Power and Energy Loss Evaluation and Reduction Study

1. Executive Summary

Keentel Engineering Solutions was retained by the host utility to determine why its reported total system losses had drifted from 5.9 percent to 6.8 percent of energy input over five reporting years, to decompose those losses into components an engineer could act on, and to produce a loss model defensible enough to support both a capital program and a regulatory filing. The Client had a top-down energy balance and nothing else. It could state that 232,560 MWh disappeared each year between the wholesale delivery meters and the customer bills; it could not state where, why, or what any of it cost to fix. The engagement was structured as a reconciliation problem rather than a modeling exercise. Two independent estimates of loss were built and forced to agree. The top-down estimate came from an energy balance across 6 bulk delivery points, 4 embedded generators, 292,000 customer meters and an estimated unbilled quantity. The bottom-up estimate came from an 8,760-hour time-series simulation of all 168 distribution feeders in CYME as part of Keentel’s power system studies workflow, cross-checked against the Client's legacy Milsoft models, with the 138 kV subtransmission layer carried in PSS®E and the whole reconciliation orchestrated by a Python framework that ingested AMI, CIS billing, SCADA and wholesale settlement data. Agreement between the two estimates — a mean absolute monthly deviation of 0.3 percentage points after billing calendarization — was the study's principal validation test, and the only credible basis for splitting technical from non-technical loss. The answer was that 5.1 percentage points of the 6.8 were technical and 1.7 were not. Within the technical portion, 62 percent was load-dependent and 38 percent was fixed, and the single largest component in the entire system was distribution transformer no-load loss at 1.4 percent of system energy — the product of 70,400 units, a third of them predating any efficiency standard, with a median peak loading of 22 percent of nameplate. Three secondary drivers followed: severe current unbalance on the worst 20 feeders, nine surviving 4.16 kV circuits carrying 3.1 percent of energy but 8.4 percent of technical losses, and 118 of 386 capacitor banks not switching as designed. The recommended nine-measure program costs $14.34 million, reduces technical losses from 5.1 percent to 3.9 percent, saves approximately 41,000 MWh and 8.6 MW of coincident peak annually, returns a net present value of $10.4 million at a 7 percent real discount rate, and pays back in a blended 6.4 years. The worst feeder in the system was taken from 9.2 percent losses to 4.1 percent. Avoided emissions are approximately 17,200 metric tons of CO2 per year. Figures presented are representative of the delivered study and have been generalized to protect client confidentiality. Study at a Glance

2. Background and Study Drivers

The Client is a vertically integrated utility serving approximately 292,000 customers through 412 miles of 138 kV subtransmission, 26 distribution substations and 3,140 miles of primary distribution, predominantly 12.47 kV with nine legacy 4.16 kV circuits fed from three older substations. Energy enters the system at six 230/138 kV bulk delivery points and from four embedded generators connected at 138 kV, including two 20 MW solar facilities. Behind-the-meter photovoltaic penetration stood at 61 MW nameplate at the time of study. Three pressures converged to trigger the engagement. The first was arithmetic. Reported losses had risen from 5.9 percent to 6.8 percent over five reporting years. At a marginal energy cost of $38 per MWh, the 0.9 percentage point drift alone represented approximately $1.17 million per year of energy the Client purchased and could not bill. Nobody could say whether the drift was real, a metering artifact, or a billing calendar effect. The second was regulatory. The Client's line loss factors, used to gross up retail sales to wholesale purchase quantities and embedded in every rate class, had been derived from a study performed decades earlier and updated since only by scaling. In the Client's most recent general rate case, intervenor testimony challenged the loss factors directly, arguing they were stale, unsupported by any current engineering analysis, and allocated in a manner that shifted cost between rate classes without evidentiary basis. The Client had no adequate answer. Commission staff signaled that the next filing would require a contemporaneous loss study with an auditable methodology. The third was capital planning. The Client's distribution planning group had a standing list of candidate loss-reduction projects — reconductoring, voltage conversion, capacitor additions — but no consistent economic framework with which to rank them against each other or against reliability-driven work. Projects were justified individually, with inconsistent energy prices, inconsistent discount rates and inconsistent treatment of peak capacity value. Two proposals under active consideration used loss valuations differing by a factor of three. The commercial stake was therefore not only the energy itself but the credibility of the Client's loss reporting and the integrity of its capital ranking. A defensible loss number was worth more than the megawatt-hours it described.

