Position Adjustment Using the Artificial Neural Network Backpropagation Technique

Nowadays, the usual adjustment methods of the surveying networks are typically addressed through the principle of Least Squares (L.S.) methodologies. One of the most important field data that needs to be adjusted is the observed coordinates of the stations. This kind of adjustment is commonly performed by using the L.S. criterion through solving a set of complex nonlinear functions. Therefore, this research aims to present an approach for implementing position adjustment in surveying networks based on the Artificial Neural Network Backpropagation (ANNB) concept. The proposed method is built on forming one or more geometrical conditions that consider the misclosure error of observations depending on the network’s field measurement circumstances. The initial weights of input data are determined based on their variances, which are then trained and updated. The weights updating are obtained according to the used activation function. Thus, the main advantage of the ANNB method lies in its adoption of a set of linear functions that vanish the misclosure error existing in each geometrical condition according to the network’s situation. The desired residuals are determined based on the updated weights for departures and latitudes of all formed geometrical conditions. Therefore, all misclosure errors in all geometrical conditions will be eliminated perfectly. The evaluation of the proposed ANNB method is done according to the well-known, Least Squares adjustment. The obtained results of the evaluation reveal a logic convergence with minor differences between the outcomes of the ANNB method and those of the parametric L.S. method.