Intelligent Control and Optimal Operation of Complex Electric Power Systems Using Hierarchical Neural Networks

This chapter is aimed at developing a unified neural network based framework that can be utilized in prediction and control of complex dynamic system behaviors. In particular, in power systems, accurate prediction of system load behavior provides vital information to allow for optimal planning and most economic operation of power systems; on the other hand, the real-time system stability must be maintained against various random factors, disturbances and contingencies. The hierarchical neural networks are studied in depth in the context of prediction, optimization and control; and unified design techniques are developed for providing control robustness, optimality and prediction accuracy as well. The unified methodology builds upon hierarchical neural networks, and may be utilized and extended for other practical applications.

Stochastic Methods for Hard Optimization

This chapter aims at solving difficult optimization problems arising in many engineering areas. To this end, a brief review of the main stochastic methods which can be used for solving continuous non-convex constrained optimization problems is presented i.e.: Simulated annealing (SA), Genetic algorithm (GA), and Particle swarm optimization (PSO). In addition to that, we will present a recently developed optimization method called Heuristic Kalman Algorithm (HKA) which seems to be, in some cases, an interesting alternative to the conventional approaches. The performance of these methods depends dramatically on the feasible search domain used to find out a solution as well as the initialization of the various user defined parameters. From this point of view, some practical indications concerning these issues will be given. Another objective of this chapter is to show that the stochastic methods, notably HKA, can be efficiently used to solve robust synthesis problems in the area of structured control and fault diagnosis systems. More precisely, we will deal with the following problems: the synthesis of a robust controller with a given fixed structure and the design of a robust residual generator. Some numerical experiments exemplify the resolution of this kind of problems.

An Adaptive Framework for Power Components Dynamic Loadability

In electrical competitive markets, where deregulation and privatisation have determined changes in the organizational structures of the electricity supply industry as well as in the operation of power systems, utilities necessitate to change dynamically the loadability rating of power components without penalizing their serviceability. When assessing network load capability, the prediction of the Hot Spot Temperature (HST) of power components represents the most critical factor since it is essential to assess the thermal stress of the components, the loss of insulation life and the consequent risks of both technical and economical nature. In this chapter a general adaptive framework for power components dynamic loadability is proposed. In order to estimate the effectiveness of the adaptive framework, based on grey-box modelling, a specific case study, concerning the problem of forecasting the HST of a mineral-oil-immersed transformer, is presented.

Technical Analysis and Implementation Cost Assessment of Sigma-Point Kalman Filtering and Particle Filtering in Autonomous Navigation Systems

The chapter provides technical analysis and implementation cost assessment of Sigma-Point Kalman Filtering and Particle Filtering in autonomous navigation systems. As a case study, the sensor fusion-based navigation of an unmanned aerial vehicle (UAV) is examined. The UAV tracks a desirable flight trajectory by fusing measurements coming from its Inertial Measurement Unit (IMU) and measurements which are received from a satellite or ground-based positioning system (e.g. GPS or radar). The estimation of the UAV’s state vector is performed with the use of (i) Sigma-Point Kalman Filtering (SPKF), (ii) Particle Filtering (PF). Trajectory tracking is succeeded by a nonlinear controller which is derived according to flatness-based control theory and which uses the UAV’s state vector estimated through filtering. The performance of the autonomous navigation system which is based on the aforementioned state estimation methods is evaluated through simulation tests. Implementation cost assessment shows that PF requires more sample points than SPKF to approximate the state distribution. Therefore PF is a computationally more demanding method which needs more costly computing machines. However, the PF is a nonparametric filter which can be applied to any kind of state distribution, while the SPKF state estimators are still based on the assumption of a Gaussian process and measurement noise.

Fault Detection, Isolation and Characterisation for Discrete Event Systems Based on Petri Nets Models

Petri net models are used to detect and isolate faults in case of discrete event systems as manufacturing, robotic, communication and transportation systems. This chapter addresses two problems. The first one is the structure designs and parameters identification of the Petri net models according to the observation and analysis of the sequences of events that are collected. Deterministic and stochastic time Petri nets are concerned. The proposed method is based on a statistical analysis of data and has a practical interest as long as sequences of events are already saved by supervision systems. The second problem concerns the use of the resulting Petri net models to detect, isolate and characterize faults in discrete event systems. This contribution includes the characterization of intermittent faults. This issue is important because faults are often progressive from intermittent to definitive and early faults detection and isolation improve productivity and save money and resources.