3. Study Objectives and Scope of Work

1. Establish a top-down energy balance for the whole system at monthly and annual resolution, correcting for billing cycle misalignment and unbilled energy, and quantify its uncertainty.
2. Build a bottom-up engineering model of technical losses covering every voltage level from the 138 kV delivery points to the customer service entrance, and decompose losses by component and by load-dependent versus no-load classification.
3. Reconcile the bottom-up model against the top-down energy balance, and derive the technical and non-technical split as the residual of a validated reconciliation rather than as an assumption.
4. Establish loss factors and loss allocations by delivery voltage and rate class suitable for regulatory filing.
5. Identify, quantify and rank loss reduction measures on a consistent economic basis, using loss capitalization factors derived specifically for the Client.
6. Deliver a repeatable annual loss reporting process, with the data pipeline, model update procedure and documentation required to run it without further consultant involvement.
7. Support field validation of implemented measures and reconcile measured savings against predicted savings.

Scope and Deliverables

Explicitly excluded and documented as such: harmonic loss contributions beyond a screening assessment, transformer stray and eddy loss temperature correction beyond IEEE Std C57.12.90 test conditions, loss effects of conductor operating temperature above the modeled seasonal profile, and any determination of the cause of individual non-technical losses. The residual identified as non-technical was quantified and characterized but not investigated in the field; that work was recommended as a separate engagement. The Client accepted these exclusions in writing at Month 1.

4. Regulatory and Standards Basis

Loss studies sit at the junction of engineering practice and regulatory accounting, and the standards basis reflects that. There is no NERC reliability standard that governs distribution loss performance; the compliance-adjacent obligations are the modeling data requirements of the MOD family for the subtransmission layer, and the transformer efficiency requirements imposed by the U.S. Department of Energy at 10 CFR Part 431 Subpart K, which are mandatory for new distribution transformers and which set the efficiency floor against which the replacement program was evaluated. The predecessor voluntary NEMA TP-1 levels were used only to characterize the installed legacy population. The regulatory driver is the state commission of jurisdiction, before which the Client reports line loss factors and books purchased power. The FERC Uniform System of Accounts and Form 1 reporting conventions determine how the energy balance quantities are booked and therefore how the study's results must be expressed to be reconcilable with filed data. This constrained the study more than any technical standard: the engineering model had to produce numbers in the same units, on the same calendar and on the same accounting boundary as the Client's filings, or it would be useless for its principal purpose. Standards and Regulatory Register

5. System Modeling and Data Development

5.1 Data sources and gap closure

The study consumed more effort in data remediation than in analysis, which is normal and was budgeted for. The Client held a modern GIS with full connectivity for most of the system, a CIS containing every billed account, an AMI deployment covering 78 percent of meters at 15-minute resolution, SCADA at every substation bank and feeder head, and settlement-quality metering at the six bulk delivery points. None of these systems agreed with each other. Model Data Sources and Gap Closure

The connectivity repairs mattered enormously. The Client's first-pass model, run before remediation, returned technical losses of 3.9 percent — comfortably below the energy balance and comfortably wrong. The 1,240 orphaned sections carried 18.4 MW of peak load that the model simply did not serve, and the missing phasing caused the solver to distribute single-phase lateral load evenly across three phases, erasing the unbalance that turned out to be one of the study's principal findings. After remediation the model returned 5.1 percent. The difference between those two numbers is the entire value of the data work, and it is the reason a loss study cannot be run directly off an unaudited GIS export.

5.2 Model build and benchmarking

Each of the 168 feeders was built in CYME from the remediated GIS, with distribution transformers represented individually rather than aggregated, secondary and service conductor represented by a calibrated equivalent per transformer, and every regulator, capacitor bank and sectionalizing device represented with its actual control settings as recovered from field records and controller interrogation. The Client's incumbent Milsoft models were retained and run in parallel for 22 feeders as an independent cross-check; agreement on annual feeder losses was within 2.8 percent, and the residual differences traced to secondary representation, where the two tools use different default assumptions. The 138 kV subtransmission layer, the 26 substation banks and the four embedded generators were carried in PSS®E, with the network checks aligned to Keentel’s load flow analysis services, with the distribution system represented at each substation low-side bus as a time-varying load derived from the CYME feeder solutions. The two layers were coupled iteratively: the distribution solution set the substation load, PSS®E set the substation low-side voltage, and the loop was run to convergence for each of a set of representative hours before the full 8,760-hour sweep. Benchmarking used SCADA. Modeled feeder head real power at system peak matched metered values within 2.1 percent across the fleet; modeled per-phase current matched within 3.4 percent; modeled annual energy at the feeder head matched metered within 1.3 percent. Voltage profiles were validated against AMI voltage data at 4,900 service points, with mean absolute deviation of 1.1 V on a 120 V base.

