Weather data analysis based on typical weather sequence analysis. Application: energy building simulation

TL;DR

This paper introduces the Runeole software for analyzing typical weather sequences to improve building energy simulation accuracy, using statistical and classification methods on French climate data.

Contribution

It presents a novel C++ tool that analyzes multivariable climatic data to generate representative weather sequences for building simulations.

Findings

Effective selection of weather sequences for energy modeling.

Validation of methodology on Reunion Island climate data.

Potential for generating hourly meteorological years with neural networks.

Abstract

In building studies dealing about energy efficiency and comfort, simulation software need relevant weather files with optimal time steps. Few tools generate extreme and mean values of simultaneous hourly data including correlation between the climatic parameters. This paper presents the C++ Runeole software based on typical weather sequences analysis. It runs an analysis process of a stochastic continuous multivariable phenomenon with frequencies properties applied to a climatic database. The database analysis associates basic statistics, PCA (Principal Component Analysis) and automatic classifications. Different ways of applying these methods will be presented. All the results are stored in the Runeole internal database that allows an easy selection of weather sequences. The extreme sequences are used for system and building sizing and the mean sequences are used for the determination of the annual cooling loads as proposed by Audrier-Cros (Audrier-Cros, 1984). This weather analysis was tested with the database of the French weather forecast utility Meteo France. Reunion Island experiences a lot of different micro-climates due to the high altitude (3069m), specific relief, and geographic situation (Tropic of Capricorn). Furthermore Reunion Island has the densest meteorological network in France and is an ideal place to validate the methodology with different climates. To test the efficiency of such analysis, simulations using the resulting weather sequences were carried out with the building simulation software CODYRUN. This analysis is the first step of a more global research concerning weather data generation. Future work will permit whole hourly typical meteorological year generation using neural networks.