Reliability, Resilience, Planning and Asset Management

Introduction

Modern electric power systems must deliver energy continuously despite equipment failures, uncertain demand, variable renewable generation, cyber-physical threats and extreme weather. Reliability, resilience, planning and asset management provide complementary perspectives on this obligation.

Reliability concerns the probability that a system performs its intended function under stated conditions for a specified period. Resilience addresses the ability to anticipate, withstand, adapt to and recover rapidly from high-impact disturbances. Planning determines economically and technically appropriate future investments. Asset management coordinates the life-cycle performance, risk and cost of physical infrastructure.

These disciplines increasingly require integrated treatment. A highly reliable system under routine conditions may remain vulnerable to correlated failures during floods, wildfires or cyberattacks. Conversely, excessive redundancy can be economically inefficient and environmentally burdensome. Sound engineering therefore balances service quality, affordability, safety, sustainability and risk.

Learning Objectives

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

• distinguish reliability, adequacy, security, resilience and asset-management concepts;
• calculate and interpret generation and distribution reliability indices;
• construct and use a capacity outage probability table;
• explain probabilistic, deterministic and risk-based planning methods;
• formulate resilience and capacity-expansion metrics;
• apply condition-, risk- and reliability-centred asset-management principles;
• evaluate investments using life-cycle cost, failure probability and consequence;
• recognize regulatory, ethical, data-governance and climate-related challenges.

1. Reliability Foundations

Power-system reliability is commonly divided into adequacy and security.

• Adequacy is the ability of installed generation, networks and supporting resources to meet aggregate demand, including scheduled and reasonably expected outages.
• Security is the ability to withstand sudden disturbances, such as short circuits, generator trips or transmission-line outages, without violating unacceptable operating limits.

Reliability analysis may be conducted hierarchically:

1. Hierarchical Level I: generation adequacy only;
2. Hierarchical Level II: composite generation and transmission adequacy;
3. Hierarchical Level III: the complete system, including distribution.

A component is often represented by an operating state and a failed state. If failures and repairs follow stationary exponential processes, with failure rate λ and repair rate μ, the steady-state availability is

and unavailability is

Here, MTTF is mean time to failure and MTTR is mean time to repair. All time quantities must use consistent units. These equations assume a repairable two-state component, constant transition rates and statistical stationarity. They are not generally valid for ageing components with nonconstant hazard rates.

When λ × MTTR ≪ 1, and λ is expressed in failures per unit time while MTTR uses the same time unit, the approximation U ≈ λ × MTTR is useful.

1.1 Forced-Outage Rate

For a generating unit, the forced-outage rate may be estimated from observed operating history as

where FOH is forced-outage hours and SH is service hours. Depending on the reporting standard, reserve shutdown hours and outage classifications may require separate treatment. The equivalent forced-outage rate on demand extends the concept to partial deratings and demand periods.

1.2 Independent and Common-Cause Failures

For independent components, joint-state probabilities are products of individual probabilities. For two components having unavailabilities q₁ and q₂, the probability that both are unavailable is q₁q₂.

This multiplication is invalid when failures share causes such as flooding, substation fires, protection misoperation, fuel interruption or communication loss. Common-cause and spatially correlated hazards must be modelled explicitly through conditional probabilities, event trees, scenario models, copulas or hazard-dependent fragility curves.

2. Reliability Indices

2.1 Distribution-System Indices

Let N_i be the number of customers interrupted in event i, r_i its interruption duration in hours, and N_T the total customers served.

The System Average Interruption Frequency Index, SAIFI, has units of interruptions per customer over the reporting period.

The System Average Interruption Duration Index, SAIDI, is usually reported in hours or minutes per customer per year.

The Customer Average Interruption Duration Index, CAIDI, measures average restoration time per customer interruption, provided SAIFI > 0.

Here, T is the number of hours in the reporting period. ASAI is dimensionless. Momentary interruptions are normally reported separately using indices such as MAIFI because mixing sustained and momentary events can distort interpretation.

2.2 Energy-Based Indices

Energy not supplied is

where P_i^uns is curtailed power in MW and Δt_i is duration in hours. ENS is measured in MWh. Average energy not supplied is AENS = ENS/N_T, usually expressed in kWh per customer.

Customer-average indices may conceal unequal impacts. Hospitals, low-income households, remote communities and industrial plants can experience consequences that are not represented adequately by a system average.

