PV power forecast using a nonparametric PV model

Forecasting the AC power output of a PV plant accurately is important both for plant owners and electric system operators. Two main categories of PV modeling are available: the parametric and the nonparametric. In this paper, a methodology using a nonparametric PV model is proposed, using as inputs several forecasts of meteorological variables from a Numerical Weather Forecast model, and actual AC power measurements of PV plants. The methodology was built upon the R environment and uses Quantile Regression Forests as machine learning tool to forecast AC power with a confidence interval. Real data from five PV plants was used to validate the methodology, and results show that daily production is predicted with an absolute cvMBE lower than 1.3%.

Learning Bayesian networks from data by the incremental compilation of new network polynomials

Probabilistic graphical models are a huge research field in artificial intelligence nowadays. The scope of this work is the study of directed graphical models for the representation of discrete distributions. Two of the main research topics related to this area focus on performing inference over graphical models and on learning graphical models from data. Traditionally, the inference process and the learning process have been treated separately, but given that the learned models structure marks the inference complexity, this kind of strategies will sometimes produce very inefficient models. With the purpose of learning thinner models, in this master thesis we propose a new model for the representation of network polynomials, which we call polynomial trees. Polynomial trees are a complementary representation for Bayesian networks that allows an efficient evaluation of the inference complexity and provides a framework for exact inference. We also propose a set of methods for the incremental compilation of polynomial trees and an algorithm for learning polynomial trees from data using a greedy score+search method that includes the inference complexity as a penalization in the scoring function.

Expressive power of binary relevance and chain classifiers based on Bayesian Networks for multi-label classification

Bayesian network classifiers are widely used in machine learning because they intuitively represent causal relations. Multi-label classification problems require each instance to be assigned a subset of a defined set of h labels. This problem is equivalent to finding a multi-valued decision function that predicts a vector of h binary classes. In this paper we obtain the decision boundaries of two widely used Bayesian network approaches for building multi-label classifiers: Multi-label Bayesian network classifiers built using the binary relevance method and Bayesian network chain classifiers. We extend our previous single-label results to multi-label chain classifiers, and we prove that, as expected, chain classifiers provide a more expressive model than the binary relevance method.

Multi-dimensional classification of GABAergic interneurons with Bayesian network-modeled label uncertainty

Abstract Interneuron classification is an important and long-debated topic in neuroscience. A recent study provided a data set of digitally reconstructed interneurons classified by 42 leading neuroscientists according to a pragmatic classification scheme composed of five categorical variables, namely, of the interneuron type and four features of axonal morphology. From this data set we now learned a model which can classify interneurons, on the basis of their axonal morphometric parameters, into these five descriptive variables simultaneously. Because of differences in opinion among the neuroscientists, especially regarding neuronal type, for many interneurons we lacked a unique, agreed-upon classification, which we could use to guide model learning. Instead, we guided model learning with a probability distribution over the neuronal type and the axonal features, obtained, for each interneuron, from the neuroscientists’ classification choices. We conveniently encoded such probability distributions with Bayesian networks, calling them label Bayesian networks (LBNs), and developed a method to predict them. This method predicts an LBN by forming a probabilistic consensus among the LBNs of the interneurons most similar to the one being classified. We used 18 axonal morphometric parameters as predictor variables, 13 of which we introduce in this paper as quantitative counterparts to the categorical axonal features. We were able to accurately predict interneuronal LBNs. Furthermore, when extracting crisp (i.e., non-probabilistic) predictions from the predicted LBNs, our method outperformed related work on interneuron classification. Our results indicate that our method is adequate for multi-dimensional classification of interneurons with probabilistic labels. Moreover, the introduced morphometric parameters are good predictors of interneuron type and the four features of axonal morphology and thus may serve as objective counterparts to the subjective, categorical axonal features.

Genetic algorithms and Gaussian Bayesian networks to uncover the predictive core set of bibliometric indices

The diversity of bibliometric indices today poses the challenge of exploiting the relationships among them. Our research uncovers the best core set of relevant indices for predicting other bibliometric indices. An added difficulty is to select the role of each variable, that is, which bibliometric indices are predictive variables and which are response variables. This results in a novel multioutput regression problem where the role of each variable (predictor or response) is unknown beforehand. We use Gaussian Bayesian networks to solve the this problem and discover multivariate relationships among bibliometric indices. These networks are learnt by a genetic algorithm that looks for the optimal models that best predict bibliometric data. Results show that the optimal induced Gaussian Bayesian networks corroborate previous relationships between several indices, but also suggest new, previously unreported interactions. An extended analysis of the best model illustrates that a set of 12 bibliometric indices can be accurately predicted using only a smaller predictive core subset composed of citations, g-index, q2-index, and hr-index. This research is performed using bibliometric data on Spanish full professors associated with the computer science area.

