BESS Test Matrix

Reference engine: EnergyPlus 26.1 ElectricLoadCenter:Storage:Simple plus ElectricLoadCenter:Storage:Converter (required at SimpleFixed 1.0 even on AC storage) plus ElectricLoadCenter:Distribution with AlternatingCurrentWithStorage buss type and TrackChargeDischargeSchedules operation scheme. NREL SAM Battery as a documented analytical reference for the bulk round-trip efficiency check. Matched-handshake math: CogenS flat PCS curves (1.0 at every PLR) plus EnergyPlus nominal_eff_charge = nominal_eff_discharge = sqrt(battery_dc_rte) gives round-trip = battery_dc_rte exactly on both engines.

Reference engine: EnergyPlus 26.1 ElectricLoadCenter:Storage:Simple driven via Schedule:File reading per-scenario hourly fraction CSVs over the full TMY3 weather file run period. The same per-hour command profile drives the CogenS engine via an explicit EEP profile array; the same Chicago O'Hare TMY3 outdoor dry-bulb file drives both engines. Four dispatch families: a deterministic daily peak-shave cycle plus three DOE Commercial Reference Building archetype-driven dispatches (Large Hospital, Large Office, Standalone Retail) whose hour-by-hour shape comes from real building electric loads.

Reference engine: EnergyPlus 26.1 ElectricLoadCenter:Storage:Simple as a matched-handshake against the engine's cubic PCS curve evaluated at the dispatch PLR (CogenS applies the cubic PCS coefficient to each step; EP holds nominal_eff_charge = nominal_eff_discharge constant at the PCS-evaluated value for that PLR, so the integral over the horizon matches). NREL SAM Battery as a documented analytical reference for the bulk round-trip efficiency: SAM uses a constant nominal_eff while CogenS uses a per-step cubic PCS, and the cross-engine RTE delta lands inside the 25% architectural-difference band. SOC and power envelope respect are scored as engine contract checks.

Reference engine: NREL SAM Battery model (PySAM 7.1.1) running the Custom Lifetime degradation matrix at the OEM-published 2 %/yr linear rate over a full 8760-hour year. The CogenS engine's capacity_available_kwh and the PySAM Battery batt_capacity_percent are sampled at every hour, with end-of-year values cross-validated against both the spec (CogenS internal arithmetic) and the OEM envelope (PySAM Custom Lifetime).

Reference engine: NREL SAM Battery (PySAM 7.1.1) running dispatch_choice=0 (auto-dispatch optimization) over a DOE Reference Building Large Office 8760-hour electric load profile with the LADWP-style demand-charge tariff, plus SAM Utilityrate5 for the demand-charge bill. CogenS runs the same OEM tier and same tariff through its daily LP dispatch in modules/microgrid/optimization_lp. Four implemented strategies covered: TOU energy arbitrage; self-consumption (allow_grid_charging = False); outage ride-through (LP outage_mask + outage_reserve_margin); demand-charge management. Frequency regulation, ancillary services revenue, and multi-day rolling arbitrage are not implemented in the engine today and are out of scope.

Reference engine: NREL SAM Battery (PySAM 7.1.1) at the combined nameplate (N x single-unit rated power and energy) as a documented analytical reference for the bulk round-trip efficiency. Plus engine-contract regression: each scenario also runs at num_units = 1 (unit reference) and at num_units in {2, 5, 10}, so extensive metrics scale linearly and intensive metrics stay identical.

Reference engine: EnergyPlus 26.1 ElectricLoadCenter:Storage:Simple run independently on side A, side B, and the combined-spec equivalent (kW_a + kW_b nameplate at arithmetic-mean RTE), plus NREL SAM Battery as the analytical bulk RTE reference for each side and the combined. The composition gates score CogenS sum-of-two-specs against the combined-spec engine run, and also confirm EnergyPlus shows the same blend behavior so the math is consistent across engines. FEMP M&V Tier 2 +/-5% relaxed band absorbs the small RTE-blend artifact between two specs at different battery_dc_rte values.

Reference engine: EnergyPlus 26.1 ElectricLoadCenter:Storage:Simple run independently on LFP-parameter, NMC-proxy parameter, and the combined-parameter blended equivalent. The composition gates score CogenS sum-of-two-parameter-sets against the combined-parameter engine run, AND cross-engine on EnergyPlus — proving the CogenS 0.1005% discharge-balance artifact is parameter-averaging math (not an engine bug): EnergyPlus shows the corresponding -0.18% on the same composition, and the cross-engine blend consistency lands at +0.28% architectural difference. NREL SAM Battery as the analytical bulk RTE reference per parameter set.

