12. System Modelling, Optimization and Digital Intelligence

Introduction and context

Renewable-energy systems are governed by variability, uncertainty, spatial diversity and interactions among multiple physical assets. A photovoltaic plant depends on irradiance, temperature, soiling and inverter behaviour; a wind farm depends on atmospheric conditions, wake effects and turbine availability; a hybrid microgrid must coordinate generation, storage, flexible loads and the utility grid. Consequently, design and operation cannot rely only on nominal ratings or isolated component calculations.

System modelling provides a structured representation of these interactions. Optimization selects an appropriate design or operating policy from competing alternatives, while digital intelligence enables continuous observation, prediction and adaptation. Together, these capabilities support capacity sizing, economic dispatch, reliability assessment, predictive maintenance, energy trading and carbon reduction.

The engineering challenge is not simply to build a sophisticated algorithm. A useful model must represent the relevant physics, use trustworthy data, expose its assumptions and remain computationally practical. An optimization result is meaningful only when its objective, constraints and uncertainty treatment reflect the actual decision problem. Similarly, machine-learning predictions are valuable only when they are validated, explainable and integrated with operational practice.

This chapter develops an integrated framework for physics-based, statistical and hybrid modelling; single- and multi-objective optimization; digital twins; IoT and SCADA architectures; GIS-supported planning; and machine-learning applications. It also considers validation, cybersecurity, data governance and professional responsibility.

Learning objectives

After studying this chapter, the reader should be able to:

1. Distinguish physics-based, statistical, data-driven and hybrid models for renewable-energy applications.

2. Formulate sizing and dispatch problems using objective functions, constraints and decision variables.

3. Apply multi-objective methods and interpret trade-off solutions using Pareto analysis.

4. Explain the architecture and value of SCADA, IoT platforms, GIS and digital twins.

5. Select suitable machine-learning methods for forecasting, anomaly detection and predictive maintenance.

6. Design validation procedures that address accuracy, uncertainty, drift and operational relevance.

7. Evaluate cybersecurity, explainability, privacy and data-governance requirements.

8. Integrate digital intelligence into technically robust and economically defensible energy decisions.

12.1 The modelling hierarchy: from equipment to energy system

A model is a purposeful simplification of a real system. The appropriate level of detail depends on the decision being made, the available data and the time scale of interest. A converter model containing switching transients may be necessary for power-quality studies but unnecessary for annual capacity planning. Conversely, a planning model that ignores minimum battery state of charge may produce an apparently economical but infeasible design.

At the component level, models describe conversion and degradation. A simplified photovoltaic power model may be written as

\[

P_{PV,t}=P_{rated}\frac{G_t}{G_{ref}}[1-\gamma(T_{c,t}-T_{ref})]\eta_{sys},

\]

where \(G_t\) is irradiance, \(T_{c,t}\) is cell temperature, \(\gamma\) is the temperature coefficient and \(\eta_{sys}\) represents aggregate system losses. Wind-power modelling commonly uses a piecewise turbine power curve: output is zero below cut-in speed, rises over the operating region, remains near rated power and falls to zero above cut-out speed. These relationships are useful, but they must be corrected for availability, wake losses, icing, curtailment and electrical losses.

At the subsystem level, energy balances become central. For a battery, with charging efficiency \(\eta_c\), discharging efficiency \(\eta_d\), energy capacity \(E_B\), and time interval \(\Delta t\), one representation is

\[

SOC_{t+1}=SOC_t+\frac{\eta_c P_{c,t}\Delta t}{E_B}-\frac{P_{d,t}\Delta t}{\eta_d E_B}.

\]

Constraints may include \(SOC_{min}\leq SOC_t\leq SOC_{max}\), limits on charge and discharge power, and the prohibition of simultaneous charging and discharging. Battery ageing can be represented by equivalent full cycles, calendar ageing, temperature dependence or an empirical replacement-cost term.

At the system level, the power balance for a grid-connected hybrid system can be expressed as

\[

P_{PV,t}+P_{W,t}+P_{grid,t}+P_{d,t}=P_{load,t}+P_{c,t}+P_{curt,t}.

\]

The equation is accompanied by network, reserve, import/export and reliability constraints. Models may also include thermal loads, hydrogen production, electric vehicles, flexible industrial processes and demand-response actions.