3. Generation Adequacy and the COPT

A capacity outage probability table (COPT) lists possible unavailable-capacity states and their probabilities. For independent two-state units, it can be constructed recursively by convolving each unit's available and unavailable states. Multi-state units require derated states as well.

At time interval t, available capacity is C_t and demand is L_t. The shortage is

The loss-of-load probability is

For intervals of duration Δt, expected loss-of-load duration is

LOLE is measured in hours per study period. In daily peak-load models it may instead be reported in days per year; these formulations should not be mixed without conversion.

Expected energy not served is

EENS has units of MWh when shortage is in MW and time is in hours. Unlike LOLE, it reflects shortage magnitude.

Worked Example: Correct COPT-Based EENS

An aggregated COPT for a 100 MW system is shown below. It may represent an empirically derived multi-state generating portfolio. Demand remains at 90 MW for 100 hours. State probabilities are mutually exclusive and sum to one.

Thus, the expected shortage power is

For the 100-hour period,

The probability of any shortage is 0.20676 + 0.08000 + 0.05000 + 0.00500 = 0.34176. Therefore, if every state persists for the full interval in this simplified model, expected loss-of-load duration is 0.34176 × 100 = 34.176 h. Chronological models are required when storage, ramping, weather-dependent renewable output or outage duration matters.

4. Reliability Versus Resilience

Reliability analysis traditionally emphasizes relatively frequent, statistically characterizable events. Resilience analysis emphasizes disruptive events that may be rare, severe, evolving and correlated.

Figure: A resilience curve showing disturbance absorption and recovery.

If Q(t) is a dimensionless performance measure and Q₀(t) is the target performance, resilience loss over [t₀,t₁] can be represented by

where [x]_+ = max(0,x). Smaller R_L indicates better performance for the specified event and horizon. This metric is scenario-dependent and should not be treated as a universal resilience score.

Resilience interventions include substation flood barriers, undergrounding selected circuits, vegetation management, mobile transformers, black-start resources, microgrids, controlled islanding, diversified communications, spare-equipment strategies and mutual-assistance agreements.

5. Power-System Planning

Planning connects forecasts and policy objectives to investment decisions over horizons ranging from one year to several decades.

5.1 Deterministic Criteria

The N−1 criterion requires the system to withstand the credible loss of one specified component without unacceptable consequences, following allowed corrective actions. N−1 does not guarantee zero interruption, nor does it represent simultaneous or common-cause failures. More severe criteria, such as N−1−1 or selected N−2 assessments, may be used for critical systems.

A traditional planning reserve margin is

where C_firm is firm capacity and L_peak is forecast peak demand. Installed renewable nameplate capacity should not automatically be treated as firm capacity.

5.2 Capacity Credit and ELCC

The effective load-carrying capability (ELCC) of a resource is the additional load that can be supplied after adding that resource while retaining the original reliability criterion. If baseline load is L and the resource permits load L + ΔL at the same LOLE or EENS target, its ELCC is ΔL. ELCC depends on resource mix, demand profile, weather correlation, storage duration and network constraints.

5.3 Capacity-Expansion Optimization

A simplified planning model minimizes discounted investment and operating costs:

subject to power balance, capacity, reserve, emissions, reliability, construction and network constraints. Here, x_j,y is investment in technology j during year y, I_j,y is unit investment cost, u represents operating decisions, O_y is annual operating cost, d is the real discount rate and Y is the planning horizon.

The objective is valid only when all cash flows are expressed consistently in real or nominal terms and occur at the assumed timing. Reliability may be imposed as a constraint, such as LOLE ≤ LOLE^max, or monetized using the value of lost load:

Here, VOLL is expressed in currency per MWh and EENS in MWh. VOLL varies by customer, event duration, warning time and season; a single value is a simplification.

5.4 Uncertainty and Risk

Planning uncertainties include demand growth, fuel prices, technology costs, policy, climate, outages and renewable production. Suitable methods include:

• scenario analysis for structurally different futures;
• Monte Carlo simulation for probabilistic inputs;
• stochastic programming for decisions before and after uncertainty is revealed;
• robust optimization for solutions that remain feasible over an uncertainty set;
• real-options analysis for the value of flexibility and deferred commitment.

Expected cost alone may underweight catastrophic outcomes. Conditional value at risk (CVaR) can represent average loss in the worst specified probability tail and may be combined with expected cost in a risk-averse objective.