Multi-dimensional classification of GABAergic interneurons with Bayesian network-modeled label uncertainty

Interneuron classification is an important and long-debated topic in neuroscience. A recent study provided a data set of digitally reconstructed interneurons classified by 42 leading neuroscientists according to a pragmatic classification scheme composed of five categorical variables, namely, of the interneuron type and four features of axonal morphology. From this data set we now learned a model which can classify interneurons, on the basis of their axonal morphometric parameters, into these five descriptive variables simultaneously. Because of differences in opinion among the neuroscientists, especially regarding neuronal type, for many interneurons we lacked a unique, agreed-upon classification, which we could use to guide model learning. Instead, we guided model learning with a probability distribution over the neuronal type and the axonal features, obtained, for each interneuron, from the neuroscientists’ classification choices. We conveniently encoded such probability distributions with Bayesian networks, calling them label Bayesian networks (LBNs), and developed a method to predict them. This method predicts an LBN by forming a probabilistic consensus among the LBNs of the interneurons most similar to the one being classified. We used 18 axonal morphometric parameters as predictor variables, 13 of which we introduce in this paper as quantitative counterparts to the categorical axonal features. We were able to accurately predict interneuronal LBNs. Furthermore, when extracting crisp (i.e., non-probabilistic) predictions from the predicted LBNs, our method outperformed related work on interneuron classification. Our results indicate that our method is adequate for multi-dimensional classification of interneurons with probabilistic labels. Moreover, the introduced morphometric parameters are good predictors of interneuron type and the four features of axonal morphology and thus may serve as objective counterparts to the subjective, categorical axonal features.

Predicting EQ-5D from the Parkinson's disease questionnaire PDQ-8 using multi-dimensional Bayesian network classifiers

The impact of the Parkinson's disease and its treatment on the patients' health-related quality of life can be estimated either by means of generic measures such as the european quality of Life-5 Dimensions (EQ-5D) or specific measures such as the 8-item Parkinson's disease questionnaire (PDQ-8). In clinical studies, PDQ-8 could be used in detriment of EQ-5D due to the lack of resources, time or clinical interest in generic measures. Nevertheless, PDQ-8 cannot be applied in cost-effectiveness analyses which require generic measures and quantitative utility scores, such as EQ-5D. To deal with this problem, a commonly used solution is the prediction of EQ-5D from PDQ-8. In this paper, we propose a new probabilistic method to predict EQ-5D from PDQ-8 using multi-dimensional Bayesian network classifiers. Our approach is evaluated using five-fold cross-validation experiments carried out on a Parkinson's data set containing 488 patients, and is compared with two additional Bayesian network-based approaches, two commonly used mapping methods namely, ordinary least squares and censored least absolute deviations, and a deterministic model. Experimental results are promising in terms of predictive performance as well as the identification of dependence relationships among EQ-5D and PDQ-8 items that the mapping approaches are unable to detect

Neuronanatomy, neurology and Bayesian networks

Bayesian networks are data mining models with clear semantics and a sound theoretical foundation. In this keynote talk we will pinpoint a number of neuroscience problems that can be addressed using Bayesian networks. In neuroanatomy, we will show computer simulation models of dendritic trees and classification of neuron types, both based on morphological features. In neurology, we will present the search for genetic biomarkers in Alzheimer's disease and the prediction of health-related quality of life in Parkinson's disease. Most of these challenging problems posed by neuroscience involve new Bayesian network designs that can cope with multiple class variables, small sample sizes, or labels annotated by several experts.

