Dong Hyawn Kim

Modified Probabilistic Neural Network Considering Heterogeneous Probabilistic Density Functions in the Design of Breakwater

In this study, a modified probabilistic neural network approach is proposed. The global probability density function (PDF) of variables is reflected by summing the heterogeneous local PDFs automatically determined in the individual standard derivation of each variable. The proposed modified probabilistic neural network (MPNN) is applied to predict the stability number of armor bøcks of break waters using the experiental data of van der Meer, and the estimated results of the MPNN are compared with those of conventional probabilistic neural network. The MPNN shows improved results in predicting the stability number of armor blocks of breakwaters and in providing the promising reliability for stability numbers estimated by using the individual standard deviation in a variable.

Earthquake Response Reduction of Building Structures Using Learning-Based Lattice Pattern Active Controller

An active vibration control technique for building structures using a learning-based lattice pattern controller (LBLPC) is proposed in this article. The training pattern of the LBLPC is composed of a lattice form for the control force and a state vector. The training pattern was trained by a learning rule using the gradient descent method (GDM) in earthquakes. The LBLPC calculates the control force using only the adjacent input information, thus making the corresponding calculation process much faster. A three-story building in the El Centro earthquake was used to train the LBLPC. And the California and Northridge earthquakes were used to verify the performance of the proposed method. In order to prove the control capability of the LBLPC, the control results of the LBLPC were compared with those of a lattice type probabilistic neural network (LPNN) in a numerical example. The results demonstrated that the proposed LBLPC algorithm reduces the response of the building structure during earthquakes more effectively than the LPNN.

Capability enhancement of probabilistic neural network for the design of breakwater armor blocks

In this study, the capability of probabilistic neural network (PNN) is enhanced. The proposed PNN is capable of reflecting the global probability density function (PDF) by summing the heterogeneous local PDF automatically determined in the individual standard deviation of variables. The present PNN is applied to predict the stability number of armor blocks of breakwaters using the experimental data of van der Meer, and the estimated results of PNN are compared with those of empirical formula and previous artificial neural network (ANN) model. The PNN showed better results in predicting the stability number of armor blocks of breakwaters and provided the promising probabilistic viewpoints by using the individual standard deviation in a variable.