A model for short-term semi-submersible platform motion prediction based on linear optimization considering error

Accurate short-term prediction of offshore structural motions is imperative for ensuring maneuver safety. Research has indicated that a highly effective approach to time series prediction is intergration of forecasts. A brand-new linear integration framework based upon neural networks and their errors herein is proposed for short-term motion forecasting of semi-submersible platforms. The framework is founded on three primary aspects. (a) Two neural network models have been selected for the sun-models: neural hierarchical interpolation for time series (N-HITS) models, and DLinear. Superior performance of these models when utilized in conjunction with the proposed framework has been demonstrated, but it maintains operational diversity. (b) The framework under consideration incorporates a swarm intelligence optimization algorithm for determining the weighting coefficients for each model. Moreover, algorithm considers errors series of the two models. To determine errors’ weights of the two model, the same swarm intelligence optimization algorithm is utilizeds. (c) A dynamic weight combination method was implemented, with weights assigned to the two sub-models. This method was used to generate the relatively optimal combination weights. Research findings suggest that the proposed hybrid prediction model exhibits superior performance regarding to accuracy and stability for semi-submersible platform predictions under varying conditions.