Super-heater steam temperature in power plant is the strong nonlinearity system. Though neural networks have the ability to approximate nonlinear functions with arbitrary accuracy, good generalization results are obta...
详细信息
ISBN:
(纸本)1424400600
Super-heater steam temperature in power plant is the strong nonlinearity system. Though neural networks have the ability to approximate nonlinear functions with arbitrary accuracy, good generalization results are obtained only if the structure of the network is suitably chosen. Therefore, selecting the "best" structure of the neural network is more difficulty. Sparse Least squares support vector networks (SLSVN) are proposed to model the superheated steam of power plant in this paper. The structure of the SLSVN is obtained by equality-constrained minimization. Under the condition of modeling approximating to performance, the pruning algorithm gets the sparse modeling. The merits-of the algorithm are conforming to the least structural risk in training process and hardly leading to over-fitting. The simulation of a superheating system, in a 600MW supercritical concurrent boiler, is taken. The result shows that the proposed SLSVN model can adapt to the strong nonlinear super-heater steam temperature process.
In recent years, it has been recognized that the there is a gapbetween artificial neural network and the brain neural network. The biological neural network is neither a random network nor a regular network, it is a n...
详细信息
ISBN:
(纸本)9781538630228
In recent years, it has been recognized that the there is a gapbetween artificial neural network and the brain neural network. The biological neural network is neither a random network nor a regular network, it is a network structure between the two. While the small world network has both a larger clustering coefficient of a regular network and a smaller average path length of a random network [1], so the superiority of the small world network has aroused people's attention. As the BP algorithm converges slowly in the process of error back propagation, it is easy to fall into the local minimum point in the modified weight stage, so this paper Optimizes the BP algorithm to improve the convergence rate of the network and improve the problem that the network is easy to fall into the local minimal. The connection between nodes is too close because there are too many hidden nodes, as a result, the problem of overfitting is arisen. In other words, as for data that are not in the training sample, the learning ability of the network is not strong, resulting in a decrease in the practical value of the network. So we need to find a suitable network structure [2]. Therefore, this paper proposes a e-exponential information entropy multilayer feedforward small world network pruning algorithm based on Shannon entropy principle. By constantly training the network, the network is trimmed according to the entropy of the hidden nodes until the network tends to be stable. The experimental results show that the it has been improved obviously in terms of calculated error and test accuracy for trimmed network, which improved the problem of over-fitting to some extent.
The railway switch failure prediction for railway signal equipment maintenance plays an important role. The paper put forward railway switch failure prediction algorithm based on least squares support vector machine, ...
详细信息
ISBN:
(纸本)9783038352105
The railway switch failure prediction for railway signal equipment maintenance plays an important role. The paper put forward railway switch failure prediction algorithm based on least squares support vector machine, and chose five characteristic indexes composed of railway switch failure prediction models characteristic input vectors. It reduces the dimension of input vectors, shorten the least squares support vector machine training time, and use a pruning algorithm to accelerate the computing speed maintaining a good regression performance at the same time. The experiment proved that railway switch failure prediction algorithm has strong self-learning ability and higher prediction accuracy based on least squares support vector machine. And it can accelerate the speed of switch failure prediction and improve the accuracy and reliability of railway switch failure prediction.
In this paper, an Efficient Adaptive Fuzzy Neural Network (EAFNN) model is proposed for electric load forecasting. The proposed approach is based on an ellipsoidal basis function (EBF) neural network, which is functio...
详细信息
ISBN:
(纸本)9783642132070
In this paper, an Efficient Adaptive Fuzzy Neural Network (EAFNN) model is proposed for electric load forecasting. The proposed approach is based on an ellipsoidal basis function (EBF) neural network, which is functionally equivalent to the TSK model-based fuzzy system. EAFNN uses the combined pruning algorithm where both Error Reduction Ratio (ERR) method and a modified Optimal Brain Surgeon (OBS) technology are used to remove the unneeded hidden units. It can not only reduce the complexity of the network but also accelerate the learning speed. The proposed EAFNN method is tested on the actual electrical load data from well-known EUNITE competition data. Results show the proposed approach provides the superior forecasting accuracy when applying in the real data.
Real-time detection and tracking of vehicles is very important in the field of automatic *** this paper,a YOLOv3 network based on pruning algorithm is proposed to solve the problem of real-time vehicle *** reducing th...
详细信息
Real-time detection and tracking of vehicles is very important in the field of automatic *** this paper,a YOLOv3 network based on pruning algorithm is proposed to solve the problem of real-time vehicle *** reducing the number of channels and layers in the backbone network,the computation of the model is *** thus,the rate of detection is greatly *** the basis of vehicle detection,the real-time tracking of multiple vehicle targets is completed by using Kalman filter algorithm for prediction and Hungarian algorithm for data *** experimental results show that compared with the original YOLOv3 network,the model size is compressed by 95% to11.25 MB and the detection rate is doubled to 128.1 frames/s while the average accuracy is basically *** detection and tracking frame rate of the whole algorithm is 18 fps,and the recall is 98.5%.The algorithm also has strong robustness on complex traffic roads,and can basically realize real-time detection and tracking of road vehicles.
Convolutional neural network (CNN) training is computationally intensive, requiring a great deal of time and resources. Exploiting data sparsity is a promising method to accelerate CNN training. In this work, we propo...