Bayesian scene analysis for multi-camera 3D tracking and camera positioning

Los sistemas de seguimiento mono-cámara han demostrado su notable capacidad para el análisis de trajectorias de objectos móviles y para monitorización de escenas de interés; sin embargo, tanto su robustez como sus posibilidades en cuanto a comprensión semántica de la escena están fuertemente limitadas por su naturaleza local y monocular, lo que los hace insuficientes para aplicaciones realistas de videovigilancia. El objetivo de esta tesis es la extensión de las posibilidades de los sistemas de seguimiento de objetos móviles para lograr un mayor grado de robustez y comprensión de la escena. La extensión propuesta se divide en dos direcciones separadas. La primera puede considerarse local, ya que está orientada a la mejora y enriquecimiento de las posiciones estimadas para los objetos móviles observados directamente por las cámaras del sistema; dicha extensión se logra mediante el desarrollo de un sistema multi-cámara de seguimiento 3D, capaz de proporcionar consistentemente las posiciones 3D de múltiples objetos a partir de las observaciones capturadas por un conjunto de sensores calibrados y con campos de visión solapados. La segunda extensión puede considerarse global, dado que su objetivo consiste en proporcionar un contexto global para relacionar las observaciones locales realizadas por una cámara con una escena de mucho mayor tamaño; para ello se propone un sistema automático de localización de cámaras basado en las trayectorias observadas de varios objetos móviles y en un mapa esquemático de la escena global monitorizada. Ambas líneas de investigación se tratan utilizando, como marco común, técnicas de estimación bayesiana: esta elección está justificada por la versatilidad y flexibilidad proporcionada por dicho marco estadístico, que permite la combinación natural de múltiples fuentes de información sobre los parámetros a estimar, así como un tratamiento riguroso de la incertidumbre asociada a las mismas mediante la inclusión de modelos de observación específicamente diseñados. Además, el marco seleccionado abre grandes posibilidades operacionales, puesto que permite la creación de diferentes métodos numéricos adaptados a las necesidades y características específicas de distintos problemas tratados. El sistema de seguimiento 3D con múltiples cámaras propuesto está específicamente diseñado para permitir descripciones esquemáticas de las medidas realizadas individualmente por cada una de las cámaras del sistema: esta elección de diseño, por tanto, no asume ningún algoritmo específico de detección o seguimiento 2D en ninguno de los sensores de la red, y hace que el sistema propuesto sea aplicable a redes reales de vigilancia con capacidades limitadas tanto en términos de procesamiento como de transmision. La combinación robusta de las observaciones capturadas individualmente por las cámaras, ruidosas, incompletas y probablemente contaminadas por falsas detecciones, se basa en un metodo de asociación bayesiana basado en geometría y color: los resultados de dicha asociación permiten el seguimiento 3D de los objetos de la escena mediante el uso de un filtro de partículas. El sistema de fusión de observaciones propuesto tiene, como principales características, una gran precisión en términos de localización 3D de objetos, y una destacable capacidad de recuperación tras eventuales errores debidos a un número insuficiente de datos de entrada. El sistema automático de localización de cámaras se basa en la observación de múltiples objetos móviles y un mapa esquemático de las áreas transitables del entorno monitorizado para inferir la posición absoluta de dicho sensor. Para este propósito, se propone un novedoso marco bayesiano que combina modelos dinámicos inducidos por el mapa en los objetos móviles presentes en la escena con las trayectorias observadas por la cámara, lo que representa un enfoque nunca utilizado en la literatura existente. El sistema de localización se divide en dos sub-tareas diferenciadas, debido a que cada una de estas tareas requiere del diseño de algoritmos específicos de muestreo para explotar en profundidad las características del marco desarrollado: por un lado, análisis de la ambigüedad del caso específicamente tratado y estimación aproximada de la localización de la cámara, y por otro, refinado de la localización de la cámara. El sistema completo, diseñado y probado para el caso específico de localización de cámaras en entornos de tráfico urbano, podría tener aplicación también en otros entornos y sensores de diferentes modalidades tras ciertas adaptaciones. ABSTRACT Mono-camera tracking systems have proved their capabilities for moving object trajectory analysis and scene monitoring, but their robustness and semantic possibilities are strongly limited by their local and monocular nature and are often insufficient for realistic surveillance applications. This thesis is aimed at extending the possibilities of moving object tracking systems to a higher level of scene understanding. The proposed extension comprises two separate directions. The first one is local, since is aimed at enriching the inferred positions of the moving objects within the area of the monitored scene directly covered by the cameras of the system; this task is achieved through the development of a multi-camera system for robust 3D tracking, able to provide 3D tracking information of multiple simultaneous moving objects from the observations reported by a set of calibrated cameras with semi-overlapping fields of view. The second extension is global, as is aimed at providing local observations performed within the field of view of one camera with a global context relating them to a much larger scene; to this end, an automatic camera positioning system relying only on observed object trajectories and a scene map is designed. The two lines of research in this thesis are addressed using Bayesian estimation as a general unifying framework. Its suitability for these two applications is justified by the flexibility and versatility of that stochastic framework, which allows the combination of multiple sources of information about the parameters to estimate in a natural and elegant way, addressing at the same time the uncertainty associated to those sources through the inclusion of models designed to this end. In addition, it opens multiple possibilities for the creation of different numerical methods for achieving satisfactory and efficient practical solutions to each addressed application. The proposed multi-camera 3D tracking method is specifically designed to work on schematic descriptions of the observations performed by each camera of the system: this choice allows the use of unspecific off-the-shelf 2D detection and/or tracking subsystems running independently at each sensor, and makes the proposal suitable for real surveillance networks with moderate computational and transmission capabilities. The robust combination of such noisy, incomplete and possibly unreliable schematic descriptors relies on a Bayesian association method, based on geometry and color, whose results allow the tracking of the targets in the scene with a particle filter. The main features exhibited by the proposal are, first, a remarkable accuracy in terms of target 3D positioning, and second, a great recovery ability after tracking losses due to insufficient input data. The proposed system for visual-based camera self-positioning uses the observations of moving objects and a schematic map of the passable areas of the environment to infer the absolute sensor position. To this end, a new Bayesian framework combining trajectory observations and map-induced dynamic models for moving objects is designed, which represents an approach to camera positioning never addressed before in the literature. This task is divided into two different sub-tasks, setting ambiguity analysis and approximate position estimation, on the one hand, and position refining, on the other, since they require the design of specific sampling algorithms to correctly exploit the discriminative features of the developed framework. This system, designed for camera positioning and demonstrated in urban traffic environments, can also be applied to different environments and sensors of other modalities after certain required adaptations.

