Gains:
- Ability to establish an end-to-end Python analysis workflow from data to report with the help of artificial intelligence
- Ability to verify the generated code in terms of unit compatibility, leakage and reproducibility
- Ability to put ethics, confidentiality, cyber security and engineer responsibility limits into practice in energy projects
Module Exam
1. Which of the following is the basic principle that determines the risk of an AI output in energy systems engineering?
- A) The risk of an output is equal to the harm it would cause if that output were faulty; validation scales accordingly ✔
- B) AI output is generally safe as it is written fluently and confidently
- C) Validation is unnecessary when the most current model is used
- D) If the output refers to a standard, no additional control is required
Explanation: The risk of an output is equal to the harm it will cause if that output is faulty. While a minor error in a presentation text may be harmless, an error in a relay protection setting or transformer loading decision can cause equipment damage, outage and life safety risks; so verification intensity scales with potential harm.
2. What is the most common pitfall when preparing energy time series data with artificial intelligence and which distorts the results the most?
- A) Incorrect selection of graphic colors
- B) Time zone, daylight saving time and timestamp label confusion ✔
- C) The file name is too long
- D) Keeping data in Excel instead of CSV
Explanation: Start/end tag confusions of time zone, daylight saving time (DST), and timestamp are very common in energy data; a one-hour shift distorts the entire day-type and peak-hour analysis. Therefore, the time axis should be standardized before each analysis and the AI output should be tested against physical anchors such as sunrise/peak time.
3. Which situation shows that while the MAPE (Mean Absolute Percent Error) in a short-term load forecast appears low, the forecast may still be unreliable?
- A) MAPE is always a complete and adequate measure of forecast quality
- B) Even though the average is low, there may be systematic deviation in peak hours and that is where the grid is critical ✔
- C) MAPE is only meaningful in long-term forecasting
- D) Low MAPE proves that the model is physically correct
Explanation: MAPE is an average error; Even if the total looks good, there may be systematic deviations in peak hours and seasonal transitions. Network balance is critical precisely at these peak moments, so just because the average metric is good cannot be considered safe to predict; The error should be analyzed separately on the basis of time and regime.
4. What is the most robust physical upper bound to test the reasonableness of a solar plant production forecast?
- A) Last year's total bill
- B) Physical production ceiling limited by radiation, panel area and installed power (including zero at night) ✔
- C) Percentage of confidence given by the model
- D) Wind speed measurement
Description: A solar power plant cannot produce more than the solar radiation and panel capacity allow; Production is expected to be zero at night, peaking around noon, and limited by the installed power and module area. It is physically impossible if the model gave a value exceeding the night production or installed capacity. This check eliminates the prediction without looking at the statistics.
5. What is the filter that an AI optimization proposal must pass before being directly implemented in the smart grid?
- A) It is enough for the suggestion to provide the lowest cost alone.
- B) It is sufficient for the model to be trained to high accuracy
- C) Safety constraints (voltage, thermal, N-1) control and authorized operator approval ✔
- D) It is sufficient that the recommendation fits historical data
Description: The network is safety-critical: any recommendations that violate voltage, current, thermal and N-1 (element failure) constraints cannot be implemented. The AI output is first passed through these security constraints and then presented to the authorized operator for approval; Optimization never gets in the way of a security constraint.
6. Which cost is frequently overlooked and misleads the profit when establishing the arbitrage (buy cheap, sell high) strategy with AI in the battery storage business?
- A) Internet connection fee
- B) Degradation (wear) and efficiency loss cost per cycle ✔
- C) Coffee break duration of the staff
- D) Color of battery
Description: Each charge-discharge cycle wears out the battery; degradation (loss of health) is a real cost and exhausts the warranty limits. An arbitrage analysis that only looks at the price difference will overestimate the profit and shorten the life of the battery if it ignores the degradation cost and efficiency loss per cycle.
7. What is the key balance when determining the threshold value of an anomaly detection model in predictive maintenance?
- A) Conscious balance between false alarm and missed failure costs ✔
- B) The threshold should always be kept as low as possible
- C) The threshold is determined solely by looking at the accuracy percentage of the model
- D) Once the threshold is set, it should never be changed.
Explanation: If the threshold is lowered, false alarms (alarms when there is no fault) increase and the team becomes distrustful of alarms; If the threshold is increased, the risk of a missed fault (missing the actual fault) increases. The correct threshold is a conscious balance between the cost of a false alarm and the cost of a missed fault (e.g. transformer explosion) and is adjusted according to the criticality of the equipment.
