Measurement data is not a simple collection of numbers. A value at one site is comparable with a value at another only when the same measurement and assessment approach has been used — so data must be read in relation to its full context:
\[ \text{measurement method} + \text{monitoring period} + \text{statistical index} + \text{site selection} + \text{voltage level} + \text{network characteristics} \]
Measurement data must be normalised before it can be compared.
Key idea
- Surveys differ in method, period, index, voltage level and site selection — normalise before comparing.
- The workhorse comparison indices: \(U_{h,sh95}\), \(P_{st}/P_{lt}\), \(U_{\text{neg},sh95}\), SARFI, and SAIFI/SAIDI.
- Report spread, not just the mean — few surveys plus wide dispersion give weak conclusions.
- An index is meaningful only when its measurement basis is defined.
Section 1
Why measurement data needs careful interpretation
Each phenomenon is shaped by different practical factors. Harmonic results depend on time aggregation, voltage-transformer type, background distortion, network impedance and connected loads. Flicker depends on load location, lighting technology, time of day, transfer between levels and the chosen percentile. Unbalance depends on single-phase loading, configuration, level and correct negative-sequence calculation. Dips depend on fault rate, clearing time, earthing, monitor connection and aggregation. Interruptions depend heavily on how the utility defines an event and what it includes. Before using any value, ask:
- What exactly was measured?
- Over how long, and at which voltage level?
- Which statistical index was calculated?
- At how many sites, and how were they selected?
- Are the data actually suitable for comparison?
Section 2
Harmonic measurement data and indices
Harmonic surveys are compared using individual harmonic voltages and THD. The preferred long-term index is \(U_{h,sh95}\): the weekly 95% value of the 10-minute RMS harmonic voltage for order \(h\) — reading the subscript as “harmonic \(h\), short-time (10-minute), 95th percentile”. The full family is:
Table 1 — Statistical harmonic voltage indices.
| Index | Meaning |
|---|
| \(U_{h,vs95}\) | Daily 95% value of the 3-second harmonic voltage |
| \(U_{h,vs99}\) | Daily 99% value of the 3-second harmonic voltage |
| \(U_{h,vsmax}\) | Maximum daily 3-second harmonic voltage |
| \(U_{h,sh95}\) | Weekly 95% value of the 10-minute harmonic voltage |
| \(U_{h,sh99}\) | Weekly 99% value of the 10-minute harmonic voltage |
| \(U_{h,shmax}\) | Maximum weekly 10-minute harmonic voltage |
In the reviewed surveys the 3-second very-short-time values were often missing, especially at MV and EHV, so practical comparison usually falls back on \(U_{h,sh95}\) — a reminder that comparison relies on the index that is actually available, not the full set proposed in standards.
Section 3
Criteria for comparing harmonic surveys
Not every survey is fit for comparison — one may be excluded for too short a period, too few sites or an incompatible method:
Table 2 — Filtering criteria for harmonic surveys.
| Criterion | Reason |
|---|
| At least one week of monitoring | Captures normal weekly variation |
| Sufficient number of sites | Avoids over-interpreting one location |
| Consistent statistical index | Allows comparison between surveys |
| Suitable voltage transformers | Avoids measurement distortion |
| Comparable harmonic orders | Enables like-for-like comparison |
| Same voltage-level category | MV, HV and EHV behave differently |
Capacitive voltage transformers are a particular trap: their frequency response may not represent the true harmonic voltage, so measurements made with unsuitable transformers should be treated with caution.
Do not compare harmonic surveys unless the measurement basis is comparable.
Section 4
What the harmonic data shows
Harmonic levels vary with voltage level and location. At MV, valid data came from only a few surveys, so general conclusions are weak even where many sites were monitored. At HV, more valid surveys were available and planning levels were exceeded in some cases, especially for low-order harmonics. At EHV the data was again limited, but the 5th harmonic remained important throughout:
Low-order harmonics — especially the 5th — are often dominant.
so harmonic studies should pay particular attention to the low odd orders, and to THD:
\[ h = 5,\ 7,\ 11,\ 13 \]
though the exact dominant orders depend on the connected equipment, network frequency response and background distortion. A further caution: with only a few surveys, a mean value can mislead. If two surveys give very different 5th-harmonic results, the average represents neither network well:
Few surveys + large dispersion = a weak general conclusion.
