Shinya Sekizaki

Structural optimization of neural networks and training data selection method for prediction

The neural networks are well known as that they have an ability of approximation of any nonlinear function, and they are applied for data prediction in many fields. The parameters of neural networks, the thresholds and the weights between nodes, are updated by using given data. The performance of a neural network, for example prediction accuracy, is evaluated by the degree of the amount of the prediction error for multiple kinds of unknown data. To increase the performance, the appropriate structure of the neural network should be determined including the parameters. Cross-validation is one solution for this problem, and it is often applied in the existing literature. Through the cross-validation, the neural network may minimize not only error for the data for training, but also any unknown data, in other words, the neural network obtains generality capability by using cross-validation. The training data and the performance verification data are randomly selected in the procedure of the cross-validation, therefore, this paper proposes another method for the selection. In particular, the given data is mapped on 2 dimensional surface by using SOM(Self-Organization Map), and the mapped data are clustered into k clusters by using k-means method. From each cluster, certain fraction of the data are selected as the training data, and the rest data are selected as the performance verification data. Additionally, the proposed method includes the structural optimization method of a feedforward neural network consists of 4 layers, an input layer, a dimensional compressing layer, a hidden layer, and an output layer, by using tabu search. From the experimental result, it is observed that the appropriate structure of neural networks are obtained by using the proposed method.

Development of interactive multi-objective reinforcement learning considering preference structure of a decision maker

Reinforcement learning which is applied to multiobjective optimization problem is called multi-objective reinforcement learning. Related works in the study field of the multiobjective Reinforcement Learning indicate that multi-objective reinforcement learning with a choice procedure based on Hypervolume is effective for finding Pareto optimal solution of multiobjective optimization problems. However, a selected Pareto optimal solution based on Hypervolume does not always match the preference of a decision maker. This study proposes interactive multi-objective reinforcement learning which can reflect the preference structure of a decision maker using scalarization method and interactive method after discovering Pareto optimal solution.

Aging simulation of SiC-MOSFET in DC-AC converter

This investigation focuses on the aging simulation of a DC-AC converter during the stress situation of a lightning strike. The newly developed compact model HiSIM_HSiC for high-voltage SiC MOSFETs, which considers the carrier-trap increase, is applied for the simulation. Simulation results reproduce the measured converter characteristics during the conventional earth fault protection of the DC-AC converter. It is verified that the device aging occurs in spite of this protection. The modeled device aging after the converter has endured several lightning strikes predicts enhanced efficiency reduction by more than 10% in addition to the ordinary usage.

Distance-based clustering of population and intergroup Cooperative Particle Swarm Optimization

Sun and Li (2014) have proposed TCPSO(Two-swarm Cooperative Particle Swarm Optimization) that the swarms are divided into two groups with different migration rules. TCPSO has higher performance for high-dimensional nonlinear optimization problems. This study revises TCPSO to avoid inappropriate convergence of the swarms. The quite feature of the proposed method is that the population have same migration rules. However, through that the swarms are divided into some clusters based on distance measure, k-means clustering method, both diversity and centralization of search process are maintained, and it increases the potential of attainment to the global optimal solution. This study conducts numerical experiments using several types of functions, and the experimental results indicate that the proposed method has higher performance than the TCPSO for large-scale optimization problems.

A development of pseudo-synchronizing power VSCs controller for grid stabilization

In order to constitute the resilient power system against disturbances including natural disaster, various new technologies for microgrid operations have been proposed recently. Although major Renewable Energy Sources (RESs) are usually connected to single-phase feeders, there exist few research works on single-phase microgrid operations. This paper investigates the realization of the single-phase microgrid by using Voltage Source Converters (VSCs) which are widely used to connect DC-base RES systems to conventional distribution feeders. We first study effective virtual inertia design in three-phase VSC in order to realize strong synchronizing operation as well as the enhancement of transient stability. Then, we propose a control framework for single-phase VSC, in which a new synchronizing mechanism that we call DC side measurement is developed. A governor and an Automatic Voltage Regulator (AVR) are also implemented into the controller to contribute to stable and reliable operation of the microgrid. It is shown that the independent use of the proposed VSC in each phase can also work successfully in a three-phase feeder. The effectiveness of the proposed method is verified in numerical simulations in single-phase and three-phase distribution systems where multiple VSCs are successfully operated in normal and islanding modes.

A study for retailer's risk hedge considering responses of consumers in electricity deregulation

Recently in Japan, a deregulation of electricity market is being extended and retailers are expected to supply the electricity to consumers. Moreover, an energy management system enables the consumers to easily manage their energy consumption according to the electricity prices. This means that a large number of decision makers who manages electricity energy will be in the electricity market. In order to operate stably the next generation power system, in which the structure of the system is complicated, it is necessary to appropriately design the market based on a detail market analysis with a mathematical model of the decision makers. Thus, to formulate a mathematical model that adequately represents the behavior of a large number of decision-makers in the power market is indispensable. From this background, we formulates a Stackelberg game model as the bilevel programming problem which represents the statistical decision making in a trade between the retailer and the consumers considering a forward contract and day-ahead market, real time trade. The computational experiment shows that the proposed decision making model can represents the behavior of the market players.

The multi-objective optimization of Distribution System management in deregulated electricity market

In the electricity deregulation, the electricity consumption of consumers depending on the electricity prices will change because the electricity prices are expected to fluctuate according to the market conditions. Therefore, the fluctuation of the electricity consumption can cause difficulty of distribution system operation, e.g. minimizing the distribution line losses, improving the voltage profile, and so on. Previous studies show that Distribution System Reconfiguration (DSR) is effective to minimize distribution line losses and improve voltage profile on the distribution lines. However, the DSR problems in the literatures considering the electricity deregulation are not studied sufficiently. In this paper, we formulate a multi-objective optimization problem about the distribution system operation with DSR, and search for quasi-Pareto optimal solutions using Non-dominated Sorting Genetic Algorithm-II (NSGA-II).