Waste incineration plants are expected to contribute to the realization of the appropriate energy mix. In order to increase the profitability of such plants, automatic operation technology is needed. Since the fuel supply of waste incineration plants is greatly affected by variations in fuel characteristic, a prospective control technology for the fuel feeding device has not been established. We have developed a fuel feeding control technology based on reinforcement learning, which is expected to enable flexible selection of appropriate operational controls considering changes in fuel characteristic. The use of this technology on a fuel supply device simulator resulted in a reduction of 56% in the number of over-supply and 54% in the number of under-supply compared to the conventional operating conditions. This report presents the development of this technology.
