The first step is to configure your forecast and allow the system to do some calculations based on temporal series algorithms. To do so, you should go to Settings > Forecasting > and click on "New Single Forecast". Here you can define the device you want to use to train and calculate its corresponding forecasting. Just select the associated location, the device and the corresponding parameter. Below you have an example:
The app is available for the following parameters:
- Active Energy (402)
- Gas Energy (420)
- Water (901)
- Exported Active Energy (452)
- Thermal Energy (802)
Artificial Intelligence Model
The algorithm calculates a consumption forecast for each future hour, based on similar periods considering similar days/hours/months from the last year of available historical data.
A forecast is calculated for a given year (365 days), which allows to calculate a high level estimate of forecasted consumption and costs.
Once you have configured your forecast information, you should wait a bit. Forecasted data is calculated every day (03:00h GMT) for one year in advance. The system takes between 1 week and 3 months to provide accurate models. 6 months of historical data is needed
- Energy source selector. Select the energy source you want to analyse
- Dates range picker. Select the period of time for the analysis. You can use the presets available to move quickly (today, yesterday, last 7 days, last month...)
- Location. Select the location for analysis.
- Device. Select the device you want to analyse
- Frequency. Select the frequency between daily, weekly and monthly
- Compare against these Goals and Budgets. Here you can select the baselines previously configured to compare them against the forecasted data.
- Cost. Convert the forecasting directly into Euros. We take the average price for the selected period to do the calculations
- Export data to Excel. Export the selected data to an Excel file.
- Bars chart. View your data in bars
- Accumulated view. Analyse your forecasting aggregating all the consumption for the selected period.
- Summary table. Comparison between the real consumption, the forecast, and all the baselines selected if any. Here you can see the total consumption and the difference with the real consumption.
This part of the article explains the Forecasting Reliability Metric available in the Forecasting App.
The Forecast Reliability Metric allows you to assess how much you can rely on your forecast. It consists in a tag with 5 possibles values. When you hover your mouse over the tag, a tooltip pops up with information relative to the metric.
What is the Forecasting Reliability Metric?
The difference between your forecast and your data is less than 10%. You can rely on this forecast.
The difference between your forecast and your data is between 10 and 20%. Is not a very good forecast but it is still acceptable. Please, check for any gaps or missing data.
The difference between your forecast and your data is greater than 20%. We do not advise you to trust in this forecast.
There is not enough data to calculate your forecast or your profile has too much variability.
Something went wrong during the calculation of the forecast. Please, contact our support team.
How does it work?
The Forecasting Reliability Metric is a mark based on statistical indicators that asses the quality of your data. Some of the analyzed variables are:
- How many zero values have your data
- How many N/A values have your data
- Chaotic pattern changes
The better the quality of your data is, the more grounded the forecasts will be. The score obtained determines the value of the Forecasting Reliability Metric.
For reliable predictions, it is recommended to have 1 year of historical data. This way, the algorithm will have the reference of the behaviour in summer and winter. With fewer data it is possible that the prediction does not correspond to the specific behaviour in those periods.