Due to its ability to deal with non-determinism and partial observability, represent goals as an immediate reward function and find optimal solutions, planning under uncertainty using factored Markov Decision Processes (FMDPs) has increased its importance and usage in power plants and power systems. In this paper, three different applications using this approach are described: (i) optimal dam management in hydroelectric power plants, (ii) inspection and surveillance in electric substations, and (iii) optimization of steam generation in a combined cycle power plant. For each case, the technique has demonstrated to find optimal action policies in uncertain settings, present good response and compilation times, deal with stochastic variables and be a good alternative to traditional control systems. The main contributions of this work are as follows, a methodology to approximate a decision model using machine learning techniques, and examples of how to specify and solve problems in the electric power domain in terms of a FMDP.

The Mexican Center for Innovation in Wind Energy (CEMIE-Eólico) designed a wind turbine diagnostic system based on turbine behavior models using the signals of the Supervisory Control and Data Acquisition system (SCADA). The system provides a pattern of variables that exhibit abnormal behavior in the presence of a fault. The patterns are formed with the detection of the abnormal behavior of the variables during a time window in which the failure manifests itself. This paper presents the application of machine learning techniques for the identification of faults in wind turbines after the diagnostic system. The training and validation data were obtained from the simulation of six different faults in the wind turbine using the Mexican Wind Machine (MEM) designed at the National Institute of Electricity and Clean Energy (INEEL). The diagnostic system was applied, profiles of abnormal behavior were generated and experiments were carried out for the multiclass classification of fault patterns using the "Random Forest" algorithm. Finally, the algorithm performance was evaluated using accuracy and precision metrics achieving 91% in the classification of patterns to identify the root failure.

In this paper the SPI system is presented. SPI is a software tool for planning under uncertainty based on learning Markov Decision Processes. A brief review of some similar tools as well as the scientific basis of factored representations and some of its variants are included. Among these variants are qualitative representations and hybrid qualitative-discrete representations that are the core of the software tool. The functional structure of SPI, which is composed of four main modules, is also described. These modules are: the compiler, the policy server, a format translator and a didactic simulator. The experimental results obtained when testing SPI in a robot navigation domain using different types of representations and different state partitions demonstrated its capability to reduce state spaces.

The aim of this paper is the neutronic flux prognosis in a nuclear reactor for faults in the measurement of local power range monitors (LPRMs) in real time using differential probabilistic space-temporal model (DPSTM). The LPRMs provide inputs to the average power range monitor (APRM). The LPRM houses a fission chamber and their associated signal cables. The failure of one or more chains of LPRMs is common during the operational cycle. The circuit averages only LPRM signals that are operational and the output from the averaging circuit for each APRM channel is the route to the process computer. The DPSTM allows a reliable reconstruction in real time signal of those LPRMs that are out of order. The DPSTM is evaluated in terms of predictive accuracy for different time horizons and compared to a time series. The DPSTM based prognosis methodology was developed and validated with real signals of Ringhals stability benchmarks.

Condition monitoring systems have been used to determine the need for equipment maintenance and successful results have been achieved, however, these systems in combination with fuzzy logic have not been sufficiently explored, and one area of application is wind turbines. In this paper, the state of the art of wind turbine condition monitoring is briefly reviewed and it is proposed to use fuzzy logic for the diagnosis of wind turbine condition, in order to detect the abnormal behavior of the signals through a system of fuzzy inference that can be used as a fundamental element in the diagnosis of turbine condition. The system is based on data from a Komai wind turbine whose specifications were used to model the system. The results of the tests indicate that this system can be used to represent human knowledge and the diagnosis is reliable.

Data integration and data cleansing are particularly relevant for military applications where trustable data can make a difference in life-threatening conditions. This chapter proposes a novel and robust mechanism for information validation and amendment in databases where certainty in information is critical. It describes the set of experiments conducted for the comparison of performance between different methods on different datasets. The chapter includes a description of some available methods to complete databases and a more complete description of the autoregressive Bayesian network. The knowledge in a process using Bayesian networks can be represented with two elements: the structure of the network, and the parameters. Dynamic Bayesian networks (DBNs) are an attempt to add the temporal dimension into the Bayesian network model. Autoregressive Bayesian networks are a simplified variant of DBNs. Results suggest that the interplay between the variable's characteristics in the dataset dictates the most beneficial reconstruction option.

The detection and subsequent reconstruction of incongruent data in time series by means of observation of statistically related information is a recurrent issue in data validation. Unlike outliers, incongruent observations are not necessarily confined to the extremes of the data distribution. Instead, these rogue observations are unlikely values in the light of statistically related information. This paper proposes a multiresolution Bayesian network model for the detection of rogue values and posterior reconstruction of the erroneous sample for non-stationary time-series. Our method builds local Bayesian Network models that best fit to segments of data in order to achieve a finer discretization and hence improve data reconstruction. Our local multiscale approach is compared against its single-scale global predecessor (assumed as our gold standard) in the predictive power and of this, both error detection capabilities and error reconstruction capabilities are assessed. This parameterization and verification of the model are evaluated over three synthetic data source topologies. The virtues of the algorithm are then further tested in real data from the steel industry where the aforementioned problem characteristics are met but for which the ground truth is unknown. The proposed local multiscale approach was found to dealt better with increasing complexities in data topologies.

