Rule-based auto-design can size a 5 MW hybrid system in 90 seconds with results within 8% of an engineer spending 3 days on manual iteration. For $200/kWh battery cost scenarios, the LP optimizer finds solutions that rule-based methods miss by 12-15% in NPV.

These numbers raise a practical question that every engineering team faces: when should you trust the algorithm, and when should you roll up your sleeves and iterate by hand? The answer is not a simple rule of thumb — it depends on the project stage, the cost sensitivity, the complexity of the multi-objective tradeoffs, and the bankability requirements of the financing partners.

Energy Optima's auto-design engine supports two fundamentally different sizing methods — RULE_BASED and LP_OPTIMIZATION — plus the option to run fully manual designs. This article walks through how each method works, compares them head-to-head on the same hybrid system, and presents a practical decision framework for choosing the right approach at each project phase.

What You'll Learn

The Two Auto-Design Methods: RULE_BASED vs LP_OPTIMIZATION

Auto-design is not a single monolithic algorithm. It is a family of approaches, each with distinct mathematical foundations, computational profiles, and output characteristics. Energy Optima implements two parallel auto-design tracks, and understanding the difference between them is essential to interpreting their outputs.

RULE_BASED applies a deterministic set of engineering heuristics — over 120 rules stored in the platform's design rule database — to produce a sized system from project inputs. It fires sequentially: evaluate the solar resource, determine inverter loading ratios, size the BESS for autonomy targets, select the diesel generator if applicable, verify cable losses stay within thresholds, and apply economic defaults. The entire sequence completes in under 90 seconds for typical systems up to 50 MW.

LP_OPTIMIZATION formulates the sizing problem as a constrained linear program. The solver searches across the feasible design space — every combination of PV capacity, BESS MW and MWh, and optional diesel capacity — to find the optimal configuration that minimizes levelized cost of energy (LCOE) subject to reliability, budget, and component constraints. The LP track typically runs in 3-12 minutes depending on system complexity and constraint density, producing a single minimum-LCOE result plus a ranked list of near-optimal alternatives.

Both methods share the same underlying component database — 112 batteries, 200+ inverters, 165 PCS units — but they arrive at their sizing recommendations through fundamentally different paths.

Rule-Based Architecture: 120+ Engineering Rules

The rule-based engine encodes decades of hybrid system design experience into deterministic logic. Each rule represents a bounded engineering judgment that an experienced designer would apply intuitively. The rules fall into seven principal categories:

Battery autonomy hours. The default autonomy target is 4 hours for grid-connected hybrid systems, with 6-8 hours for critical-load or island-mode applications. The engine adjusts autonomy based on the solar fraction, backup requirements, and the local grid reliability profile.

Depth of discharge limits. Default DoD limits are 90% for LFP chemistries and 80% for NMC, with tighter bounds applied when the system serves critical or mission-sensitive loads. The rules adjust DoD limits downward when the cycling frequency exceeds 2 full-equivalent cycles per day.

C-rate ranges. The engine enforces C-rate constraints that vary by application: 0.2-0.5C for solar shifting, 0.3-0.75C for energy arbitrage, and up to 2.0C for frequency regulation. These bounds are drawn from manufacturer warranty terms and field performance data.

Diesel sizing ratios. When the design includes a backup generator, the rule engine sizes the diesel at capacity ratios of 0.5 to 1.5 relative to the PV capacity. The ratio is determined by the critical-load fraction and the maximum allowed diesel runtime.

Cable loss defaults. The engine applies default cable loss assumptions of 1.5-3% for the DC collection system and 0.5-1.5% for the AC collection system, scaled by system voltage level. Higher losses are assumed for 480V systems; lower losses for 33 kV and above.

PV performance defaults. Performance ratio defaults range from 0.70 to 0.80 depending on geographic location, soiling assumptions, and module type. The engine applies bifacial uplift of 5-15% when bifacial modules are specified.

Economic discount rates by project scale. Default discount rates range from 6% for utility-scale projects (>50 MW) to 10% for C&I-scale projects (<5 MW), reflecting the risk premium associated with smaller, less diversified project portfolios.

