In chemical plants and industrial plants, "soft sensors" is known as a technology to estimate state quantities that cannot be measured directly. Their application for purposes such as control and monitoring are progressing. Recent technological developments in deep learning are remarkable. While their employment in soft sensors has enabled highly accurate estimation, the problem lies in the necessity of a longer training time. To address this problem, we built a soft sensor using "Reservoir Computing," with which training can be done in an extremely short time while retaining the high accuracy of estimation. This report presents an overview of Reservoir Computing and its application using a case study on the prediction of calcium carbonate concentration in flue gas desulfurization equipment.