Jia-Wei Zhang

SVM optimized scheme based PSO in application of engineering industry process

Aimed to the problem that it is hardship to get real-time and on-line measuring parameters in wood drying process, a novel PSO-SVM model that hybridized the particle swarm optimization (PSO) and support vector machines (SVM) to improve the nonlinearity caused by ambient temperature and other disturbance factors is presented. Support vector machines (SVM) based on statistical learning theory and structural risk minimization is proposed to deal with these problems. However, the model complexity and generalization performance of support vector machines (SVM) depend on a good setting of the three parameters (ε,c,γ). In this paper, the particle swarm optimization is applied to optimize the parameters (ε,c,γ) at the same time. Based on the proposed method, both PSO-SVM and SVM models are established and implemented to estimate lumber moisture content value in wood drying process. The result of comparative analysis is given. Experimental results show that solutions obtained by PSO-SVM training seem to be more robust and better generalization performance compared to SVM training.

A spatial-temporal fusion algorithm based support degree and self-adaptive weighted theory for multi-sensor

Due to the differ sensors distributing position, operation performance and some uncertainty factors effect in the industrial process, the measured parameters excursion inevitably is caused in the real world. To obtain the accurate measuring value, a spatial-temporal fusion algorithm based support degree and self-adaptive weighted theory is put forward in this paper. Considering the temporal and special domain feature, the architecture of spatial-temporal fusion modeling is built. The temporal fusion method based support degree and recursive estimation is proposed to determine consistent and reliable estimation of measured variables with setting the support degree function. The data of the n moment from the one sensor are estimated by temporal fusion method. The spatial fusion based on the adaptive weighted method is determined by Lagrange multiplier method to solving the optimal weighted factors. The simulation results show that the spatial-temporal fusion algorithm is effective. Then, the algorithm is applied for the detecting lumber moisture content in the real world. It is verified by the accuracy and reliability for the measured parameter.

Support degree and adaptive weighted spatial-temporal fusion algorithm of multisensor

According to the noise contained in many sensor data, an algorithm based on support degree and adaptive weighted spatial-temporal fusion of the multi-sensor is proposed. Spatial-temporal data fusion algorithm is to decompose data fusion into space and time, conducting the first fusion with adapting supporting degree and the recursive estimation based on time for the first time, then conducting the second fusion in space through adaptive weighted estimation. In this algorithm, there is no need to know any prior knowledge the of sensors measurement data, only minimizing variance of the fusion system all the time by using sensor variance changes in space position and adjusting weighting coefficients of each sensor fusion of weighted fusion system. The computer simulation results show further that the algorithm can improve the effectiveness of the measurement accuracy.

Time Series Forecasting for Density of Wood Growth Ring using ARIMA and Neural Networks

Wood density is one of the most important wood characteristics which determine final wood product qualities and properties. In this article, ARIMA, multilayer perceptron (MLP), and particle swarm optimization BP (PSO-BP) network models are considered along with various combinations of these models for forecasting density of wood growth ring. The forecasting principle and procedure of these three methods are presented. Measurement experiments are carried out to get the time series data of wood density. Simulation comparison of forecasting performances shows that the neural network models with particle swarm optimization give a better performance in solving the wood density forecasting problem.

Dynamic compensation method on temperature drift in Pt-resistance temperature online measuring system

The Pt-resistance temperature online measuring circuit is described in detail in the paper. Four-wire connection mode is introduced to improve effectively measurement precision of system. This paper analyses the causes of temperature drift in Pt-resistance temperature online measuring system, and puts forward a dynamic compensation method on temperature drift during the period of measuring process to solve the problems of traditional measuring methods, which are complicated to operate and has low reliability. Combined with this compensation method, a micro-controller is embedded in the online measuring system. Programs are compiled to realize automatic calibration and zero adjustment. In this method, two ordered precise standard resistors (temperature coefficient is much less than 1/spl times/10/sup -6//spl deg/C) are employed as intermediate transfers. Without measuring environmental temperature and pre-calibration, the measured variables can be compensated dynamically during measurement process. Experimental results show that the relative error of a measuring system that has adopted this dynamic compensation method is less than 0.2%.

