Algorithm for parametric identification of fuel consumption characteristics by a vessel using neural network technology

The purpose of the study is to enhance methods for computer monitoring and parametric identification of models describing vessels' fuel consumption characteristics. These improvements aim at analyzing and forecasting energy efficiency indicators of water transport facilities and optimizing the operational modes of diesel generator units. The paper proposes an algorithm for parametric identification of input-output characteristics across various technological processes and systems (technical, biological, economic, social, environmental, etc.) based on measurement data using approximate motor (regression) neural networks. The algorithm enables quantitative error assessment of parametric optimization using the Euclidean norm. Unlike traditional methods relying on statistical series for model fitting, the proposed approach trains a multilayer neural network with backpropagation to minimize deviations in output signal values from reference values by adjusting synaptic weight coefficients. The study demonstrates that radial neural networks with fixed structures—comprising one hidden layer with nonlinear activation functions and one output layer with linear activation functions—are suitable for solving problems in this domain. These networks ensure accurate image mapping based on the Euclidean metric while simplifying training modes and maintaining acceptable approximation and identification accuracy. The algorithm has been implemented to estimate parameters of a vessel's fuel consumption model based on statistical series with a predefined initial approximation. It can also be applied to identify energy consumption characteristics in the inland water transport sector when calculating target indicators and developmental metrics.