Application of Bayesian approach in geotechnical engineering

En esta Tesis Doctoral se emplean y desarrollan Métodos Bayesianos para su aplicación en análisis geotécnicos habituales, con un énfasis particular en (i) la valoración y selección de modelos geotécnicos basados en correlaciones empíricas; en (ii) el desarrollo de predicciones acerca de los resultados esperados en modelos geotécnicos complejos. Se llevan a cabo diferentes aplicaciones a problemas geotécnicos, como es el caso de: (1) En el caso de rocas intactas, se presenta un método Bayesiano para la evaluación de modelos que permiten estimar el módulo de Young a partir de la resistencia a compresión simple (UCS). La metodología desarrollada suministra estimaciones de las incertidumbres de los parámetros y predicciones y es capaz de diferenciar entre las diferentes fuentes de error. Se desarrollan modelos "específicos de roca" para los tipos de roca más comunes y se muestra cómo se pueden "actualizar" esos modelos "iniciales" para incorporar, cuando se encuentra disponible, la nueva información específica del proyecto, reduciendo las incertidumbres del modelo y mejorando sus capacidades predictivas. (2) Para macizos rocosos, se presenta una metodología, fundamentada en un criterio de selección de modelos, que permite determinar el modelo más apropiado, entre un conjunto de candidatos, para estimar el módulo de deformación de un macizo rocoso a partir de un conjunto de datos observados. Una vez que se ha seleccionado el modelo más apropiado, se emplea un método Bayesiano para obtener distribuciones predictivas de los módulos de deformación de macizos rocosos y para actualizarlos con la nueva información específica del proyecto. Este método Bayesiano de actualización puede reducir significativamente la incertidumbre asociada a la predicción, y por lo tanto, afectar las estimaciones que se hagan de la probabilidad de fallo, lo cual es de un interés significativo para los diseños de mecánica de rocas basados en fiabilidad. (3) En las primeras etapas de los diseños de mecánica de rocas, la información acerca de los parámetros geomecánicos y geométricos, las tensiones in-situ o los parámetros de sostenimiento, es, a menudo, escasa o incompleta. Esto plantea dificultades para aplicar las correlaciones empíricas tradicionales que no pueden trabajar con información incompleta para realizar predicciones. Por lo tanto, se propone la utilización de una Red Bayesiana para trabajar con información incompleta y, en particular, se desarrolla un clasificador Naïve Bayes para predecir la probabilidad de ocurrencia de grandes deformaciones (squeezing) en un túnel a partir de cinco parámetros de entrada habitualmente disponibles, al menos parcialmente, en la etapa de diseño. This dissertation employs and develops Bayesian methods to be used in typical geotechnical analyses, with a particular emphasis on (i) the assessment and selection of geotechnical models based on empirical correlations; on (ii) the development of probabilistic predictions of outcomes expected for complex geotechnical models. Examples of application to geotechnical problems are developed, as follows: (1) For intact rocks, we present a Bayesian framework for model assessment to estimate the Young’s moduli based on their UCS. Our approach provides uncertainty estimates of parameters and predictions, and can differentiate among the sources of error. We develop ‘rock-specific’ models for common rock types, and illustrate that such ‘initial’ models can be ‘updated’ to incorporate new project-specific information as it becomes available, reducing model uncertainties and improving their predictive capabilities. (2) For rock masses, we present an approach, based on model selection criteria to select the most appropriate model, among a set of candidate models, to estimate the deformation modulus of a rock mass, given a set of observed data. Once the most appropriate model is selected, a Bayesian framework is employed to develop predictive distributions of the deformation moduli of rock masses, and to update them with new project-specific data. Such Bayesian updating approach can significantly reduce the associated predictive uncertainty, and therefore, affect our computed estimates of probability of failure, which is of significant interest to reliability-based rock engineering design. (3) In the preliminary design stage of rock engineering, the information about geomechanical and geometrical parameters, in situ stress or support parameters is often scarce or incomplete. This poses difficulties in applying traditional empirical correlations that cannot deal with incomplete data to make predictions. Therefore, we propose the use of Bayesian Networks to deal with incomplete data and, in particular, a Naïve Bayes classifier is developed to predict the probability of occurrence of tunnel squeezing based on five input parameters that are commonly available, at least partially, at design stages.

How should the sustainability of the location of dry ports be measured? A proposed methodology using Bayesian networks and multi-criteria decision analysis

The global economic structure, with its decentralized production and the consequent increase in freight traffic all over the world, creates considerable problems and challenges for the freight transport sector. This situation has led shipping to become the most suitable and cheapest way to transport goods. Thus, ports are configured as nodes with critical importance in the logistics supply chain as a link between two transport systems, sea and land. Increase in activity at seaports is producing three undesirable effects: increasing road congestion, lack of open space in port installations and a significant environmental impact on seaports. These adverse effects can be mitigated by moving part of the activity inland. Implementation of dry ports is a possible solution and would also provide an opportunity to strengthen intermodal solutions as part of an integrated and more sustainable transport chain, acting as a link between road and railway networks. In this sense, implementation of dry ports allows the separation of the links of the transport chain, thus facilitating the shortest possible routes for the lowest capacity and most polluting means of transport. Thus, the decision of where to locate a dry port demands a thorough analysis of the whole logistics supply chain, with the objective of transferring the largest volume of goods possible from road to more energy efficient means of transport, like rail or short-sea shipping, that are less harmful to the environment. However, the decision of where to locate a dry port must also ensure the sustainability of the site. Thus, the main goal of this article is to research the variables influencing the sustainability of dry port location and how this sustainability can be evaluated. With this objective, in this paper we present a methodology for assessing the sustainability of locations by the use of Multi-Criteria Decision Analysis (MCDA) and Bayesian Networks (BNs). MCDA is used as a way to establish a scoring, whilst BNs were chosen to eliminate arbitrariness in setting the weightings using a technique that allows us to prioritize each variable according to the relationships established in the set of variables. In order to determine the relationships between all the variables involved in the decision, giving us the importance of each factor and variable, we built a K2 BN algorithm. To obtain the scores of each variable, we used a complete cartography analysed by ArcGIS. Recognising that setting the most appropriate location to place a dry port is a geographical multidisciplinary problem, with significant economic, social and environmental implications, we consider 41 variables (grouped into 17 factors) which respond to this need. As a case of study, the sustainability of all of the 10 existing dry ports in Spain has been evaluated. In this set of logistics platforms, we found that the most important variables for achieving sustainability are those related to environmental protection, so the sustainability of the locations requires a great respect for the natural environment and the urban environment in which they are framed.

