Our customer is a large global company in the Energy and Utilities Industry.
In order to meet the business challenge, we have delivered a pilot project for four months with the goal to optimize forecasting activities.
The Business Challenge
Our customer is a large global company in the Energy and Utilities Industry .
The solution we have implemented answers the following business requirements:
- Granular measuring unit and consumption forecast: MWh;
- Customer clustering according to consumption following parameters: meteorological data, the weekend effect, field, consumption profile, seasonality, etc.;
- Individual consumer forecast;
- Forecast using aggregated synthetic profiles determined based on customer segmentation;
- Regional aggregated forecast per reseller, county or other criteria. The individual consumption forecast aggregation was designed to obtain an overview of the entire portfolio consumption curve;
- Take into consideration meteorological information in adjusting the short-term consumption forecast for the next day, next week and other time frames;.
- Forecast adjustment and refining based on the actual consumption taken from SAP;
- Allows manual changes to individual forecasts or by import from Excel or other formats;
- Identify and eliminate error entries from the source files;
- Reduced processing time of large databases (historical data from the past three years).
The purpose of the delivered solution, based on Microsoft Azure and Machine Learning, was to maximize the potential of data regarding the consumption points of all existing customers by creating a usage forecast for the short term at a very high granularity level. This takes into account variables such as meteorological data, consumption history, and data obtained from the company systems in real-time.
In terms of technology, the solution was deployed using the data from the ERP system SAP-ISU (mining & data entry), ZaiNet applications (data entry), the Argus platform/www100 (data mining) and meteorological data providers (data mining).
To obtain an accurate consumption forecast, we have taken the historic data of the customer for over 5 years regarding national consumption and using Microsoft Azure Machine Learning and Predictive Analytics. We reached a 3% variation which was a significant improvement compared to the 15% variance obtained in previous forecasting efforts based on the processing of excel macros.
This insight helped the customer to reduce their costs within the energy stock market by bidding for the right volumes of energy at the national level.