Time scales and model fidelity

Renewable-energy decisions occur over nested time scales. Resource assessment and capacity planning may use hourly or sub-hourly data over several years. Economic dispatch may operate at five-minute or fifteen-minute resolution. Inverter control and protection require millisecond-level models. Maintenance planning may use weeks or months, whereas asset degradation develops over years.

A practical approach is to use a hierarchy: detailed models for calibration and critical operating studies, reduced-order models for optimization, and surrogate models for repeated simulation. Aggregating data can reduce computation, but it may conceal ramps, short-duration peaks and correlation among resources. The chosen time step should therefore be tested through sensitivity analysis rather than selected solely for convenience.

12.2 Physics-based, statistical and hybrid representations

Physics-based models derive relationships from conservation laws, component characteristics and engineering principles. Their strengths are interpretability, extrapolation within known physical regimes and lower dependence on large historical datasets. Their weaknesses include parameter uncertainty, model-form error and the difficulty of representing unobserved operational effects such as soiling or abnormal control behaviour.

Statistical models infer relationships from observations. Time-series methods such as autoregressive models, exponential smoothing and state-space models can forecast load, irradiance or wind power. Their parameters may be estimated from historical data and updated as conditions change. Statistical models are often effective for short-term forecasting when the data-generating process is reasonably stable, but they may perform poorly during structural changes or rare events.

Machine-learning models—including random forests, gradient-boosting models, support-vector regression and neural networks—can represent nonlinear relationships. Deep learning is particularly useful for high-dimensional sequences, images and sensor streams, although it requires careful data preparation and validation. Greater predictive complexity does not automatically imply better engineering performance.

Hybrid models combine physical constraints with statistical learning. For example, a weather-driven photovoltaic model can provide a physically meaningful baseline, while a learning model estimates residual error caused by temperature bias, soiling and inverter clipping. A digital twin may use a state estimator to reconcile a first-principles model with live measurements. Such combinations often improve robustness and interpretability.

Important modelling concepts include:

• State variables: quantities describing the system condition, such as battery state of charge or component temperature.
• Inputs: external drivers and control actions, such as irradiance, wind speed, price and dispatch commands.
• Outputs: measured or calculated responses, such as power, voltage, temperature and alarms.
• Parameters: relatively stable quantities estimated through calibration, such as efficiencies and thermal coefficients.
• Uncertainty: lack of complete knowledge concerning inputs, parameters, model structure or future events.
• Identifiability: the extent to which available data can uniquely determine model parameters.

A model with many parameters may fit historical data closely while predicting poorly. This is overfitting. Regularization, cross-validation, physical constraints and out-of-sample testing help control it. In engineering, a slightly less accurate but transparent model may be preferable if operators must understand and trust its recommendations.

12.3 Optimization for design and operation

Optimization translates engineering choices into a mathematical decision problem. A general formulation is

\[

\min_{x} F(x) \quad \text{subject to } g(x)\leq 0,\; h(x)=0,

\]

where \(x\) contains design and operating variables, \(F\) is an objective function, and \(g\) and \(h\) represent inequality and equality constraints.

For system sizing, design variables may include photovoltaic capacity, wind capacity, battery energy and power ratings, electrolyser size and grid-interconnection capacity. Typical objectives include annualized cost, net present cost, levelized cost of energy, renewable-energy fraction, emissions and loss-of-load probability. Capital cost, replacement cost, salvage value, fuel cost, electricity purchase and discount rate must be treated consistently.

For dispatch, decision variables vary by time: generator output, battery charging and discharging, grid exchange, curtailment and flexible-load schedules. A cost-minimizing dispatch objective might include energy prices, start-up costs, battery degradation and penalties for unserved load. Constraints enforce power balance, ramp rates, minimum up and down times, reserve requirements and network limits.

Single-objective and multi-objective decisions

Renewable-energy design rarely has one universally accepted objective. A lower-cost system may have higher emissions or greater land use; a larger battery may improve reliability but remain underutilized. Multi-objective optimization makes these conflicts explicit. Common objectives are

\[

\min [C(x),\; E(x),\; LPSP(x),\; A(x)],

\]

where cost, emissions, loss-of-power-supply probability and environmental impact are minimized.

A weighted-sum method converts several objectives into one:

\[

J=\sum_i w_i\hat f_i(x),

\]

where normalized objectives \(\hat f_i\) and weights \(w_i\) reflect preferences. The method is simple, but results depend strongly on scaling and weights and may miss non-convex portions of the trade-off surface. Constraint-based methods optimize one primary objective while imposing limits on others. Evolutionary methods, such as multi-objective genetic algorithms, generate a set of non-dominated solutions known as a Pareto front.