Bayesian network modeling of the consensus between experts: an application to neuron classification

Neuronal morphology is hugely variable across brain regions and species, and their classification strategies are a matter of intense debate in neuroscience. GABAergic cortical interneurons have been a challenge because it is difficult to find a set of morphological properties which clearly define neuronal types. A group of 48 neuroscience experts around the world were asked to classify a set of 320 cortical GABAergic interneurons according to the main features of their three-dimensional morphological reconstructions. A methodology for building a model which captures the opinions of all the experts was proposed. First, one Bayesian network was learned for each expert, and we proposed an algorithm for clustering Bayesian networks corresponding to experts with similar behaviors. Then, a Bayesian network which represents the opinions of each group of experts was induced. Finally, a consensus Bayesian multinet which models the opinions of the whole group of experts was built. A thorough analysis of the consensus model identified different behaviors between the experts when classifying the interneurons in the experiment. A set of characterizing morphological traits for the neuronal types was defined by performing inference in the Bayesian multinet. These findings were used to validate the model and to gain some insights into neuron morphology.

Contribución al estudio de las metodologías de cálculo realista con incertidumbre (BEPU), dentro del análisis determinista de seguridad de plantas nucleares

El análisis determinista de seguridad (DSA) es el procedimiento que sirve para diseñar sistemas, estructuras y componentes relacionados con la seguridad en las plantas nucleares. El DSA se basa en simulaciones computacionales de una serie de hipotéticos accidentes representativos de la instalación, llamados escenarios base de diseño (DBS). Los organismos reguladores señalan una serie de magnitudes de seguridad que deben calcularse en las simulaciones, y establecen unos criterios reguladores de aceptación (CRA), que son restricciones que deben cumplir los valores de esas magnitudes. Las metodologías para realizar los DSA pueden ser de 2 tipos: conservadoras o realistas. Las metodologías conservadoras utilizan modelos predictivos e hipótesis marcadamente pesimistas, y, por ello, relativamente simples. No necesitan incluir un análisis de incertidumbre de sus resultados. Las metodologías realistas se basan en hipótesis y modelos predictivos realistas, generalmente mecanicistas, y se suplementan con un análisis de incertidumbre de sus principales resultados. Se les denomina también metodologías BEPU (“Best Estimate Plus Uncertainty”). En ellas, la incertidumbre se representa, básicamente, de manera probabilista. Para metodologías conservadores, los CRA son, simplemente, restricciones sobre valores calculados de las magnitudes de seguridad, que deben quedar confinados en una “región de aceptación” de su recorrido. Para metodologías BEPU, el CRA no puede ser tan sencillo, porque las magnitudes de seguridad son ahora variables inciertas. En la tesis se desarrolla la manera de introducción de la incertidumbre en los CRA. Básicamente, se mantiene el confinamiento a la misma región de aceptación, establecida por el regulador. Pero no se exige el cumplimiento estricto sino un alto nivel de certidumbre. En el formalismo adoptado, se entiende por ello un “alto nivel de probabilidad”, y ésta corresponde a la incertidumbre de cálculo de las magnitudes de seguridad. Tal incertidumbre puede considerarse como originada en los inputs al modelo de cálculo, y propagada a través de dicho modelo. Los inputs inciertos incluyen las condiciones iniciales y de frontera al cálculo, y los parámetros empíricos de modelo, que se utilizan para incorporar la incertidumbre debida a la imperfección del modelo. Se exige, por tanto, el cumplimiento del CRA con una probabilidad no menor a un valor P0 cercano a 1 y definido por el regulador (nivel de probabilidad o cobertura). Sin embargo, la de cálculo de la magnitud no es la única incertidumbre existente. Aunque un modelo (sus ecuaciones básicas) se conozca a la perfección, la aplicación input-output que produce se conoce de manera imperfecta (salvo que el modelo sea muy simple). La incertidumbre debida la ignorancia sobre la acción del modelo se denomina epistémica; también se puede decir que es incertidumbre respecto a la propagación. La consecuencia es que la probabilidad de cumplimiento del CRA no se puede conocer a la perfección; es una magnitud incierta. Y así se justifica otro término usado aquí para esta incertidumbre epistémica: metaincertidumbre. Los CRA deben incorporar los dos tipos de incertidumbre: la de cálculo de la magnitud de seguridad (aquí llamada aleatoria) y la de cálculo de la probabilidad (llamada epistémica o metaincertidumbre). Ambas incertidumbres pueden introducirse de dos maneras: separadas o combinadas. En ambos casos, el CRA se convierte en un criterio probabilista. Si se separan incertidumbres, se utiliza una probabilidad de segundo orden; si se combinan, se utiliza una probabilidad única. Si se emplea la probabilidad de segundo orden, es necesario que el regulador imponga un segundo nivel de cumplimiento, referido a la incertidumbre epistémica. Se denomina nivel regulador de confianza, y debe ser un número cercano a 1. Al par formado por los dos niveles reguladores (de probabilidad y de confianza) se le llama nivel regulador de tolerancia. En la Tesis se razona que la mejor manera de construir el CRA BEPU es separando las incertidumbres, por dos motivos. Primero, los expertos defienden el tratamiento por separado de incertidumbre aleatoria y epistémica. Segundo, el CRA separado es (salvo en casos excepcionales) más conservador