Pu Han

Wind Speed Forecasting Based on Support Vector Machine with Forecasting Error Estimation

An approach of a mean hourly wind speed forecasting in wind farm is proposed in this paper. It applies support vector regression as well as forecasting error estimation. Firstly, support vector regression is applied to the mean hourly wind speed forecasting. Secondly, a support vector classifier is trained to estimate the forecasting error. Finally, the forecasting results can tailor themselves to the estimated forecasting error, and thus improve the forecasting precision. To test the approach, three-year data from a wind farm is given as a support vector regression process, and a support vector classifier is trained in addition to estimate the forecasting error. Experimental results show that the proposed approach can achieve higher quality of mean hourly wind speed forecasting; also it has lower mean square error compared with the traditional support vector regression forecasting.

Modeling the circulating fluidized bed boiler using RBF-NN based on immune genetic algorithm

Combining the excellence of immune algorithm and Genetic Algorithm(GA), a modeling. method using parallel Radial Basis Function Neural Network(RBF-NN) is proposed based on immune GA optimization. And this method is applied to modeling of a Circulating Fluidized Bed(CFB) boiler so as to realize neural network modeling of CFB boiler. The established neural network model is very useful for both characteristic research and advanced control strategy development of CFB boiler.

SMO Algorithm Applied in Time Series Model Building and Forecast

As a novel learning machine, the support vector machine (SVM) based on statistical learning theory can be used for regression: support vector regression (SVR). SVR has been applied successfully to time-series analysis, but its optimization algorithm is usually built up from certain quadratic programming (QP) packages. Therefore, for small datasets this is practical and QP routines are the best choice, but for large datasets, data processing runtimes become lengthy, which limits its application. Sequential minimal optimization (SMO) algorithm can improve operation speed and reduce this long runtime. In this paper, SVR that is based on the SMO algorithm is used to forecast two typical time series models: Wolfer sunspot number data and Box and Jenkins gas furnace data. The results of simulation prove that the operational speed of SVR using the SMO algorithm is improved in comparison to SVR employing QP optimization algorithm; moreover, the forecasting precision is better than that of neural network and SVR using QP optimization algorithm.

Complete Compensation for Time Delay in Networked Control System Based on GPC and BP Neural Network

A new framework is proposed to cope with the uncertain time delay of networked control system. Event-clock-driven controller nodes, together with clock-driven sensor nodes and actuator nodes are required in this framework. Queuing Strategy is introduced both in controller nodes and actuator nodes while the time delay between controller node and actuator node is compensated by multi-step control increment given by the algorithm of General Predictive Control. An output error prediction model is built using BP neural network to deal with the time delay between sensor node and controller node. The principle of this model is to revise the predictive output of general predictive control model using predictive error signal; if the value of time delay exceeds the upper limit, controller nodes will immediately produce the control strategies adopting the revised predictive output, and thus the compensation for time delay between sensor nodes and controller nodes would be accomplished. Simulation experiments are practiced over Ethernet network which embraces both kinds of time delay. It is proved that the scheme of complete compensation remains a good control performance.

Boiler Flame Image Classification Based on Hidden Markov Model

The classification is an important domain in boiler flame image processing and is a preliminary step toward detection, recognition and understanding of combustion condition. In this paper, a hidden Markov model (HMM) approach is introduced into boiler flame image classification. Firstly, we define a feature vector for each flame image including 5 feature elements, which are the brightness of flame, the area of the high temperature flame, the brightness of high temperature flame, the rate of area of the high temperature flame, the offset of the flame centroid respectively. Next, for classification and recognition of the flame image, a method of the maximum posterior marginal (MPM) based on the hidden Markov random field model, which is described as a probabilistic framework for learning probability distribution defined on the sample space, is introduced. Then, we construct a sample space including 63 flame images, parts of which are used to train the model. Finally, the entire samples are recognized and classified. Experiments prove this method is effective for classification of boiler flame images

