Neural network inference of biomass fuel moisture during combustion process evaluating of directly unmeasurable variables

There are discussed various approaches to the evaluation of variables whose values are for any reason impossible to be measured directly. For moisture evaluation of combusted fuel, several formula were previously proposed. In the investigations reported in the paper they have been examined which of them is the most suitable for the moisture inference gained in small-scale biomass fired boilers. In the proposed neural network based on two neurons, the back propagation method has been used for derivation of the adaptation rule. Results of the evaluation are based on real data obtained in the experiments carried on a prototype 100 kW of Fiedler biomass boiler. The boiler has a special instrumentation making possible to check correctness of obtained results not only in the values of moisture but also in the other parameters occurring in the used formula.