A Pareto solution is one for which no objective can be improved without worsening at least one other objective. The final choice remains an engineering and organizational decision, not an automatic consequence of the algorithm. Stakeholder preferences, regulatory limits, resilience requirements and qualitative risks must be considered.

Uncertainty and robust decisions

Using a single weather year and a point forecast can make a design appear more reliable than it is. Uncertainty may be represented through scenarios, probability distributions, intervals or robust constraints. Stochastic optimization assigns decisions across scenarios, while robust optimization seeks solutions that remain feasible under a defined uncertainty set. Monte Carlo simulation can estimate reliability indices, but the number of samples and dependence among variables require attention.

Forecast error, correlated solar and load behaviour, extreme weather and asset failure should not be treated as independent by default. Scenario reduction may lower computational burden while preserving important statistical properties. Sensitivity analysis should identify which assumptions most strongly affect the decision.

12.4 Simulation, validation and verification

Simulation experiments test how a proposed system behaves under alternative resources, prices, failures and control policies. Discrete-time simulation is common for annual energy studies; dynamic simulation is required for transients, voltage regulation and control interaction. Co-simulation may link electrical-network, building, market and communication models.

Verification asks whether the model has been implemented correctly: equations, units, algorithms and numerical procedures are checked. Validation asks whether the model represents the real system sufficiently for its intended use. Validation is therefore purpose-dependent. A model validated for annual energy yield may not be valid for fault-current analysis or battery thermal safety.

A sound validation process includes:

1. Defining the intended use and acceptable error.

2. Checking data quality, time alignment, missing values and sensor calibration.

3. Calibrating parameters using a training period.

4. Testing on independent periods and, where possible, different operating regimes.

5. Comparing relevant metrics such as mean absolute error, root-mean-square error, bias, ramp error and calibration of prediction intervals.

6. Examining residuals, extreme events and failure cases rather than only average performance.

7. Documenting assumptions, version, data provenance and known limitations.

Prediction intervals are often more useful for operational decisions than a single forecast. A probabilistic forecast can support reserve scheduling, whereas a point forecast may conceal the risk of a rapidly developing cloud event. Validation must also be repeated after equipment changes, sensor replacement, software updates or changes in operating policy.

12.5 Digital architecture: SCADA, IoT, GIS and the digital twin

SCADA systems provide supervisory monitoring, data acquisition, alarms, control commands and historical records. They typically contain field sensors and remote terminal units, communication networks, servers, historian databases and operator interfaces. SCADA is designed for operational reliability and deterministic control, whereas IoT platforms often emphasize device connectivity, scalable data services, analytics and integration with enterprise applications. The two may coexist, with carefully controlled interfaces.

An IoT-enabled renewable plant commonly includes sensors for irradiance, module temperature, wind speed, vibration, oil condition, voltage, current, power quality and weather. Edge computing can filter data, detect immediate anomalies and continue local operation when cloud communication is unavailable. Cloud or central platforms support long-term storage, fleet-level analytics and model training. A layered architecture should separate safety-critical control from non-critical analytics.

GIS adds spatial intelligence. Resource maps, terrain, land ownership, protected areas, roads, transmission lines, population centres and environmental constraints can be combined to identify candidate sites. Multi-criteria GIS analysis must distinguish hard exclusions from preferences and should account for uncertainty in resource layers and planning boundaries. Spatial resolution and map age can materially affect siting decisions.

A digital twin is a continuously updated digital representation of a physical asset or system. It differs from a static simulation model because it is linked to the operational asset through data and is used for monitoring, prediction, diagnosis or control. A wind-turbine twin may combine a drivetrain model, vibration measurements, weather data and maintenance history to estimate bearing condition. A microgrid twin may forecast demand, test dispatch policies and evaluate the effect of changing battery capacity.

The twin should have a defined purpose, data interface, update frequency, uncertainty representation and action pathway. A visually impressive dashboard is not necessarily a digital twin. If its model is not calibrated, its data are not traceable or its recommendations cannot be acted upon, it provides limited engineering value.