que el CRA combinado. El CRA BEPU no es otra cosa que una hipótesis sobre una distribución de probabilidad, y su comprobación se realiza de forma estadística. En la tesis, los métodos estadísticos para comprobar el CRA BEPU en 3 categorías, según estén basados en construcción de regiones de tolerancia, en estimaciones de cuantiles o en estimaciones de probabilidades (ya sea de cumplimiento, ya sea de excedencia de límites reguladores). Según denominación propuesta recientemente, las dos primeras categorías corresponden a los métodos Q, y la tercera, a los métodos P. El propósito de la clasificación no es hacer un inventario de los distintos métodos en cada categoría, que son muy numerosos y variados, sino de relacionar las distintas categorías y citar los métodos más utilizados y los mejor considerados desde el punto de vista regulador. Se hace mención especial del método más utilizado hasta el momento: el método no paramétrico de Wilks, junto con su extensión, hecha por Wald, al caso multidimensional. Se decribe su método P homólogo, el intervalo de Clopper-Pearson, típicamente ignorado en el ámbito BEPU. En este contexto, se menciona el problema del coste computacional del análisis de incertidumbre. Los métodos de Wilks, Wald y Clopper-Pearson requieren que la muestra aleatortia utilizada tenga un tamaño mínimo, tanto mayor cuanto mayor el nivel de tolerancia exigido. El tamaño de muestra es un indicador del coste computacional, porque cada elemento muestral es un valor de la magnitud de seguridad, que requiere un cálculo con modelos predictivos. Se hace especial énfasis en el coste computacional cuando la magnitud de seguridad es multidimensional; es decir, cuando el CRA es un criterio múltiple. Se demuestra que, cuando las distintas componentes de la magnitud se obtienen de un mismo cálculo, el carácter multidimensional no introduce ningún coste computacional adicional. Se prueba así la falsedad de una creencia habitual en el ámbito BEPU: que el problema multidimensional sólo es atacable desde la extensión de Wald, que tiene un coste de computación creciente con la dimensión del problema. En el caso (que se da a veces) en que cada componente de la magnitud se calcula independientemente de los demás, la influencia de la dimensión en el coste no se puede evitar. Las primeras metodologías BEPU hacían la propagación de incertidumbres a través de un modelo sustitutivo (metamodelo o emulador) del modelo predictivo o código. El objetivo del metamodelo no es su capacidad predictiva, muy inferior a la del modelo original, sino reemplazar a éste exclusivamente en la propagación de incertidumbres. Para ello, el metamodelo se debe construir con los parámetros de input que más contribuyan a la incertidumbre del resultado, y eso requiere un análisis de importancia o de sensibilidad previo. Por su simplicidad, el modelo sustitutivo apenas supone coste computacional, y puede estudiarse exhaustivamente, por ejemplo mediante muestras aleatorias. En consecuencia, la incertidumbre epistémica o metaincertidumbre desaparece, y el criterio BEPU para metamodelos se convierte en una probabilidad simple. En un resumen rápido, el regulador aceptará con más facilidad los métodos estadísticos que menos hipótesis necesiten; los exactos más que los aproximados; los no paramétricos más que los paramétricos, y los frecuentistas más que los bayesianos. El criterio BEPU se basa en una probabilidad de segundo orden. La probabilidad de que las magnitudes de seguridad estén en la región de aceptación no sólo puede asimilarse a una probabilidad de éxito o un grado de cumplimiento del CRA. También tiene una interpretación métrica: representa una distancia (dentro del recorrido de las magnitudes) desde la magnitud calculada hasta los límites reguladores de aceptación. Esta interpretación da pie a una definición que propone esta tesis: la de margen de seguridad probabilista. Dada una magnitud de seguridad escalar con un límite superior de aceptación, se define el margen de seguridad (MS) entre dos valores A y B de la misma como la probabilidad de que A sea menor que B, obtenida a partir de las incertidumbres de A y B. La definición probabilista de MS tiene varias ventajas: es adimensional, puede combinarse de acuerdo con las leyes de la probabilidad y es fácilmente generalizable a varias dimensiones. Además, no cumple la propiedad simétrica. El término margen de seguridad puede aplicarse a distintas situaciones: distancia de una magnitud calculada a un límite regulador (margen de licencia); distancia del valor real de la magnitud a su valor calculado (margen analítico); distancia desde un límite regulador hasta el valor umbral de daño a una barrera (margen de barrera). Esta idea de representar distancias (en el recorrido de magnitudes de seguridad) mediante probabilidades puede aplicarse al estudio del conservadurismo. El margen analítico puede interpretarse como el grado de conservadurismo (GC) de la metodología de cálculo. Utilizando la probabilidad, se puede cuantificar el conservadurismo de límites de tolerancia de una magnitud, y se pueden establecer indicadores de conservadurismo que sirvan para comparar diferentes métodos de construcción de límites y regiones de tolerancia. Un tópico que nunca se abordado de manera rigurosa es el de la validación de metodologías BEPU. Como cualquier otro instrumento de cálculo, una metodología, antes de poder aplicarse a análisis de licencia, tiene que validarse, mediante la comparación entre sus predicciones y valores reales de las magnitudes de seguridad. Tal comparación sólo puede hacerse en escenarios de accidente para los que existan valores medidos de las magnitudes de seguridad, y eso ocurre, básicamente en instalaciones experimentales. El objetivo último del establecimiento de los CRA consiste en verificar que se cumplen para los valores