The rule-based engine also applies defaults for EMS parameters — state-of-charge operating windows (10-90% SOC for most applications), ramp rate limits, and minimum reserve margins — as well as electrical standards for AC bus voltage selection (480V to 33 kV based on system scale).

Key insight: Rule-based auto-design produces a design that is fast and reasonable — typically within 8% of a manually optimized solution for standard configurations — but it cannot discover novel solutions outside the bounds encoded in its rule set. For projects with unconventional constraints, non-standard component costs, or complex multi-objective requirements, the rule-based result is a starting point, not a final answer.

LP Optimization: Solving the Capacity Problem

Where the rule-based engine applies judgment, the LP optimizer applies mathematics. It formulates the capacity optimization problem as a linear program with a clear objective and a set of hard and soft constraints.

Objective function: Minimize levelized cost of energy (LCOE) over the project life, defined as the ratio of total lifecycle costs (capital + operating + replacement + fuel) to total lifetime energy output. The optimizer can also maximize NPV when revenue streams are specified.

Decision variables: PV capacity (MW DC), BESS power rating (MW), BESS energy rating (MWh), diesel generator capacity (kVA), and inverter loading ratio (DC/AC).

Constraints:

  • Reliability constraint: Loss of load probability (LOLP) must not exceed 1% for grid-connected systems or 0.1% for island/microgrid systems
  • Budget constraint: Total installed CAPEX must not exceed the specified project budget
  • Land constraint: PV capacity must not exceed the available land area at the specified packing factor
  • Duration constraint: BESS duration (MWh / MW) must fall within application-specific bounds (e.g., 2-6 hours for solar shifting)
  • Component availability: Selected components must be available from the database or user-supplied catalog
  • Regulatory constraint: System must meet interconnection requirements for the target grid code

The LP solver explores the feasible solution space using the simplex or interior-point method, then returns the optimal design plus a ranked set of near-optimal alternatives within 5% of the optimum. This ranked set is often more valuable than the single optimal solution, because it reveals the shape of the cost surface and identifies configurations that are nearly optimal but offer better operational flexibility, faster construction timelines, or simpler supply contracts.

For a deeper treatment of the LP formulation itself, see our guide on BESS capacity sizing optimization.

A/B Comparison: Three Approaches, One System

The best way to understand the differences between manual, rule-based, and LP-optimized design is to apply all three to the same project. We sized a hybrid system with the following parameters:

System specification: 10 MW AC solar PV, 5 MW / 20 MWh BESS, grid-connected hybrid in Texas ERCOT market. 25-year project life. LFP battery at $135/kWh DC, $40/kW PCS. Inverter loading ratio: 1.25. Discount rate: 8%. Target: maximize after-tax NPV.

Metric Manual (3 days) Rule-Based Auto (90 sec) LP Auto (8 min)
PV capacity (MW) 10.0 10.5 9.6
BESS power (MW) 5.0 5.0 4.5
BESS energy (MWh) 20.0 22.5 19.0
BESS duration (hours) 4.0 4.5 4.2
Total CAPEX ($M) $8.90 $9.55 $8.43
LCOE ($/MWh) $47.80 $49.20 $46.40
NPV ($M) $5.20 $4.82 $5.45
IRR (%) 11.4% 10.9% 11.8%
Energy yield (GWh/yr) 17.5 18.2 17.0
Time to result 3 days 90 seconds 8 minutes

The manual design represents an experienced engineer iterating through candidate configurations, running hourly simulations for each, and converging on a solution judged to be the best practical tradeoff. It produces an LCOE of $47.80/MWh and an NPV of $5.20M after 3 days of focused work.

The rule-based auto-design completes in 90 seconds and produces a design within 7% of the manual LCOE ($49.20 vs $47.80) and 7.3% below the manual NPV ($4.82M vs $5.20M). It oversizes the system slightly — 10.5 MW PV and 22.5 MWh BESS — because the autonomy rules default to conservative margins. For a fast proposal, this margin is acceptable. For final investment decisions, the 0.9 percentage point IRR gap matters.