12.6 Machine learning for forecasting and predictive maintenance

Forecasting applications include solar irradiance, photovoltaic power, wind power, load, electricity prices and reserve requirements. Feature engineering may use lagged power, numerical weather predictions, time of day, day of year, cloud indicators and neighbouring plant data. Models should be compared with meaningful baselines, such as persistence for very-short-term solar forecasting or seasonal averages for demand.

Predictive maintenance aims to identify degradation before functional failure. The workflow normally includes condition monitoring, data cleaning, feature extraction, anomaly detection, diagnosis, remaining-useful-life estimation and maintenance decision-making. Supervised classification requires labelled failure events, which are often scarce and imbalanced. Unsupervised or semi-supervised methods can learn normal operating behaviour and flag deviations. However, an anomaly is not automatically a fault; changing weather, curtailment or sensor failure may produce similar patterns.

A maintenance decision should consider consequence, confidence, inspection cost, spare-part availability and the risk of unnecessary intervention. A model that predicts a fault accurately but provides no lead time or actionable recommendation has limited operational value. Data leakage—using information that would not have been available at prediction time—is a frequent source of falsely impressive results.

Explainable AI is important where predictions affect safety, warranty, investment or maintenance. Feature importance, partial-dependence analysis, counterfactual explanations and physically meaningful indicators can help operators assess a recommendation. Explanations should not be treated as proof of causality; they are aids to review and diagnosis.

12.7 Applied case: digital optimization of a remote hybrid microgrid

Background

The following is an illustrative case involving a remote community supplied by diesel generators, photovoltaic generation and a battery energy-storage system. Fuel delivery is costly and outages affect water pumping and refrigeration. The community authority is considering additional PV and storage, but its budget and land availability are limited.

Situation and problem

Historical data contain fifteen-minute load, generator output, battery measurements and weather observations, but there are missing intervals and inconsistent timestamps. Operators report that the battery occasionally reaches a low state of charge during cloudy periods. The design question is to select PV capacity, battery energy capacity and an operating policy that reduce fuel use without compromising reliability.

Analysis

A component model was developed for PV output, generator minimum loading, battery state of charge and charge/discharge efficiencies. The load forecast used calendar variables and recent demand, while weather scenarios represented clear, variable and overcast conditions. Data quality checks identified sensor bias and periods during which the battery-management system was unavailable.

The optimization minimized annualized cost and fuel consumption subject to a reliability constraint, a battery state-of-charge range, generator ramp limits and a reserve requirement. A second objective represented emissions. Rather than selecting a single weighted score immediately, the study generated a Pareto set. Solutions with small batteries were inexpensive but required more diesel during overcast periods. Larger batteries reduced fuel consumption but increased capital cost and replacement exposure.

Approach and solution

The recommended design was selected after a workshop involving operators, the utility, community representatives and the finance team. The final choice was not the mathematically extreme solution; it was a compromise with adequate reserve, manageable battery cycling and a control policy understandable to operators. A rule-based controller was retained as a fallback, while model-predictive dispatch was tested in simulation before deployment.

An edge gateway collected measurements, performed plausibility checks and transmitted summarized data to a central platform. The digital twin estimated battery state and compared expected with measured PV output. An anomaly detector generated advisory alarms, but all protection and emergency shutdown functions remained within certified local controls.

Outcome

Simulation under independent weather and load periods showed lower expected diesel operation and improved reserve management relative to the existing policy. The study also revealed that sensor calibration and data completeness affected the result almost as much as the choice of optimization algorithm. The implementation plan therefore included metering upgrades, operator training, staged commissioning and periodic model review.

Learning points

The case demonstrates that optimization is inseparable from data quality, stakeholder preferences and operational constraints. A Pareto analysis supports transparent trade-offs, while a digital twin is useful only when it is connected to trustworthy data and an actionable workflow. Retaining a simple fallback controller reduces dependence on communication networks and predictive models.

12.8 Risks, governance and professional responsibility

Model risk arises from incorrect equations, inappropriate assumptions, poor calibration and misuse outside the validated domain. Operational risk increases when automated recommendations are accepted without human review, especially during abnormal conditions. Cybersecurity risk includes unauthorized access, credential theft, malware, manipulation of sensor values and disruption of control communications.

A defence-in-depth architecture should include network segmentation, least-privilege access, multifactor authentication where appropriate, secure remote access, encryption, asset inventories, patch management, backups, logging, anomaly monitoring and tested incident-response procedures. Safety functions should fail safely and should not depend on an unsecured cloud connection. Cybersecurity design must respect the availability and real-time requirements of operational technology.

