Lessons

TOPIC 60 · APPLIED CAPSTONE

Final capstone: Australian energy market decision

Integrate engineering, markets, law, finance, data, safety and people.

APPLIED CAPSTONE · TOPIC 60

Defend a resilient operating decision

Use the worked data-centre project as an evidence pack. A proposed staged load reaches the normal 42 MW import ceiling, but a new planning scenario reduces available import to 35 MW for the hot two-hour interval. Critical load remains 16 MW. Backup remains 8 MWh usable AC energy with a 12 MW power limit. Prepare a decision pack that connects physical feasibility, commercial exposure, control-system evidence and accountable escalation.

1. Establish the evidence pack

  • Use the published worked-site inputs in the practical project; preserve that version in your pack.
  • Hot interval: proposed staged IT load 30 MW × 1.40; stress import ceiling 35 MW.
  • Critical load: 16 MW; backup: 8 MWh and 12 MW; nominal duration target: 30 minutes.
  • Do not assume backup can serve all critical load, or count reserved backup as routine peak shaving. This is a planning exercise, not switching or emergency-response instructions.
Open the worked data-centre case

2. Produce the deliverables

  1. A boundary diagram and workbook comparing original, staged and stressed load cases; calculate the stress-compliant IT ceiling at the stated ratio.
  2. An updated commercial reconciliation with unchanged contract cover, explicitly excluding the unpriced loss of computing service.
  3. A control and assurance register: evidence, accountable role, decision gate and escalation trigger for each material physical, cyber, safety or commercial risk.
  4. A two-page decision memo plus a five-minute defence outline; reconcile or supersede assumptions carried forward from Topics 30 and 50 rather than treating their unrelated synthetic figures as this site's data.

3. Review the work, not just the answer

For each dimension, record 0 — Not yet, 1 — With guidance, or 2 — Independent. Explain your rating. Correct material physical or numerical errors before sharing; do not average them away.

Accuracy

Reconcile the 35 MW stress boundary, IT/facility ratio and both backup constraints; do not claim safe operation from arithmetic alone.

Traceability

Retain the source case version, changed inputs and linked workbooks; distinguish this scenario from historical or operational data.

Judgement

Separate screening, engineering validation and operating authority; resolve contradictions and specify missing evidence.

Communication

Present a conditional decision with owners, go/no-go gates and review triggers, ready for qualified review.

Use the Self-review tab to rate your evidence against all four dimensions. A quiz does not complete this capstone. No automatic marking, tutor review or credential is provided; arrange qualified feedback separately.

Check the numerical anchors after doing your own working

30 MW IT × 1.40 = 42 MW facility demand. Against 35 MW import, that is a 7 MW gap. At the same ratio, 35 ÷ 1.40 = 25 MW IT is the arithmetic import ceiling, not an operating approval.

8 MWh ÷ 16 MW is 30 minutes on energy alone, but 12 MW backup cannot serve 16 MW of critical load. The 4 MW power shortfall remains. A demand or contract change does not resolve it.

Supporting lesson and optional data practice

Background study is separate from the decision pack. Do not combine unrelated teaching datasets as if they describe one real site.

The lesson

0 chapters
Browse the chapters
    GUIDED ANALYST LAB · TOPICS 31 / 32 / 35

    From raw intervals to a decision

    Clean a small dataset, test a forecast and explain a cost exposure. Use Excel or your preferred analysis tool; this page checks selected results, not your code or judgement.

    Your inputs and memo stay here when you switch sections, topics or resources in this open page. Export before reloading, closing or opening another page. This is a temporary draft, not an account-saved submission.

    New to data analysis? Start with one row

    An interval is a fixed period of time. This lab uses five-minute intervals. A revision is an updated version of a record; keep the latest valid version at the stated cutoff rather than adding both versions.

    1. Read the demand table: find the first clean interval with 60 MW average demand.
    2. Convert five minutes to hours: 5 ÷ 60. Energy = 60 × 5 ÷ 60 = 5 MWh.
    3. Match that interval and region to its A$50/MWh price. Cost = 5 × 50 = A$250.
    4. Repeat for the other intervals. Add MWh and dollars separately; never add MW readings and call the sum energy.

    MAE means mean absolute error: average the size of each forecast error, ignoring its sign. A zero average signed error can hide errors that cancel out. First reproduce one row; then follow the full method below.

    1. Download and inspect

    Synthetic site demand in NSW1: six five-minute intervals ending 09:05–09:30 AEST on 18 September 2026. Average MW, not instantaneous SCADA or dispatch targets. Extract cutoff: 10:00 AEST. All listed revisions are available by the cutoff. Spot energy cost only: exclude losses, network and retail charges, taxes, hedges and credit. Forecasts are assumed issued at 08:55 using only earlier information; they are invented, not fitted models.

