Browsing by Author "Shen, Liang-Chi"

Now showing 1 - 8 of 8

Higher Order Neural Networks for Well Log Data Inversion
(2008)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
Chen, Chun-Yu
;
資訊工程學系
;
Department of Computer Science
Multilayer perceptron is adopted for well log data inversion. The input of the neural network is the apparent resistivity (Ra) of the well log and the desired output is the true formation resistivity (Rt). The higher order of the input features and the original features are the network input for training. Gradient descent method is used in the back propagation learning rule. From our experimental results, we find the expanding input features can get fast convergence in training and decrease the mean absolute error between the desired output and the actual output. The multilayer perceptron network with 10 input features, the expanding input features to the third order, 8 hidden nodes, and 10 output nodes can get the smallest average mean absolute error on simulated well log data. And then the system is applied on the real well log data.
Multilayer Perceptron Learning with Particle Swarm Optimization for Well Log Data Inversion
(2012)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
Chen, Kai-Ju
;
Huang, Ming-Che
;
資訊工程學系
Well log data inversion is important for the inversion of true formation. There exists a nonlinear mapping between the measured apparent conductivity (C-a) and the true formation conductivity (C-t). We adopt the multilayer perceptron (MLP) to approximate the nonlinear input-output mapping and propose the use of particle swarm optimization with mutation (MPSO) algorithm to adjust the weights in MLP. In the supervised training step, the input of the network is the measured C-a and the desired output is the C-t. MLP with optimal size 10-9-10 is chosen as the model. We have experiment in simulation and real data application. In simulation, there are 31 sets of simulated well log data, where 25 sets are used for training, and 6 sets are used for testing. After training the MLP network, input Ca, then C-t' can be inverted in testing process. Also we apply it to the inversion of real field well log data. The result is acceptable. It shows that the proposed MPSO algorithm in MLP weight adjustments can work on the well log data inversion.
MULTILAYER PERCEPTRON WITH GENETIC ALGORITHM FOR WELL LOG DATA INVERSION
(2013-01-01)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
Chen, Kai-Ju
;
Huang, Ming-Che
;
資訊工程學系
Two-layer multilayer perceptron (MLP) learning by genetic algorithm (GA) is used to approximate the nonlinear mapping between the input and the desired output. The GA is a global optimization method that can avoid the local minimum during the training in MLP and is implemented in binary and real number calculations. We have experiments on 31 simulated well log data and real data application. In the supervised training step, the input of the network is the apparent conductivity (Ca) and the desired output is the true formation conductivity (Ct). The best size of two-layer MLP is chosen as 10-9-10 by theorem and experiments. And we get the best parameters of binary GA and real number GA by sequential method. After getting the best MLP network in training, the corresponding true formation conductivity can be inverted for each input Ca pattern in testing process. From comparison of errors in experiments of simulated data, the real number GA has less error than that of binary GA. That is because the bit string in binary GA limits the range of weighting coefficient and has higher error. We also apply the best 10-9-10 MLP model to the inversion of real field well log data. It shows that this method can work on well log data inversion and is feasible.
MULTILAYER PERCEPTRON WITH PARTICLE SWARM OPTIMIZATION FOR WELL LOG DATA INVERSION
(2012)
Huang, Kou-Yuan
;
Chen, Kai-Ju
;
Huang, Ming-Che
;
Shen, Liang-Chi
;
資訊工程學系
A nonlinear mapping exists between the measured apparent conductivity (C-a) and the true formation conductivity (C-t). We adopt the multilayer perceptron (MLP) to approximate the nonlinear input-output mapping and propose the use of particle swarm optimization with mutation (MPSO) algorithm to adjust the weights in MLP. In the supervised training step, the input of the network is the measured Ca and the desired output is the C-t. MLP with optimal size 10-9-10 is chosen as the model. We have experiments in simulation and real data application. In simulation, there are 31 sets of simulated well log data, where 25 sets are used for training, and 6 sets are used for testing. After training the MLP network, input C-a, then C-t' can be inverted in testing process. Compared with radial basis function (RBF) networks and particle swarm optimization (PSO) method, the error of MPSO is the smallest. Also we apply it to the inversion of real field well log data. The result is acceptable. It shows that the proposed MPSO algorithm in MLP weight adjustments can perform the well log data inversion.
