How much electricity to produce? A neural network predicts it

Vi-POC is a tool designed for the specific needs of renewable energy producers, traders and operators in the "day-ahead market," energy utilities and power grid operators.

To get to the point of predicting as accurately as possible the amount of electricity to be produced, historical production data and weather data were collected with which a neural network could be trained.

IThis project was financed by MIUR funds with participation from MISE, in the area of accounting/prediction of energy produced by photovoltaic parks, and was developed by GFM Integration in a Temporary Association of Purpose with Università di Bari, CNR-ICAR of Rende (CS), SunElectrics Srl and ISKRA Srl.

How VI-POC Works
All data from the sample plants and forecast calculations are stored on the database so that the forecast curve can later be compared with the actual and theoretical production curves.

Database fields allow determination of individual plant availability/efficiency. For smaller plants-whose daily production data is known only-some essential information, such as module temperatures and irradiances, is reconstructed so that it can be fully used in the forecast models.
In order to obtain a more accurate forecast, different predictive models are used:
  • Clustering from data streams;
  • Classifiers (k-NN, Bayesian, etc.);
  • Mining of frequent patterns from sequences;
  • Neural networks.
The prediction takes into account spatial self-correction aspects typical of the context.

The distributed approach allows the analysis task to be divided over a cluster of nodes, resulting in an efficient system with high scalability and availability.

Technologies used
Vi-POC uses the Hadoop framework, an open source technology recognized as the platform of choice in Big Data management and distribution.

The system is mainly based on these technologies:
  • Apache HBase (Columnar NoSQL Database);
  • Apache HDFS (Hadoop File System);
  • Postgres DB (Relational Database);
  • AngularJS.

The languages used are Java, Javascript, CSS and HTML.

Technologies used
Vi-POC was created to make the forecasting system available from the Manager of Energy Services more efficient, effective and reliable this enables it to have positive effects on the supply side of the power exchange and improves the purchase forecast.

In addition to large players, Vi-POC is also designed for smaller entities, which can use it for better purchase planning for their customers.

The most modern renewable energy plants can rely on the most advanced technologies for data storage and analysis, with continuous monitoring of production and analysis, both in real time and afterwards, aimed at extracting very useful strategic information to improve the quality of forecasts significantly.

#BigData #AI #DataAnalytics #energyefficiency