6. Asset Management

Power-system asset management is the coordinated activity through which an organization realizes value from generators, transformers, switchgear, cables, lines, protection systems and digital infrastructure. Value incorporates performance, risk, expenditure, safety and sustainability throughout the life cycle.

6.1 Asset Life-Cycle Activities

Figure: Closed-loop, risk-based asset-management process.

Policies include:

• corrective maintenance: intervention after failure;
• time-based preventive maintenance: intervention at scheduled intervals;
• condition-based maintenance: intervention triggered by measured condition;
• predictive maintenance: forecasting future condition or failure using models;
• reliability-centred maintenance: selecting tasks according to functions, failure modes and consequences;
• risk-based maintenance: prioritizing interventions according to probability and consequence.

6.2 Ageing and Weibull Analysis

For assets whose time to failure follows a Weibull distribution, reliability is

and the hazard rate is

where t is age, η is the scale parameter and β is the shape parameter. For β > 1, hazard increases with age; for β = 1, hazard is constant; and for β < 1, hazard decreases. These interpretations require a homogeneous population and appropriate treatment of censored data, refurbishment and operating conditions.

Calendar age alone is rarely a sufficient condition indicator. Transformer health, for example, may combine dissolved-gas analysis, moisture, oil quality, thermal history, loading, insulation ageing and bushing condition.

6.3 Risk and Annual Cost Consistency

A basic asset-risk measure is

The probability horizon must be specified. If p_f,yr is the probability of at least one relevant failure during one year, a simplified expected annual cost is

where C_planned,yr is annual planned expenditure and C_f is the monetary consequence per annual failure event. This approximation is appropriate when at most one consequential failure per year is modelled or when the probability-consequence structure explicitly represents the annual event.

If failures follow a Poisson process with annualized rate λ_yr and each failure has mean consequence C_f, the expected annual failure cost is instead

The probability of at least one failure is then p_f,yr = 1 - exp(-λ_yr). Substituting this probability for the expected number of failures would understate cost when multiple failures per year are possible.

Worked Asset Decision

An ageing transformer has annual planned maintenance cost of $20,000, annual probability of a single consequential failure of 0.04, and failure consequence of $2.0 million. Its expected annual cost is

A monitoring and refurbishment programme costs $60,000 per year, reduces annual failure probability to 0.015 and reduces consequence to $1.4 million through earlier intervention:

Under the stated single-annual-event model, the programme reduces expected annual cost by $19,000. The decision must still consider uncertainty, safety, environmental damage, regulatory obligations and the possibility of common-mode station outages.

7. Integrated Illustrative Case: Coastal Grid Modernization

Background

A fictional coastal utility supplies 420,000 customers through two major substations, radial coastal feeders and an ageing transformer fleet. Storm-related outages have increased, while electrification is raising winter peak demand.

Problem

The utility's routine SAIFI is acceptable, but major storms produce prolonged interruptions. Flood mapping indicates that both major substations can be affected by one event, invalidating independent-failure assumptions. Several transformers also show elevated dissolved gases.

Analysis

The utility combines four studies:

1. a chronological adequacy model incorporating load, wind, solar, storage and forced outages;
2. flood-conditioned substation fragility curves and spatial storm scenarios;
3. feeder-level critical-load mapping for hospitals, water systems and shelters;
4. transformer health indices linked to annual failure probabilities and consequences.

The analysis finds that adding generation alone reduces routine LOLE but has little effect when flooded substations disconnect load. Selective flood hardening and network reconfiguration reduce both EENS and recovery time. A mobile transformer provides flexibility across multiple contingencies, while a community microgrid preserves essential services during upstream outages.

Approach

The adopted portfolio includes raised control buildings, flood walls, automated feeder switches, one mobile transformer, condition-based transformer maintenance, a critical-services microgrid and annual restoration exercises. Investments are staged so that updated climate and condition information can influence later commitments.

Outcome

Scenario simulation indicates lower expected annual EENS and a substantial reduction in critical-load interruption during severe flood scenarios. The portfolio costs more than a generation-only option but performs better against correlated infrastructure failure and uncertain hazard escalation.

Learning Points

• Adequacy investment cannot compensate for every network vulnerability.
• Correlated hazards must not be represented as independent outages.
• Critical-load continuity and restoration time complement average reliability indices.
• Flexible and staged investments have option value under uncertainty.
• Asset condition, network planning and emergency management should share data and objectives.