Identification of Linear Time-Varying Systems

This chapter deals with identification of time-varying systems using Kalman filter approach. Most physical systems exhibit some degree of time-varying behaviour for many reasons. These systems cannot effectively be modelled using time invariant models. A time-varying autoregressive with exogenous input (TVARX) model is good to model these time-varying systems. The Kalman filter approach is a superior way to estimate the system parameters. This approach can track the time-varying parameters and is suitable for recursive estimation. It works well even when there are abrupt changes in the system parameters. Kalman filter is known to be an optimal estimator even when there is significant noise. In the proposed approach, for the purpose of simulation, we employ first order TVARX model and its parameters are estimated using recursive Kalman filter method. The system parameters are varied in continuous and abruptly changing manner to reveal the physical situation. To show the efficacy of the proposed approach, the time-varying parameters are estimated for different noise conditions. The performance is evaluated by calculating error performance measures. The results are found to be satisfactory with reasonable accuracy for noisy conditions even for fast changing parameters. The numerical examples illustrate efficacy of the proposed Kalman filter based approach for identification of time-varying systems.

Component Models Based Approach for Failure Diagnosis of Discrete Event Systems

This chapter addresses the problem of diagnosing Discrete Event Systems (DESs), specifically manufacturing systems with discrete sensors and actuators. Manufacturing systems are generally composed of several components which can evolve with the course of time (new components, new technologies …). Their diagnosis requires the computation of a global model of the system. This is not realistic due to the great number of components. In this chapter, we propose to perform the diagnosis by using component models. Each component model is constructed using different information sources represented by sensor-actuator spatial structure (plant model), controller specifications (desired behaviour) and temporal information about the actuators reactivity. In addition, components’ technological constraints and characteristics are considered for this construction. For each model, a local diagnoser is computed. Its complexity is polynomial because the diagnosis is computed only for the faults that it can diagnose. Limited information about the global system functioning is required to synchronize the functioning of local diagnosers. This synchronisation is considered using a set of expert rules representing the symbolic information about the global desired behaviour. The local diagnosers are then used to perform diagnosis online. They validate, in the case of normal functioning, the transmission of control signals and incoming sensor data between the controller and the plant.

An Advanced IDE for Designing Transparent Fuzzy Agents

Recently, computational agents received significant attention in computer science research community. In fact, intelligent agents methodology is a powerful artificial intelligence technology showing considerable promise as a new paradigm for mainstream software development and able to offer new ways of abstraction, decomposition, and organization that fit well with our natural view of the world. However, despite their promise, intelligent agents are still scarce in the market place. A key reason for this is that developing intelligent agent software requires significant training and skill. Artificial Intelligence methodologies and computer networking tools represent the necessary basic knowledge to design and implement advanced agents oriented systems. This paper introduces an integrated development environment supporting the agents developers to design fuzzy-based agents in a simple and fast way. Proposed framework has been realized by integration of theoretical methodologies as fuzzy logic and labeled tree, together with Open Source Software tools as JaxMe2.

Monitoring of Non Stationary Systems Using Dynamic Pattern Recognition

The monitoring of non stationary systems permit to follow online the evolutions and changes which occur in the course of time. In Pattern Recognition (PR) the functioning modes are represented by a set of similar patterns, called classes. These patterns are obtained by observation of the most informative parameters of the system. To realize the monitoring of a system functioning PR methods uses a classifier which determines at each instant the class of a new incoming pattern. In this paper, we propose to develop the classification method Incremental Fuzzy Pattern Matching (IFPM) to be operant in the case of dynamic classes and to be used for the online monitoring of evolving systems. IFPM gives good results for static classes and its classification time is constant according to the size of the database. However, with non stationary systems, the classifier parameters must be adapted in order to take into account the temporal changes of classes’ characteristics. These temporal changes can be represented for example by a displacement, a rotation, a splitting, or a fusion of classes. Therefore, the classification method must be able to forget the information which is no more representative of classes and it must adapt its parameters based only on the recent and useful information. This development is based on the use of an incremental algorithm allowing to follow the accumulated gradual changes of classes’ characteristics after the classification of each new pattern. When these changes reach a suitable predefined threshold, the classifier parameters are adapted online using the recent and useful patterns. The developed method is applied on several simulations and on a two tanks benchmark.

Robust Fault Detection Based on State Observers for Networked Control Systems

With the development of network and control technology, networked control systems (NCS) have been widely studied recently, especially in the area of complex industrial systems. The system model and states observer based approach is very important for the fault detection (FD) and diagnosis of NCS. This chapter focuses on robust fault detection methods based on states observer. States observers on NCS with short time-delay and uncertain time-delay are both discussed and designed without changing the structure of the systems. The corresponding theorems are systematically given and proved. The methods of robust fault detection on NCS are proposed and some typical examples are demonstrated to test the presented methods.