Results should therefore be presented with the minimum, maximum and mean — and, where possible, the 95% site value — never a single average without checking the spread.
Section 5
Flicker measurement data
Flicker data is less widely available than harmonic data and is often gathered near known sources — arc furnaces or large fluctuating loads — so it may be biased toward worst locations. The indices are \(P_{st}\) (10-minute short-term) and \(P_{lt}\) (2-hour long-term), reported as weekly statistical values such as \(P_{st95}\), \(P_{st99}\), \(P_{lt95}\) and \(P_{lt99}\). Worst-location surveys can exceed planning or standard values, but interpretation is not simple, because perception depends on lighting, time of day and transfer between levels:
Measured flicker above a limit does not always mean a known customer disturbance — but it does demand careful assessment.
Table 3 — Context needed to interpret flicker data.
| Factor | Why It Matters |
|---|
| Lighting technology | Different lamps respond differently to fluctuation |
| Time of day | Less relevant during daylight or unoccupied periods |
| Customer type | Residential and industrial sensitivity differ |
| Source location | Worst-location surveys may not represent the system |
| Transfer between levels | Flicker may attenuate or amplify downstream |
| Background flicker | Existing disturbance reduces the planning margin |
\(P_{st}\) and \(P_{lt}\) are necessary indices, but they should never be read without this context — the flickermeter was standardised on a reference incandescent lamp, and modern lighting can be more or less sensitive.
Section 6
Unbalance measurement data
Unbalance data uses the negative-sequence factor, reported as the weekly 95% value of the 10-minute measurement, \(U_{\text{neg},sh95}\):
\[ U_{\text{neg}}=\frac{U_2}{U_1}\times 100 \]
- \(U_2,U_1\)
- negative- and positive-sequence voltage (fundamental)
Table 4 — Reviewed unbalance observations by voltage level.
| Voltage Level | General Observation |
|---|
| MV | Reported sites generally below 2% (mostly one survey) |
| HV | Some sites reached or exceeded 1% |
| EHV | Very few sites exceeded about 1.1% |
HV and EHV unbalance is usually lower than at MV, but site conditions still matter — and the equipment concern is what counts. Even a small negative-sequence voltage drives significant negative-sequence current in machines:
\[ U_2\;\rightarrow\;I_2\;\rightarrow\;\text{machine heating} \]
- \(U_2\)
- negative-sequence voltage
- \(I_2\)
- negative-sequence current
so for motor-heavy sites, unbalance should be checked at the equipment terminals as well as the PCC: a network can comply at the PCC while a large motor still needs derating. A complete result states \(U_{\text{neg},sh95}\), the assessment period, the location, the voltage level and the calculation method.
Section 7
Voltage dip measurement data
Dips are harder to compare than harmonics, flicker or unbalance, because they are event-based and depend on faults, protection, topology, earthing, weather and monitor configuration. A survey is best expressed as events per year in a magnitude-duration table, each cell holding a count:
\[ N(U_{\text{retained}},\,t) \]
- \(N\)
- number of dips per year in that retained-voltage and duration range
Voltage dip performance is highly site-dependent.
Table 5 — Factors affecting dip survey results.
| Factor | Effect on Dip Results |
|---|
| Fault rate | How often dips occur |
| Protection clearing time | Dip duration |
| Fault location | Retained voltage |
| Earthing system | Phase voltages during faults |
| Monitor connection | Star vs delta changes the recorded count |
| Overhead vs underground | Fault rate and annual variation |
Monitor connection matters especially: star-connected monitors may record more events than delta-connected ones in some earthing arrangements, so a survey must state whether it is line-to-neutral or line-to-line.
Dip results cannot be compared unless the monitor settings and network context are known.
Section 8
SARFI and dip statistics
Dip results are often summarised with SARFI indices — \(\mathrm{SARFI}_{90}\), \(\mathrm{SARFI}_{70}\), \(\mathrm{SARFI}_{40}\) — the events per year with retained voltage below 90%, 70% and 40% respectively. They are simple and good for comparing sites, but they do not capture duration: a 70% dip of 60 ms and one of 1 second fall in the same threshold yet affect equipment very differently. SARFI should therefore sit alongside magnitude-duration tables or tolerance curves where process performance matters.
