Digitalization, Modeling and Artificial Intelligence

Introduction and context

Renewable-energy systems are increasingly operated as cyber-physical systems: physical equipment produces, converts, stores, and consumes energy, while digital networks measure conditions, predict behavior, optimize decisions, and issue control commands. A modern wind farm, photovoltaic plant, battery system, microgrid, or district-energy network therefore combines mechanical and electrical engineering with sensing, communications, software, statistics, and cybersecurity.

Digitalization is not simply the replacement of paper records by computers. It involves converting physical processes into observable data, connecting assets through communication networks, creating mathematical or computational models, and using the resulting information to improve safety, reliability, efficiency, flexibility, and economic performance. Artificial intelligence (AI) is one group of methods within this broader digital system. It can identify patterns and support decisions, but it does not eliminate the need for physical understanding, engineering judgment, or accountable human oversight.

The central engineering challenge is to connect four layers:

1. Physical layer: turbines, modules, inverters, generators, transformers, batteries, pumps, buildings, and loads.
2. Measurement and control layer: sensors, actuators, programmable logic controllers (PLCs), supervisory control and data acquisition (SCADA), and protection systems.
3. Data and model layer: databases, communication protocols, digital twins, forecasting models, optimization, and analytics.
4. Decision and governance layer: dispatch, maintenance planning, market participation, cybersecurity, privacy, safety, and regulatory compliance.

A technically impressive model can still fail if sensors are poorly calibrated, timestamps are inconsistent, communication is unreliable, or operators cannot interpret its recommendations. Successful digital energy engineering is therefore a systems-engineering discipline.

Learning objectives

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

• distinguish digitalization, automation, data analytics, machine learning, and artificial intelligence;
• describe the architecture and function of SCADA, IoT sensing, and industrial communication systems;
• formulate simplified physical, statistical, and hybrid models of renewable-energy assets;
• explain digital twins and their relationship to real-time data and simulation;
• apply forecasting, optimization, predictive-maintenance, and fault-detection methods;
• evaluate model accuracy using appropriate metrics and validation procedures;
• identify limitations involving uncertainty, bias, explainability, cybersecurity, and data governance;
• recommend a responsible implementation strategy for an energy asset or microgrid.

From measurement to intelligent action

Sensors, IoT, and SCADA

A sensor converts a physical variable into a measurable signal. Typical renewable-energy measurements include irradiance, module temperature, wind speed and direction, rotor speed, vibration, bearing temperature, electrical current and voltage, state of charge, pressure, flow rate, and ambient conditions. Sensor selection must consider range, resolution, accuracy, response time, environmental rating, calibration, drift, and failure behavior.

The Internet of Things (IoT) describes connected devices that collect, communicate, and sometimes process data. In energy systems, edge devices may filter or aggregate data locally before sending it to a supervisory platform. This reduces bandwidth and can maintain operation when communication with a cloud service is interrupted.

SCADA is an industrial supervisory system. It normally includes field sensors and actuators, remote terminal units or PLCs, a communication network, a historian for time-series data, human-machine interfaces, alarms, and supervisory commands. SCADA is designed for dependable monitoring and control, whereas a cloud analytics platform may be designed primarily for storage and large-scale computation. They should not be treated as interchangeable.

A robust data record should include at least a value, unit, timestamp, asset identifier, quality flag, and sometimes calibration or provenance information. Time synchronization is essential. If wind speed is recorded at 10:00:00 and power at 10:00:08, a naïve correlation may suggest a false physical relationship.

Data quality and preprocessing

Common data-quality problems include missing values, duplicated records, outliers, stuck sensors, range violations, communication gaps, and changes in sampling frequency. Preprocessing should be based on engineering knowledge rather than blind deletion. A negative solar irradiance value may indicate a sensor or timing fault, while a negative power value may be physically valid during auxiliary consumption.

For a measured variable x, a simple standardized score is:

where x_i is an observation, μ is the sample mean, and σ is the sample standard deviation. This score is useful for screening approximately symmetric data, but it is not a universal fault detector. Nonstationary operation, seasonal changes, and heavy-tailed distributions can make fixed thresholds misleading.

