LSTM-GA-SVR based launch time window prediction algorithm

Abstract

The strike-type unmanned underwater vehicle (UUV) is susceptible to external factors such as ocean currents during attack missions, resulting in unavoidable non-linear attitude fluctuations that have an adverse impact on the weapon hit rate of the UUV. To meet the requisite accuracy for target hits of strike-type UUVs in the presence of attitude fluctuations, this paper proposes an LSTM-GA-SVR launch time window prediction algorithm for short-term attitude prediction of UUVs. This algorithm combines a Long Short-Term Memory (LSTM) neural network, a Genetic Algorithm (GA) and Support Vector Regression (SVR) to address the non-linear motion characteristics of the UUV and enhance the generalization ability of the prediction. The Genetic Algorithm (GA) is employed to optimize the Support Vector Regression (SVR) model parameters, thereby enhancing the accuracy of the fitted data. The LSTM prediction model is introduced to capture long-term dependencies in the data and extract complex features, thus further optimizing short-term attitude prediction. The loss function is optimized using the least squares method, thereby achieving optimal weighting of the model, and improving prediction accuracy. The efficacy of the algorithm is then corroborated through simulation and experimentation. The results demonstrate that, under specific circumstances, the algorithm can achieve the predetermined target of deep displacement error <0.1m and yaw angle error <0.05rad. The model prediction time window reaches 88% of the actual window, thereby providing a relatively accurate launch time window based on measured data.