CLASIFICACIÓN DE PATRONES SINÓPTICOS CON ALGORITMOS DE SOFT COMPUTING SYNOPTIC PATTERNS CLASSIFICATION WITH SOFT-COMPUTING ALGORITHMS Sancho Salcedo Sanz Departamento de Teoría de la Señal y Comunicaciones Universidad de Alcalá 1
ÍNDICE Introduction Synoptic pattern classification for wind analysis Relationship wind pressure fields. Clustering problem in pressure. Peculiarities: large dimension of input data. Evolutionary clustering. Objective functions. Results. Alternative Applications. Temperature prediction. 2
INTRODUCCIÓN Wind Energy is currently the most extended clean energy in the world. Spain is one the world s leaders, together with Germany and USA. Currently, about 15% of energy consumed in Spain comes from wind farms. The objective is to reach near 20% by 2020. Wind energy booming: Increasing number of companies and public research funding. Increasing number of wind farms and facilities. Increasing research works in the area. 3
INTRODUCCIÓN Research in wind farm facilities (also in photovoltaic plants). Short-term prediction (wind, energy). Long-term prediction and analysis. Efficient design of wind farms (micrositting, wind resource). Wind farm management (production analysis and fail evaluation). Wind turbine design. Etc. Difficult problems, usually classic methodologies yield poor results. Problems related to meteorology in many cases. Soft-Computing and machine learning algorithms are quite useful in this field. 4
INTRODUCCIÓN In this talk: Wind series reconstruction, wind series analysis and long-term prediction in wind farms using synoptic situation classification. How to perform such a classification using evolutionary algorithms. Applications of the synoptic situation classification in the management of wind farms. Alternative applications in other prediction problems. 5
SYNOPTIC SITUATION CLASSIFICATION 6
A LITTLE BIT OF PREVIOUS WORK Many works in the literature have study the relationship between Synoptic Situation and Meteorological/Climate applications: Precipitation: R. Trigo and C. DaCamara, Circulation weather types and their impact on the precipitation regime in Portugal Intern. Journ. Climatology, vol. 20, pp. 1559-1581, 2000. Pollution: Z. Chen et al. Relationship between atmospheric pollution processes and synoptic pressure patterns in northern China, Atmospheric Environment, vol. 42, pp. 6078 6087, 2008. Wind: C. Soriano et al. Objective synoptic classification combined with high resolution meteorological models for wind mesoscale studies, Meteorology and Atmospheric Physics, vol. 91, pp. 165-181, 2006. There are many studies of circulation patterns associated to extreme atmospheric processes. There are even studies of synoptic situations and their relationship with car crashes and other human activities/problems. 7
PROBLEM DEFINITION 8
PROBLEM DEFINITION Data: T patterns, consisting of daily mean wind measures in a given point (wind farm). Each wind measure has associated pressure values in a grid. Problem: setting a pressure clustering which minimizes a measure of dispersion in the system. The number of classes (synoptic situations must be set in advance). 9
EVOLUTIONARY ALGORITHMS Inspired by Darwin s Theory of Evolution Applied to Computer Science problems I. Rechenberg, 1960, Evolutionary Strategies Lawrence Fogel, 1960, Evolutionary Programming Genetic Algorithms (GAs) John Holland, 1975, "Adaptation in Natural and Artificial Systems. David Goldberg (1989) Genetic Algorithms in Search Optimization and Machine Learning. John Koza, Genetic Programming, 1992. 10
EVOLUTIONARY ALGORITHMS Elements of an evolutionary algorithm: Population. Individuals that represent a given solution to the problem, encoded in a appropriate way. Operators. Operations carried out to evolve the individuals of a population. Selection Crossover Mutation Fitness function. Value used to measure the adaptation of an individual (solution) to its environment (it measures the good or poor that a given solution is for a problem). The algorithm runs in an iterative way, where the best individuals have more chances of surviving for the next generations. At the end of the process, the individual with best fitness is considered the solution to the problem. 11
EVOLUTIONARY ALGORITHMS General scheme of an evolutionary algorithm: Generación de Población Inicial G(i=0) Calculo de Fitness de G(i=0) While (Criterios de Parada =FALSE) i=i+1 Generación de G(i) Procedimento de Selección Procedimiento de Cruce Procedimiento de Mutación Calculo de Fitness de G(i). End While 12
APPLICATION TO SYNOPTIC SITUATIONS Solution proposed: we consider an evolutionary clustering algorithm. Encoding: 4 pressure differences in the grid (feature selection). 26 points in the differences space (clusters). The algorithm evolves both the differences and the clusters. The objective function is the dispersion of the wind measures in each cluster. 13
APPLICATION TO SYNOPTIC SITUATIONS Weather types algorithm: pure physics-based algorithm, based on the calculation of total flux (F) and vorticity (Z). The algorithm compares F with Z, selecting a type of flux (strait, cyclonic, anticyclonic or hybrid). The final number of clusters of WT is 26 classes. Comparison AE WT: AE: 3.73 m/s WT: 4.67 m/s 14
APPLICATION TO SYNOPTIC SITUATIONS 15
APPLICATION TO SYNOPTIC SITUATIONS 16
APPLICATION TO SYNOPTIC SITUATIONS 17
TEMPERATURE PREDICTION 18
TEMPERATURE PREDICTION 19
TEMPERATURE PREDICTION 20
TEMPERATURE PREDICTION 21
FUTURE WORK The proposed synoptic classification system can be applied to different variables related to energy, such as radiation. This system can also be used to improve short-term wind speed prediction systems. Fuzzy clustering techniques can improve the results obtained so far. These techniques have not been studied up until now in this specific problem. 22
TO KNOW A LITTLE MORE L. Carro-Calvo, S. Salcedo-Sanz, N. Kirchner-Bossi, A. Portilla-Figueras, L. Prieto, R. García-Herrera and E. Hernández-Martín, Extraction of synoptic pressure patterns for long-term wind speed estimation in wind farms using evolutionary computing, Energy, vol. 36, pp. 1571-1581, 2011. L. Carro-Calvo, S. Salcedo-Sanz, N. Kirchner-Bossi, L. Prieto, A. Portilla- Figueras and S. Jiménez-Fernández, Wind speed reconstruction from synoptic pressure patterns using an evolutionary algorithm, Applied Energy, vol. 89, no. 1, pp. 347-354, 2012. A. Paniagua-Tineo, S. Salcedo-Sanz, C. Casanova-Mateo, E. G. Ortiz-García, M. A. Cony and E. Hernández-Martín, Prediction of Daily Maximum Temperature using a Support Vector Regression Algorithm, Renewable Energy, vol. 36, no. 11, pp. 3054-3060, 2011. E. G. Ortiz-García, S. Salcedo-Sanz, C. Casanova-Mateo, A. Paniagua-Tineo and A. Portilla-Figueras, Accurate local very short-term temperature prediction based on synoptic situation support vector regression banks, Atmospheric Research, accepted for publication, 2011. 23
PROBLEMAS DE PREDICCIÓN EN ENERGÍAS RENOVABLES Y CAMBIO CLIMÁTICO: APROXIMACIONES DE APRENDIZAJE MÁQUINA Sancho Salcedo Sanz Departamento de Teoría de la Señal y Comunicaciones Universidad de Alcalá 24