Ship Stability Prediction via RBFNN and Random Forest Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ship stability prediction methods are hindered by numerous input parameters, complex calculations, and time-consuming processes, which fail to accurately determine dynamic stability failures in varying sea conditions.
Innovation Solution
A method and system utilizing weighted fusion of Radial Basis Function Neural Network (RBFNN) and Random Forest based on Gradient Descent (GD) for ship stability prediction, which involves selecting critical ship parameters, determining optimal hyperparameters, and performing weighted fusion of output results to improve prediction accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ship stability prediction methods are used, then prediction accuracy for dynamic stability failures can be achieved, but the calculation process becomes cumbersome and time-consuming
Solution Approach 1:
The patent segments the ship stability prediction task into two distinct models: a lightweight model for rapid prediction and a precise model for accurate results. The lightweight model handles initial assessment quickly, while the precise model provides detailed analysis only when needed, thereby reducing overall time consumption while maintaining prediction accuracy.
Solution Approach 2:
The patent introduces a classification model as an intermediary between the lightweight prediction model and the precise prediction model. This intermediary classifies input cases to determine which model should process them, enabling the system to achieve both speed and accuracy by routing simple cases to the fast model and complex cases to the precise model.
2Reliability
If traditional ship stability prediction methods are used, then comprehensive stability assessment can be achieved, but the number of input parameters increases and complexity increases
Solution Approach 1:
The patent segments the prediction system into multiple specialized models (lightweight model and precise model) with different levels of complexity. Each model handles specific aspects of stability assessment, allowing the system to maintain comprehensive assessment capability while reducing the complexity burden on any single model or operation.
Solution Approach 2:
The patent creates a unified prediction system that can handle both simple and complex stability assessment cases through a single integrated architecture. The system universally processes all input cases by automatically selecting the appropriate model based on case characteristics, eliminating the need for separate assessment procedures for different case types.
3Productivity
If lightweight prediction models are used, then calculation speed improves, but prediction accuracy for complex failure modes deteriorates
Solution Approach 1:
The patent introduces a classification model as an intermediary that assesses the complexity of each prediction case. Based on this classification, the system routes simple cases to the lightweight model for fast prediction and complex cases to the precise model for accurate results, thereby maintaining both calculation speed and prediction accuracy across different case types.
Solution Approach 2:
The patent implements a dynamic model selection mechanism that adapts the prediction approach based on the characteristics of each input case. The system dynamically switches between the lightweight model and precise model according to the complexity and requirements of the specific stability assessment task, optimizing the balance between speed and accuracy for each case.
Data Source
AI summary
A method and a system for ship stability prediction by weighted fusion of RBFNN and random forest based on GD are provided. Firstly, input characteristics when predicting failure probabilities under different failure modes are determined through prior knowledge. Secondly, a mean square error of k-fold cross-validation is used as performance evaluation criterion of the RBFNN and the RF to search for model capacities of the RBFNN and the RF. Then, network parameters of the RBFNN are updated. Multiple random sample sets are generated using a bootstrap sampling method and are parallelly trained to generate multiple regression trees. A Gini index is used as an attribute division index, and a prediction result of the random forest is obtained. Finally, weight coefficients are introduced for weighted fusion of prediction results of the RBFNN and the RF. The weight coefficient is obtained by solving through iterative optimization of the gradient descent.