详细信息
ISBN:
(数字)9781665409964
ISBN:
(纸本)9781665409964
Convolutional neural network (CNN) training is computationally intensive, requiring a great deal of time and resources. Exploiting data sparsity is a promising method to accelerate CNN training. In this work, we propose a novel algorithm for sparse training processes in which the weight matrices are pruned in a fine-grained block-wise manner. Both the forward propagation (FP) and backward propagation (BP) phases use the identical data layout. It can eliminate the matrix transposition procedure, reducing storage space and training time. Based on this pruning approach, we developed an FPGA-based accelerator for CNN training using a systolic array. The architecture can effectively skip the zero values calculation without incurring the imbalance between different processing elements (PEs). Our experimental results indicate that our design achieves 1.024 TOPS and 118.4 GOPS/VV in terms of computational throughput and energy efficiency. Our design is 1.41x similar to 4.93x more energy efficient than the state-of-the-art training accelerator.
Background: Examining the distribution of variation has proven an extremely profitable technique in the effort to identify sequences of biological significance. Most approaches in the field, however, evaluate only the...
详细信息
Background: Examining the distribution of variation has proven an extremely profitable technique in the effort to identify sequences of biological significance. Most approaches in the field, however, evaluate only the conserved portions of sequences - ignoring the biological significance of sequence differences. A suite of sophisticated likelihood based statistical models from the field of molecular evolution provides the basis for extracting the information from the full distribution of sequence variation. The number of different problems to which phylogeny-based maximum likelihood calculations can be applied is extensive. Available software packages that can perform likelihood calculations suffer from a lack of flexibility and scalability, or employ error-prone approaches to model parameterisation. Results: Here we describe the implementation of PyEvolve, a toolkit for the application of existing, and development of new, statistical methods for molecular evolution. We present the object architecture and design schema of PyEvolve, which includes an adaptable multi-level parallelisation schema. The approach for defining new methods is illustrated by implementing a novel dinucleotide model of substitution that includes a parameter for mutation of methylated CpG's, which required 8 lines of standard Python code to define. Benchmarking was performed using either a dinucleotide or codon substitution model applied to an alignment of BRCA1 sequences from 20 mammals, or a 10 species subset. Up to five-fold parallel performance gains over serial were recorded. Compared to leading alternative software, PyEvolve exhibited significantly better real world performance for parameter rich models with a large data set, reducing the time required for optimisation from similar to10 days to similar to6 hours. Conclusion: PyEvolve provides flexible functionality that can be used either for statistical modelling of molecular evolution, or the development of new methods in the field. The
This research work proposes a fuzzy neural network (FNN) for pattern classification. The proposed network is the modified version of the Radial basis function neural network (RBFNN). FNN uses supervised fuzzy clusteri...
详细信息
This research work proposes a fuzzy neural network (FNN) for pattern classification. The proposed network is the modified version of the Radial basis function neural network (RBFNN). FNN uses supervised fuzzy clustering and pruning algorithm to determine the precise number of clusters with proper centroid and width to form the processing nodes in the hidden layer. These clusters represent fuzzy set hyperspheres (FSHs), which are defined by the fuzzy membership function. The training between the hidden layer to output layer which is done by using the LMS algorithm in RBFNN is avoided, and the output is determined by using the fuzzy union operation. The fuzzy membership function shields the clustered patterns resulting in 100% accuracy for the data set used during training. Unlike other clustering algorithms used to construct the hidden layer of RBFNN, the proposed clustering algorithm is independent of tuning parameters and is fast in training and retrieval. Thus FNN reduces the computation time, guarantees 100% accuracy for any training set, and provides superior and comparable recognition accuracy for the datasets with the precise number of FSHs in the hidden layer. Hence the proposed FNN can be used for pattern classification.
The thinning algorithm is one of the fastest approaches to extract skeletons from an object, especially when adopting the parallel strategy. Skeletons are very useful descriptors and can be applied in many recognition...
详细信息
The thinning algorithm is one of the fastest approaches to extract skeletons from an object, especially when adopting the parallel strategy. Skeletons are very useful descriptors and can be applied in many recognition fields. However, one of the drawbacks that limits the use of these techniques is that thinning algorithms are not robust against inner noise and outer noise, which may produce many unwanted branches. To alleviate the influence of noise and increase the robustness, pruning methods and scale-space methods have been proposed in the past, in which pruning methods are aimed at suppressing the outer noise (boundary noise) and scale-space methods are aimed at suppressing the inner noise (such as scratch noise and dithering noise). In this paper, we proposed an improved framework that can deal with both inner noise and outer noise. The experiment proved that the proposed framework has better visual effects than the existing pruning method and existing scale-space method. In addition, the proposed framework is an adaptive framework that does not require manual tuning of parameters.
Tikhonov regularized SVM is a kind of new SVM which can convert multi-class problems to be single optimized problems. Since SVM has some limitations in disposition of big data collection, this paper puts forward a new...
详细信息
ISBN:
(纸本)9781424421138
Tikhonov regularized SVM is a kind of new SVM which can convert multi-class problems to be single optimized problems. Since SVM has some limitations in disposition of big data collection, this paper puts forward a new reduction Tikhonov regularized SVM by utilizing pruning algorithm to gain reduction data collection. Meanwhile, the paper applies genetic algorithm to make automatic selection from the balance parameter and kernel function parameter of Tikhonov regularized SVM. The experiment proves this newly improved Tikhonov Regularized SVM is more advantageous for classifying precision and train rate.
暂无评论