6. Study Methodology and Assumptions

6.1 The two-estimate structure

The method rests on a single principle: an energy balance can measure total loss but cannot decompose it, and an engineering model can decompose loss but cannot measure it. Neither is sufficient alone. Used together, with the engineering model constrained to reproduce the measured total, they yield a decomposition that carries the credibility of a measurement. The top-down energy balance is stated as: energy into the system at all sources, minus billed energy, minus estimated unbilled energy, equals total loss. Every term carries uncertainty. Source energy is the best-measured quantity, from settlement-class metering at the bulk delivery points and revenue metering on the embedded generators, with an assessed uncertainty of ±0.15 percent. Billed energy is precise as an accounting quantity but misaligned in time, since the Client reads meters across 21 cycle groups over 21 business days and each cycle bill spans a different window than the calendar month. Unbilled energy — company use, street lighting, traffic signals, cabinet and equipment loads billed on fixed schedules rather than metered — was the weakest term, estimated from a table last revised decades earlier. The bottom-up model computes technical loss directly from currents, impedances and transformer characteristics for all 8,760 hours of the study year. It cannot see theft, meter under-registration or billing error, and so the difference between the measured total and the modeled technical quantity is, by construction, the non-technical residual plus every modeling error the study failed to catch. That is the honest description of what a non-technical loss figure is, and it is why the reconciliation quality determines how much weight the figure can bear.

6.2 Loss factor, load factor and the 8,760-hour calculation

Load-dependent losses vary with the square of current, so annual energy loss cannot be obtained by multiplying peak loss by hours. The classical approach uses the loss factor, the ratio of average power loss to peak power loss, estimated from the load factor by the widely used empirical relationship: Loss factor = 0.15 × (load factor) + 0.85 × (load factor)² For the Client's system, with an annual load factor of 0.558, this yields a loss factor of 0.348. The 8,760-hour time-series calculation yielded 0.371 — the empirical estimate understated load-dependent energy losses by 6.2 percent system-wide. The approximation was retained as a sanity check only, never as the basis of a result, and the study documented where it fails. It fails on peaky load shapes: on a residential feeder with a load factor of 0.42 the empirical relation gives 0.213 against a computed 0.246, an error of 15 percent. It fails in the opposite direction, though mildly, on flat industrial shapes. It fails badly wherever the load shape is bimodal, wherever significant distributed generation reshapes net load, wherever a feeder is reconfigured seasonally, and wherever regulator and capacitor switching changes the voltage at which a given power is delivered. All four conditions were present somewhere on this system. The 8,760-hour calculation removes the need to assume any of it away.

6.3 Load allocation

Energy must be assigned to each of 70,400 distribution transformers before the model can compute anything. Two methods were used and compared. Allocation by connected kVA distributes feeder energy in proportion to transformer nameplate, which is simple, requires no billing data, and is systematically wrong wherever transformers are oversized — which on this system was almost everywhere. Allocation by billed kWh assigns each customer's metered consumption to the transformer serving it, shaped by the class load curve for that customer's rate class, and is correct wherever the customer-to-transformer linkage is correct. Billed-kWh allocation was used as primary, with connected-kVA allocation as fallback for the 4.6 percent of energy that could not be linked. For the 78 percent of meters with AMI, actual interval data replaced the class shape entirely, which removes the coincidence assumptions that dominate uncertainty in traditional load research. The two allocation methods differed by up to ±0.8 percentage points on individual feeders and by ±0.1 percentage points system-wide — a useful demonstration that allocation method matters greatly for ranking feeders and hardly at all for the system total.