Data governance covers ownership, access, retention, quality, lineage, versioning and permissible use. Sensor data may reveal commercial operations or information about consumers. Personal or commercially sensitive data should be minimized, protected and shared only for legitimate purposes. Data dictionaries and common naming conventions reduce ambiguity when combining SCADA, maintenance, GIS and market data.

Ethically, engineers must disclose uncertainty and avoid presenting a forecast as certainty. Automated maintenance systems should not conceal adverse evidence to meet performance targets. Decisions affecting communities should account for land use, environmental justice, local employment and reliability consequences. Professional accountability remains with the organization and engineers deploying the system; an algorithm is not a substitute for responsibility.

12.9 Current developments and best practice

Emerging work includes probabilistic digital twins, physics-informed neural networks, federated learning across asset owners, graph-based models for networked energy systems, reinforcement learning for dispatch and foundation models for multimodal operational data. These approaches are promising but require careful control of generalization, computational cost, cyber risk and explainability. Reinforcement learning, in particular, should not be allowed to explore unsafe actions on a live plant without a verified simulator, constraint mechanism and supervisory safeguards.

Effective practice is characterized by the following principles:

• Begin with the decision and its consequences, not with an attractive algorithm.
• Establish a traceable data pipeline and a documented model purpose.
• Use physical constraints and engineering baselines wherever possible.
• Validate on independent periods, rare events and changing operating regimes.
• Quantify uncertainty and communicate it to decision-makers.
• Separate advisory analytics from safety-critical automatic control.
• Use Pareto analysis or explicit stakeholder weighting for conflicting objectives.
• Monitor model drift, recalibrate when required and retain version history.
• Design cybersecurity and data governance from the beginning.
• Include operators in development, testing and acceptance.

Chapter Summary

System modelling, optimization and digital intelligence form an integrated basis for advanced renewable-energy engineering. Physics-based models preserve conservation laws and component meaning; statistical and machine-learning models capture empirical behaviour; hybrid models combine both strengths. Optimization supports capacity sizing and dispatch through objective functions, constraints, uncertainty treatment and Pareto analysis. Simulation must be verified and validated for its intended purpose.

SCADA and IoT provide operational connectivity, GIS supports spatial planning, and digital twins link live data with models for diagnosis and decision support. Machine learning is valuable for forecasting, anomaly detection and predictive maintenance, but only when data leakage, imbalance, drift and uncertainty are addressed. Cybersecurity, explainability, privacy, data lineage and human accountability are essential engineering requirements. The most reliable solutions are not necessarily the most complex: they are transparent, validated, secure and operationally actionable.

Key Terms

Physics-based model: A model derived from physical laws, component characteristics and conservation principles.

Statistical model: A model that estimates relationships and patterns from observed data.

Hybrid model: A model combining physical equations with statistical or machine-learning components.

State variable: A quantity describing the condition of a dynamic system at a given time.

Digital twin: A data-connected digital representation of a physical asset or system used for monitoring, prediction or control.

SCADA: Supervisory Control and Data Acquisition system for monitoring, recording and controlling industrial equipment.

IoT: Networked sensing and communication architecture enabling devices to exchange data and services.

GIS: Geographic Information System used to analyse spatial data and support site and infrastructure decisions.

Objective function: Mathematical measure that an optimization process seeks to minimize or maximize.

Constraint: Equality or inequality condition limiting feasible decisions.

Pareto front: Set of non-dominated solutions representing trade-offs among competing objectives.

Stochastic optimization: Optimization incorporating uncertain outcomes or scenarios.

Robust optimization: Optimization seeking decisions that remain acceptable under defined uncertainty sets.

Model calibration: Estimation or adjustment of model parameters using observations.

Model validation: Assessment of whether a model is sufficiently representative for its intended use.

Overfitting: Learning historical data so closely that performance on new data deteriorates.

Predictive maintenance: Maintenance based on predicted equipment condition or failure risk.

Model drift: Decline in model performance caused by changes in data, equipment or operating conditions.

Explainable AI: Methods that make the basis or influence of algorithmic predictions more understandable.

Data governance: Policies and controls governing data quality, ownership, access, security and lifecycle.

