Verdigris systems measure electricity use, and then enrich that data to forecast electricity consumption. Forecasting has many potential benefits and revenue savings for operations, planning and decision making. Verdigris forecasted data is accessible via our API: https://app.verdigris.co/docs/v3/buildings/forecast.html

What is demand forecasting

Electricity demand forecasting uses statistical models to predict the consumption patterns of equipment. Common electrical loads in commercial buildings include lighting, space conditioning, and machinery for the production of goods and services.

How Verdigris forecasts

Verdigris uses deep learning models to provide day-ahead forecasts of the whole building load in 15 minute intervals. The model is regularly retrained to continually learn from your data.  The forecast includes the mean and a confidence interval around the mean. The confidence levels are calculated for every 5th percentile between 5% and 95% (Figure 1). The model incorporates inputs such as historical energy use, local weather forecasts, as well as time of day and day of week to account for periodicity. We also benchmark our models on synthetic tests and with open source data. 

Figure 1: Forecasting showing confidence interval at 90% and 80%

Forecasting as a service

Accurate forecasting in commercial and industrial buildings can serve many purposes. It can detect anomalous behavior that deviates beyond an expected range (Figure 2). This activity may signal the need to inspect equipment or an unexpected outage. It also enables the ability to create dynamic setpoints for building operation. Demand management is a method to manage peak demand costs and is triggered by on demand forecasts. The forecast anticipates high demand events so that corrective action can be predetermined (Figure 3).  Some examples include throttling non-critical ventilation, dynamic chiller operation, and efficiency optimization. 

Figure 2: Detect anomalous behavior beyond anticipated forecast at 23:30

Figure 3: Demand management through forecasting

Did this answer your question?