6.4 Loss capitalization: the A-factor and B-factor method

Every economic comparison in the study used a single pair of loss capitalization factors derived for this Client, so that no two projects could be justified on different energy prices. The A-factor is the present value of one kilowatt of no-load loss, which runs continuously for the equipment life. The B-factor is the present value of one kilowatt of load loss at rated load, which runs at a duty determined by the loss factor and the equipment's peak loading ratio. A-factor derivation: 1 kW continuous consumes 8.76 MWh per year, valued at a levelized $48 per MWh over a 30-year life at 7 percent real, plus a levelized capacity value of $95 per kW-year, giving $515 per year and a present value of $6,400 per kW using an annuity factor of 12.409. B-factor derivation: 1 kW of rated load loss on a unit peaking at 65 percent of nameplate produces 0.4225 kW at peak; at a transformer loss factor of 0.30 this is $221 per kW-at-peak per year, giving $93 per year per kW of rated load loss and a present value of $1,160 per kW. The corresponding value per kW of actual loss at peak is $2,744. Total owning cost is then purchase price plus A × no-load loss plus B × load loss. The ratio A/B of 5.5 is typical and expresses the essential asymmetry of the problem: a watt of core loss is worth roughly five and a half times a watt of winding loss, because it never stops.

6.5 Cases and acceptance criteria

Case and Scenario Matrix

Acceptance Criteria

7. Analysis and Results

7.1 The top-down energy balance

Annual energy into the system totaled 3,420,000 MWh. Billed energy totaled 3,175,440 MWh. Estimated unbilled energy totaled 12,000 MWh. Total loss was therefore 232,560 MWh, or 6.80 percent of input. The monthly series was initially uninterpretable. Reported monthly losses swung between 4.9 percent and 8.7 percent with no physical explanation — a swing of ±1.9 percentage points that no combination of temperature, load and configuration could produce. The cause was billing cycle misalignment. Source energy is measured on the calendar month; billed energy is accumulated across 21 cycle groups whose read windows straddle month boundaries. A month containing a read-date shift, a holiday, or an estimated-read batch accumulates a different number of customer-days than the calendar month contains, and the difference appears entirely in the loss line. The Python framework calendarized the billing data by prorating each cycle bill across the calendar days it spanned, using AMI interval data where available and the applicable class shape where not. The residual monthly variation after calendarization was ±0.4 percentage points, and it tracked temperature and load in the direction physics predicts. This correction did not change the annual number by more than 0.02 percentage points. It changed the study's ability to use monthly data at all, and it is the reason the Client's previous internal attempts to track losses month to month had produced nothing but noise.

7.2 Loss taxonomy and decomposition

Losses were classified first as technical or non-technical, then within technical as load-dependent (series, variable, I²R) or no-load (shunt, fixed). Load-dependent losses arise from current flowing through series impedance and vary with the square of that current. No-load losses arise from magnetizing and dielectric phenomena and are, to first order, present whenever the equipment is energized regardless of load. Technical Loss Decomposition (percent of annual energy input)

Load-dependent components total 3.16 percent, or 108,072 MWh, which is 62.0 percent of technical loss. No-load components total 1.94 percent, or 66,348 MWh, which is 38.0 percent. Non-technical loss is the residual 1.70 percent, or 58,140 MWh, giving the 6.80 percent total. Two features of this decomposition drove everything that followed. First, distribution transformer no-load loss is the largest single line item in the system, larger than every substation transformer's total loss combined and larger than the entire 138 kV network. This is a common finding and it is arithmetic, not pathology: no individual 25 kVA transformer wastes much, but 70,400 of them energized 8,760 hours a year is 5.47 MW of continuous draw. Second, the no-load share of 38 percent is high enough that a program targeting only conductor losses would have left more than a third of the technical loss untouched. The non-technical residual was characterized, though not investigated: approximately 0.55 percentage points attributable to systematic meter under-registration, 0.60 to diversion and unauthorized use, 0.35 to unmetered and inadequately estimated load, and 0.20 to residual billing estimation error.

7.3 Load unbalance

Load unbalance was the study's largest surprise. In a balanced three-phase four-wire system the neutral carries no current and contributes no loss. In an unbalanced system the neutral and earth return path carries the vector sum of the phase currents, and that current produces loss in a conductor typically smaller than the phase conductors and in the earth return, on top of the increased loss in the heavily loaded phases. Because loss varies with the square of current, the penalty is not proportional to the unbalance — it is quadratic, and it accrues even when total feeder load is unchanged. Current unbalance, expressed as maximum phase deviation from the average, ranged from 3 percent to 41 percent across the fleet with a fleet average of 22 percent. On the worst 20 feeders the unbalance-attributable loss term, including the neutral component, accounted for between 0.6 and 2.9 percent of feeder energy — that is, on the worst feeder, unbalance alone was costing more than the entire technical loss target for the system. The causes were mundane and cumulative. Single-phase laterals had been added to whichever phase was convenient at the time of construction. Transformer change-outs had been re-phased for crew convenience. Load had grown unevenly along feeders originally balanced at construction. No process existed to detect drift, because feeder head SCADA reported total current and the per-phase values were not trended.