Aggregating data also changes its meaning. The average of power measurements over an interval estimates average power, while the sum of energy-meter readings or the integral of power over time estimates energy. If power is in kilowatts and time is in hours, energy is in kilowatt-hours:

For discrete samples at equal interval Δt, an approximation is:

The integral assumes a consistent definition of the sign convention and compatible units. If Δt is measured in seconds, power in watts gives joules; conversion is required before reporting kilowatt-hours.

Modeling renewable-energy systems

A model is a purposeful representation of a system. No model is universally best: a model suitable for real-time control may be too simple for design studies, while a detailed computational-fluid-dynamics model may be too slow for dispatch.

Physical and engineering models

Physical models use conservation laws and constitutive relationships. For a photovoltaic array, a simplified power estimate may be written as:

where G is irradiance in W/m², A is active area in m², η_ref is reference efficiency, γ is the temperature coefficient of power per °C, T_cell is cell temperature, and T_ref is the reference temperature. The expression is an approximation: it neglects detailed electrical behavior, shading, mismatch, inverter limits, soiling, and wiring losses. The sign of γ is usually negative for crystalline silicon modules.

The mechanical power available in wind is:

where ρ is air density in kg/m³, A is swept area in m², v is wind speed in m/s, and C_p is the power coefficient, dependent on tip-speed ratio λ and blade pitch angle β. This is not the electrical output: gearbox, generator, converter, and availability losses must also be considered. The cubic dependence on wind speed explains why measurement error and wind-speed forecasting error can strongly affect power estimates.

Battery models range from an equivalent circuit to an electrochemical model. A simplified state-of-charge balance is:

where SOC is dimensionless, P_c is charging power, P_d is discharging power, η_c and η_d are charging and discharging efficiencies, Δt is the time step, and E_nom is nominal usable energy capacity. Charging and discharging powers are defined as nonnegative in this expression. Constraints must prevent simultaneous charging and discharging unless the model explicitly allows it. Real batteries also exhibit temperature dependence, rate dependence, degradation, hysteresis, and cell imbalance.

Statistical and machine-learning models

A forecasting model estimates an unknown or future quantity from available information. A linear regression can be expressed as:

where y_t is the target, x variables are predictors, β coefficients are estimated from data, and ε_t is the unexplained error. Regression is interpretable but may be inadequate for nonlinear relationships.

Time-series models use temporal dependence. A basic autoregressive model is:

where c is an intercept, φ coefficients describe dependence on previous observations, and ε_t is an error term. Such a model assumes that the selected lag structure is appropriate and that the process is sufficiently stable after any required detrending or seasonal treatment.

Machine-learning methods include decision trees, random forests, gradient boosting, support-vector methods, neural networks, clustering, and autoencoders. Supervised learning requires labelled examples; unsupervised learning seeks structure without target labels; reinforcement learning learns actions from rewards, but safety-critical energy applications require carefully constrained exploration.

Hybrid models and digital twins

A hybrid model combines physical equations with data-driven correction. For example, a physical PV model may estimate expected output while a learned residual model estimates systematic effects of soiling or shading. This approach can improve generalization and preserve physical plausibility, provided the residual model is trained on representative data.

A digital twin is a continuously updated digital representation of a physical asset or process. It normally contains:

• an asset model and configuration data;
• live or near-real-time measurements;
• state estimation and parameter updating;
• simulation or prediction capability;
• comparison of expected and observed behavior;
• a defined purpose, such as condition monitoring or dispatch support.

A static CAD model is not automatically a digital twin. Nor is a dashboard containing sensor values. A twin must have a meaningful relationship between the physical asset and the computational representation, including clear update rules and validation.

Forecasting, optimization, and energy management

Forecasting

Forecast horizons should match decisions. Seconds-to-minutes forecasts support control and ramp management; hours-ahead forecasts support unit commitment and storage scheduling; day-ahead forecasts support market bids and maintenance coordination; seasonal forecasts support planning.

Important metrics include mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE):

Here y_i is the observed value, ŷ_i is the forecast, and n is the number of observations. MAE and RMSE have the same units as the target; RMSE penalizes large errors more strongly. MAPE is undefined when y_i = 0 and unstable near zero, so it is often inappropriate for low-output periods. Forecasts should also provide prediction intervals or quantiles when uncertainty matters.

A model must be compared with a baseline, such as persistence forecasting, in which the next value is assumed equal to the latest observed value. A complex model that does not outperform a credible baseline under out-of-sample testing may not justify its cost or risk.

