No abstract

This paper is concerned with the short-term forecasting of the time variation of the total output of photovoltaics over a wide area. Thermal power generation is essential for adjusting unstable output fluctuations of photovoltaic power output. Meanwhile, thermal power plants are forced to operate in no-load standby mode to cope with unforecastable power fluctuations. To reduce wasteful standby operation, we developed a new forecasting techniques using a neural network to forecast short-term changes in the total output of wide-area solar power. We applied the developed model to the northern part of Kyushu in Japan and forecasted solar power output 20 minutes ahead. It was evaluated by the amount of solar radiation itself and the clear-sky index. The forecasting reproduced the actual data well, especially when the clear-sky index was applied.

This paper is concerned with the multi-objective optimization of MOST, a new optimization method proposed by the author. Optimization techniques have become essential tools for efficient and rational engineering design and development. To meet this need, many optimization methods have been proposed and tested for practical use. In engineering design, there are often multiple objective functions, each with a trade-off. Therefore, multi-objective optimization techniques are essential. The optimization method MOST proposed by the author is highly convergent and has been applied to the optimization of neural networks and verified in previous papers. In this paper, we apply MOST to each of multi-objective functions and find the optimal value for each. Furthermore, MOST is applied to minimize the sum of the error of each objective function and its optimal value. This approach yields a Pareto solution. The improved MOST was applied to the benchmark problem of Zitzler et al., and compared with existing methods. As a result, MOST was able to reproduce the theoretical Pareto solution.