Electricity Load Shapes

KiNESYS employs a sophisticated methodology for representing electricity demand temporal patterns, combining globally consistent climate-driven load estimates with actual measured data where available. This page provides comprehensive technical documentation of the load shape generation, sectoral disaggregation, and timeslice aggregation processes.

Overview

Electricity demand exhibits strong temporal variation driven by:

  • Seasonal factors: Temperature-driven heating and cooling, daylight hours, economic activity cycles

  • Weekly patterns: Weekday vs. weekend activity, industrial production schedules

  • Diurnal variation: Sector-specific time-of-day patterns (business hours, shift work, household routines)

Capturing these patterns is essential for:

  • Resource adequacy: Ensuring sufficient generation capacity to meet peak demands

  • Renewable integration: Valuing flexibility and storage to manage variable generation

  • System optimization: Identifying cost-effective technology mixes across different load conditions

  • Policy analysis: Assessing demand response, electrification, and load management strategies

KiNESYS represents these temporal patterns through:

  1. Hourly load profiles (8760 hours/year) for base year calibration and detailed analysis

  2. Timeslice aggregation (12-24 slices) for computational efficiency in long-term optimization

  3. Sectoral decomposition (industrial, commercial, residential) to capture different consumption patterns

Methodology Workflow

The complete process follows a multi-stage pipeline:

+--------------------------------------------------------------------+
|                    STAGE 1: DATA ACQUISITION                       |
+--------------------------------------------------------------------+
|                                                                    |
|   ERA5 Climate        Actual Load         IEA Energy              |
|   Reanalysis    +     Data (where    +    Balances                |
|   (211 countries)     available)          (sector shares)         |
|                                                                    |
|   ↓                   ↓                   ↓                        |
|                                                                    |
|   Country hourly      China: WuHaochi    Annual TWh by            |
|   total load          Europe: ENTSO-E    sector (IND/COM/RES)     |
|                                                                    |
+--------------------------------------------------------------------+
                                ↓
+--------------------------------------------------------------------+
|              STAGE 2: REGIONAL AGGREGATION                         |
+--------------------------------------------------------------------+
|                                                                    |
|   Country hourly     Region mapping      Regional hourly          |
|   profiles      +    (configurable)  →   total load               |
|                      aggregation                                  |
|                                                                    |
|   Examples: KiNESYS_ANL_Europe (34 regions)                        |
|            KiNESYS_S100D (108 regions)                             |
|                                                                    |
+--------------------------------------------------------------------+
                                ↓
+--------------------------------------------------------------------+
|           STAGE 3: SECTORAL DISAGGREGATION                         |
+--------------------------------------------------------------------+
|                                                                    |
|   Regional total    Sector shares    Sectoral hourly profiles     |
|   hourly load   +   (IEA) + Logic → (IND, COM, RES)               |
|                     assumptions                                    |
|                                                                    |
|   Key: Industrial damping factors, commercial hour factors,        |
|        residential as residual                                     |
|                                                                    |
+--------------------------------------------------------------------+
                                ↓
+--------------------------------------------------------------------+
|            STAGE 4: TIMESLICE AGGREGATION                          |
+--------------------------------------------------------------------+
|                                                                    |
|   Sectoral 8760    Timeslice      COM_FR, COM_PKFLX, G_YRFR       |
|   hourly profiles + definition →  by region/sector/timeslice      |
|                                                                    |
|   Validation: Mass balance, non-negativity, consistency checks     |
|                                                                    |
+--------------------------------------------------------------------+
                                ↓
                      MODEL-READY PARAMETERS

Data Sources

ERA5 Climate Reanalysis

Source: ECMWF (European Centre for Medium-Range Weather Forecasts)

Coverage: Global, 211 countries, hourly resolution

Weather Year: 2013 (standard), multiple years available

Methodology: Temperature-driven electricity demand modeling calibrated to historical consumption patterns

Advantages:

  • Globally consistent methodology

  • Complete spatial coverage

  • Temporally coherent (captures weather-driven patterns)

  • Multiple weather years for sensitivity analysis

Limitations:

  • Modeled data, not actual measured consumption

  • May not fully capture country-specific behavioral patterns

  • Industrial load patterns less accurate than residential/commercial

Validation: Cross-validated against ENTSO-E actual data for European countries, showing strong correlation (R² > 0.85) for total load patterns.