Optimization and dispatch

Energy management seeks feasible decisions that balance cost, emissions, reliability, degradation, and user requirements. A general optimization formulation is:

Here u represents decisions such as generator output, storage power, or flexible demand; x represents system states; w represents uncertain disturbances such as weather; f is the state-transition model; g represents operational constraints; and J is the objective function. The formulation is meaningful only when units, bounds, ramp rates, efficiencies, network limits, and reserve requirements are represented consistently.

For a battery microgrid, a simplified objective may include energy cost and a degradation proxy:

where c_t is grid-energy price, c_deg is an assumed degradation cost per unit throughput, c_curt is a curtailment penalty, and P_curt is curtailed renewable power. This is an engineering-economic approximation, not a universal battery-life model. If degradation is important, a more defensible model should represent temperature, depth of discharge, cycle history, and manufacturer evidence.

Deterministic optimization uses one forecast. Stochastic or robust optimization explicitly considers uncertainty. Model-predictive control repeatedly solves an optimization problem over a moving horizon, implements the first action, measures the new state, and solves again. This feedback makes it useful for changing conditions, but computational speed and model reliability are essential.

Predictive maintenance and fault detection

Corrective maintenance repairs equipment after failure. Preventive maintenance uses time or usage intervals. Predictive maintenance uses condition information and forecasts of degradation or failure risk. Its purpose is not to predict every failure perfectly; it is to improve decisions about inspection, spare parts, crew scheduling, and safe operation.

A practical workflow is:

1. define failure modes and their consequences;
2. identify measurable precursors, such as vibration, temperature, oil particles, or electrical signatures;
3. establish healthy operating envelopes;
4. clean and label historical data;
5. train and validate a model without leakage;
6. set alarm and escalation thresholds with operators;
7. verify findings through inspection or independent evidence;
8. monitor model drift and update the system under change control.

Fault detection identifies abnormal behavior; fault diagnosis identifies the likely cause; prognosis estimates future condition or remaining useful life. These are different tasks and should not be conflated.

For an anomaly score s, a threshold can produce an alarm when s exceeds τ. However, τ should reflect the costs of false positives and false negatives, operating regime, and safety consequences. A highly sensitive alarm may overwhelm operators, while a conservative threshold may miss dangerous degradation. Confusion-matrix measures such as precision and recall are useful, but class imbalance can make accuracy misleading.

Data leakage is a major error. If measurements collected after a known failure are accidentally included in training features for an earlier prediction, apparent performance will be unrealistically high. Randomly splitting time-series data can also leak future conditions into training. Chronological or blocked validation is usually more appropriate.

Cybersecurity, data governance, and responsible AI

Energy infrastructure is a safety- and security-critical environment. Threats include credential theft, ransomware, denial of service, manipulation of sensor data, malicious control commands, insecure remote access, and supply-chain compromise. An attacker need not shut down a plant; subtly falsified measurements can cause poor dispatch, equipment stress, or unsafe maintenance decisions.

Important controls include network segmentation between corporate IT, operational technology, and safety systems; least-privilege access; multifactor authentication; secure configuration; patch and vulnerability management; encrypted communication where appropriate; signed software; tested backups; incident response; logging; and manual-safe operating modes. Cybersecurity must be designed around availability and safety, not only confidentiality.

Data governance specifies who may collect, access, modify, retain, share, and delete data. It should address ownership, metadata, quality, lineage, retention, interoperability, privacy, and contractual responsibilities. High-frequency operational data may reveal commercially sensitive production patterns or information about building occupants. Data minimization and access control are therefore relevant even when data are not personally identifiable in an obvious form.

Responsible AI requires:

• Validity: the model measures the intended engineering quantity or decision objective.
• Reliability: performance remains acceptable under expected disturbances and degradation.
• Transparency: users can understand inputs, uncertainty, limitations, and escalation paths.
• Accountability: a named person or organization remains responsible for decisions.
• Fairness: impacts on different users, sites, or communities are not hidden by aggregated metrics.
• Human oversight: automated actions have appropriate approval, bounds, and override mechanisms.
• Security and privacy: models and data are protected from unauthorized manipulation or disclosure.