8. Digitalization, Analytics and AI

Online sensors, phasor measurement units, smart meters, drones, digital substations and weather services generate data for asset and system decisions. Machine-learning applications include anomaly detection, vegetation assessment, outage prediction, load forecasting and remaining-life estimation.

However, predictive accuracy does not automatically imply operational value. Models may fail under changing climate, maintenance practices or sensor configurations. Utilities should control data quality, document training populations, quantify uncertainty, test for distribution shift and retain human accountability. Cybersecurity is essential because compromised sensors or maintenance recommendations can create physical risk.

Digital twins can combine network models, condition data and operational states, but their usefulness depends on calibration and model governance. A visually sophisticated digital representation is not necessarily a validated engineering model.

9. Regulation, Governance and Ethics

Reliability standards specify planning and operating obligations, but compliance is not identical to optimal risk management. Governance should establish risk appetite, decision authority, data ownership, auditability and escalation procedures.

Important ethical and regulatory issues include:

• equitable allocation of reliability and resilience investments;
• transparent outage reporting and index exclusions;
• privacy protection for smart-meter and customer data;
• worker and public safety;
• environmental consequences of equipment failure and replacement;
• cybersecurity responsibilities across utilities and vendors;
• avoidance of discriminatory load shedding;
• recognition of vulnerable and medically dependent customers.

Performance incentives require careful design. A narrow focus on SAIDI can encourage gaming through event exclusions or prioritization of easily restored customers. Balanced regulation should consider routine performance, extreme events, critical services, long-term asset condition and affordability.

10. Limitations and Best Practices

Reliability and asset models are abstractions. Historical failure data may be sparse, censored or nonstationary. Climate change challenges assumptions based on past hazards. Interdependencies with telecommunications, fuel, transport and water systems can amplify disruption. Monetary consequence estimates may omit social and health impacts.

Good professional practice includes:

1. defining the decision, horizon and performance criteria before modelling;
2. maintaining consistent time and monetary units;
3. separating aleatory variability from epistemic uncertainty where practical;
4. testing common-cause and extreme-event scenarios;
5. validating models against observations and engineering judgement;
6. reporting sensitivity ranges rather than only point estimates;
7. combining system, customer and critical-service measures;
8. coordinating planning, operations, emergency management and asset teams;
9. documenting assumptions and maintaining auditable data lineage;
10. reviewing decisions as conditions and evidence change.

Chapter Summary

Reliability describes the ability of a power system to perform its required function and encompasses adequacy and security. Distribution reliability is measured using indices such as SAIFI, SAIDI, CAIDI and ENS, while generation adequacy commonly uses LOLP, LOLE and EENS. Capacity outage probability tables connect outage states to shortage probabilities and magnitudes.

Resilience extends conventional reliability by considering preparation, absorption, adaptation and recovery under severe, often correlated disturbances. Planning combines deterministic criteria such as N−1 with probabilistic adequacy, optimization and uncertainty analysis. Asset management coordinates condition, performance, risk and life-cycle expenditure. Annual cost comparisons must use annual failure probabilities or annualized event rates consistently. Integrated decision-making is increasingly important as ageing infrastructure, climate hazards, digitalization and distributed resources transform power systems.

Key Terms

Adequacy: Ability of available resources and networks to meet aggregate demand.

Asset criticality: Importance of an asset based on the consequences of its failure.

Availability: Long-run proportion of time that a repairable component is operational.

Capacity credit: Contribution of a resource to meeting system adequacy requirements.

COPT: Table of capacity-outage states and their probabilities.

Common-cause failure: Multiple failures produced by a shared initiating cause.

Condition-based maintenance: Maintenance initiated using observed asset condition.

EENS: Expected energy not served over a specified period.

ELCC: Additional load that can be served after adding a resource at unchanged reliability.

Failure rate: Expected failure-event frequency per unit time under specified assumptions.

Fragility curve: Conditional probability of damage or failure as a function of hazard intensity.

LOLE: Expected duration for which demand exceeds available capacity.

N−1 criterion: Requirement to withstand one specified credible component outage.

Reliability: Probability of satisfactory performance under stated conditions and time.

Resilience: Ability to anticipate, withstand, adapt to and recover from disruption.

Risk-based maintenance: Maintenance prioritized using failure likelihood and consequence.

SAIDI: Average sustained interruption duration per customer over a reporting period.

SAIFI: Average number of sustained customer interruptions over a reporting period.

