumericalModel

Error

StaticData

Context

Approach

Knowledge modelling

Prototype

SimulationCode

Dedicated engine

Parameter

Perspectives

4

Operative Generic Knowledge: Model Calibration

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

5

Operative Generic Knowledge: Model Calibration ModelCalibration

Preprocessing

ParameterDetermination [else] ParameterDefinition DataAllocation ParameterInitialisation

[parameter definition = ok]

SimulationRunning

OutputComparison

ResultEvaluation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

5

Operative Generic Knowledge: Model Calibration ModelCalibration

Key points: Preprocessing

ParameterDetermination [else] ParameterDefinition DataAllocation ParameterInitialisation

[parameter definition = ok]

SimulationRunning

OutputComparison

ResultEvaluation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

5

Operative Generic Knowledge: Model Calibration ModelCalibration

Key points: Preprocessing

• Data allocation: Which data for which role?

ParameterDetermination [else] ParameterDefinition DataAllocation ParameterInitialisation

[parameter definition = ok]

SimulationRunning

OutputComparison

ResultEvaluation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

5

Operative Generic Knowledge: Model Calibration ModelCalibration

Key points: Preprocessing

• Data allocation: Which data for which role?

ParameterDetermination [else] ParameterDefinition DataAllocation ParameterInitialisation

[parameter definition = ok]

• Parameter definition: How many parameters? SimulationRunning

OutputComparison

ResultEvaluation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

5

Operative Generic Knowledge: Model Calibration ModelCalibration

Key points: Preprocessing

• Data allocation: Which data for which role?

ParameterDetermination [else] ParameterDefinition DataAllocation ParameterInitialisation

[parameter definition = ok]

• Parameter definition: How many parameters? SimulationRunning

OutputComparison

• Output comparison and result evaluation: Which method? Context

Approach

Knowledge modelling

ResultEvaluation

Prototype

Dedicated engine

Perspectives

5

Descriptive 1-D Hydraulic Knowledge

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

6

Descriptive 1-D Hydraulic Knowledge

GraphicalObject

Curve

GivenReachCurve

EnvelopCurve

Reach

Point

Section

GivenSectionCurve

TimeLimitedPoint

TimeExtendedPoint

GivenTimeCurve

EnvelopWaterSurfaceProfile

WaterSurfaceProfile

DischargeSpatialEvolution

Floodmark

WaterLevel

RatingCurve

GaugingPoint

StageHydrograph

DischargeHydrograph

Discharge

Hierarchy of graphical objects used in hydraulic calibration Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

6

Development of a Prototype Knowledge-based System with an Existing Engine

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique Conclusions:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique Conclusions: • Need for a calibration-domain language

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique Conclusions: • Need for a calibration-domain language ⇒ notion of Model

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique Conclusions: • Need for a calibration-domain language ⇒ notion of Model • Hierarchical planning not entirely satisfactory

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Development of a Prototype Knowledge-based System with an Existing Engine Tools: • User-friendly knowledge representation language • Inference engine based on a hierarchical planning technique Conclusions: • Need for a calibration-domain language ⇒ notion of Model • Hierarchical planning not entirely satisfactory ⇒ additional reasoning steps: DataAllocation and ParameterDefinition Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

7

Towards a New Engine for Simulation Validation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

• Simulation validation concepts:

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

• Simulation validation concepts: – Events occurring on a physical System

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

• Simulation validation concepts: – Events occurring on a physical System – Two types of data: Static Data and Dynamic Data

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

• Simulation validation concepts: – Events occurring on a physical System – Two types of data: Static Data and Dynamic Data – Simulation Code → simulating generic behaviour of real systems

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Towards a New Engine for Simulation Validation • Objective: dedicated inference engine associated with generic knowledge about model calibration • Existing concepts: – Functionality and Request → expressing purpose – Criteria → expressing decision knowledge – Operator → encapsulating codes

• Simulation validation concepts: – – – –

Context

Events occurring on a physical System Two types of data: Static Data and Dynamic Data Simulation Code → simulating generic behaviour of real systems Numerical Model → simulating a particular system thanks to a simulation code

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

8

Generic Knowledge for the Engine

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

9

Generic Knowledge for the Engine

Request

Functionality

SimulationCode

AuxiliaryCode PreAndPostProcessors

uses

Error

uses Criteria

NumericalModel

Parameter

OutputData EventData

InputData

DynamicData

ModelData

StaticData

SystemData

System

Event

Concepts to be instantiated in the knowledge base Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

9

Perspectives • Improvement of a language for simulation validation knowledge representation

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

10

Perspectives • Improvement of a language for simulation validation knowledge representation

• Adaptation of the existing engine to get an operational calibration assistant

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

10

Perspectives • Improvement of a language for simulation validation knowledge representation

• Adaptation of the existing engine to get an operational calibration assistant

• Experiments with real-life hydraulic model calibration

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

10

Perspectives • Improvement of a language for simulation validation knowledge representation

• Adaptation of the existing engine to get an operational calibration assistant

• Experiments with real-life hydraulic model calibration – Application on river systems (Ard`eche river)

Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

10

Perspectives • Improvement of a language for simulation validation knowledge representation

• Adaptation of the existing engine to get an operational calibration assistant

• Experiments with real-life hydraulic model calibration – Application on river systems (Ard`eche river) – Comparison with “expert” calibration Context

Approach

Knowledge modelling

Prototype

Dedicated engine

Perspectives

10