Explainability tools can show feature importance or local explanations, but an explanation is not proof of causation. A model may associate low power with high temperature because both occur during a particular season, not because temperature alone caused the power loss. Engineers should combine statistical explanation with physical tests and domain knowledge.

Evaluating models in practice

Model evaluation must reflect deployment conditions. A useful evaluation plan includes:

• a clearly defined target and operating envelope;
• a physically and economically meaningful baseline;
• separation of training, validation, and test data;
• chronological testing for time-dependent data;
• testing across seasons, weather regimes, equipment types, and failure modes;
• uncertainty and calibration assessment;
• sensitivity to missing, delayed, or manipulated data;
• computational latency and communication requirements;
• consequences of wrong decisions;
• post-deployment monitoring for drift.

The statistical error is only one part of value. A forecast with slightly lower RMSE may be less useful than a calibrated forecast that correctly identifies high-risk ramp events. Similarly, a maintenance classifier with high accuracy may have little operational value if it detects only common healthy cases.

Concept drift occurs when the relationship between inputs and outputs changes, for example after inverter replacement, climate change, control-software updates, or altered operating policy. Retraining should be governed, documented, tested, and reversible. Continuous automatic retraining without review can embed corrupted data or an attacker’s influence.

Worked example: battery dispatch and forecast error

Consider a battery with usable energy capacity E_nom = 100 kWh, initial SOC = 0.50, charging efficiency η_c = 0.95, and a one-hour charging command P_c = 20 kW. Ignore self-discharge and assume no simultaneous discharge. Using the SOC balance:

Thus, the predicted next state is 0.69, or 69%. The stored energy increase is 19 kWh rather than 20 kWh because of charging loss. The result is valid only if the command is within the battery’s power and SOC limits and if the usable capacity is correctly defined for the operating temperature and condition.

Suppose the actual renewable forecast errors over four intervals, in kW, are 2, −4, 1, and −3. The absolute errors are 2, 4, 1, and 3 kW. Therefore:

The example shows why storage scheduling should not rely only on expected renewable output. A robust schedule may reserve energy or power headroom for forecast errors, subject to the economic cost of holding that reserve.

Illustrative case: digital twin for a wind-farm substation

Background. A 60-MW wind farm experiences repeated transformer temperature alarms. Operators cannot determine whether the alarms indicate genuine thermal stress, sensor drift, or unusually high ambient temperature.

Problem. The existing SCADA system records oil temperature, load, ambient temperature, fan status, and transformer alarms at one-minute intervals. Sensor calibration records are incomplete, and the farm’s maintenance contractor holds some historical data in a separate database.

Analysis. Engineers first define a data dictionary, synchronize clocks, identify missing intervals, and compare sensor values with physical limits and redundant measurements. They create a thermal model based on load and ambient temperature, then compare predicted temperature with observations. A residual alarm is considered only when the residual persists for a defined period and the operating regime is comparable. This prevents a single transient from generating an unnecessary work order.

Approach and solution. The project implements an edge gateway that preserves local SCADA control while securely forwarding selected data to an analytics platform. The digital twin contains transformer ratings, cooling logic, thermal parameters, calibration metadata, and a residual-monitoring model. A dashboard displays measured temperature, expected temperature, confidence bands, alarm status, and data quality. The model does not directly trip the transformer; protective relays remain independent. When a persistent residual is detected, the system requests inspection of the sensor and cooling fans before recommending load derating.

Outcome. In this illustrative case, the first investigation identifies a drifting temperature sensor rather than transformer insulation failure. A later event reveals a fan-control problem. The twin therefore supports both false-alarm reduction and earlier detection, but only because model outputs are linked to inspection procedures and independent protection.

Learning points. Digital twins require data governance, physical validation, clear operating boundaries, and integration with work processes. Separating advisory analytics from safety-critical protection reduces the consequence of model error. The value comes from better decisions, not from the label “AI.”

Contemporary developments and professional practice

Edge AI is increasingly used where latency, bandwidth, or privacy makes cloud processing unsuitable. Physics-informed machine learning attempts to include conservation laws or operating constraints in training. Probabilistic forecasting and scenario generation support storage and flexibility decisions under uncertainty. Federated learning can train models across organizations without centralizing all raw data, although it does not automatically eliminate privacy or poisoning risks.

Interoperability is also important. Open data models and standardized interfaces can reduce vendor lock-in, but interoperability does not guarantee semantic consistency. Two systems may both report “power” while using different signs, time bases, or measurement points.

