Artificial neural network (ANN) has outstanding characteristics in machine learning, fault, tolerant, parallel reasoning and processing nonlinear problem abilities. But BP training algorithm is based on the error grad...
详细信息
Artificial neural network (ANN) has outstanding characteristics in machine learning, fault, tolerant, parallel reasoning and processing nonlinear problem abilities. But BP training algorithm is based on the error gradient descent mechanism that the weight inevitably fall into the local minimum points. In this paper, a hybrid genetic algorithms(HGA) was proposed to solve the problem. The proposed HGA incorporates simulated annealing into a basic genetic algorithm that enables the algorithm to perform genetic search over the subspace of local optima. The two proposed solution methods were compared on Shaffer function global optimal problems, and the results indicated that HGA was successful in evolving ANNs.
This paper focuses on space-time non-linear intelligent modeling for regional data, researches how to apply back-propagation neural network (BPN) into analysis of regional data. Thinking about sectional instability of...
详细信息
This paper focuses on space-time non-linear intelligent modeling for regional data, researches how to apply back-propagation neural network (BPN) into analysis of regional data. Thinking about sectional instability of spatial pattern, this paper divided space units of researching regions into different subregions by improved K-means algorithm based on spatial adjacency relationship. Then build a space-time model with BPN. To solve the problem that deviation is too large in determining boundary of zonings by BPN model when data dimension increased, this paper bring forward a modular BPN model which named regional space-time neural network (RSTNN) model, modeling and predicting respectively based on every zoning with BPN. At last, compare the abilities of modular BPN model and global BPN model by means of analysis of an example.
Artificial neural network is a nonlinear dynamic system, which can attain the reflection of nonlinear relations among variables within any precision, possessing the ability of solving nonlinear problems, therefore als...
详细信息
Artificial neural network is a nonlinear dynamic system, which can attain the reflection of nonlinear relations among variables within any precision, possessing the ability of solving nonlinear problems, therefore also meeting requirements of economic forecasting. Taking advantages of the nonlinear and dynamic characteristics, by adjusting weights, we can approach any continuous functions by enough precision, therefore being able to approach the function in which the stock price changes with time, so we can imitate and learn the trading model of stock market. However, the traditional BP algorithm has low convergent speed. By proposing the deviation rate, this paper improves the convergent speed and it is tested in the forecasting of stock market.
In this paper, nonlinear functions generated by randomly initialized multilayer perceptrons (MLPs) and simultaneous recurrent neural networks (SRNs) are learned by MLPs and SRNs. Training SRNs is a challenging task an...
详细信息
In this paper, nonlinear functions generated by randomly initialized multilayer perceptrons (MLPs) and simultaneous recurrent neural networks (SRNs) are learned by MLPs and SRNs. Training SRNs is a challenging task and a new learning algorithm - DEPSO is introduced. DEPSO is a standard particle swarm optimization (PSO) algorithm with the addition of a differential evolution step to aid in swarm convergence. The results from DEPSO are compared with the standard backpropagation (BP) and PSO algorithms. It is further verified that functions generated by SRNs are harder to learn than those generated by MLPs but DEPSO provides better learning capabilities for the functions generated by MLPs and SRNs as compared to BP and PSO. These three algorithms are also trained on several benchmark functions to confirm results.
At the dawn of the 3 rd millennium, Human Handwriting Recognition is emerging from its infancy and set to become a mature technique. We shall probably see in the near future a number of mixed systems able to read bot...
详细信息
At the dawn of the 3 rd millennium, Human Handwriting Recognition is emerging from its infancy and set to become a mature technique. We shall probably see in the near future a number of mixed systems able to read both online and off-line handwriting. In this study we propose a simple yet robust structural solution for performing character recognition in Gujrati, the official language of Gujarat. Pursued by the preprocessing techniques, we suggest a method called template matching where a character is identified by analyzing its shape and comparing its features that distinguish each character. The algorithm appears to be very robust against stroke order variations and large shape variations. The results seem encouraging.
