A microgrid is a system consisting of interconnected loads and distributed energy resources with well‑defined electrical boundaries. It can connect to or disconnect from the utility grid, enabling operation in both grid‑tied and islanded modes. Resource conditions, load characteristics, power‑supply reliability requirements and economic constraints vary substantially across scenarios, leading to vastly different technical combinations. Typical technologies deployed in microgrids include renewable energy sources, energy storage, conventional generating units and loads. When multiple technologies are coupled, the variable space for capacity sizing expands dramatically. Combined with flexible loads, interruptible loads and dynamic electricity tariffs, traditional spreadsheet calculations and empirical rules can hardly yield a true global optimum.
In engineering practice, deviations in capacity configuration are often attributed to optimization algorithms. More commonly, however, deficiencies occur in upstream stages: ambiguous project nature, insufficient data accuracy, and failure to incorporate constraints into modelling. Such deviations propagate downstream and cannot be remedied by optimization algorithms in subsequent steps.
The credibility of microgrid sizing results hinges on four stages: scenario definition, data preparation, model solving, and result delivery. The methodological framework described herein also underpins EcoSizing, an intelligent microgrid planning and sizing platform.
The following sections elaborate technical key points and common pitfalls for each of the four stages.
1.Scenario Definition
Microgrid capacity sizing essentially converts an engineering optimization problem into a mathematical one for solution. To ensure the mathematical model accurately represents the real‑world problem, operating scenarios and technical mixes of the microgrid must be clearly defined to determine parameters, variables, constraints and objective functions. Before commencing capacity calculation, the following four items shall be established.
1.1 Scenarios and Technologies
Scenario and technology definition constitutes the first step of microgrid sizing. It clarifies where the project will be built, who it serves, what resources supply power, and what constraints apply, laying foundations for subsequent parameter setting, variable definition, constraint modelling and objective‑function formulation. Definition is completed at three levels:
First, clarify project nature and power‑supply objectives. Distinguish between grid‑tied and islanded configurations, greenfield new builds versus retrofits of existing assets, and primary objectives such as economic optimality, reliability priority or renewable‑energy penetration targets. For instance, remote island projects prioritise autonomous off‑grid power supply; industrial parks focus more on peak‑valley arbitrage and demand management; critical‑infrastructure projects rank power continuity above economics. Project nature directly guides objective‑function selection and sets the stringency of subsequent constraints.
Second, define resource conditions and load characteristics. Resource conditions include local solar irradiance, wind‑speed distribution and fuel prices. Load characteristics cover load profiles and proportions of flexible / interruptible loads. For projects incorporating charging‑swapping facilities, EV charging timelines and V2G participation potential should additionally be considered. These inputs feed into data preparation, and their accuracy directly shapes sizing credibility.
Third, determine the technology mix. Select technologies for optimisation according to scenario requirements. Common options include photovoltaic generation, wind power, battery energy storage, diesel generators, gas turbines, charging‑swapping facilities and flexible loads. Once the technology mix is fixed, sets of parameters, variables and constraints are established, determining the mathematical scale and complexity of the sizing problem.
1.2 Business Models
Both business models target full‑lifecycle economic optimality, but differ in accounting entities and revenue components.
Under the PPA (Power Purchase Agreement) model, the project developer invests, builds and sells electricity to end‑users. Revenues stem from power sales at contracted tariffs. Calculations centre on capital investment, contract tariffs, power‑sale income and operating costs.
Under the self‑investment model, the property owner builds assets on‑site within its own park. Benefits take the form of reduced electricity expenditure. Analysis focuses on capital outlay, electricity‑cost savings and long‑run operating expenses.
Identical projects yield different optimal configurations under the two models. Sizing carried out without confirming the business model will produce configurations that diverge significantly from the actual investor’s objectives.
1.3 Project Phase
Planning for new projects falls under design optimisation: equipment capacities and quantities are decision variables to be solved, with calculations delivering inputs for preliminary investment assessment and scheme selection.
Evaluation of existing configurations belongs to operational optimisation: equipment layouts are fixed, while operating strategies are solved variables. Analysis assesses reliability and economics of given schemes.
1.4 Constraints
Constraints should be embedded during modelling rather than checked only after results are generated. Solving first and validating afterwards is a major source of rework. Typical constraints include: ‑ Upper limit of capital budget ‑ Maximum installable PV capacity and available site area ‑ Solar‑to‑storage ratio and grid‑forming constraints ‑ Required power‑supply hours and renewable‑energy penetration targets ‑ Grid‑connection capacity ceiling and anti‑reverse‑power‑flow limits ‑ Power balance ‑ Energy balance
2.Data Preparation
Data quality sets the lower bound of calculation errors. Deviations in this stage are extremely prevalent in engineering practice.
2.1 Limitations of Typical‑Day Approaches
To reduce computational burden, traditional planning widely adopts typical‑day methods: representative load and generation profiles are extracted and extrapolated to full‑year performance. This approach discards temporal dynamics and creates three categories of bias:
Economic weights of extreme‑value periods are diluted. Demand charges are determined by the single peak demand within each billing cycle. When typical‑day profiles are extrapolated by day‑weighting, marginal weights of peak‑demand instants are reduced, and the required power capacity for peak‑shaving is underestimated.