Actual Load Data Integration

China Provincial Load Data (WuHaochi Dataset)

Source: WuHaochi et al., Zenodo repository

Coverage: 31 Chinese provinces, 2016-2020 (5 years)

Resolution: Hourly

Data Points: 43,800 hours per province, ~1.4 million total hourly observations

Why China Replacement Necessary:

ERA5 modeled data for China showed unrealistic load curve characteristics:

  • Minimum load occurring in evening hours (18:00-20:00) instead of overnight

  • Insufficient diurnal variation for a manufacturing-heavy economy

  • Negative sectoral loads when disaggregating to industrial/commercial/residential

Actual provincial data shows:

  • Realistic overnight minimum (3:00-5:00 AM)

  • Clear industrial shift patterns (elevated daytime loads)

  • Strong seasonal variation (winter heating, summer cooling)

  • Peak loads 1.5-2.0x higher than minimum, matching empirical load factors

Processing:

Provincial loads are summed to create national aggregate. The 2016-2020 average pattern is used to create a representative annual profile, mapped to standard months/days/hours for consistency with ERA5 temporal structure.

ENTSO-E Transparency Platform

Source: European Network of Transmission System Operators for Electricity

Coverage: 35+ European countries, transmission system operator data

Resolution: Hourly actual load, real-time and historical archives

Usage in KiNESYS:

  • Validation of ERA5 load patterns for Europe

  • Quality assessment of sectoral disaggregation methodology

  • Benchmark for evaluating adaptive industrial formula accuracy

ENTSO-E data confirms that ERA5 captures seasonal and diurnal patterns well for European countries, validating its use for regions without measured data.

IEA Energy Balances

Purpose: Annual electricity consumption by sector

Sectors Covered:

  • TOTIND: Total industry (including construction, mining)

  • COMMPUB: Commercial and public services

  • RESIDENT: Residential sector

Data Quality: High-quality, internationally harmonized energy statistics

Usage: Sectoral shares (\(s_{ind}\), \(s_{com}\), \(s_{res}\)) used to disaggregate total hourly load into sectoral components, ensuring annual energy totals match reported statistics.

Sectoral Disaggregation Methodology

Mathematical Framework

Total hourly electricity load for a region is decomposed into three sectors:

\[L_{\text{total}}(h) = L_{\text{ind}}(h) + L_{\text{com}}(h) + L_{\text{res}}(h) \quad \forall h \in [1, 8760]\]
where:

  • \(L_{\text{total}}(h)\) = total regional load at hour h (from ERA5 or actual data)

  • \(L_{\text{ind}}(h)\) = industrial sector load

  • \(L_{\text{com}}(h)\) = commercial sector load

  • \(L_{\text{res}}(h)\) = residential sector load (computed as residual)

Key Constraint: Annual totals must match IEA sectoral consumption:

\[\sum_{h=1}^{8760} L_s(h) = E_s^{\text{annual}} \quad \forall s \in \{\text{ind}, \text{com}, \text{res}\}\]

where \(E_s^{\text{annual}}\) is the annual sectoral electricity consumption from IEA balances.

Industrial Sector: Adaptive Methodology

Industrial loads exhibit variation that depends strongly on regional industrial composition. The adaptive formula uses region-specific damping factors that scale with industrial electricity share:

\[L_{\text{ind}}(h) = \bar{L} \cdot s_{\text{ind}} + (\bar{L}_{\text{month}(h)} - \bar{L}) \cdot s_{\text{ind}} \cdot \alpha_s + (\bar{L}_{\text{day}(h)} - \bar{L}_{\text{month}(h)}) \cdot s_{\text{ind}} \cdot \alpha_d(s_{\text{ind}}) + (L_{\text{total}}(h) - \bar{L}_{\text{day}(h)}) \cdot s_{\text{ind}} \cdot \alpha_h(s_{\text{ind}})\]
where:

