> For the complete documentation index, see [llms.txt](https://docs.myrspoven.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.myrspoven.com/measure-savings/ai-on-off-savings.md).

# AI ON vs. OFF savings

The savings shown in myPortal 2.0 compare what each energy meter used with myCoreAI in control against what it would have used without myCoreAI in the same weather. The method draws two curves from the building's own meter data, one for days without the AI and one for days with it, and reads the gap between them at each day's outdoor temperature. Each day's saving is noted once and never revised afterwards. The method follows ASHRAE Guideline 14 and ISO 50015.

The **AI ON vs. OFF** page in myPortal 2.0 shows the curves, the days behind them and each day's saving. The savings totals on the overview pages add up the same daily savings.

## Why the weather is taken out first

A cold winter needs more heating and a hot summer needs more cooling, whatever the control system does. Comparing this year with last year mixes the effect of the weather with the effect of the AI. So every day is placed on a chart with its outdoor temperature along the bottom, and compared with what the building normally uses at that temperature. See [Climate compensation](https://docs.myrspoven.com/measure-savings/climate-compensated.md) for the general principle.

## Two curves and the gap between them

<figure><img src="https://2190044209-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FkvWb2bblPg0JMbwweyz7%2Fuploads%2Fgit-blob-deb22de09bbe6ed4f4a17b2a8a371e4e4fc00141%2Fai-on-off-two-curves.png?alt=media" alt="Scatter plot of daily energy use against daily mean outdoor temperature, with a grey baseline curve for days without the AI and an orange curve for days with the AI. At 1.4 °C the gap between the curves is 782 kWh a day."><figcaption><p>One dot per day. The gap between the baseline curve and the AI curve is the saving at that temperature.</p></figcaption></figure>

For each energy meter, myCoreAI fits two curves of daily energy use against daily mean outdoor temperature:

* **Baseline curve:** drawn from days when the AI was not in control.
* **AI curve:** drawn from days when the AI was in control.

The curves follow the shape of the data rather than a straight line, because energy use falls steeply in cold weather and flattens out in summer. At any temperature, the gap between the baseline curve and the AI curve is the saving for a day at that temperature.

Each meter gets its own pair of curves for weekdays and another pair for weekends, because most buildings use less energy at weekends at the same outdoor temperature. Meters that report monthly totals instead of daily readings are included too, with each month's total spread over its days.

### Which days are used for the curves

Every hour, the building records whether myCoreAI had control. The share of hours with the AI in control is the day's **uptime**.

| Day                                         | Used for                                   |
| ------------------------------------------- | ------------------------------------------ |
| AI in control for more than 90 % of the day | The AI curve                               |
| AI in control for less than 10 % of the day | The baseline curve                         |
| Anything in between (a mixed day)           | Neither curve, but its saving still counts |

Mixed days are left out of the curves because they would pull the two curves towards each other and shrink the gap.

Buildings change over time, so the curves favour recent days. In each outdoor temperature range, at most the 30 newest days without the AI and the 15 newest days with the AI are kept. A temperature range with fewer days keeps them all, so rare cold or warm days are not lost.

## From the gap to a daily saving

For every day, myCoreAI reads both curves at that day's outdoor temperature. The saving counts only for the hours the AI was in control:

| Figure in myPortal 2.0 | What it is                                                                                                                   |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| **Observed savings**   | The gap at the day's temperature, times the day's uptime                                                                     |
| **Forfeited savings**  | The gap times the share of the day the AI was not in control: what the AI would have saved if it had been in control all day |
| **Potential savings**  | The gap at 100 % uptime: observed plus forfeited                                                                             |
| **Average uptime**     | The share of time the AI was in control over the period                                                                      |

Adding up the days gives the monthly bars and the totals.

### Days colder or warmer than the curves cover

Each curve only covers the temperatures it has data for. On a day colder or warmer than the range both curves cover, the saving uses the gap from the nearest end of that range. The curves themselves are never extended.

## Quality checks before a saving is shown

A meter shows a saving only when both of its curves pass the quality checks. Each curve needs:

* At least 13 days of energy data.
* Days covering at least 8 °C of outdoor temperatures.
* Energy use that changes with outdoor temperature. A curve that stays nearly flat while the other curve clearly follows the temperature is rejected.
* A clear consumption pattern that the curve follows.

If either curve fails a check, the AI ON vs. OFF page shows the reason instead of a saving for that meter. A lot of good data on one side cannot make up for weak or missing data on the other. Days with missing meter readings or missing outdoor temperature are marked in the data availability view.

## Noted savings: set once, never revised

Every morning, myCoreAI scores the previous day for each energy meter. Once the meter's curves have passed their checks, the day's saving is **noted**. A noted saving is never revised: when the curves are refitted later on more data, the improvement applies to later days only, so a saving you have seen in one report stays the same in the next.

If a day's saving cannot be noted the morning after, for example because the curves have not passed their checks yet, it is tried again on the following mornings, for up to a year. Until then the day is shown as **in progress** and is left out of the savings totals.

{% hint style="info" %}
On the AI ON vs. OFF page you can change settings such as the day filter or the recency filter. The savings are then recalculated with the current curves and are no longer the noted values. **Reset to noted savings values** brings back the noted figures.
{% endhint %}

### When savings are paused

When a meter has had no valid reading for more than 30 days, new savings for it are paused and shown as **Savings paused: meter offline**. Savings already noted stay included. When the meter reports again, the missing days from the past year are filled in.

## Standards

ASHRAE Guideline 14 and ISO 50015 are the two common standards for measuring energy savings; they set how good the curves must be before a saving counts as proven. The method follows both. Next to a saving, the AI ON vs. OFF page can show a PASS or FAIL verdict for each standard over the selected period, together with a range for the saving. The range comes from refitting both curves 200 times on reshuffled whole weeks of days, and covers the middle 68 % of the results.

| Standard            | A saving passes when                                                                                                                             |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
| ASHRAE Guideline 14 | Every curve has a CV(RMSE) of at most 25 % and an NMBE within ±5 %, and the uncertainty of the saving is at most 50 %                            |
| ISO 50015           | Every curve passes its quality checks, a range exists for the period, and at most 25 % of the days rely on the gap held at the end of the curves |

A FAIL does not mean the saving is wrong. It means the saving is not yet proven to the standard, usually because there are too few days on one side or the saving is small compared with its range. The page says what would change the verdict, for example collecting more days without the AI.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.myrspoven.com/measure-savings/ai-on-off-savings.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
