Catálogo de publicaciones - libros

This article presents an algorithm that combines a FAST-based algorithm (Flexible Adaptable-Size Topology), called ARM, and Q-learning algorithm. The ARM is a self organizing architecture. Dynamically adjusting the size of sensitivity regions of each neuron and adaptively pruning one of the redundant neurons, the ARM can preserve resources (available neurons) to accommodate more categories. The Q-learning is a dynamic programming-based reinforcement learning method, in which the learned action-value function, Q, directly approximates Q*, the optimal action-value function, independent of the policy being followed. In the proposed method, the ARM acts as a cluster to categorize input vectors from the outside world. Clustered results are then sent to the Q-learning architecture in order that it learns to present the best actions to the outside world. The effect of the algorithm is shown through computer simulations of the well-known control of balancing an inverted pendulum on a cart.

The local asymptotical stability of bi-directional associative memory (BAM) neural networks with generalized saturation output functions is studied. By adopting the method of decomposing the state space to sub-regions and by using the technique of matrix norm, some delay-independent stability algebraic criteria are obtained, and the attractive domains are estimated. The results obtained in this paper need only to compute the norm of some matrices constructed by the parameters of the neural networks which are very convenient to verify in system synthesis.

This paper applies a methodology to structure the field development schemes using artificial neural network in conjunction with numerical simulation for Sirri oil field. In this method, a few field development scenarios are studied using a numerical simulator. The results of these studies are used to train the neural network. The trained neural network model is then used as a predictive tool for field development purposes.

In order to improve short term load forecasting accuracy, a novel particle swarm optimization (PSO) based support vector machine (SVM) model that combined input variable and similar day selection technique is proposed. Among so many load influential factors, rough set theory is used to select most relevant ones in order to make the input neurons representative. Next, Euclidean norm based similar day selection is used both for the forecasting day and for the days in the training set. After the preprocessing finished, PSO based support vector machine is used to establish the forecasting model. PSO is applied to train support vector machine to solve quadratic programming problem which is an effective method with better convergence and stability. The presented model is applied in certain area; and the experiment showed satisfactory results.

We present in this paper some results on the temporal segmentation and retrieval of stored memories or patterns using neural networks composed of the widely used model neurons in the neuroscience society, the bursting Hindmarsh-Rose neurons. For an input pattern which is an overlapped superposition of several stored patterns, it is shown that the proposed neuronal network model is capable of segmenting out each pattern one after another as synchronous firings of a subgroup of neurons, and if a corrupted input pattern is presented, the network is shown to be able to retrieve the perfect one, that is it has the function of associative memory. And we also notice some phenomena in our simulation that still have not been reported elsewhere in our knowledge.

In this paper, a new learning algorithm which encodes additional constraints into feedforward neural networks is proposed for function approximation problem. The algorithm incorporates two kinds of constraints into single hidden layered feedforward neural networks from a priori information of function approximation problem, which are architectural constraints and connection weight constraints. On one hand, the activation functions of the hidden neurons are a class of specific polynomial functions based on Taylor series expansions. On the other hand, the connection weight constraints are obtained from the first-order derivative information and the second-order one of the approximated function. The new algorithm has been shown by theoretical justifications and experimental results to have better generalization performance than other traditional learning ones.

The minimization of finite Automata model is deeply analyzed, and a parallel algorithm of minimization based on distinguishable state table is proposed. The parallel processing of algorithm is described in detail with an example, and its feasibility is verified.

The LBDCM is greatly affected by the uncertainties of the plant Large uncertainties may be imposed on the system and input parameters, so that the performance deviates far from the nominal design. In order to improve the performance of LBDCM, the paper presents a method of PID parameter optimization based on modified linear quadratic (LQ) optimal control. Time domain analysis and frequency domain analysis are the basic approaches in designing control system and they have close relationship to each. The simulation results verified the validity of the scheme.

The expert recommending process is highly specialized and requires specific domain knowledge as well as years of experience. Case- based reasoning (CBR), as a methodology of Artificial Intelligence, is widely applied in knowledge sharing and experience reuse area. In this paper we try to induce CBR into expert recommending system, and present a new approach to recommend expert, which is based on CBR and text mining, and describe the process of our methodology and some related algorithms. This paper uses the CBR as the main frame, and the knowledge and experience are well organized under this frame.

A lot of activities inside human body are carried out intelligently without the explicit intervention of human itself, e.g. various actions of nervous systems, blood circulation system etc. Inspired from these natural systems, autonomic computing is an emerging concept which promises to enable such kind of self-management capabilities inside software systems. Case-based reasoning (CBR) is a methodology to solve current problems using the solutions of past problems of the similar nature. In this paper, we propose to use CBR to achieve self-configuration in autonomic systems. We introduce a new similarity measure to find nearest neighbors. We have also suggested the case preparation, case retrieval and case reuse and refinement methods to enable self-configuration in autonomic systems. To support our proposed methodology, we illustrate a case-study of Autonomic Forest Fire Application.