Design of longwall face support by use of neural network models

Powered supports of insufficient capacity may be unable to prevent roof fai lure or falls at the longwall face or may result in structural damage under intense roof activity, whereas a system of support that is heavier than ne cessary can increase costs substantially. The design of a powered support s ystem that will be appropriate under specific mining and roof conditions pr esupposes knowledge of the relationship between support performance paramet ers and parameters representative of the mining and roof conditions. Since this relationship is characterized by uncertainty, nonlinearity and depende nce on a multiplicity of factors, an approach founded on the theory of arti ficial neural networks was selected. A computer program based on a back-pro pagation training algorithm was written in C++ to derive the weighting fact ors between inputs and outputs from field data; once these stabilized weigh ting factors had been established two mathematical models were developed to describe the relationship, The back-calculation of field data indicates th at the results generated from the neural network models are much more accur ate than those derived by traditional methods and can be used with a higher level of confidence. The support density, possible yield frequency of face supports and interval of periodic roof weighting under specific mining and roof conditions can be determined by application of the artificial neural network models that have been developed.