Hybrid Intelligentalgorithm Based On Machine Learning Parameter Optimization Problem

In generally,using intelligent algorithm with parameter optimization problem,we should establish the corresponding mathematical optimization model.And about the choice of the objective function,we usually based on the least square principle.That is calculate the sum of the subtraction squares of experimental data and measured value.In this method without considering the possible measurement data itself or statistical error.This article is based on matrix factorization of the step length of machine learning methods put forward by Meng-deyu,to improve the stability of the optimization model.Main approach is add a parameters that between 0 and 1 to all the data,the parameters used to weaken the each data on overall predicted results.And then,add a regulation function in the whole formula to adjust the whole model,prevent excessive fitting model training data and enhance the overall generalization performance of the model.In this condition,the parameters in the model is more,the complexity is very high.A fitting is to ensure the training error is very small,but the actual predictive power is not necessarily good.First,the paper puts forward the thought prototype machine learning method to improve the traditional objective function and the related theory of machine learning algorithm.And then,as the hydro-geological parameters and river water quality problem solving process an example.Respectively based on machine learning improved simulated annealing algorithm and improved particle swarm algorithm.Compared the calculated results about general intelligent optimization algorithm and hybrid algorithm,fully illustrates the proposed machine learning method was applied to the optimization model of the idea is feasible.Under the same conditions is higher than other algorithm's efficiency,and the degree of fitting of the actual data better.In addition,by adding disturbance to the original data and the improving method of the data packet,further verify the stability and he universal applicability of the method.