Multi-layer perceptron soft-measurement algorithm based on improved nonnegative garrote
Aiming at the issues of multivariable,strong coupling and nonlinearity in data-driven modeling of complex processes,a multi-layer perceptron(MLP)soft measurement algorithm combining sensitivity analysis with nonnegative garrote(NNG)is proposed.Firstly,the sensitivity analysis based on variance decomposition is used to quantize relational degree between each input variable and the target variable,and the total sensitivity index of each variable is calculated.Secondly,the total sensitivity index is embedded into NNG algorithm and then combines with MLP neural network to realize input variable selection.Finally,the validity of the algorithm is verified by Friedman dataset and the prediction of gasoline octane number in a petrochemical enterprise.Experimental results show that the proposed algorithm overcomes the shortcoming of biased coefficients estimations of NNG algorithm,effectively reduces the model complexity and improves the model prediction precision.
NETL
NSTL
万方数据
