Comparison of the neural net training algorithms for the emergencies forecasting of technological processes

International Workshop "Advanced Technologies in Material Science, Mechanical and Automation Engineering – MIP: Engineering – 2019"

材料科学;机械制造;原子能学

Tkalich, S.A.^1 ; Burkovsky, V.L.^1 ; Kravets, O Ja^1

Voronezh State Technical University, Moscow ave 14, Voronezh, Russia^1

关键词: Chemical water;    Composite modeling;    Controlled parameter;    Initial functions;    LM algorithm;    Neural networks toolboxes;    Technological process;    Training algorithms;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/537/3/032040/pdf
DOI  :  10.1088/1757-899X/537/3/032040

学科分类：材料科学（综合）

来源: IOP

【 摘 要 】

Composite model of emergencies forecasting of technological process of chemical water purification for the nuclear power plant (NPP) is considered. To create a neural network component of this model the Neural Networks Toolbox MATLAB package is used. In the process of neural net training the gradient of error functionality in three controlled parameters is calculated: viz., specific electric conductivity, hydrogen indicator pH, concentration of silicon acid. A comparison was made of a training algorithm of CGF realizing Fletcher-Reeves method with LM algorithm of Levenberg-Markvardt. The conclusion is drawn that a sufficiently exact repetition of a type of initial function of the proximity degree to an emergency occurs when the LM algorithm is used.