7.4 Power factor and reactive flow

Reactive current produces the same I²R loss as real current and delivers no energy. System power factor at the substation buses at peak was 0.94 lagging as found, and the reactive component of current accounted for 11.3 percent of load-dependent losses. The capacitor fleet was substantially non-functional: interrogation and field inspection of all 386 pole-mounted banks found 118 not operating as designed — 24 stuck open, 61 stuck closed, and 33 with one or more failed capacitor units producing an unbalanced and reduced output. The 61 stuck-closed banks were the more insidious problem, because they leave the feeder overcompensated at light load, producing leading power factor, raised voltage and reactive current flowing back toward the substation, which is loss with the sign of the reactive flow reversed but the magnitude undiminished.

7.5 Distribution transformer population

The transformer population was analyzed unit by unit against CIS load data. Median peak loading was 22 percent of nameplate. Thirty-one percent of units predated both the DOE efficiency standards and the voluntary NEMA TP-1 levels that preceded them; their no-load losses run typically 40 to 70 percent above a current-compliant unit of the same rating. Approximately 19 percent of units were oversized to the degree that a unit one or two standard sizes smaller would carry the measured peak with adequate margin under IEEE Std C57.91 loading criteria. Right-sizing is counterintuitive and worth stating precisely. Moving a load from an oversized transformer to a smaller one reduces no-load loss substantially and increases load loss, because the load loss at a given kVA rises as the square of the loading ratio. For a representative case — an 11 kVA peak load on a legacy 50 kVA unit with 145 W no-load and 620 W rated load loss, moved to a compliant 25 kVA unit with 68 W no-load and 360 W rated load loss — no-load loss falls by 77 W while load loss at peak rises by 40 W. Applying the A and B factors gives a present value benefit of $493 against a present value penalty of $109, a net $384 per unit. Across the recommended program the average net present value was $500 per unit, and the average net energy saving 861 kWh per unit per year.

7.6 Subtransmission and substation losses

The 138 kV layer, modeled in PSS®E, contributed 0.42 percent of system energy in line losses and a further 0.58 percent across substation transformers, split 0.36 load and 0.22 no-load. Two substations were found operating both banks in parallel year-round at an aggregate loading of 18 percent of firm capacity, a configuration adopted for reliability but never re-examined. Parallel operation of two lightly loaded banks doubles no-load loss while reducing load loss by half, and at 18 percent loading the no-load term dominates by a wide margin. Seasonal single-bank operation at six substations, with automatic transfer retained for contingency, recovers 0.21 MW of continuous loss with no reduction in firm capacity and no change in N-1 performance. Substation transformer replacement economics were evaluated using the A and B factors. A representative comparison of three bids for a 25 MVA 138/12.47 kV replacement unit illustrates the method: Substation Transformer Total Owning Cost Comparison

Bid C carries neither the lowest first cost nor the lowest losses, and wins on total owning cost by $26,000 over the cheapest unit. The margins among modern units are narrow, which is the market working as intended. The method earns its keep elsewhere: applied to the question of whether to continue operating a legacy pre-standard bank against replacing it, the evaluated loss differential is large enough to move the answer.

7.7 Distributed generation

Distributed generation reduces losses when it displaces current that would otherwise flow from the substation through feeder impedance, and increases them when it produces reverse flow large enough that the net current exceeds what the load alone would have drawn. The transition point depends on penetration, location along the feeder, and the coincidence of generation with load. At the as-found 61 MW of behind-the-meter photovoltaic, distributed generation reduced system technical losses by approximately 3,100 MWh per year on net. The benefit was concentrated on feeders with dispersed rooftop systems and daytime-peaking commercial load. At the modeled 120 MW penetration the net benefit fell to 1,800 MWh as reverse flow began on lightly loaded residential feeders in spring shoulder months. At 180 MW with the clustered siting pattern the Client's interconnection queue implied, six feeders turned net adverse, adding 1,900 MWh per year and eroding the system benefit to approximately 400 MWh. The engineering conclusion is that distributed generation is not a loss reduction measure and should not be counted as one in a capital plan; it is a loss modifier whose sign depends on conditions the utility does not control.

