Example

This example demonstrates how to use this AI & Machine Learning Agent to pinpoint new trends in the average power usage of a device.

Refer to configuration to understand all configuration options of this Agent.

Step 1: Add the Agent

Drag the Anomaly Detection Agent onto the canvas. Link the input endpoint to the consumption data and the output to the filter. Rename the Agent and save the Data Stream.

Step 2: Configure General

Select the Agent and click Configure. Keep the default Collection.

Step 3: Configure Training

Set the Learning Algorithm, Training File, Has Headers, and Separator Character.

In this case, use the SSA Spike algorithm and load the provided training file. Keep the default Has Headers and Separator Character.

Next, select the Input Field and the Input Map. The Input Field (Consumption Diff Normalized), is populated from the Training File, while the Input Map (also Consumption Diff Normalized) is populated from the parent agent.

Step 4: Configure Advanced Options

Set the Sensitivity, History Length, Training Window, Seasonality, and Spike Direction.

In this case, set the Sensitivity to 50, History Length to 10, and Training Window Size to 90. Keep the default Seasonality and Spike Direction.

Step 5: Configure Model Options

Optionally, set the Determinism seed. This example uses a value of 5.

Step 6: Results

Apply the changes, save the Data Stream, and publish it.

Let's look at the Live Data View. Observe the Alert column on the predicted events: 1 indicates a suspected spike in consumption and 0 indicates the opposite. In this case, the 0 records are removed by the Filter Agent. Observe that metadata about the decision is appended too - Score, PValue, and for some models, Martingale.

Files

File	Link	Security Key
Data Stream	Anomaly Detection Example.xuc	C0mp|ex123
Training Data	power training.csv
Test Data	power-export_min.csv

See the Import, Export, and Clone - XMPro article for steps to import a Data Stream.