A responsible implementation typically follows a staged path:

1. articulate the operational decision and its value;
2. verify instrumentation and data quality;
3. build a transparent baseline model;
4. test a pilot without automatic control authority;
5. validate under representative and adverse conditions;
6. define cybersecurity, governance, and accountability;
7. integrate with maintenance and control procedures;
8. monitor performance, drift, and realized benefits.

For mechanical engineers, employability increasingly requires the ability to interpret sensor data, understand asset physics, communicate with data specialists, and judge whether a digital recommendation is physically plausible. Coding is useful, but disciplined problem formulation and safety-aware engineering remain fundamental.

Challenges and limitations

Digital systems have structural limitations. Sensors cannot measure what has not been instrumented, and historical labels often record only failures that were noticed. Renewable resources are variable and partly unobservable. Models trained at one site may not transfer to another because of different climates, equipment, maintenance histories, or control policies.

AI can be confidently wrong outside its training distribution. Correlation can be mistaken for causation. Black-box models may be difficult to certify or explain during an incident. Optimization can produce mathematically optimal but operationally unacceptable schedules if constraints or degradation costs are omitted. Connectivity creates attack surfaces, and automation can amplify a bad measurement rapidly.

These limitations imply a hierarchy of trust: use hard protection and verified physical constraints for safety-critical boundaries; use validated models and human review for operational recommendations; use exploratory analytics only where errors have controlled consequences. Every deployment should specify what the model may do, what it may not do, and who can override it.

Chapter Summary

Digitalization connects renewable-energy equipment, sensors, industrial control, data platforms, mathematical models, and organizational decisions. SCADA provides dependable supervision and control, while IoT and edge computing extend sensing and analytics. Data quality, time synchronization, metadata, and provenance are prerequisites for meaningful analysis.

Physical, statistical, machine-learning, and hybrid models serve different purposes. Digital twins combine an asset representation with live data, state estimation, and validated prediction. Forecasting supports control, markets, and planning; optimization supports dispatch and energy management; predictive maintenance and fault detection support reliability when their labels, thresholds, and validation are carefully designed.

Metrics such as MAE, RMSE, and MAPE must be selected with awareness of units and zero values. Chronological validation, credible baselines, uncertainty assessment, and drift monitoring are essential. Cybersecurity, data governance, explainability, accountability, human oversight, and safe fallback modes are not optional additions: they are engineering requirements. AI can improve renewable-energy systems, but responsible deployment depends on physical insight, explicit constraints, secure infrastructure, and continuous evaluation.

Key Terms

Digitalization: Integration of digital data, computation, communication, and control into physical operations.

SCADA: Supervisory control and data acquisition system for industrial monitoring and control.

IoT sensor: Connected sensing device that measures and communicates physical data.

Edge computing: Processing data near the source rather than exclusively in a remote cloud.

Data provenance: Record of a datum’s origin, transformations, timing, and ownership.

Digital twin: Dynamic digital representation linked to a physical asset or process.

Physical model: Model based on engineering laws and known system relationships.

Machine learning: Methods that infer patterns or relationships from data.

Hybrid model: Model combining physical equations with data-driven components.

Forecasting: Estimation of future values or events using current and historical information.

Optimization: Selection of feasible decisions to minimize or maximize an objective.

Model-predictive control: Receding-horizon optimization with repeated measurement and re-solving.

Predictive maintenance: Maintenance planning based on condition information and predicted risk.

Fault detection: Identification of behavior inconsistent with expected operation.

Data leakage: Use of information in model development that would not be available at deployment.

Concept drift: Change over time in the relationship represented by a model.

Cyber-physical system: Integrated physical process, computation, communication, and control system.

Data governance: Rules and responsibilities for data quality, access, security, retention, and use.

Explainability: Ability to communicate how or why a model produced an output.

Responsible AI: Development and use of AI with attention to validity, safety, fairness, accountability, security, and human oversight.

Review Questions

Short Answer Questions

1. What are the main functional layers of a digital renewable-energy system?
2. Why are timestamps and data-quality flags important in SCADA data?
3. Distinguish fault detection, fault diagnosis, and prognosis.
4. Why can MAPE be inappropriate when the measured value is zero or close to zero?
5. State two cybersecurity controls suitable for operational technology.

