Today, artificial neuralnetworks (ANN) are used in various areas of application, in order to reproduce interdependencies between input and output information, similar to the information processing of the human brain....
详细信息
Today, artificial neuralnetworks (ANN) are used in various areas of application, in order to reproduce interdependencies between input and output information, similar to the information processing of the human brain. Especially where analytic modelling and calculation is impossible, or in case of unacceptable calculation times, they present a promising way of problem solving. In this project, the evaluation of the thermal capacity of energy cable, with load curves as input data, is discussed by using an ANN-approach. The presented work deals with the calculation of the temperature during normal and faulty operation. The configuration of the ANN is chosen according to this problems and in addition, the input data, especially for training purposes, is analysed. The results underling sufficient exactness for load limit calculations concerning the thermal behaviour of energy cables.
The learning algorithms and structures which have become popular in the neural network community over the last few years have been successfully applied to various challenging modelling problems. In contrast to the lin...
详细信息
The learning algorithms and structures which have become popular in the neural network community over the last few years have been successfully applied to various challenging modelling problems. In contrast to the linear modelling methods, there is still no clearly defined engineering process from which a model can reliably be created from measurements taken from a physical system. Problems for the practical application of neuralnetworks involve the interpretation of the trained models, and the explicit introduction of a priori models into the learning system, as well as the use, in many cases, only basic ad hoc validation and experiment design techniques. This talk will discuss several methods which can improve the engineering aspects of model identification with neural nets.< >
The ability to learn is the cornerstone of current neural technology. It was responsible for their decline in the late sixties when it was shown that the perceptron training algorithm could not be easily extended to m...
详细信息
The ability to learn is the cornerstone of current neural technology. It was responsible for their decline in the late sixties when it was shown that the perceptron training algorithm could not be easily extended to multilayered networks, and also the revival of interest in these techniques was initiated by the discovery of a multilayer network adaptation rule. Artificial neuralnetworks attempt to mimic a human's information processing capabilities by building neuronally inspired systems which learn to interact with their environment in a desirable fashion. There are many engineering problems which could benefit from the use such of such systems, although there are also problems with applying these adaptive networks. Most artificial neuralnetworks can be regarded as "black box" learning systems; they are difficult initialise as knowledge is stored in an opaque fashion and validation can only performed using input/output data; their internal structure provides little information to an engineer. In addition, the generalisation, modelling and learning abilities of these networks are generally poorly understood, although such results are necessary when these adaptive systems are applied online in safety critical situations.< >
In this paper we prove the stability of a certain class of nonlinear discrete MIMO systemscontrolled by a multilayer neural net with a simple weight adaptation strategy. The proof is based on the Lyapunov formalism. ...
详细信息
In this paper we prove the stability of a certain class of nonlinear discrete MIMO systemscontrolled by a multilayer neural net with a simple weight adaptation strategy. The proof is based on the Lyapunov formalism. The stability statement is, however, only valid if the initial weight values are not too far from their optimal values that allow perfect model matching. We therefore propose to initialize the weights with values that solve the linear problem. This extends our previous work (Renders, 1993; Saerens, Renders and Bersini, 1993), where SISO systems were considered.< >
Batch processes may constantly be changing, sometimes in unpredictable ways. For this reason, the operators need to keep a close watch on things. Batch-plant control stations have traditionally been located very close...
详细信息
Batch processes may constantly be changing, sometimes in unpredictable ways. For this reason, the operators need to keep a close watch on things. Batch-plant control stations have traditionally been located very close to the process, because operators can do a better job of sensing abnormalities if they can see, hear and smell what is going on. The reaction process can be considered as unstable, the instability being understood as the characteristic making the self abandoned system evolve towards unpermitted temperature values with the associated danger of accidents. The temperature also plays a decisive role in the final product quality, each time the reactor changes from the desired temperature the quality diminishes, the appearance of undesired species increases, and the overall performance of the reaction decreases which directly affects the economic performance.< >
Many authors have shown by simulated studies, that a great number of non-linear dynamical systems could be identified and controlled by using neural network models. The authors applied these results to a real process ...
