Demand Forecast (Tech Preview)

Use the Demand Forecast tool, currently in Tech Preview, to predict system demand for a 24-hour horizon. What is a Tech Preview?

To leave feedback on the Demand Forecast tool, go to our Feedback Portal.

The tool uses machine learning techniques to recognize patterns and provide a prediction of system demand. It can be useful for model calibration, operations planning, and system optimization.

To create a demand forecast:

1. Go to Tools Demand Forecast and select New Forecast.
2. Select the Flow Rate Sensor for which you want to create the forecast.
3. Give the forecast a Name and Description.
4. If the sensor has a location configured, the Latitude and Longitude will show up automatically. If it doesn't, the location will default to your organization's address. You can edit the Latitude and Longitude as required.

   This location is used to determine the additional input data (temperature, precipitation, holidays) used in the forecast model.

5. Set the Training Time Window. This determines the data used to train the forecast model.

   We recommend including at least one year's worth of data.

   Note: The model will remove any days with incomplete data from the data training set. So, the first timestamp with data (in the selected time window) will effectively become the start date. The training data chart shows the actual data that will be used.
6. Select Run. You can continue working while the model runs.
7. When the forecast status is ready, you can view the results of the different models run and their performance metrics. See below for an explanation of the metrics and how they can help you decide which model variation to use for your forecast.

   You can also use the Available Charts to visually compare the models with the actual sensor data and see which one aligns best. You can adjust the zoom or the date range in the chart to visualize the data better.

8. Select Deploy and select the model variation you want to use from the drop-down list.
9. You can view the Forecast duration (how far into the future), Forecast interval (data points given), and Forecast frequency (how often the forecast is updated) that will be used. The model defaults are used and cannot be modified.
10. Select Deploy. This will create the forecast for your sensor.
11. Go back to the Demand Forecast list and once the forecast status is Deployed, you can add a gauge chart to a workspace to view the forecast data. See Create a Gauge Chart.

Performance Metrics

You can use the models' performance metrics to help you choose which one to deploy.

• In-sample: The model's performance when using the training data (data the model has seen before).
• Cross-validation: The model's performance when using test data (data the model has not seen before). In general, you should focus more on the cross-validation columns.

• MAPE (Mean Absolute Percentage Error): A measure of the prediction accuracy of a forecasting method in statistics. The smaller the MAPE, the better the model's performance. If your priority is to minimize the percentage error, this is the metric to focus on.
• R2 (R-Squared): A statistical measure that represents the proportion of variance for a dependent variable that's explained by an independent variable or variables in a regression model. If the R-squared of a model is 0.50, then approximately half of the observed variation can be explained by the model's inputs. For this metric, a higher value (closer to 1) is better.
• RMSE (Root Mean Squared Error): A risk metric that corresponds to the expected value of squared (quadratic) error loss. The smaller the RMSE, the closer you are to finding the line of best fit. If avoiding large errors is your main concern, focus on this metric.