    Download raw demand CSVDownload regional prices CSVDownload forecast comparison CSV
    Raw demand · 8 raw rows
    Raw demand — synthetic teaching inputs
    interval_end_aestsiteregionrevisionaverage_mw
    2026-09-18 09:05SITE1NSW1160
    2026-09-18 09:10SITE1NSW11100
    2026-09-18 09:10SITE1NSW12120
    2026-09-18 09:15SITE1NSW11180
    2026-09-18 09:20SITE1NSW11120
    2026-09-18 09:20SITE1NSW11120
    2026-09-18 09:25SITE1NSW1160
    2026-09-18 09:30SITE1NSW11180
    Regional prices · 7 raw rows
    Regional prices — synthetic teaching inputs
    interval_end_aestregionaud_per_mwh
    2026-09-18 09:05NSW150
    2026-09-18 09:10NSW1-20
    2026-09-18 09:15NSW1100
    2026-09-18 09:15VIC140
    2026-09-18 09:20NSW1200
    2026-09-18 09:25NSW10
    2026-09-18 09:30NSW180
    Forecast comparison · 6 raw rows
    Forecast comparison — synthetic teaching inputs
    interval_end_aestbaseline_mwcandidate_mw
    2026-09-18 09:056065
    2026-09-18 09:10100115
    2026-09-18 09:15160170
    2026-09-18 09:20100130
    2026-09-18 09:258065
    2026-09-18 09:30160175
    2. Build the analysis
    1. Preserve raw sheets. Import CSVs into separate sheets. Store timestamps consistently as AEST; this exercise uses a fixed UTC+10 convention, not your computer's local daylight-saving conversion.
    2. Resolve records. Identify the exact duplicate. For each interval and site, retain the highest revision available at the stated cutoff. Conflicting rows at the same revision need investigation, not an arbitrary choice.
    3. Validate the join. Filter prices to NSW1 or join on interval AND region. Count price matches before a lookup: zero or multiple matches must stop the calculation. Six expected intervals should become six matched rows.
    4. Calculate. On your clean sheet use columns A=interval, B=actual MW, C=price, D=baseline MW, E=candidate MW. F2=B2*5/60; G2=F2*C2; H2=B2-D2; I2=B2-E2; J2=ABS(H2); K2=ABS(I2). Fill down six rows. SUM(F2:F7) gives MWh; SUM(G2:G7) gives cost; AVERAGE(J2:J7) and AVERAGE(K2:K7) compare MAE.
    5. Visualise. Make a line chart of actual, baseline and candidate MW against interval end. Label MW, AEST and the synthetic boundary. Use a separate chart for cost; do not put A$/MWh on the MW axis.
    6. Stress, then explain. Recalculate 09:20 at A$300/MWh; then also increase its demand by 20%. Compare deltas, name exclusions and propose what to test next.

    No Excel or SQL experience? Start with Topic 1's unit conversions and a spreadsheet basics course. This lab teaches an energy workflow, not every feature of an analysis package.

    3. Check your results

    Enter all six values before checking. Changing a value resets the check.

    Worked reconciliation

    Clean demand is 60, 120, 180, 120, 60, 180 MW. Interval energy is 5, 10, 15, 10, 5, 15 MWh: 60 MWh total. Interval costs are A$250, −A$200, A$1,500, A$2,000, A$0 and A$1,200: A$4,750 total. The volume-weighted price is A$79.17/MWh; the simple interval-price average is A$68.33/MWh. The two calculations use different weights.

    Baseline MAE is 16.67 MW; candidate MAE is 6.67 MW. Candidate signed mean error is zero but individual errors remain. A price-only stress adds A$1,000 (total A$5,750); the joint stress adds A$1,600 (total A$6,350). Six synthetic points do not establish model superiority in production or a probabilistic loss measure.

    4. Test transfer on a full day

    Download the independent full-day CSV

    A separate synthetic site-day starting 19 September 2026: 288 five-minute intervals ending from 00:05 AEST to 00:00 the following day. The raw file includes one exact duplicate and one missing candidate prediction. All power values are interval averages. Forecasts are invented teaching outputs assumed issued before the day, not evidence of a fitted model. Prices, load and errors are deliberately simplified.

    1. Preserve the original, resolve duplicates and report coverage. Do not convert the missing prediction to zero.
    2. Calculate daily MWh, energy-only cost and a volume-weighted price. Compare forecasts on identical valid intervals and report excluded records.
    3. Plot the full day. Compare all-day MAE with errors for intervals ending 17:05–20:00. Would an all-day score alone justify using the candidate for peak procurement?
    4. Raise peak-interval prices by A$100/MWh, then also raise peak demand by 20%. Reconcile both scenarios and their interaction.
    5. Deliver the clean workbook, chart and conditional recommendation. The six-input check above does not grade this independent task; use the reviewer checklist below.
    5. Prepare your work product

    Attach your clean table, formulas or script and labelled chart to a one-page memo. State the decision, evidence, exclusions, alternative and next action. Export before reloading or leaving this page; nothing is submitted to a reviewer.

    Reviewer checklist

    • Reproducibility: keep the raw files, cutoff, field definitions, formulas or script and a clean six-row output.
    • Data quality: explain the duplicate, revision and extra region; prove that the join neither loses nor multiplies intervals.
    • Analysis: reconcile energy and cost, compare both forecasts on the same intervals and separate price-only from joint stress.
    • Judgement: state what six invented intervals cannot establish, the evidence needed and a conditional recommendation.
    • Communication: submit one labelled chart, the workbook or script and a one-page memo; have a reviewer reproduce the results.

    Numerical checks do not approve the chart, data treatment or recommendation. Obtain a competent review before representing this as workplace capability.

    4. Draft the decision pack

    Keep the decision memo within 800 words, or the shorter limit in your brief. State the recommendation, three decision-driving numbers with sources or assumptions, one rejected alternative, and the largest risk with its owner and review trigger. Separate facts, modelled results and missing evidence. Keep your calculation workbook separately and record its filename, version and key results here. Do not enter confidential information.

    Draft notes are saved in this browser when storage is available; they are not submitted or synced.