Conductor · WEM forecasting and dispatch by Weaponised Nerds

Tomorrow

Loading tomorrow's forecast…

Forecast by Ngangk from WEM market history, live demand and the BOM weather forecast.

Daily plan · draft for approval

Charge when power is cheapest, hold, and sell into the highest prices.

–Average charge price

–Average discharge price

–Spread captured

Rule of thumb: charge in the cheapest 4 hours below the daily median price and discharge in the dearest 3 above it. It does not yet use Zenith's battery size or state of charge.

WEM forecast, midnight to midnight

48 half-hours, 00:00 to 23:30 AWST, driven by the BOM ACCESS weather model

GET /api/ngangk/forecast?horizon=24&day=tomorrow

Peak demand -- MW At --

Lowest demand -- MW At --

Peak rooftop solar -- MW At --

Price range -- -- negative-price intervals

Demand and rooftop solar

Operational demand in MW, per 30-minute trading interval

Demand, most likely Likely range (P10 to P90) Rooftop solar

Wholesale price and battery schedule

$/MWh per trading interval, with the plan's battery action underneath

Price Negative price Charge Discharge Hold

Trading Interval Forecast Curves (24h)

P10/P50/P90 Operational Demand, Rooftop Solar, and Reference Trading Price

P50 Demand

P10-P90 Range

Rooftop Solar DPV

Wholesale Price ($/MWh)

48-Interval Detailed Data Table

Half-hourly, midnight to midnight (AWST)

Interval	Temp (°C)	Solar GHI	Demand P10	Demand P50	Demand P90	Solar DPV (MW)	Wholesale RTP	Duck Curve %

7-Day Multi-Horizon Outlook

336 Intervals

Macro multi-day weather wave prediction for fuel, maintenance, and storage planning

GET /api/ngangk/forecast?horizon=168&day=today

7-Day Rolling Diurnal Wave (336 Intervals)

Captures multi-day weather cycles, weekday industrial peaks vs weekend solar gluts

All 7 Days Interval Records (336 Intervals)

Standard 30-min Resolution

Timestamp	Interval	Temp (°C)	Demand P50 (MW)	Solar DPV (MW)	Wind (MW)	Wholesale RTP	Duck Severity

Physical Asset Management & Dispatch Solvers

Select customer tier to model co-optimized dispatch, peak shaving, and energy sinks

SWIS Grid Outage & Maintenance Simulator

145 Network Facilities

Select any generating unit or grid battery across the South West Interconnected System to take offline for scheduled or forced maintenance. The Ngangk ML model dynamically shifts the wholesale clearing price upwards, allowing you to evaluate your portfolio's dispatch response and revenue upside.

Offline Capacity: 0 MW

0 offline

Loading 145 SWIS network generators...

Your Generating Portfolio

Configure your asset, cost curve & bidding strategy

Asset Type

Asset Size (Capacity) 100 MW

MW

Production Cost ($/MWh)

WEM Avg: $95.00/MWh

$ /MWh

$0 Adjust slider to reflect actual plant cost $250

Target Profit Margin ($/MWh)

Desired margin above production cost

+$20.00 / MWh

$0 (Break-even bid) Adjusts market bid per interval +$150/MWh

Resulting Market Bid

Cost ($95) + Margin ($20)

$115.00 / MWh

Dispatches when Spot Price ≥ Bid

Total Profit for the Day Profitable

+$0 AUD

Net profit across 48 trading intervals after production costs

Total 24h Revenue

$0 AUD

Total Production Cost

$0 AUD

Dispatched Energy

0 MWh

Capacity Factor / Run

0% (0 hrs)

Daily Co-Optimized Dispatch Schedule

Merit order dispatch against predicted wholesale RTP and FCESS prices

Projected 24h Profit

$0 AUD

Gross Revenue: $0 AUD

Interval	Spot Price	Market Bid	Dispatched	Energy	Revenue	Cost	Interval Profit

Facility Load & DER Sizing

Configure behind-the-meter assets and operating shift hours

Base Industrial Load (MW)

12 MW

Operating Hours Profile

Behind-the-Meter Solar (MWp)

6 MWp

On-Site BESS Storage (MWh)

16 MWh

Peak Shaving Strategy

Western Power RT4/RT7 network tariffs penalize maximum coincident peak demand between 17:00 and 20:00. BESS discharges to ensure net grid import never exceeds baseload during this window.

Peak Grid Demand Reduced

-- MW

Avoided coincident peak

Monthly Network Charge Saved

-- AUD

Based on $18.50/kVA Western Power RT4

Solar Self-Consumption Rate

92.5%

Zero solar export curtailment

Behind-the-Meter 24h Load & BESS Trajectory

Interval	Base Load (MW)	Solar Gen (MW)	BESS Action	BESS SoC	Net Grid Import	Peak Shaving Status

SUPER OFF-PEAK SOLAR SOAK ZENITH MICROGRID SIGNAL

Rooftop solar generation is peaking across the SWIS. Wholesale electricity price is negative (-$14.50/MWh). Plug in your EV and turn on high-draw appliances now to consume free solar energy.

Current Grid Price

-$14.50 / MWh

VPP signal console

Override the broadcast for each asset class and target a tranche of subscribers. Gateways outside a tranche stay on their normal schedule.

Subscribers – Mode –

Market signal Hold for Reason shown to homes

Asset class	Directive	Setting	Tranche (% of cohort)	Reach

Frame preview

Automated Appliance Energy Sinks HomeAssistant Control Targets

Estimated Daily Savings: $6.45 AUD

Household Solar & Battery State-of-Charge Profile

Interval	RTP Price	Rooftop Solar (kW)	Home Load (kW)	Home Battery (kW)	Battery SoC	Active Smart Sinks

Model Accuracy & Retrospective Evaluation

Closed-Loop MLOps

Continuous backtesting against realized AEMO WEM settlement data & BOM observations with online adaptive bias correction

Demand MAPE (7d)

--%

Benchmark: ≤ 3.5%

Demand RMSE (7d)

-- MW

MAE: -- MW

Price MAE (7d)

$--

Wholesale RTP $/MWh

P10-P90 Coverage

--%

Target: ~80.0% Band

Spike / Negative F1

--%

Directional Accuracy

Online Bias Correction

--

Active Edge Adjustment

Select Evaluation Date:

Predicted (P50)

Confidence Band (P10 - P90)

Realized Actual Outcome

Day MAPE: --% | --

48-Interval Residual Error & Weather Attribution Audit

Granular interval comparison of forecast errors against actual temperature, solar irradiance, and dispatch outcomes

Intervals for Selected Date

Interval (AWST)	Forecast P50	Actual Outcome	Residual Error (Δ)	P10 - P90 Band	Confidence Status	Weather Δ (Temp / GHI)

Fast Loop: Online Edge Bias Correction

The Cloudflare Worker continuously tracks the rolling exponential moving average of 48-interval residuals. If systematic seasonal under- or over-prediction is detected, the inference engine automatically recalibrates diurnal intercepts in memory at edge execution time.

Active Correction: ΔDemand = -- MW | ΔRTP = -- $/MWh

Slow Loop: Scheduled Offline Retraining

Nightly settlements append validated ground-truth intervals to the training warehouse. When cumulative 30-day drift triggers or on monthly cadence, local/CI Python training fits updated Ridge and GBDT models, validating out-of-sample RMSE before deploying weights.

Pipeline Status: Synchronized Total Records: 84,240+