Selective ensemble using discrete differential evolution algorithm for short-term load forecasting

In order to further improve the accuracy of short-term load forecasting, a selective neural network ensemble method using discrete differential evolution algorithm is proposed. Firstly, the individual vectors in differential evolution algorithm are dispersed. Secondly, a group of RBF neural networks with larger difference are trained independently and a binary bit string in multi-dimensional space with the value of 0 or 1 is used to describe all the possible neural network integrations. Lastly, part of individual networks is optimized selected to ensemble and an entropy method is used to determine the integrated weighted coefficient of component neural networks according to the variability of prediction error sequences. The experiments show that the proposed approach has higher accuracy and stability.

Wind speed conformal prediction in wind farm based on algorithmic randomness theory

An approach of a mean hourly wind speed conformal prediction in wind farm is proposed. Conformal prediction is a new prediction methodology. It can be used not just to make predictions but also to estimate the confidence under the usual independent and identically distributed assumption. Based on support vector regression, wind speed regions are predicted by inductive confidence machine. Wind speed regionpsilas width and confidence reflect the accuracy and reliability of the prediction. Compared to bare wind speed forecasting, the accuracy and reliability of the prediction can be used to reduce the risk of decision-making. Experimental results are given by using wind mean hourly speed measured in wind farm, and the application of the method carried out a detailed analysis and verification.

Extraction of Characteristic Parameters of Furnace Flame Based on Markov Model

The combustion of pulverized coal in furnace is a kind of complex and unstable suspension burning process. Obtaining more accurate characteristic parameters is crucial to detection of the flame, which is important to control combustion conditions, maintain economical operation and safeguard security. In this paper, first, we define the concept of the characteristic region in flame image and the characteristic parameters of the characteristic region. These characteristic parameters include the size of characteristic region; the average grey-level of characteristic region; the lessening rate of the characteristic region size and the flicker signal of the flame. Next, an algorithm of mean field approximation annealing (MFAA) based on compound Gauss-Markov random field (CGMRF) model is introduced to extract the characteristic region and these characteristic parameters. The experimentation to the sample images proves that these parameters are available to identify combustion state and this algorithm is effective to extract theses parameters.

Identification of Thermal Process Using Hammerstein Model Based on Particle Swarm Optimization Algorithm

In order to identify the controlled objects which are nonlinear time-delay processes with slow time-varying in the thermal system accurately, the Hammerstein model and particle swarm optimization (PSO) algorithm were used in this paper. For the Hammerstein model discussed in this paper, the polynomial and difference equations were used to express the nonlinear part and linear part of Hammerstein model, respectively. This study used the PSO algorithm to find the optimal solution of Hammerstein model’s undetermined parameters in the parameters space. For illustration, an example of main-steam temperature system identification was utilized to show the feasibility of the Hammerstein model based on PSO algorithm in identifying the thermal system processes. The PSO-based Hammerstein model can effectively represent the controlled objects which are nonlinear time-delay processes in the thermal system and thus a class of identification problems with nonlinearity in thermal system can be solved.

Optimization of controllers in the thermal system using initial pheromone distribution in Ant Colony Optimization

Ant Colony Optimization (ACO) , an intelligent swarm algorithm, proves effective in various fields. However, the choice of the first route and the initial distribution of pheromone are among the toughest yet most crucial factors in determining the performance of process optimization. According to the materials we referred to, almost all the existing methods of ACO set the same constant in all routes as the initial pheromone. However, in that case, the searching process might be misleading, or stick into local optimal values. In this article, a new method is proposed to optimize the parameter searching process in thermal objects particularly, implementing initial pheromone distribution according to a set of formulas concluded from many observances and practical tests. We used MATLAB as the program design platform. The experiment showed that this method is satisfactory. Moreover, it can be applied in other intelligent algorithms such as Genetic Algorithm, which is also in demand of setting initial parameters and range of values.