Large scale neural networks have many hundreds or thousands of parameters (weights and biases) to learn, and as a result tend to have very long training times. Small scale networks can be trained quickly by using seco...
详细信息
Large scale neural networks have many hundreds or thousands of parameters (weights and biases) to learn, and as a result tend to have very long training times. Small scale networks can be trained quickly by using second-order information, but these fail for large architectures due to high computational cost. Other approaches employ local search strategies, which also add to the computational cost. In this paper we present a simple method, based on opposite transfer functions which greatly improve the convergence rate and accuracy of gradient-based learning algorithms. We use two variants of the backpropagation algorithm and common benchmark data to highlight the improvements. We find statistically significant improvements in both convergence speed and accuracy.
This paper presents a novel application of the self-organised multilayer perceptrons inspired by the immune algorithm in financial time series prediction. The simulation results were compared with the multilayer perce...
详细信息
This paper presents a novel application of the self-organised multilayer perceptrons inspired by the immune algorithm in financial time series prediction. The simulation results were compared with the multilayer perceptrons and the functional link neural networks. The prediction capability of the various neural networks was tested on ten different data sets; the US/UK exchange rates, the JP/US exchange rate, the US/EU exchange rates, NASDAQO time series, NASDAQC time series, DJIAO time series, DJIAC time series, DJUAO time series, DJUAC time series and the oil price. A new training algorithm was utilized with the self-organised multilayer perceptrons neural network that is inspired by the immune using weight decay, the simulation results indicated significant improvement of the proposed training over the standard network.
A new approach based on artificial neural network (ANN) was developed in this study to determine the performance of solar collectors. The experiments were performed under the meteorological conditions of Beijing. Perf...
详细信息
A new approach based on artificial neural network (ANN) was developed in this study to determine the performance of solar collectors. The experiments were performed under the meteorological conditions of Beijing. Performance parameters obtained from the experimentation were used as training data. The backpropagation learning algorithm and logistic sigmoid transfer function were used in the ANN. Ambient temperature of collector, solar identity, declination angle, azimuth angle and tilt angle are used in the input layer and the efficiency and heating capacity are outputs. The results showed that the ANN with 10 neurons in the hidden layer is the most suitable algorithm with maximum correlation coefficient (R2), minimum root mean square error (RMSE) and low coefficient of variance (COV). Simulation results conformed that the use of ANN for performance prediction of solar collectors is acceptable.
In this paper, based on the traditional algorithm of TSK fuzzy reasoning model, a new fuzzy reasoning algorithm is proposed for two rules, two linguistic input variables and one output variable, in which the membershi...
详细信息
In this paper, based on the traditional algorithm of TSK fuzzy reasoning model, a new fuzzy reasoning algorithm is proposed for two rules, two linguistic input variables and one output variable, in which the membership functions are Gaussian-type functions. By using neural network back-propagation algorithm, the parameters in the membership functions can be adjusted on-line without changing the rules. The proposed reasoning algorithm can overcome the weak firing or non-firing cases.
A novel sample normalization algorithm based on fuzzy rough set theory is proposed to avoid the longtime training of neural network classifier caused by the smaller distances between samples of different classes. Firs...
详细信息
A novel sample normalization algorithm based on fuzzy rough set theory is proposed to avoid the longtime training of neural network classifier caused by the smaller distances between samples of different classes. Firstly, the samples are discretized based on rough set theory. Then, according to the distance differences between their discretized samples and two class samples and the energy differences between the two class samples, the original samples are extended or contracted based on fuzzy set theory. Then, the samples extended or contracted are normalized. Finally, the normalized samples are used to train the neural network. The method is analyzed with an example of faulty line detection for distribution network. The simulation results show that the training time of neural network with preprocessed samples is shorter markedly.
暂无评论