|Call||National R+D Program Oriented to Society Challenges
(Programa Estatal de I+D+i Orientada a los Retos de la Sociedad)
|Title (coordinated project)||Modeling, Forecasting and Interactive Data Visualization Techniques for the Improvement of Energy Efficiency in Buildings and Industrial Facilities (ENERGYDATA)|
|Partners||UniLeón (grupo SUPPRESS)
UniOvi (grupo GSDPI)
|Duration||1/1/2016 - 31/12/2018 (extended until 31/12/2019)|
Energy efficiency and savings policies have become a challenge of extraordinary strategic importance for Spain and the EU, as a result of their excessive current and forecasted energy dependence. Despite the huge volumes of data and information available today, these data are not presented in an intuitive way that the user can assimilate. As a result, the decisions in energy management are often taken under insufficient or ill-defined information. Thus, it becomes important to have tools based on the large amounts of data obtained from energy facilities, including both buildings and industrial facilities, which increase our energy awareness, that is, our perception of the use given to energy. These tools should hint strategies to improve energy efficiency. They also should be able to handle large volumes of changing information of different kinds, tolerate imprecision, present information in a clear and intuitive way and also provide a global view. Information visualization allows getting insight from data, through the combination of intelligent data analysis algorithms and the ability and plasticity of the human visual system, which allows the detection of interesting patterns quickly and efficiently. According to this approach, the user is part of the information visualization process and ultimately takes the decisions. The 1st research line proposes the development and application of visual analytics techniques oriented to system and process monitoring that allow the user to discover and understand the factors that affect energy efficiency. In order to improve the exploration process, these techniques combine classification,forecasting and dimensionality reduction algorithms (able to produce visual "maps" of the process states) with principles and techniques of information visualization, as well as interaction. The aim of the 2nd research line is to use the information obtained by the application of previous techniques to data from buildings, facilities and processes in order to improve efficiency in two main areas: a) the efficiency in consumption buildings and facilities, applying the proposed techniques to analyze the behavior of energy demand, and to characterize the influence of environmental variables, as well as to obtain models that enable the tuning of systems and control strategies for a more efficient operation; b) the efficiency in processes, exploring the relationships between factors (process parameters, known variables) and process performance (quality, efficiency), as well as through fault detection and efficiency monitoring. Experimentation in this project will be based on real data acquired from the energy facilities of the Universities of León and Oviedo and the León Hospital, as well as from industrial processes accessible by the partners.
Visit and seminar from prof. Jaakko Hollmén
Seminar Prof. Jaakko Hollmén at EPI Gijón about data science
Workshop for presentation of advances in project ENERGYDATA
Seminar "Analítica Visual para la Mejora de la Eficiencia en Procesos y Edificios: pasado, presente y futuro"
|2016-09-20||Seminar: "Análisis del fenómeno de chatter para su exploración visual". Daniel Pérez|
|2016-09-26||Seminar: "Posibilidades de las técnicas deep learning en analítica visual de procesos y edificios". Ignacio Díaz|
|2016-10-03||Seminar: "Análisis de Componentes Independientes (ICA): fundamentos matemáticos y su aplicación en consumos eléctricos". Diego García|
The next videos show some of the visual analytics techniques developed during the project.
|Visual analytics of process data
Power demand analysis (univ. campus)
(Díaz et al, 2017)
Air quality analysis
(Díaz et al, 2017)
|NMF + iHistograms
Power demand analysis in a hospital
(García et al, 2018)
|tSNE + iHistograms
Visual analytics of cold rolling data
(Pérez et al, 2018)
electric power demand
|Original version of the interactive data cube visualization webapp, showing electric power demand data from a university campus, published in Energy and Buildings [→ publication web page] and developed during the project.|
|iHistograms: interactive histograms||Version of iHistograms for visualization of the quality of air in Gijón city from open datasets available in datos.gijon.es from Ayuntamiento de Gijón.|
These sample visualizations show different ways to visualize demand data using specific visual encodings that take advantage of problem domain knowledge to disentangle patterns on data (structure, periodicities) and to highlight informative features.
|Spiral web visualization||Basic example of spiral web visualization for several electric power demand parameters from a hospital|
|Interactive web visualization (zoom, pan)||Basic example of interactive web visualization of electric power demand in a hospital|
|Power factor visualization||Example of power factor visualization in a rural area, aggregated by day hour and weekday|
|Electric demand forecasting in UK||Example of electric demand forecasting in july 2013, from training data between may 2011 and may 2012. The example uses an extreme learning machine (ELM) algorithm.|
In these web-based data visualizations we use an interactive version of a manifold learning algorithm (stochastic neighbor embedding (SNE)) to show the data to the user in an organized way according to criteria that he/she can change. In these visualizations, the multivariate elements are organized by similitudes of the describing vectors.
The user can tune the weights of each component in defining these similarities so the displayed elements are reorganized "on the fly" according to the current similarity metrics. This results in "animated transitions" whereby the user keeps a mental map of all the elements between different organization criteria.
|iDR interpolation demo||Demo app for paper "Interactive dimensionality reduction of large datasets using interpolation" presented in ESANN 2018.|
|visualización iDR: rendimiento de estudiantes||2D projection of academic performance dataset os students from secondary school in terms of social factors like age, sex, parents education, etc. Based on the dataset from the UCI repository Student Performance Data Set.|
|iDR visualization: map of vibration states (version using P5js)||Interactive 2D projection of vibrational states, organized by similarities in a set of frequency bands, allowing to filter values, add labels, zoom and pan, etc.|
|iDR visualization: Electic power demand||Interactive 2D map of electric power demand 24-hour patterns, organized by similarities.|
|iDR visualization: Environmental patterns||Visualize an interactive map of electric stations (anonymized), organized by similarities on their emision signatures (6 parameters: SOx, NOx, Dust, etc.)|
|iDR visualization: map of vibration states||Interactive 2D projection of vibrational states, organized by similarities in a set of frequency bands of a rotating machine (induction motor) under mechanical imbalance (eccentric mass) and electrical imbalance (different levels of asymmetric load in one of the phases).|
Note: in the iDR visualizations, the user can modify the weight of each variable in the computation of the similarity metrics, by dragging the mouse on the corresponding bar in the bottom-right corner. This has the effect of modifying the layout of the displayed elements to accommodate the user-driven changes in the similarity metrics
|Actividad realizada en el marco del proyecto DPI2015-69891-C2-2-R, financiada por el Ministerio de Economía y Competitividad (MINECO), Programa Estatal de I+D+i Orientada a los Retos de la Sociedad y por el Fondo Europeo de Desarrollo Regional (FEDER) "Una manera de hacer Europa"|