Cross‑temporal continuous operation of energy storage is broken. Typical days are independent with energy reset daily. Cross‑day energy shifting such as “charge over weekends, discharge on workdays” cannot be represented, hence undervaluing the capacity value of long‑duration energy storage.
Typical‑day selection criteria mismatch investment‑decision criteria. Typical days are selected based on profile similarity to other days. By contrast, investment scale is driven by the most costly operating instants. These two evaluation logics differ; errors can be positive or negative, and adding more typical days does not guarantee convergence toward real‑world annual performance.
2.2 Technical Requirements for Four Categories of Data
Meteorological data: Typical Meteorological Year (TMY) data covering irradiance, temperature, wind speed and other parameters shall be obtained based on project latitude‑longitude coordinates to support renewable‑generation calculation. Regional resource disparities must be reflected in the model.
Renewable‑energy output data: Full‑year hourly generation profiles capturing seasonal and intraday variations are required. Replacing time‑series curves with annual utilisation hours alone cannot support generation‑load matching analysis or energy‑storage demand assessment.
Load data: Measured 8760‑hour annual load profiles are preferred. Where measurements are unavailable, full‑year datasets can be generated via industry‑specific load templates, scaling from annual or monthly energy consumption, or combining weekday‑weekend typical profiles. Hierarchical and adjustable loads shall be defined to differentiate power‑supply priorities. This definition directly determines the feasibility of flexible load‑shedding strategies and equipment sizing for islanded scenarios.
Grid data: Comprises purchase and feed‑in tariffs, time‑of‑use rates, demand charges plus billing rules, maximum import‑export power and anti‑reverse‑power‑flow interconnection constraints. Such data shape the revenue structure.
2.3 Matching Data Accuracy with Project Stages
Incomplete data does not halt analysis. A tiered workflow is recommended: conduct detailed analysis with sufficient data; deliver preliminary assessments with limited inputs; refine outputs as more data becomes available. It is critical to explicitly state assumptions underpinning results and associated error margins.
3.Solution Methodology
It is not uncommon in industry to witness capacity outcomes differing by a factor of two for the same microgrid when solved by different parties. Solving methodologies are the primary cause of such large discrepancies.
3.1 Limitations of Heuristic Methods
Heuristic stochastic algorithms including Genetic Algorithm (GA), Particle Swarm Optimisation (PSO) and Bayesian Optimisation (BO) have been widely deployed for microgrid planning. These approaches offer modelling flexibility and accommodate non‑linear constraints. Nevertheless, when handling 8760‑hour time‑series datasets and multi‑technology coupling problems, they tend to converge to local optima and demand lengthy computation.
More importantly, result determinism suffers. Stochastic mutation mechanisms in heuristic solvers produce divergent outputs for identical input parameters across multiple runs. Non‑reproducible results are hard to accept by institutional investors during due‑diligence reviews.
Similar drawbacks apply to static Excel‑based models and rule‑based dispatch strategies. When user‑side electricity prices fluctuate hourly or even higher‑frequency under spot markets, fixed rules cannot mathematically converge toward global optimality.
3.2 Mixed‑Integer Linear Programming (MILP)
Reconstructing microgrid asset models within a Mixed‑Integer Linear Programming (MILP) framework enables co‑optimisation of equipment capacities and operational strategies.
Decision variables include integer / binary sizing variables (quantity and model selection for PV, energy storage, inverters and charging facilities) as well as continuous / binary operational variables (grid import‑export, storage charge‑discharge and state‑of‑charge, load supply status).
Constraints cover hourly power balance, equipment rated limits, cross‑period state‑of‑charge dynamics for storage, demand‑charge and grid‑interconnection boundaries, generating‑unit operating limits, charging‑swapping coupling relationships and annual service‑level requirements.
This framework delivers two key engineering benefits:
-
Solutions represent mathematically global optima rather than approximate feasible results from trial‑and‑error iteration.
-
Every charge‑discharge decision is traceable and interpretable for audit purposes.
3.3 Energy‑Storage Degradation and Replacement
After commissioning, energy‑storage assets typically operate for up to twenty years. Electrochemical storage suffers calendar ageing and non‑linear cyclic degradation, especially under frequent deep‑discharge conditions. Financial models built on static assumptions of “constant performance over full lifecycle” introduce two structural flaws:
-
Substantial usable‑capacity degradation after 5‑10 years and associated revenue erosion are not captured.
-
Timing of replacement costs is not predicted, leading to distorted full‑lifecycle IRR metrics.
Accordingly, complete models should incorporate multi‑factor lithium‑ion battery ageing models, taking depth‑of‑discharge (DOD), charge‑discharge rates, cycle life and calendar life into account. When user‑defined degradation thresholds are hit, capital expenditure for replacement or capacity replenishment is activated and new operational profiles are adopted.