The LP optimizer produces the best result: $46.40/MWh LCOE, $5.45M NPV, 11.8% IRR. The optimizer achieves this by slightly reducing the PV capacity (9.6 MW) and BESS power (4.5 MW) relative to the manual baseline, finding a configuration that hits a better CAPEX-to-revenue balance. The LP result outperforms the manual design by 4.8% in NPV and the rule-based design by 13.1% in NPV.

Key insight: For this standard grid-connected hybrid, all three approaches produce viable designs within 13% of each other in NPV. The LP optimizer delivers the best economics, the rule-based engine delivers the fastest turnaround, and the manual approach sits in the middle on both dimensions. The choice between them depends on project stage and the cost of being wrong.

When we re-run the same comparison at a $200/kWh battery cost (representing a tight-margin, high-sensitivity scenario), the differences widen significantly. The LP optimizer finds a 4.2-hour duration system that reduces BESS capacity by 18% compared to the rule-based default, shifting capital toward additional PV capacity. At this cost point, the rule-based result misses the LP solution's NPV by 14.7% — the autonomy-default rules conservatively oversize the battery, and the penalty for that oversizing is much higher when battery costs are elevated.

When Each Approach Wins

Each sizing approach has a zone of advantage. The decision framework below maps project characteristics to the recommended method.

Rule-based auto-design wins for fast proposals and early-stage screening. When you need to generate a budget-quality sizing for a client RFP response, a grant application, or a pre-feasibility study, the rule-based engine delivers a defensible result in 90 seconds. The 5-8% accuracy gap relative to manual optimization is acceptable at this stage because the project economics have not been finalized and multiple scenarios must be explored. The rule-based approach also excels for standardized system configurations — the same 5 MW solar + 2-hour BESS template that the engineering team has deployed dozens of times. In these cases, the rule engine replicates the institutional knowledge that the team would apply anyway, and does it faster.

LP optimization wins for optimal CAPEX decisions in cost-sensitive environments. When battery costs are high, land is expensive, or the project competes for capital against alternative investments, the LP optimizer's ability to search the full design space delivers measurable returns. The 12-15% NPV advantage at $200/kWh battery costs translates into real dollars — $650K-780K on a 10 MW hybrid system. The LP result also generates a ranked alternative set that helps the project team understand the cost surface: if the optimal solution requires a component with a 16-week lead time, the second-best solution at 3% higher LCOE might use an available component and close financing a quarter earlier.

Manual design (often paired with LP as a sanity check) wins for complex multi-objective projects. When the objective function includes non-financial goals — local content requirements, minimum job creation targets, grid code compliance above the regulatory minimum, aesthetic constraints on land use — the LP's single-objective optimization cannot capture the full tradeoff surface. In these cases, the best workflow is to run the LP optimizer first to establish the Pareto frontier of optimal LCOE solutions, then shift to manual iteration to adjust the selected design for non-quantified constraints. The LP result serves as a distortion-free baseline; the manual adjustments introduce project-specific preferences on top of that baseline.

For microgrid and island systems with multiple dispatchable generators, complex load profiles, and resilience requirements, the manual-plus-LP approach is particularly valuable. These projects have too many interacting constraints for rule-based heuristics and too many non-linear objectives for pure LP. See our guide on microgrid simulation software for a treatment of the multi-objective optimization problem in island systems.

Practical Workflow: Auto First, Manual Second

The most effective hybrid-sizing workflows do not choose one method over the others — they sequence all three. Here is the recommended approach, refined through 200+ design iterations on the Energy Optima platform:

Step 1 — Run the rule-based auto-design. Enter the basic project parameters: location, system scale, application type, and component preferences. The rule-based engine produces a baseline design in under 2 minutes. This gives the engineering team an immediate sense of the order-of-magnitude sizing, the component selection, and the economic ballpark. It also flags any input inconsistencies — if the land area is insufficient for the proposed PV capacity, the rule engine catches it before the LP solver wastes cycles on infeasible solutions.