Several problems requiere the combination of temporal and spatial reasoning under uncertainty, such as wind prediction for electricity generation in wind farms. In this work we pro-pose a probabilistic spatial-temporal model (PSTM) focused on prediction problems, based on two common properties of these scenarios: sparsity and multivariable mutual information. The proposed spatial-temporal model is essentially a Bayesian network that represents the dependencies between a target variable of interest and a subset of predictor variables in different times and spaces. We developed an algorithm for learning the structure of the model based on a stochastic search of the optimal subset of predictor variables. The proposed model has been applied for wind prediction at different locations in Mexico, using information from several locations at different times. The PSTM is evaluated in terms of predictive accuracy for different time horizons – 1 to 24 hours; and compared to a dynamic Bayesian network (DBN) developed for wind prediction. The performance of the PSTM is in general competitive, and in most cases superior to the DBN.

Forecasting represents a very important task for planning, control and decision making in many fields. Forecasting the dollar price is important for global companies to plan their investments. Forecasting the weather is determinant to make the decision of either giving a party outdoors or indoors. Forecasting the behaviour of a process represents the key factor in predictive control. In this paper, we present a methodology to build wind power forecasting models from data using a combination of artificial intelligence techniques such as artificial neural networks and dynamic Bayesian nets. These techniques allow obtaining forecast models with different characteristics. Finally, a model recalibration function is applied to raw discrete models in order to gain an extra accuracy. The experiments ran for the unit 1 of the Villonaco wind farm in Ecuador demonstrated that the selection of the best predictor can be more useful than selecting a single high-efficiency approach.

Behavior can be defined as combination of variable’s values according to external inputs or environmental changes. This definition can be applied to persons, equipment, social systems or industrial processes. This paper proposes a probabilistic mechanism to represent the behavior of industrial equipment and an algorithm to identify deviations to this behavior. The anomaly detection mechanisms, together with the sensor validation theory are combined to propose an efficient manner to diagnose industrial equipment. A case study is presented with the failure identification of a wind turbine. The diagnosis is conducted when detecting deviations to the turbine normal behavior.

The viscosity measurement and control of fuel oil in power plants is very important for a proper combustion. However, the conventional viscometers are only reliable for a short period of time. This paper proposes an on-line analytic viscosity evaluation based on energy balance applied to a piece of tube entering the fuel oil main heater and a new control strategy for temperature control. This analytic evaluation utilizes a set of temperature versus viscosity graphs were defined during years of analysis of fuel oil in Mexican power plants. Also the temperature set-point for the fuel oil main heater output is obtained by interpolating in the corresponding graph. Validation tests of the proposed analytic equations were carried out in the Tuxpan power plant in Veracruz, Mexico.

For the last years, the research and development in the field of Renewable Energy has been growing due to the need of Renewable Energy as an extended and reliable source of energy. However, the implementation of renewable energy has many complex problems not easily solved with conventional methods. Recently, Artificial Intelligence techniques such as Artificial Neural Networks, Fuzzy Logic and Genetic Algorithms, have been widely used to deal with these problems in the field of Renewable Energy. Nevertheless, issues with a degree of uncertainty need Bayesian Networks since this is one of the most effective theories to face them. This technique can contribute to the Renewable Energy harnessing and other open issues on this field. In this work we show the state of the art of the applications of Bayesian Networks in Renewable Energy, such as solar thermal, photovoltaic, wind, geothermal, hydroelectric energies and biomass. Additionally, we include related topics such as energy storage, smart grids and energy assessment. We classify the literature by areas considering three main subjects: resource evaluation, operation, and applications, and in each section we describe the possible directions to be taken in the research of the field. We find that the main applications are done for forecasting, fault diagnosis, maintenance, operation, planning, sizing and risk management. We conclude that Bayesian Networks are a promising tool for the field of Renewable Energy with potential applications due to their versatility.

The behavior of an equipment can be seen as the variations of some parameters when changes in the envieronment are experienced. This paper describes how Bayesian networks can be used to learn a probabilistic model of the equipment’s behavior. Using this model, it is possible to identify deviations to the normal behavior. This means that on–line diagnosis can be executed. This paper describes the behaviour model of a wind turbine and the preliminary experiments to identify deviations. The experiments to learn the model based on historical data are presented and the experiments to validate the model using a turbine failure data.

