Catálogo de publicaciones - libros

The purpose of this study was to develop a process management system to manage ingot fabrication and the quality of the ingot. The ingot is the first manufactured material of wafers. Trace parameters were collected on-line but measurement parameters were measured by sampling inspection. The quality parameters were applied to evaluate the quality. Therefore, preprocessing was necessary to extract useful information from the quality data. First, statistical methods were used for data generation, and then modeling was performed, using the generated data, to improve the performance of the models. The function of the models is to predict the quality corresponding to control parameters.

In this paper, the stable personal browsing patterns shown in Internet surfing are utilized to determine the users’ preference on specific content. To be more specific, they are used to calculate the so called implicit ratings. We performed an experiment on all possible combinations of the implicit indicators to pick out the most significant indicators— elements of user browsing patterns. A thorough analysis and comparison are carried out before four indicators are selected as the input of an Artificial Neural Network which is adopted to calculate the implicit ratings. The mechanism of the implicit rating calculation is integrated into an educational resource sharing system as a featured module and works well.

Source grey GM(1,1) model usually be used simulation and prediction of equidistant monitoring data sequent. But to non-equidistant and high growth data sequent, had to build the grey GM(1,1) model through equidistant treatment of non-equidistant data or to build directly non-equidistant grey model through complex transfermation , and usually had larger lagging error. In time sequent [ k , k +1] interval, in order to majorize and increase accuracy of background value z ^(1) ( k +1), the area of [ k , k +1] interval and GM(1,1) function curve envelope had been replaced by n small interval trapezoidal area.The GM(1,1) grey majorized model was built based on majorized grey model background value generally be used simulation and prediction of equidistant or non-equidistant and low or high growth data sequent of surrounding rock displacement in tunnel. Data sequent characters of I,II and III shape of surrounding rock displacement can be simulated and predicted better by the grey majorized model, and the model had higher simulation and prediction accuracy.

The discrete-time system of multilayer composite plate is modeled using neural network (NN) to produce a nonlinear exogenous autoregressive moving-average model (NARMAX). The model is implemented by training a NN with input-output experimental data. Each damaged sample can be modeled by a parameter governed by the propagation behaviors of the NN. A residual signal is evaluated from the difference between the output of the model and that of the real system. A threshold function is used to detect the damaged behavior of the system. The results show that a three-layer neural network can be a general type of and suitable for the nonlinear input-output mapping problems of multilayer composite system.

Support Vector Machine (SVM) is a precise and fast method for the prediction of short-term electrical load and the similar day method is a simple and direct method for load forecasting. This paper tries to combine SVM model and similar day method for next day load forecasting. The proposed method forecasts the load of next day using SVM. Then, the load curve of a similar day is selected to correct the curve forecasted by SVM, which can avoid the appearance of large forecasting error effectively. Corresponding software was developed and used to forecast the next day load in a practical power system, and the final forecasting result is accurate and reliable.

Particle swarm optimization algorithm (PSO) is applied to train artificial neural network (NN) to construct a neural network based on particle swarm optimization algorithm (PSONN). Then, PSONN is employed to construct a practical soft-sensor of gasoline endpoint of main fractionator of fluid catalytic cracking unit (FCCU). The obtained results indicate that soft-sensing model based on PSONN has better performance than soft-sensing model based on BPNN and the new method proposed by this paper is feasible and effective in soft-sensing modeling of gasoline endpoint.

Very Fast Simulated Reannealing (VFSR) is applied to the solution of an inverse electromagnetic problem. The problem is to model the distribution of dipole current sources in a finitely resistive infinite half space. Modelling is done using the observed electrical potential on the interface between the half space and the free space. This method is known as Spontaneous Potential (SP) and is widely used for geophysical prospecting. Object of the VFSR algorithm is a quadratic error function between measured and synthetic data. Method is tested on a field data.

We introduce a new method to improve web site text content by identifying the most relevant free text in the web pages. In order to understand the variations in web page text, we collect pages during a period. The page text content is then transformed into a feature vector and is used as input of a clustering algorithm (SOFM), which groups the vectors by common text content. In each cluster, a centroid and its neighbor vectors are extracted. Then using a reverse clustering analysis, the pages represented by each vector are reviewed in order to find the similar. Furthermore, the proposed method was tested in a real web site, proving the effectiveness of this approach.

Conventional Support Vector Regression (SVR) is not capable of online setting and its training algorithm is inefficient in real-time applications. Through analyzing the possible variation of support vector sets after new samples are added to the training set, and extending the incremental support vector machine for classification, an online learning algorithm for SVR is proposed. To illustrate the favorable performance of the online learning algorithm, a nonlinear system identification experiment is considered. The simulation results indicate that the learning efficiency and prediction accuracy of the online learning algorithm are higher than that of the existing algorithms, and it is more suitable for system identification.

In this paper the optimization of Purified Terephthalic Acid (PTA) crystal crystallizer based on FGMDH networks and Adaptive Differential Evolutionary (ADE) algorithm is discussed in detail. Due to the existence of many by-products and impurity in PTA continuous industry production process, it is very difficult to build mechanism models for this process. Since Artificial Neural networks have been proved to be able to approximate a wide class of functional relationships very well in modeling chemical process, we apply a kind of FGMDH networks to build PTA granularity model, which is incorporated with human experiences. To implement the control of PTA granularity, which is one of the key product quality indexes, a kind of global real-value optimization algorithm -— ADE algorithm is proposed for optimizing of PTA crystallization process. The proposed ADE is capable of find the optimal operation conditions effectively and efficiently and suitable for industrial application.