Ion concentration prediction method based on support vector machine

Abstract

The invention discloses an ion concentration prediction method based on a support vector machine. The method comprises the steps that (1) n pieces of data of different types of ion concentrations in multiple salt ponds in different time periods are collected, wherein the data comprises collection time, collection place horizontal coordinates, collection place vertical coordinates, numbers of the salt ponds subjected to collection and the ion concentrations; (2) based on the collected original data of ions, data normalization processing is performed; (3) a random number function is utilized totaken m pieces of data as a training set, and the remaining (n-m) pieces of data serve as a test set; (4) optimal parameters, namely penalty factors c and a variance g in an RBF kernel function are found, the support vector machine is trained, and the ion concentrations in the salt ponds are predicted through support vector regression; (5) a regression model which is trained and used for predicting the ion concentrations in the salt ponds is stored; and (6) support vector machine simulation prediction is performed, and training set prediction result comparison and test set prediction result comparison are obtained. Through the method, the ion concentrations can be predicted more accurately, and prediction precision is high.

Description

technical field [0001] The invention relates to ion concentration prediction technology, in particular to an ion concentration prediction method based on a support vector machine. Background technique [0002] At present, the common methods for establishing ion concentration prediction models are as follows: [0003] 1. Partial least squares method: while establishing the regression model, principal component analysis can be performed to simplify the data, and the prediction performance is better, but it is only advantageous in a few cases; [0004] 2. BP neural network algorithm: It has strong nonlinear mapping ability and self-learning ability, but the learning speed is slow and prone to "over-fitting" phenomenon; [0005] 3. K nearest neighbor classification algorithm: suitable for problems with many attributes or a large amount of data, but it takes a long time and has high requirements for the growth rate of the observation set; [0006] 4. Linear discriminant analysi...

AI Technical Summary

Problems solved by technology

[0004] 2. BP neural network algorithm: It has strong nonlinear mapping ability and self-learning ability, but the learning speed is slow and prone to "over-fitting" phenomenon;

[0005] 3. K nearest neighbor classification algorithm: suitable for problems with many attributes or a large amount of data, but it takes a long time and has high requirements for the growth rate of the observation set;

[0006] 4. Linear discriminant analysis method: belongs to supervised learning dimension reduction, not suitable for non-Gaussian distribution samples for dimension reduction, may overfit the data

Images