A Novel Fusion Technique based Functional Link Artificial Neural Network for LMC Measuring

Lumber moisture content (LMC) measuring is a key industry process of wood drying. The precise of LMC will be disturbed by many ambient factors such as temperature, equilibrium moisture content, wind speed etc. Data Fusion is a novel technique to solve the coupling problem of multi-parameters. A novel fusion technique based functional link artificial neural networks (FLANN) is put forward to remove the ambient temperature disturbance. In the FLANN, functional expansion substitutes the hidden layer of multilayer perceptron (MLP). It increases the dimension of the input signal space by polynomials. Compared with MLP, FLANN exhibits a much simpler structure, less training computation and faster convergence. The calibration tests and simulation studies show that FLANN based fusion technique can eliminate effectively the disturbance from ambient factors and realize steady, real-time, high-accuracy measurement of lumber MC.

SVM for Sensor Fusion-a Comparison with Multilayer Perceptron Networks

Sensor fusion is a method of integrating signals from multiple sources. This paper investigated the possibility of using a new universal approximator: support vector machines (SVMs), as the sensor fusion architecture for the accuracy measurement and estimation of lumber moisture content in the wood drying process. The result of comparative analysis with multilayer perceptron was given. The training algorithm of MLP may be trapped in a local minimum and has a difficult task to determine the best architecture. SVM based on structural risk minimization can overcome these disadvantages. Experimental results show that the SVM performs as well as the optimal multilayer perceptron (MLP)

Comparison methods for high accuracy nano wood fiber strain measurement with the errors analysis

For measuring nano wood fiber strain, a whole range of perturbing factors on the primary converters and circuit elements degrade the metrological characteristics. An approach to the design of measuring circuit, based to the principle of double-channel parametric measurement, is proposed to improving the accuracy of the measured result. Three measuring circuits based on the principle are designed, and errors of conversion functions of them are analyzed, which shows the approach is effective to solve the problem of the measuring strain, and it is worthwhile for other measuring circuits.

Self-calibration level fusion method based on distribution diagrams and grouping estimation algorithm

Due to the original data from homogenous sensor interfered by all kinds of noise signals in the actual industry process, it is essentially to eliminate the false senor or information. Sensor fusion method allows extracting information from several different sources to integrate them into single signal or information. The architecture of multi-sensor data fusion for detecting system in the industry process is presented in this paper firstly. According to the functional characteristic of self-calibration layer for the operating homogenous sensors, the distribution diagrams and grouping estimation method is adopted without any prior information from each sensor. Numerical studies show that using distribution diagrams and grouping estimation can eliminate successfully the missing errors of multiple information acquisition. The distribution diagrams and grouping estimation method and arithmetic averaging method are investigated respectively. Comparison the simulation results, the former can supply reliable data even if single sensor or several sensors are failed, with more precise and accuracy measured value than the arithmetic averaging method. Data fusion method in the self-calibration layer can eliminate uncertainty factors effectively. Therefore, it can improve the system performance in adaptability and robust.

A double level fusion architecture based intelligence algorithms for lumber drying parameters detection system

To solve the problem that a single model can not precisely describe the global properties of the lumber moisture content (LMC) during the wood drying process, LMC measurement based multi-modeling method is presented in this paper. The method based on double layers intelligent structure which Fuzzy C-Means clustering is classification layer to classify equivalent resistance value, the inlet ambient temperature and the outlet ambient temperature data into subsets which have different cluster centers. The RBFNN and LS-SVM are modeling layers. The deg of membership is used for weighting and meaning the output of each subset to obtain the estimated LMC value as the final output. Experimental simulation results show that multi-modeling method has strong generalization ability and prefer measuring performance.