  • \(\bar{L}\) = annual average total load

  • \(\bar{L}_{\text{month}(h)}\) = monthly average for month containing hour h

  • \(\bar{L}_{\text{day}(h)}\) = daily average for day containing hour h

  • \(s_{\text{ind}}\) = industrial sector electricity share (from IEA)

  • \(\alpha_s\) = seasonal damping factor (0.01, constant for all regions)

  • \(\alpha_d(s_{\text{ind}})\) = daily damping factor (region-specific, 0.10-0.25)

  • \(\alpha_h(s_{\text{ind}})\) = hourly damping factor (region-specific, 0.10-0.25)

Empirical Motivation

Literature evidence shows industrial load variation is substantial and varies by industry type:

Empirical Industrial Load Factors

Industry Type

Load Factor

Peak/Avg Ratio

Diurnal Variation

Continuous (chemicals, refining)

60-90%

1.1-1.7x

Minimal (5-20% variation)

Batch manufacturing (food, electronics)

40-70%

1.4-2.5x

Moderate (30-50% variation)

Shift-based operations (general mfg)

50-75%

1.3-2.0x

Systematic day/night patterns

Sources: MDPI industrial park study (Jiangsu, China, 2024); ScienceDirect load clustering (Denmark, 2021); IEEE load factor reviews

Key Observations:

  1. China Demand Response: Chinese grid operators report 38 GW industrial load reduction capability at evening peak (7 PM), with >20% demand response potential in major industrial sectors. This flexibility requires substantial baseline diurnal variation.

  2. Shift Work Patterns: Manufacturing facilities commonly operate 2-shift schedules (16 hours/day, 5-6 days/week), creating systematic day/night load differences. Three-shift operations (24/7) are less common than often assumed.

  3. Mixed Industrial Portfolios: Even in industrial-heavy regions, the mix includes continuous processes, batch operations, and shift-based manufacturing, creating aggregate load variation.

Region-Specific Damping Factors

Damping factors are adjusted based on regional industrial electricity share:

\[\alpha_d, \alpha_h = f(s_{\text{ind}})\]
Adaptive Damping Factor Schedule

Industrial Share

Daily/Hourly Factor

Load Variation

Rationale

Example Regions

< 40%

0.10

1.05-1.15x

Service economy, continuous processes dominant

USA, France, UK, Australia

40-60%

0.15

1.15-1.35x

Mixed economy, blend of continuous + batch

Germany, India, Brazil, Japan

60-70%

0.20

1.35-1.60x

Manufacturing-heavy, significant batch/shift component

Poland, Turkey, Austria, Taiwan

> 70%

0.25

1.50-2.00x

Very high industrial, shift-based operations dominant

China, Iceland

Seasonal damping remains 0.01 (minimal seasonal variation) for all regions, as industrial seasonal patterns are legitimately low.

Hour-of-Day Adjustment (High-Industrial Regions Only)

For regions with industrial share > 60%, an additional hour-of-day profile factor is applied to capture shift-based patterns:

\[L_{\text{ind}}(h) = L_{\text{ind,base}}(h) \times \phi(h, s_{\text{ind}})\]
Hour-of-Day Profile φ(h):

Night hours (1-6 AM):     φ = 0.70  (reduced 3rd shift, maintenance)
Day shifts (7 AM-6 PM):   φ = 1.05  (main production hours)
Evening (7 PM-midnight):  φ = 0.90  (wind-down, avoid peak pricing)
Rationale:

  • Reflects two-shift dominant operations in manufacturing

  • Captures time-of-use pricing response (industrial users avoid peak hours)

  • Matches empirical patterns from Jiangsu industrial park study

  • Enables ~30% overnight load reduction, matching observed industrial flexibility

Validation Results

The adaptive formula successfully produces realistic sectoral loads across all regions:

China Load Shape Validation (Example)

Metric

Value

Status

Min COM_FR (Commercial)

0.019113

✓ Non-negative

Industrial load variation

1.50-1.65x

✓ Realistic

Overnight load reduction

~25%

✓ Matches shift patterns

Mass balance (sum to 1.0)

