Baseline
Historical load with no battery. The reference case for all savings calculations.
TowerWatch
Study Methodology
Methodology
How the TowerWatch dispatch study works: from telemetry to economic analysis. All assumptions are frozen and version-controlled. Every output carries provenance labels.
1. Telemetry Foundation
Source: UC Davis PI Web API historian. 52,093 hourly readings from 2020–2026.
Validated signals: Building demand (electricity_kw), bidirectional net energy meter (electricity_net_kwh), steam, chilled water, OAT.
Derived signals: Import/export energy from net meter deltas. PV generation reconstructed from energy balance.
Coverage: Demand from 2020-03. Net meter from 2023-04. Hourly resolution after resampling.
2. Feature Generation
Raw signals are resampled to hourly and transformed into analysis features.
gross_load_kw: Site demand from MSB meter
net_load_kw: Gross load minus PV generation
pv_generation_kw: Direct inverter or reconstructed
import/export_kwh: Hourly deltas from bidirectional net meter
ramp_rate: kW/min rate of change of net load
rolling_peak_kw: 30-day trailing maximum for demand charges
3. Bill Calculation
Energy charges: Interval-by-interval at TOU rates. Peak/off-peak/off-peak-saver by season.
Demand charges: Billing-month peak × site infrastructure rate + summer peak TOU × summer peak rate.
Fixed charges: $412.90/month.
Export credit: $0.05/kWh proxy. Holidays excluded from peak classification per SMUD rules.
4. Dispatch Strategies
Peak Shaving
Discharge when load exceeds 80% of observed peak. Charge during low-load off-peak hours. Objective: reduce demand charges.
TOU Dispatch
Charge during off-peak/saver periods. Discharge during peak TOU hours. Objective: reduce energy charges via arbitrage.
Forecast-Assisted
Persistence forecast scaffold. Pre-charge before predicted peaks. Will be replaced by ML models in the next phase.
5. Battery Model & Degradation
Simulator: Interval-level charge/discharge with SOC bounds (10–95%), 87% round-trip efficiency, mutual exclusivity per interval.
Degradation: Linear additive — cycle fade (EFC / 2,000 cycle life) + calendar fade (2%/year). Replacement at 60% remaining.
Source: Simulated. No battery telemetry exists in PI for this site.
6. Sizing & NPV
Sweep: [50, 100, 200, 400] kWh × [25, 50, 100] kW. Skip power > capacity.
NPV: 20-year horizon. 5% discount rate. $150/kWh capex. 70% replacement cost. 2%/yr tariff escalation. $5/kWh-yr O&M.
Sensitivity: One-at-a-time sweeps on discount rate, capex, and calendar fade rate.
7. Limitations
Battery is simulated. Real degradation curves are nonlinear. Thermal effects and BMS constraints are not modeled.
PV is reconstructed. Energy-balance reconstruction introduces uncertainty. Direct inverter data will replace this when ingested.
Tariff is a proxy. Published SMUD rates, not the actual UC Davis contract.
Linear degradation. Conservative for the comparison structure (relative rankings hold) but absolute years-to-replacement may differ.