Forecasting represents a very important task for control and decision making in many fields. Forecasting the dollar price is important for global companies to plan their investments. Forecasting the wind speed for a day-ahead horizon allows dispatching clean energy efficiently. One technique developed by the artificial intelligence community that has proved to be efficient for forecasting is the probabilistic graphical models approach. In order to obtain accurate models for forecasting, there exist different assumptions that might be made. This paper presents these assumptions and the results of different experiments conducted to define the characteristics of good probabilistic graphical models. A performance comparison of several graphical models is also presented. The experiments were executed to forecast wind velocity and hence, wind power in wind farms.

Renewable energy is increasing its participation in power generation in many countries. In Mexico, the strategy is to generate 35% of electricity from renewable sources by 2024. Currently only 18.3% of the generated energy is obtained from renewable and clean sources. The integration of renewable energies in the energy market is a challenge due to their high variability, instability and uncertainty. Hence, energy forecast is the required service by the power generators to offer energy with certain degree of confidence. Dynamic Bayesian networks (DBNs) have proved to be an appropriate mechanism for uncertainty and time reasoning; however there is no basic tool that builds DBN using time series for a process. This paper describes the design, construction and tests for a DBNs learning tool. This tool has already been used to construct dynamic models for wind power forecast and in this paper it is used to describe the variation of the dam level caused by rainfall in a hydroelectric power plant.

In this paper, we present the conceptual model of a real-world application of factored Markov Decision Processes to dam management. The idea is to demonstrate that it is possible to efficiently automate the construction of operation policies by modelling compactly the problem as a sequential decision problem that can be easily solved using stochastic dynamic programming. We will explain the problem domain and provide an analysis of the resulting value and policy functions. We will also present a useful discussion about the issues that will appear when the conceptual model to be extended into a real-world application.

Abstract Using existing production/injection rates is convenient for analysing a mature field as the access to relevant historical records does not cause any extra cost and a large amount of significant information can be extracted from the dataset by means of good analysis methods. In the proposed work, each of the production wells is considered as an output variable while injection rates from other wells, which have an effect on the outputs, are treated as input variables. Control and System Engineering Approaches motivate such considerations. Since injection and production are time dependent variables, time delays were considered in the analysis. This consideration allows recognition of how long it takes for the injection on a certain well to affect production. The Error Reduction Ratio (ERR) algorithm, along with the Orthogonal Least Squares (OLS) method is used to determine how each of the injection wells affects production; this is achieved by sequentially orthogonalising input variables according to the order of their contribution to production. The algorithm can be customised to explain a certain variance threshold in production resulting from injection wells. For the present case study, a 19-year monthly dataset from Beatrice, an operating field in the North Sea is studied. It is observed that the historical datasets contain missing values. In order to address this problem, missing values are predicted by using an autoregressive linear model.

This paper presents the development of a novel dynamic Bayesian network (DBN) model devoted to wind forecasting. An original procedure was developed to approximate this model, based on historical information in the form of time series. The DBN structure and parameters are learned from historical data, and this methodology can be applied to any prediction problem. In contrast to previous approaches, the proposed model considers all the relevant variables in the domain and produces a probability distribution for the predictions; providing important additional information to the decision makers. The method was evaluated experimentally with real data from a wind farm in Mexico for a time horizon of 5 hours, showing superior performance to traditional time-series prediction techniques.

For many real time applications, it is important to validate the information received from the sensors before entering higher levels of reasoning. This paper presents an any time probabilistic algorithm for validating the information provided by sensors. The system consists of two Bayesian network models. The first one is a model of the dependencies between sensors and it is used to validate each sensor. It provides a list of potentially faulty sensors. To isolate the real faults, a second Bayesian network is used, which relates the potential faults with the real faults. This second model is also used to make the validation algorithm any time, by validating first the sensors that provide more information. To select the next sensor to validate, and measure the quality of the results at each stage, an entropy function is used. This function captures in a single quantity both the certainty and specificity measures of any time algorithms. Together, both models constitute a mechanism for validating sensors in an any time fashion, providing at each step the probability of correct/faulty for each sensor, and the total quality of the results. The algorithm has been tested in the validation of temperature sensors of a power plant.

Thermo-electrical power plants utilize fossil fuel oil to transform the calorific power of fuel into electric power. An optimal combustion in the boiler requires the fuel oil to be in its best conditions. One of fuel's most important properties to consider is viscosity. Viscosity has influence on the optimal combustion between fuel and air. Hardware viscosity meters for fuel oils are expensive and unreliable to operate in power plant conditions. Chemical laboratory measures viscosity accurately with special apparatus, but they cannot be used in a real time process. This paper describes the development of a virtual sensor that estimates fuel oil viscosity in the combustion process of a power plant. A virtual sensor or soft sensor is a computer program that estimates the value of a certain variable based on related measurements and a model of the process where the variable participates. In this project, a probabilistic model is constructed using automatic learning algorithms with historical data and experts' advice. The learning and validation experiments are described and discussed. The virtual sensor is installed in the Tuxpan Power Plant in Veracruz, Mexico.