7.8 Worst-feeder deep dive

Feeder 142, a 12.47 kV circuit with 14.6 miles of three-phase main on 1/0 ACSR serving a mixed rural and small-commercial load with a 9.8 MVA peak, was the worst performer at 9.2 percent annual losses against 40,200 MWh of feeder energy input — 3,698 MWh lost. Decomposition attributed 2.4 percentage points to undersized main conductor at high loading, 2.1 to current unbalance running at 34 percent, 1.1 to reactive flow at 0.91 power factor, 1.9 to distribution transformer losses at above-average density, and the remainder to secondary and service. The remediation sequence was deliberate and cheap first. Phase balancing of 34 laterals and 61 transformer connections cut unbalance from 34 percent to 8 percent. Two switched capacitor banks corrected power factor to 0.98. Transfer of 2.1 MVA of load to an adjacent feeder through an existing tie relieved the first 4 miles of main. Together these took the feeder to 5.5 percent at a cost of $186,000. Reconductoring 4.2 miles of main from 1/0 ACSR to 477 kcmil AAC then took it to 4.1 percent for a further $840,000. Total saving 2,050 MWh per year.

7.9 Loss allocation by delivery voltage

Losses were allocated to service levels for the regulatory filing. A customer served at secondary voltage bears the losses of every element between the delivery point and the service entrance; a customer metered at 138 kV bears almost none. Average Loss Factors by Delivery Voltage (percent of delivered energy)

The load-weighted average of these factors is 5.37 percent on a delivered-energy basis, which is identical to 5.10 percent expressed on an input-energy basis — the arithmetic identity 5.10 / (100 − 5.10) = 5.37 confirms the allocation conserves energy and does not create or destroy loss in the allocation step. This check was performed at every allocation stage. Marginal loss factors approximate twice the average for the load-dependent portion, which follows directly from the quadratic loss relationship, and were computed rather than assumed.

7.10 Reconciliation: the validation test

The reconciliation is the study's load-bearing element. Two independent estimates were compared monthly over 24 months and annually. Bottom-Up to Top-Down Reconciliation

The bottom-up figure combines the modeled technical loss of 5.10 percent with the independently characterized non-technical estimate of 1.80 percent developed from meter test records, unmetered load audit and diversion case history. That estimate was subsequently refined to 1.70 percent to close the annual balance — the only place in the study where a figure was fitted rather than computed, and it is identified as such in the report so that any reader can see exactly which number carries the accumulated uncertainty of every other. The fourteen substations with revenue-quality metering on both the high side and all feeder heads allowed a closed balance at a scale small enough to be meaningful and large enough to be statistically useful. Agreement within 0.3 percentage points at the worst individual substation, on a population spanning 4.4 percent urban losses to 8.6 percent legacy 4.16 kV losses, is the strongest evidence the study produced that the model represents reality.

8. Sensitivity and Scenario Analysis

Sensitivity Results

Three conclusions came out of the sensitivity work. First, the recommended program is robust: the measure ranking by benefit-cost ratio does not change across any tested discount rate or energy price, and only load growth reorders it, promoting reconductoring relative to fixed-loss measures because conductor loss grows quadratically with load while core loss does not grow at all. Second, the loss factor method choice matters most exactly where the decisions are hardest — the peaky, poorly performing feeders where the empirical relation is least accurate and where the capital is proposed. Third, the non-technical estimate, which carries the most uncertainty of any number in the study, has no bearing whatever on the engineering program. That separation is deliberate: it means an unresolved argument about theft cannot delay a defensible capital decision.

9. Findings and Root Cause Assessment

Findings Register

Finding F-2 deserves expansion, because it was the question that triggered the engagement. The 0.9 percentage point drift decomposes into three parts. Roughly 0.4 points is genuine growth in non-technical loss, consistent with the Client's own diversion case volume and with the aging of its instrument transformer population. Roughly 0.3 points is technical, and it is the predictable consequence of load growth on feeders that had not been reinforced: because loss grows with the square of current, a 12 percent load increase on an unchanged feeder raises its losses by 25 percent. Roughly 0.2 points is not a loss at all but a reporting artifact, arising from a change in billing cycle structure that shifted the average lag between energy delivery and energy billing. Finding F-7 concerns measurement, not loss. Current transformers of 0.6 accuracy class meet their rating over a defined current range and degrade outside it; below roughly 10 percent of rated primary current, ratio and phase-angle errors of 1.5 to 3 percent are ordinary. Twenty-three installations — 14 substation and 9 large-customer — were operating routinely below that threshold because the CTs had been sized for a load that never materialized or that had since declined. The energy is delivered; it is simply not counted. This appears in the energy balance as non-technical loss and is corrected by re-ratioing the CTs, not by any change to the network.

