Aeroengine NOx-emissions automatic control based on neural network model

The study solves the problem of automatic control of the nitrogen oxides emissions from the aeroengine combustion chamber. Controlling algorithm is based on the pilot fuel ratio (PFR) optimizing. Since direct measurement of NOx emissions in flight is not available, the control system uses a built-in neural network predictive model of NOx emissions generating. An experimental comparative analysis of the model accuracy for various network configurations and training methods was carried out. It is proved that the Levenberg-Marquardt optimization algorithm provides the highest accuracy of the model. In the considered case of inapplicability of analytical methods for solving the problem of optimal adaptive control of a complex stochastic system the universal Monte Carlo statistical modeling, based on a specially organized iterative procedure simulating a statistical experiment, was used. As a result of the iterative procedure implementation a representative sample of random values of the desired variables is obtained allowing to select the best of them in terms of a given optimization criterion. The accuracy of the method is determined by the uniformity of the distribution density and the non-correlation of the analyzed random points inside the multidimensional cube limiting the space of acceptable change of the optimized parameters. In the proposed study, two pseudo-random number generators are analyzed. The best in terms of control accuracy was the one-dimension R1 sequence, which is a Kronecker sequence using the golden ratio. This random number generator is the best for solving the task of combustor chamber NOx emissions controlling based on the quasi-random Monte Carlo method.