Average dip statistics also need care, because a few high-frequency sites can dominate the mean — so both median and percentile site values are useful:
\[ 50\%\ \text{site}\;\rightarrow\;\text{typical performance}\qquad\qquad 95\%\ \text{site}\;\rightarrow\;\text{poor but not extreme} \]
The 50% site describes normal performance; the 95% site shows what to expect at weaker or more exposed locations.
Section 9
Long-interruption data and benchmarking
Interruption data is reported as system-level statistics rather than per-site waveforms — \(\mathrm{SAIFI}\), \(\mathrm{SAIDI}\), \(\mathrm{SAIRI}\), system minutes, energy not supplied, availability, incident counts and restoration times. The difficulty is that utilities apply these differently: one includes planned interruptions, another excludes them; one includes major events, another reports them separately; one counts incidents, another delivery points. Direct comparison can therefore mislead, because transmission performance also depends on the network itself:
Table 6 — What shapes interruption benchmarking.
| Factor | Influence |
|---|
| Geography & environment | Exposure to storms, lightning, ice and terrain |
| Network topology | Radial and meshed systems differ in reliability |
| Voltage level | Fault exposure and protection design differ |
| Load density | Affects the consequence of outages |
| Redundancy & spare capacity | Ability to maintain supply and restore |
| Reporting definitions | Affect every index |
The most reliable benchmark is each utility against its own historical performance — it shows whether the network is improving or deteriorating.
Section 10
Practical use of measurement data
Data supports engineering decisions; it does not replace judgement. Harmonic data identifies dominant orders, background distortion and proximity to planning levels; flicker data flags worst locations and likely disturbance; unbalance data reveals phase-loading and motor-duty concerns; dip data estimates event frequency and ride-through needs; interruption data guides reinforcement and reliability investment. In every case the path is the same:
\[ \text{measurement data}\;\rightarrow\;\text{risk identification}\;\rightarrow\;\text{engineering decision} \]
Section 11
Data-quality checklist
Before using any power-quality data, run through the context that decides whether the numbers can be trusted and compared:
Table 7 — Power-quality data-quality checklist.
| Check | Why It Matters |
|---|
| Monitoring duration | Statistical reliability |
| Number of sites | Representativeness |
| Voltage level | MV, HV and EHV behave differently |
| Measurement method & instrument class | Comparability and confidence |
| Voltage transformer type | Critical for harmonics |
| Monitor connection | Critical for voltage dips |
| Time aggregation & percentile | 95%, 99% and maximum differ |
| Site selection | Worst-site surveys may not represent the network |
| Background & operating conditions | Existing distortion and network state affect results |
| Event filtering | Dips and swells can affect flicker |
| Planned / major-event treatment | Critical for interruption statistics |
Without this context, a result can be technically correct yet wrongly interpreted.
Section 12
Reporting examples, and the key message
Every result should carry its index and context. The difference is stark:
- Not “5th harmonic = 2.5%” — but \(U_{5,sh95}=2.5\%\) at the HV PCC, from one week of 10-minute values.
- Not “\(P_{lt}=0.9\)” — but \(P_{lt,95\%,\text{weekly}}=0.9\) at the MV busbar.
- Not “unbalance = 1.2%” — but \(U_{\text{neg},sh95}=1.2\%\), fundamental negative sequence, over one week.
- Not “\(\mathrm{SARFI}_{70}=5\)” — but \(\mathrm{SARFI}_{70}=5\) events/year at the PCC, line-to-line measurements.
Key message
Power-quality measurement data must be interpreted, not just collected: surveys differ in method, period, index, level and site selection, so results are rarely directly comparable. The workhorse indices — \(U_{h,sh95}\) for harmonics, \(P_{st}\) and \(P_{lt}\) for flicker, \(U_{\text{neg}}=\tfrac{U_2}{U_1}\times 100\) for unbalance, events-per-year and SARFI for dips, and \(\mathrm{SAIFI},\ \mathrm{SAIDI},\ \mathrm{SAIRI},\ \mathrm{ENS},\ \mathrm{SM}\) for interruptions — mean nothing until their basis is stated. A robust interpretation always specifies the measurement method + monitoring period + statistical index + voltage level + assessment location + data filtering + comparison objective. Only then can the data be used with confidence for planning, compliance, equipment assessment and mitigation design — the thread that runs through all six parts of this series.