Long Answer Questions

1. Explain the architecture and roles of sensors, PLCs, SCADA, historians, edge devices, and cloud analytics in a renewable-energy plant.
2. Compare physical, statistical, machine-learning, and hybrid models, including one suitable application and one limitation of each.
3. Explain how a digital twin can support predictive maintenance while remaining separate from safety-critical protection systems.
4. Discuss how chronological validation, baselines, uncertainty, and concept-drift monitoring should be used to evaluate an energy forecast.
5. Analyze the governance and responsible-AI requirements for an automated microgrid dispatch system.

Analytical / Application Questions

1. A battery has 200 kWh usable capacity, SOC = 0.40, charging efficiency 0.90, and receives a 30-kW charging command for two hours. Calculate the next SOC, ignoring limits and self-discharge, and identify two assumptions that must be checked in practice.
2. A wind-turbine anomaly model produces many alarms during high-wind operation but few during low-wind operation. Design an investigation using physical modeling, data segmentation, and validation to determine whether the alarms are meaningful.
3. Propose a secure and responsible digital architecture for a solar-plus-storage microgrid serving a hospital. Identify which functions should remain locally controllable, what data should be governed, and how AI recommendations should be supervised.

Multiple Choice Questions with Answers

Question 1

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

A. It contains a graphical user interface.

B. It is continuously related to a physical asset through data and update rules.

C. It is stored in a cloud database.

D. It uses artificial intelligence in every calculation.

Correct Answer: B. It is continuously related to a physical asset through data and update rules.

Question 2

A forecast has observed values 10, 20, and 30 kW and forecasts 12, 18, and 27 kW. What is the MAE?

A. 1 kW

B. 5 kW

C. 7 kW

D. 2.33 kW

Correct Answer: A. 1 kW

Question 3

Why is MAPE often unsuitable for photovoltaic power during night-time periods?

A. It has units of kilowatts.

B. It cannot be calculated from measured data.

C. It becomes undefined or unstable when actual power is zero or near zero.

D. It always penalizes large errors less than MAE.

Correct Answer: C. It becomes undefined or unstable when actual power is zero or near zero.

Question 4

Which practice best prevents temporal data leakage in a forecasting study?

A. Randomly shuffle all observations before splitting them.

B. Use future observations to balance the training set.

C. Split data chronologically and test on later observations.

D. Remove all observations containing missing values after testing.

Correct Answer: C. Split data chronologically and test on later observations.

Question 5

In the simplified battery SOC equation, what does η_c represent?

A. Charging efficiency

B. Discharging power

C. Nominal energy capacity

D. Sampling interval

Correct Answer: A. Charging efficiency

Question 6

Which statement about SCADA is most accurate?

A. It is only a cloud-based machine-learning service.

B. It commonly combines field measurements, communications, alarms, historians, interfaces, and supervisory control.

C. It replaces all independent protection systems.

D. It requires every sensor to be wireless.

Correct Answer: B. It commonly combines field measurements, communications, alarms, historians, interfaces, and supervisory control.

Question 7

What is concept drift?

A. A permanent failure of a sensor communication cable.

B. A change over time in the relationship represented by a model.

C. A method for encrypting SCADA traffic.

D. The conversion of alternating current to direct current.

Correct Answer: B. A change over time in the relationship represented by a model.

Question 8

Which cybersecurity measure most directly limits the effect of a compromised user account?

A. Increasing dashboard color contrast

B. Least-privilege access and multifactor authentication

C. Using a larger training dataset

D. Increasing the forecast horizon

Correct Answer: B. Least-privilege access and multifactor authentication

Question 9

Why should an AI maintenance recommendation normally be verified before a major intervention?

A. AI models cannot process numerical data.

B. Model outputs may be uncertain, biased by data quality, or outside their training conditions.

C. Physical inspections are never useful.

D. Verification always reduces equipment availability to zero.

Correct Answer: B. Model outputs may be uncertain, biased by data quality, or outside their training conditions.

Question 10

A battery starts at SOC 0.40, has capacity 200 kWh, charges at 30 kW for 2 h, and has charging efficiency 0.90. Ignoring limits, what is its next SOC?

A. 0.54

B. 0.60

C. 0.67

D. 0.94

Correct Answer: C. 0.67