PROOF OF HIDDEN NODE NUMBER AND EXPERIMENTS ON RBF NETWORK FOR WELL LOG DATA INVERSION
(2016)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
You, Jiun-Der
;
Weng, Li-Sheng
;
資訊工程學系
In the multilayer perceptron (MLP), there was a theorem about the maximum number of separable regions (M) given the number of hidden nodes (H) in the input d-dimensional space. We propose a recurrence relation to prove the theorem using the expansion of recurrence relation instead of proof by induction. We use three-layer radial basis function net (RBF) on the well log data inversion to test the number of hidden nodes determined by the theorem. The three- layer RBF has more nonlinear mapping. In the experiments, we have 31 simulated well log data. 25 well log data are used for training, and 6 are for testing. The experimental results can support the number of hidden nodes determined by the theorem.
Proof of Hidden Node Number in MLP and Experiments on Well Log Data Inversion
(2015)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
You, Jiun-Der
;
資訊工程學系
;
Department of Computer Science
In the multilayer perceptron (MLP), there was a theorem about the maximum number of separable regions (M) given the number of hidden nodes (H) in the input d-dimensional space. We propose a recurrence relation in the high dimensional space and prove the theorem using the expansion of recurrence relation instead of proof by induction. The MLP model has input layer, one hidden layer, and output layer. We use different MLP models on the well log data inversion to test the number of hidden nodes determined by the theorem. The inputs are the first order, second order, and third order features. Higher order neural network (HONN) has the property of more nonlinear mapping. In the experiments, we have 31 simulated well log data. 25 well log data are used for training, and 6 are for testing. The experimental results can support the number of hidden nodes determined by the theorem.
Radial Basis Function Network for Well Log Data Inversion
(2011)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
Weng, Li-Sheng
;
資訊工程學系
;
Department of Computer Science
We adopt the radial basis function network (RBF) for well log data inversion. We propose the 3 layers RBF. Inside RBF, the 1-layer perceptron is replaced by 2-layer perceptron. It can do more nonlinear mapping. The gradient descent method is used in the back propagation learning rule at 2-layer perceptron. The input of the network is the apparent conductivity (Ca) and the output of the network is the true formation conductivity (Ct). 25 simulated well log data are used in the training. From experimental results, the network with 10 input data, first layer with 27 nodes, second layer with 9 hidden nodes and 10 output nodes can get the smallest average mean absolute error in the training. After training in the network, we apply it to do the inversion of the real field well log data to get the inverted Ct. Result is good. It shows that the RBF can do the well log data inversion.
WELL LOG DATA INVERSION USING RADIAL BASIS FUNCTION NETWORK
(2011)
Huang, Kou-Yuan
;
Shen, Liang-Chi
;
Weng, Li-Sheng
;
資訊工程學系
;
Department of Computer Science
We use the radial basis function network (RBF) for well log data inversion. The first step of the network is the K-means clustering. For the second step, we adopt the 2-layer perceptron instead of conventional 1-layer perceptron. The 2-layer perceptron can do the more nonlinear mapping. The gradient descent method is used in the back propagation learning rule at the second step. The input of the network is the apparent conductivity (Ca) and the output of the network is the true formation conductivity (Ct). The original features are the network input for training process. According to our experimental results, the three-layer radial basis function can get smaller error between the desired output and the actual output. The network with 10 input features, first layer with 27 nodes, second layer with 10 hidden node, and 10 output nodes can get the smallest average mean absolute error on simulated well log data. After simulation, we apply the network to the real field data. The result is good. It shows that the RBF can do the well log data inversion.