Research On Thermocline Processing Method Based On Deep Learning

With the continuous exploration of the earth's resources,the rich resources and complex environment in the ocean have gradually become the research focus of various countries.The marine thermocline is a very important research direction from the strategic point of view.However,due to the complexity of the marine environment,the traditional methods for the determination of the thermocline have different standards for deep and shallow sea areas,large subjective factors,and need to be judged manually There are some shortcomings in the system.In the thermocline parameter calculation task,the traditional method is not suitable for deep sea area.The above problems have brought severe challenges to the field of thermocline determination and thermocline parameter calculation.In this paper,through studying the literature related to thermocline,summarizing the current methods and analyzing their shortcomings,an intelligent determination and parameter calculation method of marine thermocline based on deep learning is proposed.The main work of this paper is as follows.(1)In this paper,a new method based on the least square method is proposed to calculate the thermocline parameters.The vertical three-layer structure of the ocean is re modeled.The piecewise linear function is used to fit the sea surface temperature data.The position of the segmented points is continuously optimized to minimize the variance.Thus,the thermocline calculation method suitable for deep and shallow sea areas is completed.(2)A thermocline adaptive determination method based on Feature Engineering and clustering algorithm is proposed.The experimental results of PCA feature selection method and physical oceanography based feature selection method are compared.The conclusion is that the feature selection method based on physical oceanography is better.The feature engineering construction of ocean temperature data is carried out,and K-means clustering algorithm is used to process the samples in different seasons The classification results were not supervised.This method has high self-adaptive degree,uniform judgment standard and no need of manual calibration.To a certain extent,it solves the shortcomings of traditional methods,such as human factors,non-uniform standards,and consumption of human resources and computing resources.(3)According to the optimization task objectives and the characteristics of ocean data,a multi task one-dimensional convolutional neural network is proposed.The encoder decoder architecture is used to share the low-level parameters,and the residual structure is added to complete the identity mapping.It is found that the network has a good effect on the verification set,and the effect of upper and lower bounds is slightly insufficient,but the overall error is still very small.It shows that the model has certain generalization ability.The system of adaptive determination of thermocline samples,automatic calculation of parameters and online determination and calculation of thermocline are constructed.It realizes the functions of no sample to have sample,automatic calibration of sample,offline calculation to online calculation.