Step 2 — Run the LP optimizer with the rule-based result as the warm start. Feed the rule-based baseline into the LP solver as the initial feasible solution. The LP optimizer then searches around this starting point, tightening the cost surface and identifying improvements. Because the LP solver starts from a feasible (if suboptimal) solution, it converges faster than a cold start and produces the ranked alternative set within 8-12 minutes for most systems.

Step 3 — Review the LP results against the auto-design verdict thresholds. Energy Optima's auto-design engine scores each output against configurable verdict thresholds: green (within 5% of the theoretical optimum), yellow (5-12% from optimum), and red (more than 12% from optimum). The LP result should land in green for most standard configurations. If it lands in yellow or red, the project constraints may be too tight or the input data may have errors — this is the signal to pause and audit the assumptions before proceeding.

Step 4 — Apply manual refinements for non-quantified constraints. With the LP-optimized design as the starting point, apply manual adjustments for project-specific requirements that the LP cannot model: preferred vendor relationships, construction logistics, local content rules, land parcel geometry, or aesthetic approval conditions. Each manual adjustment should be tested by re-running the LP optimizer with that parameter fixed, so the cost of the design preference is quantified and visible to the project team.

Step 5 — Finalize and export. Use Energy Optima's quick defaults and form defaults endpoints to apply consistent design standards across the project portfolio. The quick defaults endpoint applies a standard rule set (autonomy targets, discount rates, performance ratios) to all designs in a portfolio, ensuring consistency across multiple sites. The form defaults endpoint pre-populates design inputs from saved templates, eliminating data entry errors when running the same system design across multiple locations.

How Energy Optima's Auto-Design Works in Practice

Energy Optima is the only hybrid system design platform that offers both rule-based and LP-optimized auto-design within the same workflow, with the ability to compare results side by side and switch between methods without re-entering project data.

The auto-design engine is accessed through the platform's design module. Users configure the following parameters before running either method:

  • System type: Grid-connected hybrid, island microgrid, or standalone PV+BESS
  • Application: Solar shifting, energy arbitrage, frequency regulation, diesel offset, or custom
  • Component preferences: Preferred manufacturers, minimum efficiency thresholds, max C-rate limits
  • Economic assumptions: Discount rate, project lifetime, escalation rates, tax incentives
  • Design defaults: Autonomy hours, DoD limits, cable loss budgets, inverter loading ratios

The verdict thresholds system provides immediate feedback on the quality of each auto-design output. For a given set of inputs, the engine computes the theoretical minimum LCOE achievable with the selected components and compares the actual output against this bound. The thresholds are calibrated from the full design space search — the LP optimizer essentially sweeps the entire feasible region and identifies the theoretical floor, so the verdict is a comparison against the actual physical and economic limits of the component set, not against an arbitrary target.

The quick defaults system allows engineering managers to define a single rule template and apply it across all projects in a portfolio. This is particularly valuable for EPC firms running 15-20 hybrid system designs simultaneously for different clients. Changing the default discount rate from 8% to 9% updates every project in the portfolio with one action.

The form defaults system saves complete project configurations — including component selections, economic assumptions, and design method preferences — as reusable templates. When a repeat configuration appears (the same 5 MW solar + 2-hour BESS design for three different ERCOT sites), the engineer loads the template, changes the irradiance file, and the rest of the inputs carry over automatically.

Key insight: The most productive design teams on Energy Optima use a 10-80-10 rule: 10% of projects use rule-based auto-design only (simple, standardized, fast-track proposals), 80% use rule-based as warm start + LP optimization (the standard workflow for bankable designs), and 10% layer extensive manual iteration on top of the LP result (complex multi-objective projects with unique constraints).

The auto-design engine generates a complete design report for each run, including the full bill of materials, the CAPEX breakdown by component, the annual energy production summary, the lifecycle economic analysis, and a comparison against the alternative designs in the ranked set. All outputs are exportable to the standard project deliverables format — the same format accepted by debt providers, tax equity investors, and EPC contractors — so the auto-design result can go directly into a financial model or a permit application without manual transcription.