reales de las magnitudes de seguridad, y no sólo para sus valores calculados. En la tesis se demuestra que una condición suficiente para este objetivo último es la conjunción del cumplimiento de 2 criterios: el CRA BEPU de licencia y un criterio análogo, pero aplicado a validación. Y el criterio de validación debe demostrarse en escenarios experimentales y extrapolarse a plantas nucleares. El criterio de licencia exige un valor mínimo (P0) del margen probabilista de licencia; el criterio de validación exige un valor mínimo del margen analítico (el GC). Esos niveles mínimos son básicamente complementarios; cuanto mayor uno, menor el otro. La práctica reguladora actual impone un valor alto al margen de licencia, y eso supone que el GC exigido es pequeño. Adoptar valores menores para P0 supone menor exigencia sobre el cumplimiento del CRA, y, en cambio, más exigencia sobre el GC de la metodología. Y es importante destacar que cuanto mayor sea el valor mínimo del margen (de licencia o analítico) mayor es el coste computacional para demostrarlo. Así que los esfuerzos computacionales también son complementarios: si uno de los niveles es alto (lo que aumenta la exigencia en el cumplimiento del criterio) aumenta el coste computacional. Si se adopta un valor medio de P0, el GC exigido también es medio, con lo que la metodología no tiene que ser muy conservadora, y el coste computacional total (licencia más validación) puede optimizarse. ABSTRACT Deterministic Safety Analysis (DSA) is the procedure used in the design of safety-related systems, structures and components of nuclear power plants (NPPs). DSA is based on computational simulations of a set of hypothetical accidents of the plant, named Design Basis Scenarios (DBS). Nuclear regulatory authorities require the calculation of a set of safety magnitudes, and define the regulatory acceptance criteria (RAC) that must be fulfilled by them. Methodologies for performing DSA van be categorized as conservative or realistic. Conservative methodologies make use of pessimistic model and assumptions, and are relatively simple. They do not need an uncertainty analysis of their results. Realistic methodologies are based on realistic (usually mechanistic) predictive models and assumptions, and need to be supplemented with uncertainty analyses of their results. They are also termed BEPU (“Best Estimate Plus Uncertainty”) methodologies, and are typically based on a probabilistic representation of the uncertainty. For conservative methodologies, the RAC are simply the restriction of calculated values of safety magnitudes to “acceptance regions” defined on their range. For BEPU methodologies, the RAC cannot be so simple, because the safety magnitudes are now uncertain. In the present Thesis, the inclusion of uncertainty in RAC is studied. Basically, the restriction to the acceptance region must be fulfilled “with a high certainty level”. Specifically, a high probability of fulfillment is required. The calculation uncertainty of the magnitudes is considered as propagated from inputs through the predictive model. Uncertain inputs include model empirical parameters, which store the uncertainty due to the model imperfection. The fulfillment of the RAC is required with a probability not less than a value P0 close to 1 and defined by the regulator (probability or coverage level). Calculation uncertainty is not the only one involved. Even if a model (i.e. the basic equations) is perfectly known, the input-output mapping produced by the model is imperfectly known (unless the model is very simple). This ignorance is called epistemic uncertainty, and it is associated to the process of propagation). In fact, it is propagated to the probability of fulfilling the RAC. Another term used on the Thesis for this epistemic uncertainty is metauncertainty. The RAC must include the two types of uncertainty: one for the calculation of the magnitude (aleatory uncertainty); the other one, for the calculation of the probability (epistemic uncertainty). The two uncertainties can be taken into account in a separate fashion, or can be combined. In any case the RAC becomes a probabilistic criterion. If uncertainties are separated, a second-order probability is used; of both are combined, a single probability is used. On the first case, the regulator must define a level of fulfillment for the epistemic uncertainty, termed regulatory confidence level, as a value close to 1. The pair of regulatory levels (probability and confidence) is termed the regulatory tolerance level. The Thesis concludes that the adequate way of setting the BEPU RAC is by separating the uncertainties. There are two reasons to do so: experts recommend the separation of aleatory and epistemic uncertainty; and the separated RAC is in general more conservative than the joint RAC. The BEPU RAC is a hypothesis on a probability distribution, and must be statistically tested. The Thesis classifies the statistical methods to verify the RAC fulfillment in 3 categories: methods based on tolerance regions, in quantile estimators and on probability (of success or failure) estimators. The former two have been termed Q-methods, whereas those in the third category are termed P-methods. The purpose of our categorization is not to make an exhaustive survey of the very numerous existing methods. Rather, the goal is to relate the three categories and examine the most used methods from a regulatory standpoint. Special mention deserves the most used method, due to Wilks, and its extension to multidimensional variables (due to Wald). The counterpart P-method of Wilks’ is Clopper-Pearson interval, typically ignored in the BEPU realm. The problem of the computational cost of an uncertainty analysis is tackled. Wilks’, Wald’s and Clopper-Pearson methods require a minimum sample size, which is a growing function of