8. What is the basic logic of the measurement and verification (M&V) framework for verifying a savings claim in an energy efficiency project?
- A) Every decrease in the invoice amount is considered a direct savings
- B) Savings are the difference between the adjusted baseline and measured consumption ✔
- C) Savings are declared only at the manufacturer's catalog value
- D) Savings equal to what the model predicts
Explanation: Savings cannot be measured directly; What is measured is the consumption after taking precautions. Savings are the difference between measured consumption and the baseline 'what would be consumed without the measure'. So the baseline model must be corrected for variables such as weather/production, the savings calculated against this adjusted reference.
9. Why is 'overfitting' particularly dangerous in the electricity price forecasting model?
- A) Because it slows down the model
- B) It memorizes the noise of the past, collapses in the future, especially when the regime changes, and misleads with high confidence ✔
- C) Just because it uses more memory
- D) Because it makes the graphics difficult to read
Explanation: Overfitting is when the model memorizes noise and patterns in historical data that will not repeat; Backtesting looks perfect but crashes in the future. As price data changes regime (fuel price shock, legislation change), the memorized pattern becomes invalid and leads to incorrect bidding with high confidence.
10. Why is the concept of hourly carbon intensity (gCO2 value of one kWh of electricity in that hour) important for load shifting?
- A) Carbon density is the same every hour, so it doesn't matter
- B) Because it only affects the invoice amount
- C) Varies hourly depending on renewable share; Shifting flexible load to low-intensity hour reduces emissions ✔
- D) Because it is only related to the night tariff
Explanation: The carbon intensity of the grid varies hourly: it is low during hours when renewable generation is high and high during peak hours when fossil power plants come online. Shifting flexible loads to low-peak hours allows consuming the same energy but producing fewer emissions. The average annual factor hides this opportunity.
11. What does 'data leakage' mean in a Python analysis code produced by artificial intelligence and why is it dangerous?
- A) Deleting data from disk
- B) The model uses future information in training that will not be available at the time of prediction; test is good but reality is bad ✔
- C) Data has no meaning other than leaking to the internet
- D) The code runs very slowly
Explanation: Data leakage is when the model accesses information that will actually be found in the future during training; for example, using the actual value of the hour to be predicted as an attribute. This results in perfect backtesting but useless in reality. The past/future boundary in the time series must be carefully maintained in the code.
12. What is the correct approach when sending energy company SCADA data to a general AI tool?
- A) Sending all data as is gives the most accurate results
- B) Anonymizing data and sharing only the necessary technical subset with an approved tool ✔
- C) The precaution is unnecessary because the data is open anyway
- D) Just changing the file name is enough
Description: SCADA and network topology data is critical infrastructure information; Sending it to a public cloud as is poses privacy, competition and cybersecurity risks. The correct approach is to anonymize site/asset names with representative tags, share only the technical subset required for the task, and opt for institution-approved/closed tools.
13. Why is the 'capacity factor' (the ratio of the energy produced to the energy that would be produced if it operated continuously at full capacity) used as a reasonableness anchor in wind production forecasting?
- A) Because it is a number for marketing purposes only
- B) Because he gave the invoice directly
- C) Because it is a physical criterion that tests the order of magnitude of the prediction, with a known typical range (approximately 25-45%) ✔
- D) Because it determines the color of the turbine
Description: The capacity factor of onshore wind farms is typically in the range of approximately 25-45 percent. A forecast is physically implausible if it predicts production on the long-term average that is well outside this range (e.g., 90 percent). This ratio allows you to test the order of magnitude of the prediction without going into detail.
14. How can the role of artificial intelligence in a security-critical energy system decision be most accurately defined?
- A) AI can replace authorized engineer approval
- B) AI generates hypotheses and outlines; Approval and responsibility for the safety-critical decision remains with the authorized engineer ✔
- C) If AI is powerful enough, human confirmation is unnecessary
- D) AI output is useless and should be ignored completely
Description: Artificial intelligence is a powerful assistant that reads quickly, catches patterns, produces drafts; but it does not replace the approval and responsibility of a competent expert in safety-critical decisions such as protection adjustment, transformer loading, network maneuvering. AI generates hypotheses and outlines; The authorized engineer gives the decision and signature.