Research on Parameter Optimization Method of v-SVRM Forecasting Model for Crane Load Spectrum

Precise load spectrum of crane is essential to its fatigue analysis and life assessment. The v-SVRM (v-support vector regression machine) correctly established is key to an undistorted load spectrum. Due to computational complexity, low accuracy, poor stability of the conventional model parameter selection method with v-SVRM, a fruit fly optimization algorithm with the characteristics of easy adjustment and high precision is applied. In order to make three kind parameters synchronously optimizing search, the fruit fly algorithm is improved in consideration of parameters characteristic of the crane load spectrum v-SVRM prediction model. Then, combining the improved fruit fly algorithm with penalty function and using anti-bound thought, a secondary optimization is carried out for three kind parameters. The results of examples engineering show that the optimal parameter group selected by the improved algorithm shortens the training time, reduces the computational complexity and improves the learning accuracy and generalization ability of v-SVRM model. The accuracy and stability of parameters is enhanced by secondary optimization, so as to a better robustness and versatility of v-SVRM predictive model. It also provides a new way for the establishment of efficient and convenient crane load spectrum v-SVRM forecast model.