the tolerance level. The sample size is an indicator of the computational cost, because each element of the sample must be calculated with the predictive models (codes). When the RAC is a multiple criteria, the safety magnitude becomes multidimensional. When all its components are output of the same calculation, the multidimensional character does not introduce additional computational cost. In this way, an extended idea in the BEPU realm, stating that the multi-D problem can only be tackled with the Wald extension, is proven to be false. When the components of the magnitude are independently calculated, the influence of the problem dimension on the cost cannot be avoided. The former BEPU methodologies performed the uncertainty propagation through a surrogate model of the code, also termed emulator or metamodel. The goal of a metamodel is not the predictive capability, clearly worse to the original code, but the capacity to propagate uncertainties with a lower computational cost. The emulator must contain the input parameters contributing the most to the output uncertainty, and this requires a previous importance analysis. The surrogate model is practically inexpensive to run, so that it can be exhaustively analyzed through Monte Carlo. Therefore, the epistemic uncertainty due to sampling will be reduced to almost zero, and the BEPU RAC for metamodels includes a simple probability. The regulatory authority will tend to accept the use of statistical methods which need a minimum of assumptions: exact, nonparametric and frequentist methods rather than approximate, parametric and bayesian methods, respectively. The BEPU RAC is based on a second-order probability. The probability of the safety magnitudes being inside the acceptance region is a success probability and can be interpreted as a fulfillment degree if the RAC. Furthermore, it has a metric interpretation, as a distance (in the range of magnitudes) from calculated values of the magnitudes to acceptance regulatory limits. A probabilistic definition of safety margin (SM) is proposed in the thesis. The same from a value A to other value B of a safety magnitude is defined as the probability that A is less severe than B, obtained from the uncertainties if A and B. The probabilistic definition of SM has several advantages: it is nondimensional, ranges in the interval (0,1) and can be easily generalized to multiple dimensions. Furthermore, probabilistic SM are combined according to the probability laws. And a basic property: probabilistic SM are not symmetric. There are several types of SM: distance from a calculated value to a regulatory limit (licensing margin); or from the real value to the calculated value of a magnitude (analytical margin); or from the regulatory limit to the damage threshold (barrier margin). These representations of distances (in the magnitudes’ range) as probabilities can be applied to the quantification of conservativeness. Analytical margins can be interpreted as the degree of conservativeness (DG) of the computational methodology. Conservativeness indicators are established in the Thesis, useful in the comparison of different methods of constructing tolerance limits and regions. There is a topic which has not been rigorously tackled to the date: the validation of BEPU methodologies. Before being applied in licensing, methodologies must be validated, on the basis of comparisons of their predictions ad real values of the safety magnitudes. Real data are obtained, basically, in experimental facilities. The ultimate goal of establishing RAC is to verify that real values (aside from calculated values) fulfill them. In the Thesis it is proved that a sufficient condition for this goal is the conjunction of 2 criteria: the BEPU RAC and an analogous criterion for validation. And this las criterion must be proved in experimental scenarios and extrapolated to NPPs. The licensing RAC requires a minimum value (P0) of the probabilistic licensing margin; the validation criterion requires a minimum value of the analytical margin (i.e., of the DG). These minimum values are basically complementary; the higher one of them, the lower the other one. The regulatory practice sets a high value on the licensing margin, so that the required DG is low. The possible adoption of lower values for P0 would imply weaker exigence on the RCA fulfillment and, on the other hand, higher exigence on the conservativeness of the methodology. It is important to highlight that a higher minimum value of the licensing or analytical margin requires a higher computational cost. Therefore, the computational efforts are also complementary. If medium levels are adopted, the required DG is also medium, and the methodology does not need to be very conservative. The total computational effort (licensing plus validation) could be optimized.

Decision boundary for discrete Bayesian network classifiers

Bayesian network classifiers are a powerful machine learning tool. In order to evaluate the expressive power of these models, we compute families of polynomials that sign-represent decision functions induced by Bayesian network classifiers. We prove that those families are linear combinations of products of Lagrange basis polynomials. In absence of V -structures in the predictor sub-graph, we are also able to prove that this family of polynomials does indeed characterize the specific classifier considered. We then use this representation to bound the number of decision functions representable by Bayesian network classifiers with a given structure.

Crowdsourcing without a crowd : Reliable online species identification using Bayesian models to minimize crowd size

Acknowledgment This research is supported by an award made by the RCUK Digital Economy program to the University of Aberdeen’s dot.rural Digital Economy Hub (ref. EP/G066051/1).