详细信息
Many authors have shown by simulated studies, that a great number of non-linear dynamical systems could be identified and controlled by using neural network models. The authors applied these results to a real process : an oven with two inputs, one for heating and one for cooling. The output to be controlled is the temperature inside the oven. The choice of the control strategy on the one hand and the choice of the neural network architecture for the oven identification on the other hand had to satisfy two main objectives. First, the control strategy should be quite insensible to random disturbances (air leaks, door openings, ...) and the neural model should be able to fit any modification of the plant dynamics during its lifetime (modification of the internal load alteration of the heating system ...). The authors chose to use the internal model control as their control strategy. They also used a radial basis function network to identify the plant. The process identification is composed of two phases, an off-line one and an online one. The off-line part consists first in training the network while determining its internal structure using an initial training set. The on-line phase is the adaptive part of the control scheme.< >
Traditionally, power system control and management functions have been performed in centralised locations, with unprocessed data being collected from several measuring points throughout the power system and returned t...
详细信息
Traditionally, power system control and management functions have been performed in centralised locations, with unprocessed data being collected from several measuring points throughout the power system and returned to the central point for analysis. Whilst many technology advances have been made in schemes such as this, including the introduction of expert systems, neuralnetworks, and parallel computing methods at power system control centres, the potential now exists to consider distribution centres, the potential now exists to consider distributing this information technology throughout the power system by realising the concept of intelligent substations. This paper will explore the possibilities for a distributed control and management system, with major transmission substations performing tasks such as alarm processing, fault diagnosis and conditions monitoring on a local basis;that is, accepting data from local and adjacent substations, processing this data, and sending concise and summarised messages to the control centre (with the potential for localised executive action in some instances). The advantages of such an arrangement will be demonstrated, including: reduction in SCADA system load, especially during critical periods;increased local autonomy, thus facilitating substation automation;faster response times due to the distributed nature of the processing throughout the power system. Examples of ongoing research into the realisation of such a system will be given, showing test results from package already developed, using data provided by utilities and manufactures. Problems with the implementation of such systems will also be covered, and ideas for the future solutions to these problems will be suggested.
One of the important problems to be solved for neural network applications is to find a suitable network structure solving the given task. To reduce the engineering efforts for the architecture design a data driven al...
详细信息
One of the important problems to be solved for neural network applications is to find a suitable network structure solving the given task. To reduce the engineering efforts for the architecture design a data driven algorithm is desirable which constructs a network structure during the learning process. There are different approaches for structure adaptation with evolutionary algorithms, growth algorithms and others. To solve large problems successfully it is necessary to divide the problem into subproblems and to solve them separately by experts. This is a fundamental principle of nature. To implement this principle in artificial neuralnetworks there are different approaches, but these algorithms yield fixed network structures. The authors propose a learning architecture for growing complex artificial neuralnetworks which tries to include both sides of the coin, structure adaptation and task decomposition. The growing process is controlled by self-observation or reflexion. The algorithm generates a feedforward network bottom up by cyclically inserting cascaded hidden layers. Inputs of a hidden layer unit are locally restricted with respect to the input space by using a new kind of activation function, combining the local characteristics of radial basis function units with sigmoid units. Contrary to the cascade-correlation learning architecture the authors introduce different correlation measures to train the network units featuring different goals. The task decomposition between subnetworks is done by maximizing the anticorrelation between the hidden layer units output and a connection routing algorithm which only connects cooperative units of different layers. These features resemble the TACOMA (TAsk decomposition, COrrelation Measures and local Attention neurons) learning architecture. Self-observation is done by transforming the errors and the network structure to the input space. So it is possible to infer from errors to structure and reverse.< >
暂无评论