10. Mitigation Options and Recommendations

Nine measures were carried to full evaluation. Each was modeled individually in the 8,760-hour case, then in combination, so that interaction effects were captured rather than assumed additive. The savings tabulated below are the in-combination values, which total 3.1 percent less than the arithmetic sum of the same measures modeled standalone, because phase balancing and capacitor correction both reduce the current that reconductoring would otherwise have to carry. Loss Reduction Options Compared

The peak megawatt figures are measure-level and non-coincident. Applying the measured coincidence factor of 0.86 between measure-level peak loss reduction and system peak gives 8.6 MW of coincident peak reduction, which is the quantity used for capacity valuation throughout. Failing to make this distinction is one of the most common errors in loss economics and typically overstates capacity benefit by 10 to 20 percent. Two measures require honest qualification. Voltage conversion returns a 14.0-year simple payback and a present value below its capital cost on loss benefit alone. It was retained in the recommended program because the nine 4.16 kV circuits are approaching end of asset life independently, because conversion relieves a capacity constraint at two substations that would otherwise require separate investment, and because those benefits are real but were not monetized in this table. Presented on loss economics alone, the measure does not stand up, and the study says so. Conservation voltage reduction returns a 17.9-year payback on loss benefit, and on that basis alone would be rejected. Its actual justification lies elsewhere. CVR reduces customer consumption on voltage-sensitive load; at a measured CVR factor of 0.72 and an average voltage reduction of 2.4 percent across the 42 candidate feeders carrying 1,180,000 MWh, it delivers approximately 20,400 MWh per year of consumption reduction. That quantity is deliberately excluded from the 41,040 MWh loss saving, because consumption reduction and loss reduction are different things and adding them is double counting of the worst kind. The measure was recommended, with its economics presented separately and correctly labeled. The recommendation was to execute all nine measures over a six-year program, sequenced by benefit-cost ratio: the fast-payback operational measures in Phase 1 (Months 12 to 24), the transformer programs continuously from Month 12, and the capital-intensive reconductoring and voltage conversion in Phase 2 from Month 30 onward as budget allowed. Sequencing cheap measures first is not only financially sensible but technically necessary: balancing and capacitor correction change the currents against which reconductoring is sized, and doing them in the reverse order would have oversized the conductor.

11. Implementation Support and Field Validation

Phase 1 was executed between Month 12 and Month 24 and comprised capacitor control repair and replacement across all 118 defective banks, phase balancing at 240 locations, feeder reconfiguration on 12 feeders, and conversion of six substations to seasonal single-bank operation. Keentel supplied the work packages, the switching order sequences, the per-location phasing instructions and the acceptance test criteria, and updated the models as work was completed. Six feeders were instrumented for validation with permanent per-phase power quality monitoring at the feeder head and additional monitoring at two mid-feeder locations, recording for three months before and twelve months after the work. Measured savings were normalized to the pre-work load level and weather using the same 8,760-hour framework, so that the comparison is of loss at equivalent load rather than of raw energy. Field Validation: Predicted Versus Measured

Individual feeder deviations of up to 8.5 percent are larger than the aggregate deviation of 0.1 percent, and this is the expected behavior of a model calibrated on population statistics. Per-feeder error is dominated by the secondary and service equivalent, which is the least directly observable element of the model; that error is close to random across feeders and cancels in aggregate. The study documented this explicitly so that no reader would take a per-feeder prediction as carrying the same confidence as a system-level one. The measured aggregate result also confirmed the loss factor treatment. Had the empirical loss factor approximation been used instead of the 8,760-hour calculation, predicted savings on these six feeders would have been 3,570 MWh against a measured 3,825, an understatement of 6.7 percent consistent with the system-wide 6.2 percent gap identified in Section 6.2.

