Fraud Detection

User Panel : Fraud Simulation

Amplitude (kW) :
100
4000
Time Period (h) :
1
48
Number of periods :
1
10

Fraud Simulation Results

Fraud Detection Results


Fraud Detection

Load curve anomalies

The complexification of the elecrical grid forces the transport and distribution system operator to have a reliable power forecast. Consumption fraud and new mechanisms of flexibility are phenomena that the power companies need to detect to ensure network reliability and limit the economical losses.

The anomalies generally take the form of positive (fraud) or negative (flexibility) offset such as presented below:

The main goal of this bot is to show the ability of a forecasting model to identify periods of abnormal network behavior.

Model

The training of the model must be performed with the available data, containing regular behavior and fraud data. With an iterative approach, the problematic points are removed from the training set until the convergence of the model to the common behavior:

The model is getting closer to the common behavior by correcting th effect of the anomly periods. The quality of the model improves, as attested by its error, estimated with the Mean Absolute Percentage Error (MAPE), that decreases with each iteration until the acceptance criteria is met.