Value of lost load: Monetary estimate of customer loss per unit of unserved energy.

Weibull distribution: Flexible lifetime distribution used to model age-dependent failure behaviour.

Review Questions

Short Answer Questions

1. Distinguish power-system adequacy from security.
2. What information does EENS provide that LOLE does not?
3. Why can independent-failure assumptions be inappropriate during extreme weather?
4. Define effective load-carrying capability.
5. Distinguish condition-based maintenance from time-based preventive maintenance.

Long Answer Questions

1. Explain the construction and application of a capacity outage probability table.
2. Compare reliability and resilience in terms of events, metrics and interventions.
3. Discuss deterministic, probabilistic, stochastic and robust power-system planning.
4. Explain how Weibull analysis, condition monitoring and criticality support asset decisions.
5. Critically examine regulatory and ethical issues in reliability-performance management.

Analytical / Application Questions

1. A utility supplies 50,000 customers. Three outages affect 2,000 customers for 1 hour, 5,000 customers for 2 hours and 1,000 customers for 4 hours. Calculate SAIFI, SAIDI and CAIDI.
2. Construct a two-unit COPT for independent 50 MW units with forced-outage probabilities of 0.05 and 0.10. For a demand of 80 MW, determine LOLP and expected shortage power.
3. Compare two transformer strategies using a consistent annual expected-cost model. Include financial, safety, environmental and common-cause considerations in your recommendation.

Multiple Choice Questions with Answers

Question 1

For a repairable two-state component with constant failure rate λ and repair rate μ, what is its steady-state availability?

A. λ/(λ+μ)

B. μ/(λ+μ)

C. 1/(λμ)

D. λμ/(λ+μ)

Correct Answer: B. μ/(λ+μ)

Question 2

In the chapter's COPT example, the expected shortage is 7.4176 MW for 100 hours. What is EENS?

A. 74.176 MWh

B. 341.76 MWh

C. 741.76 MWh

D. 7,417.6 MWh

Correct Answer: C. 741.76 MWh

Question 3

Which index measures average sustained interruption duration per customer over a reporting period?

A. SAIFI

B. SAIDI

C. LOLP

D. FOR

Correct Answer: B. SAIDI

Question 4

Which statement best distinguishes resilience from routine reliability analysis?

A. Resilience considers disturbance absorption and recovery from severe events.

B. Resilience considers only average component failure rates.

C. Reliability applies only to transmission systems.

D. Reliability and resilience are identical concepts.

Correct Answer: A. Resilience considers disturbance absorption and recovery from severe events.

Question 5

An asset has annual planned cost of $30,000, annual probability of one consequential failure of 0.02 and failure consequence of $1,500,000. Under the single-annual-event model, what is expected annual cost?

A. $30,000

B. $45,000

C. $60,000

D. $1,530,000

Correct Answer: C. $60,000

Question 6

For a Weibull lifetime model, what does β>1 indicate?

A. A hazard rate that increases with age

B. A constant hazard rate

C. A hazard rate that decreases with age

D. Zero probability of failure

Correct Answer: A. A hazard rate that increases with age

Question 7

What does the N−1 planning criterion principally require?

A. The system must withstand every pair of simultaneous outages.

B. The system must withstand one specified credible component outage within applicable limits.

C. No customer may ever experience an interruption.

D. Installed capacity must be exactly equal to peak demand.

Correct Answer: B. The system must withstand one specified credible component outage within applicable limits.

Question 8

Which expression defines expected energy not served for time intervals of duration Δt?

A. ∑_t Pr(C_t<L_t)Δt

B. ∑_t E[max(0,L_t-C_t)]Δt

C. ∑_t (C_t-L_t)Δt

D. max_t(L_t/C_t)

Correct Answer: B. ∑_t E[max(0,L_t-C_t)]Δt

Question 9

In the transformer example, the monitoring and refurbishment programme costs $60,000 annually, with annual failure probability 0.015 and consequence $1.4 million. What is its expected annual cost?

A. $21,000

B. $60,000

C. $81,000

D. $102,000

Correct Answer: C. $81,000

Question 10

Why might a planner include CVaR in addition to expected cost?

A. To eliminate all uncertainty from the model

B. To represent average losses in a specified worst-case probability tail

C. To convert SAIDI directly into SAIFI

D. To guarantee that every scenario has equal probability

Correct Answer: B. To represent average losses in a specified worst-case probability tail