For islanded systems, achievable power‑supply hours depend on real usable storage capacity. Reliability will fall below design targets if degraded storage is not replenished. Since PV and battery lifespans differ, storage units will reach end‑of‑life ahead of project completion in most cases, hence replacement investment must be included in calculations.
4.Result Delivery
Final sizing deliverables should constitute comprehensive documentation supporting scheme review and investment decision‑making. Document completeness determines whether a project can progress to subsequent phases. Deliverables fall into two categories.
4.1 Technical and Engineering Outputs
Direct simulation outputs include sizing results, operating strategies and operational profiles. BOM lists plus primary / secondary wiring diagrams are engineering documents derived from these outputs.
- Sizing results: Capacity, power rating, quantity and summary configuration of each piece of equipment, supporting equipment selection, scheme review and investment estimation.
- Operating strategies and simulation plans: Storage charge‑discharge patterns, renewable generation outputs, grid import‑export and supply‑demand matching relationships to explain system behaviour and revenue formation mechanisms. ‑
- Operational profiles: Full‑year 8760‑hour time‑series of power flows, storage states and load supply conditions, enabling performance review for critical periods and identification of operational constraints.
- BOM list: Itemised equipment and material specifications for quotation reference and delivery verification.
- Primary wiring diagram: Main‑circuit equipment and electrical connections to visualise system architecture.
- Secondary wiring diagram: Control, measurement, protection and communication interconnections to guide installation and commissioning.
4.2 Economic‑Analysis Outputs
- Investment breakdown: Itemised capital costs for equipment, construction and other upfront outlays.
- Cash‑flow statement: Annual revenues, expenditures and net cash flow. ‑ Investment performance metrics: IRR, NPV and payback period.
- Levelised Cost of Electricity (LCOE): Metric for comparative evaluation of power‑supply schemes from a cost perspective.
- Financing outputs: Cash‑flow and capital‑return analysis under loan scenarios.
Financial modelling shall cover initial capital expenditure, annual operating expenses, capacity replenishment and replacement costs, debt servicing, depreciation, taxes and residual asset value.
Technical outputs support scheme review; economic outputs inform investment decisions. Both are indispensable. Sizing results confined to spreadsheets cannot pass technical reviews or facilitate financing negotiations and will stall project advancement. Standardised documentation with defensible conclusions carries greater practical importance than capacity values precise to several decimal places.
5.Deviations between Simulation and Real‑World Operation
Even with rigorous execution of the four preceding stages, gaps persist between simulated and field performance. Cases where simulated IRR reaches 20 % yet the first‑year realised IRR stands at merely 15 % are not rare. Such discrepancies mostly stem from non‑ideal physical characteristics rather than algorithmic defects.
Two primary sources of deviation are identified:
-
Modelling deviation: PV generation and load profiles are probabilistic forecasts built on historical data. Variations in factory utilisation rates, extreme weather and unplanned outages drive actual generation and loads away from modelling assumptions, breaking pre‑computed charge‑discharge schedules.
-
Execution deviation: Simulations assume perfect execution of dispatch orders by the Energy Management System (EMS). In reality, generation‑load forecast errors, state‑of‑charge estimation bias and equipment efficiency deviations distort theoretical arbitrage actions.
Refined modelling can reduce but never eliminate these gaps. A pragmatic solution builds correction mechanisms: introduce revenue‑loss factors during sizing to offset execution‑related losses; deploy real‑time correction algorithms in operation to adjust day‑ahead dispatch dynamically according to actual operating conditions.
Deviations can be mitigated but not eliminated. Therefore, engineering‑grade sizing conclusions must explicitly describe sources and magnitudes of uncertainties.
Pre‑Sizing Checklist
| Stage | Review Checklist |
|---|---|
| Scenario Definition | Are the asset portfolio, business model, and project stage clearly defined? Have key constraints been incorporated into the modeling assumptions? |
| Data Preparation | Which of the four data categories are based on measurements, and which rely on estimates? Are the uncertainty ranges of the estimates clearly specified? |
|
Solution Methodology |
Is the solution globally optimal? Are the results reproducible? Have degradation and replacement costs been accounted for? |
|
Result Delivery |
Are the deliverables ready for direct use in project proposal reviews and financing discussions? |
Click the image to open the full spreadsheet. This checklist enables quick self‑assessment prior to commencing sizing work.
About EcoSizing
EcoSizing is an intelligent microgrid planning and sizing platform built around the four‑stage methodology above.
Adopting Mixed‑Integer Linear Programming as its core solver, the platform completes system‑optimal sizing within five minutes without sacrificing engineering rigour. It combines high‑fidelity full‑time‑series 8760‑hour simulation with full‑lifecycle financial modelling. By solving hourly dispatch across asset lifespans, it generates investment‑analysis reports suitable for financing purposes, including cash‑flow, NPV, IRR, LCOE and payback period outputs formatted to satisfy lender and institutional‑investor due‑diligence requirements.
For complete methodological descriptions and engineering case details, refer to White Paper on Intelligent Microgrid Planning and Sizing.