Review Questions

Short Answer Questions

1. Why must model fidelity be matched to the time scale and purpose of an energy decision?

2. Distinguish verification from validation in renewable-energy simulation.

3. What is a Pareto-optimal solution?

4. State two differences between SCADA and an IoT analytics platform.

5. Why is data leakage particularly dangerous when evaluating predictive-maintenance models?

Long Answer Questions

1. Explain the development of a physics-based and hybrid model for a photovoltaic-battery system, including variables, parameters, equations and validation.

2. Describe how a multi-objective optimization problem for renewable-system sizing can be formulated and interpreted.

3. Discuss the architecture and engineering value of a digital twin for a wind farm or microgrid.

4. Explain a complete workflow for machine-learning-based predictive maintenance, including data, features, validation and decision-making.

5. Analyse the cybersecurity, data-governance and explainability requirements of an IoT-connected renewable-energy plant.

Analytical / Application Questions

1. A microgrid has high PV curtailment at midday but frequent diesel operation after sunset. Propose decision variables, objectives and constraints for an optimization model, and explain two uncertainties that should be represented.

2. A wind-turbine anomaly detector reports many alarms during high-wind events but few confirmed failures. Diagnose possible causes and design a validation procedure before deployment.

3. Two designs have equal annual cost: Design A has lower emissions but greater battery cycling, while Design B has higher emissions and lower replacement risk. Explain how Pareto analysis, reliability requirements and stakeholder preferences could guide the final selection.

22. Multiple Choice Questions with Answers

Question 1

Which model type most directly enforces conservation laws while representing component behaviour?

A. Physics-based model

B. Purely statistical model

C. Unsupervised clustering model

D. Market-price model

Correct Answer: A. Physics-based model

Question 2

In a hybrid microgrid, which equation represents the essential instantaneous energy balance?

A. PV output equals rated output at all times

B. Generation and grid import plus discharge equal load, charging and curtailment

C. Battery state of charge equals irradiance

D. Fuel consumption equals battery capacity

Correct Answer: B. Generation and grid import plus discharge equal load, charging and curtailment

Question 3

A Pareto-optimal solution is one that:

A. Has the lowest value for every objective simultaneously

B. Uses no constraints

C. Cannot improve one objective without worsening at least one other objective

D. Always has the lowest capital cost

Correct Answer: C. Cannot improve one objective without worsening at least one other objective

Question 4

What is the principal purpose of independent-period testing?

A. To increase the number of model parameters

B. To assess performance on data not used for calibration

C. To eliminate all uncertainty

D. To replace engineering validation

Correct Answer: B. To assess performance on data not used for calibration

Question 5

Which feature most clearly distinguishes a digital twin from a static simulation model?

A. It always uses artificial intelligence

B. It is connected to operational data and updated for a physical asset

C. It cannot represent uncertainty

D. It is used only for annual energy calculations

Correct Answer: B. It is connected to operational data and updated for a physical asset

Question 6

Why can a point forecast be inadequate for reserve scheduling?

A. It contains too many probability distributions

B. It conceals forecast uncertainty and extreme-error risk

C. It cannot be generated from weather data

D. It automatically causes cyberattacks

Correct Answer: B. It conceals forecast uncertainty and extreme-error risk

Question 7

A predictive-maintenance model gives excellent historical accuracy but fails after a control-system change. This is most directly an example of:

A. Model drift or distribution shift

B. Pareto dominance

C. GIS exclusion

D. Battery round-trip efficiency

Correct Answer: A. Model drift or distribution shift

Question 8

Which practice best protects safety-critical operation in a cloud-connected plant?

A. Allowing cloud analytics to replace local protection

B. Using a fallback local controller and separating analytics from safety functions

C. Disabling all alarms during communication loss

D. Sharing administrator credentials among operators

Correct Answer: B. Using a fallback local controller and separating analytics from safety functions

Question 9

In GIS-based renewable siting, protected areas are normally treated as:

A. Hard exclusion constraints or restricted zones

B. Forecast-error metrics

C. Battery state variables

D. Machine-learning labels

Correct Answer: A. Hard exclusion constraints or restricted zones

Question 10

Why should a model’s intended use be defined before validation?

A. Validation accuracy has the same meaning for every application

B. Acceptable error and relevant tests depend on the decision and operating regime

C. It removes the need for calibration

D. It guarantees a Pareto-optimal design

Correct Answer: B. Acceptable error and relevant tests depend on the decision and operating regime