Passed

✓ Valid

All regions non-negative

108/108

✓ Complete

Commercial Sector

Commercial loads exhibit strong diurnal patterns driven by business hours:

\[L_{\text{com}}(h) = N_{\text{com}}(h) \times \phi_{\text{com}}(h)\]

where \(N_{\text{com}}(h)\) is the base commercial load (before hour factor), and \(\phi_{\text{com}}(h)\) is the hour-of-day factor that captures business activity patterns:

Hour-of-Day Pattern:

  • Overnight (12 AM - 6 AM): Baseline factor = 1.0 (minimal activity)

  • Morning ramp-up (6 AM - 2 PM): Linear increase to peak factor = 1.5 (business hours)

  • Evening ramp-down (2 PM - 10 PM): Linear decrease back to baseline

  • Late night (10 PM - 12 AM): Baseline factor = 1.0

Result: Commercial loads peak during business hours (peak factor ~1.5 at 2 PM), with reduced loads overnight and on weekends.

Residential Sector

Residential loads are computed as the residual after subtracting industrial and commercial:

\[L_{\text{res}}(h) = L_{\text{total}}(h) - L_{\text{ind}}(h) - L_{\text{com}}(h)\]

This approach:

  • ✓ Guarantees mass balance (sum equals total load)

  • ✓ Captures complex patterns (dual peaks, seasonal heating/cooling) without explicit modeling

  • ✓ Automatically adjusts to region-specific behaviors

Characteristics captured:

  • Morning peak (7-9 AM): Cooking, heating, getting ready for work/school

  • Evening peak (5-9 PM): Cooking, lighting, appliances, entertainment

  • Seasonal variation: Heating (winter) and cooling (summer) demands

  • Weekend patterns: Elevated mid-day loads due to home occupancy

Transport Sector Load Shape

Road transport electricity (EV charging) uses a separate load shape based on travel survey data.

Profile: 24 hourly fractions representing typical daily driving patterns:

  • Low overnight (12 AM - 5 AM): 0.5-1.0% per hour

  • Morning peak (7-9 AM): 5-8% per hour (commute to work)

  • Midday moderate (10 AM - 4 PM): 3-5% per hour

  • Evening peak (5-7 PM): 4-6% per hour (commute home)

  • Evening decline (8 PM - 11 PM): 2-4% per hour

Assumption: Uniform across all days of year (no seasonal variation). Applied uniformly to all regions, as charging behavior is driven by mobility patterns rather than local electricity system characteristics.

Timeslice Aggregation

Hourly load shapes (8760 hours) are aggregated into model timeslices (typically 12-24 per year) that balance temporal resolution with computational efficiency. Three key parameters are computed:

COM_FR (Commodity Fraction): Fraction of annual energy consumed in each timeslice

\[\text{COM\_FR}_{r,s,ts} = \frac{\sum_{h \in ts} L_{r,s}(h)}{\sum_{h=1}^{8760} L_{r,s}(h)}\]

COM_PKFLX (Peak Flexibility): Ratio of peak-to-average load within each timeslice (total demand only)

\[\text{COM\_PKFLX}_{r,ts} = \left( \frac{\max_{h \in ts} L_r(h)}{\text{avg}_{h \in ts} L_r(h)} \right) - 1\]

G_YRFR (Year Fraction): Fraction of annual hours in each timeslice

\[\text{G\_YRFR}_{ts} = \frac{n_{\text{hours in } ts}}{8760}\]

These parameters ensure the model properly values capacity, storage, and flexible resources based on temporal demand patterns.

Literature Foundation

The adaptive industrial formula is grounded in peer-reviewed research and industry studies:

Empirical Load Factor Studies

Industrial Load Characteristics:

  1. MDPI - Processes Journal (2024): “Load Profile Analysis for Industrial Park” (Jiangsu, China)

    • Three distinct industrial load types identified

    • Type A (food processing): 10-17× peak/trough variation

    • Type B (smelting): 2-3× variation (night-heavy operations)

    • Type C (chemicals): 1.2-1.5× variation (continuous process)

    Conclusion: Industrial loads range from very flat (continuous) to highly variable (batch manufacturing)