Reference engine: NREL SAM Pvwattsv8 (PySAM 7.1.1) generates the 8760-hour PV profile from a TMY3 weather file, and PySAM Battery in dispatch_choice=2 (behind-the-meter auto-dispatch) runs the BESS against the building electric load with a LADWP-style demand-charge tariff. CogenS runs the same OEM tier and same PV profile through its daily LP dispatch. Gates compare annual grid-import and PV self-consumption fraction against SAM, and disclose architectural-difference gates on annual charge/discharge where LP daily and SAM BTM auto-dispatch disagree on when to cycle the BESS - both engines pass tariff-driven self-consumption logic.

Reference engine: Analytical electric-load-following CHP (CHP profile that varies hour-by-hour with the building load) PLUS PySAM Battery in dispatch_choice=2 (behind-the-meter auto-dispatch) cycling against the (load - chp) residual with a LADWP-style demand-charge tariff. CogenS runs the same OEM BESS tier and same CHP+load profile through its daily LP dispatch. The CHP engine itself is independently verified by the CHP module validation report (709 in-scope scenarios across 12 stages, 100% pass rate). Role-swap from CHP Stage 9 where the CHP was the primary unit under test.

Reference engine: NREL SAM Pvwattsv8 (PySAM 7.1.1) generates the 8760-hour PV profile, analytical electric-load-following CHP supplies the (load - pv) residual, and PySAM Battery in dispatch_choice=2 (behind-the-meter auto-dispatch) cycles against the (load - pv - chp) residual against a LADWP-style tariff. CogenS runs the same full microgrid composition through its daily LP dispatch. Gates compare on-site self-supply fraction and grid import against SAM, and disclose architectural-difference gates on annual charge/discharge where LP daily and SAM BTM multi-day auto-dispatch disagree on when to cycle the BESS within the multi-source composition.

Reference engine: PySAM 7.1.1 Battery (StandaloneBatteryCommercial) in dispatch_choice=3 price-signal-aware look-ahead optimization - both CogenS LP and SAM see the SAME problem (TOU arbitrage + monthly demand-charge shave) under an SCE GS-3 style commercial tariff ($40/kW-month demand + TOU $0.08/$0.20/$0.42 per kWh) with grid-charging enabled on both sides. DERs right-sized so the BESS is the primary economic driver (PV+CHP supplies only ~36% of building load). Scoring is OUTCOME-based (cycles, capacity retention, 20-yr NPV with battery replacements priced when cumulative EFCs hit the lifetime limit, IRR, payback) rather than dispatch-delta - the question being answered is what value does the customer realize, not whether the dispatch paths match. Tolerance bands are ONE-SIDED for advantage metrics (CogenS doing BETTER than SAM PASSES at any magnitude; CogenS doing worse by more than the regression band FAILS). Additionally Layer B cross-validated against REAL PySAM Singleowner 7.1.1 on the 20-year financial waterfall itself (the bess_financial_kpi.compute_engine_kpis engine that drives the workbook): 3/3 representative scenarios PASS at exactly 0.0000% NPV deviation, exercising the FULL after-tax chain (21% federal tax + MACRS 5-yr GDS depreciation + 7% discount + 20-year horizon). PySAM driver: scripts/run_validation_stage10b_bess_pysam_xval.py.

Reference engine: REAL PySAM Singleowner 7.1.1 (NREL SAM C++ kernel via PySAM bindings) running the actual SAM kernel - not just numpy_financial's closed-form equivalents. The CogenS Building Owner NPV/IRR/payback is cross-validated against PySAM Outputs.cf_project_return_aftertax_npv, analysis_period_irr, and the cumulative cashflow zero-crossing payback. All 8 scenarios PASS at exactly 0.0000% deviation - bit-for-bit cross-engine match. The PySAM driver explicitly zeroes the FlatPlatePVSingleOwner default PTC ($0.03/kWh x 10 yr at 2.5% escalation), ITC percentages, and CBI/IBI/PBI incentives to match the CogenS simple pre-tax Building Owner model. numpy_financial 1.0.0 Newton-Raphson IRR retained as a supplementary closed-form cross-implementation. IRS Publication 946 MACRS 5-year GDS half-year convention provides the depreciation schedule. PySAM driver: scripts/run_validation_stage11_bess_pysam_xval.py.

Reference engine: REAL PySAM Singleowner 7.1.1 (NREL SAM C++ kernel) running in pre-tax validation mode - all 8 scenarios PASS at 0.0000% deviation across Year-1 revenue, Year-N geometric escalation (1.03^19), 20-year revenue sum, AND PySAM NPV at the target provider IRR = $0 (the solver's definitional contract). PySAM Singleowner exactly confirms that the CogenS-solved EaaS rates produce zero NPV at the target IRR. numpy_financial 1.0.0 Newton-Raphson IRR retained as a supplementary closed-form cross-implementation. The CogenS BESS EaaS model is RATE-BASED (capacity charge in $/kW-month plus throughput charge in $/kWh, both escalating annually), NOT benefit-sharing - the CHP Stage 12 EaaS validation established the rate-based pattern correction; this stage validates the analogous two-stream BESS solver against real PySAM. PySAM driver: scripts/run_validation_stage12_bess_pysam_xval.py.