12. Results and Value Delivered

Outcomes Scorecard

Beyond the numbers, four deliverables carried value the table does not capture. The reconciled loss model became the Client's evidentiary basis for its line loss factors, and the filing support package — methodology narrative, data lineage, allocation derivation and sensitivity record — was constructed so that every number in the filing can be traced to a source measurement. The prioritized capital program replaced nine competing project proposals with a single ranked list built on one set of economic assumptions. The A and B factors were adopted as the Client's standard transformer purchasing evaluation, applied to every subsequent procurement. And the Python reconciliation pipeline was handed over with documentation and training, so that the annual loss report is now produced internally in approximately three staff-weeks rather than being reconstructed from scratch each time it is demanded. The CO2 figure uses a marginal emission rate of 0.42 metric tons per MWh applied to the 41,040 MWh of avoided loss energy. Over the 30-year evaluation horizon this is approximately 516,000 metric tons. The consumption reduction from conservation voltage reduction would add a further 8,570 metric tons per year and is again reported separately rather than combined.

13. Lessons Learned and Engineering Insights

Data quality is the study. The difference between a 3.9 percent and a 5.1 percent modeled loss on the same system was entirely data remediation — connectivity, phasing, conductor coding and transformer records. Any loss study quoted without an explicit data remediation phase should be treated as an estimate of the GIS, not of the network. A useful diagnostic is that an unremediated model almost always understates loss, because missing connectivity removes load and missing phasing removes unbalance, and both errors point the same way. The cheap measures are the good measures, and they are the ones nobody does. Phase balancing, capacitor control repair and feeder reconfiguration together cost $2.45 million of the $14.34 million program and deliver 12,050 MWh of the 41,040 MWh saving — 17 percent of the capital for 29 percent of the benefit. They are unglamorous, they generate no asset, and they are consistently deferred in favor of visible construction. A loss study's most reliable contribution is to make the economics of the unglamorous work explicit enough to survive a budget review. No-load loss is invisible and enormous. Distribution transformer core loss does not show up on any feeder loading report, does not vary with load, and is not attributable to any single asset large enough to attract attention. It was nonetheless the largest single component in this system. Utilities systematically underweight it because their planning processes are organized around peak loading, and no-load loss is precisely the quantity that peak loading does not reveal. Separate the arguments you can win from the arguments you cannot. The non-technical loss figure is the most contestable number in any loss study, and if the technical program depends on it the program becomes hostage to that argument. Structuring the study so the engineering conclusions are independent of the non-technical residual was a deliberate choice, and it allowed the capital program to proceed while the revenue protection question remained open. Calendarize before you conclude anything from a monthly series. The Client had spent years trying to interpret month-to-month loss variation that was almost entirely a billing-cycle artifact. A one-time prorating correction, costing a few days of work, turned an uninterpretable series into a usable one and would have saved considerable prior effort had it been done first. Predict at the level you validated at. The model reproduced aggregate savings within 0.1 percent and individual feeder savings only within 8.5 percent. Both are legitimate results of the same model, and communicating the difference clearly is part of the deliverable. A client who is allowed to believe that a per-feeder prediction carries system-level confidence will eventually be disappointed by a correct model.

14. Keentel Capability Summary

• System-wide technical loss studies across transmission, subtransmission, primary and secondary distribution
• Top-down energy balance construction, billing calendarization and unbilled energy assessment
• Bottom-up 8,760-hour time-series loss modeling in CYME and Milsoft, with PSS®E for the transmission and subtransmission layer
• GIS and CIS data quality assessment, connectivity and phasing remediation, and model-to-field benchmarking
• Load research design, class load shape development and AMI interval data integration
• Loss allocation by delivery voltage and rate class, average and marginal loss factor derivation
• Loss capitalization factor development (A-factor and B-factor) and transformer total owning cost evaluation
• Distribution transformer fleet analysis, right-sizing assessment and DOE efficiency compliance evaluation
• Phase balancing, feeder reconfiguration, capacitor placement and volt-var optimization studies
• Conservation voltage reduction assessment and CVR factor measurement
• Distributed generation loss impact assessment across penetration and siting scenarios
• Economic ranking of loss reduction programs, NPV and benefit-cost analysis, and sensitivity assessment
• Regulatory filing support packages, methodology documentation and data lineage
• Python-based reconciliation and reporting frameworks delivered for client self-service
• Field instrumentation planning, measurement and verification of realized savings

15. Frequently Asked Questions

16. Glossary of Terms and Abbreviations

17. Confidentiality and Use Statement

This case study has been prepared for informational purposes. All client identities, project locations, contract details, and proprietary data have been withheld or generalized. Technical parameters, study results, and figures presented are representative of work performed by Keentel Engineering Solutions and have been adapted so that no individual project, owner, or facility can be identified. Nothing in this document constitutes a design recommendation for any specific installation. Any reuse of the methodologies described requires project-specific engineering analysis by a qualified professional engineer.