  2. ScienceDirect - Applied Energy (2021): “Load Profile Clustering for C&I Consumers” (Denmark)

    • Eight distinct clusters identified from smart meter data

    • Cluster 5 (daytime manufacturers): Strong shift-based patterns (7 AM-6 PM high)

    • Cluster 8 (night-shift operations): Inverted pattern (preferential night scheduling)

    Conclusion: Shift-based manufacturing has systematic (not random) diurnal variation

  3. IEEE Load Factor Literature Review (2023):

    • Continuous processes: 60-90% load factor (1.1-1.7× variation)

    • Batch manufacturing: 40-70% load factor (1.4-2.5× variation)

    Conclusion: Current formula’s 0.10 damping produces only 1.05-1.1× variation, suitable only for continuous processes

China Demand Response Evidence

ScienceDirect - Energy Policy (2022): “Demand Response during Peak Load Period in China”

  • 38 GW industrial load reduction capability at evening peak (7 PM)

  • >20% DR potential in major industrial sectors

  • Peak periods (>95% of peak load) represent only 1.6% of annual hours (~140 hours)

Conclusion: If China’s industrial sector can reduce load by 38 GW during peaks, industrial loads must have substantial diurnal variation to flex from. The “99% flat” assumption contradicts observed DR capability.

Rocky Mountain Institute (2023): “Unlocking Demand-Side Flexibility in China”

  • Industrial flexibility potential exceeds 20% of capacity in key sectors

  • Time-of-use pricing (3-5 level TOU) creates systematic load shifting

  • Many operations: 2-shift (16 hours/day), not continuous 24/7

Conclusion: Industrial loads actively managed to avoid peak hours, creating systematic diurnal patterns.

Sector Disaggregation Validation Studies

ScienceDirect - Energy and Buildings (2023): “Building Sector Load Disaggregation” (US, NYISO/ERCOT/CAISO)

  • Method: Regression-based decomposition (heating/cooling/non-thermal)

  • Validation: Re-aggregation MAPE of 3-6% across multiple balancing areas

  • Peak error: < 5% in timing and magnitude

Conclusion: High-quality disaggregation is achievable and measurable with proper methodology.

arXiv - Machine Learning (2024): “Blind Source Separation for Sectoral Load Decomposition” (Italy, 2021-2023)

  • Method: Non-negative matrix factorization on national load curves

  • Result: Hourly residential/industrial/services profiles

  • Validation: Consistent sectoral profiles across years

Conclusion: Multiple independent methodologies converge on similar sectoral patterns, validating approach.

Conclusion

KiNESYS electricity load shape methodology combines global consistency (ERA5 climate-driven modeling) with local accuracy (actual data integration where available). The adaptive industrial formula enhancement exemplifies the KAIZEN philosophy: when empirical data contradicted original assumptions, the methodology evolved to reflect observed reality based on peer-reviewed research and industry studies.

References

Climate Data:

  • Hersbach, H., et al. (2020). “The ERA5 global reanalysis.” Quarterly Journal of the Royal Meteorological Society, 146(730), 1999-2049.

Industrial Load Studies:

  • MDPI Processes (2024). “Load Profile Analysis for Industrial Park.” Jiangsu, China case study.

  • ScienceDirect Applied Energy (2021). “Load Profile Clustering for C&I Consumers.” Denmark smart meter analysis.

  • IEEE (2023). “Industrial Load Factor Literature Review.” Global survey of industrial electricity patterns.

China Demand Response:

  • ScienceDirect Energy Policy (2022). “Demand Response during Peak Load Period in China.”

  • Rocky Mountain Institute (2023). “Unlocking Demand-Side Flexibility in China.”

Disaggregation Methodologies:

  • ScienceDirect Energy and Buildings (2023). “Building Sector Load Disaggregation” (US).

  • arXiv Machine Learning (2024). “Blind Source Separation for Sectoral Load Decomposition” (Italy).

Actual Load Data:

  • WuHaochi, et al. (2021). “China Provincial Hourly Load Data 2016-2020.” Zenodo repository.

  • ENTSO-E Transparency Platform. Real-time and historical load data for Europe.