Multi-dimensional Support Vector Regression Optimized By Complexity-based Genetic Algorithm And Its Application In Predicting The Formation Of Hydrates

As a new seawater desalination technology, gas hydrate method has got people’s wideattention. While, how to determine the best combination of the process parameters in thedesalination is one of the key factors influencing the treatment effect. Traditional method isnot only expensive, complex, but hard to ensure the optimal processing data. Thus, how tosolve this difficulty has become a hot topic. The SVM algorithm is proposed in this paperto fitting the formation of gas hydrate, and we can obtain the mapping relationship betweenparameters and performance of the algorithm. The main work includes:（1）In the process of regression prediction, we need to choose appropriate parametersto obtain better running result. Now, the determination of parameters often relies onexperience, and that is random. To solve this problem, the genetic algorithm (GA) withglobal stochastic optimization is proposed to optimize parameters and to achieveautomation of the algorithm. On the basis of completion of regression model with geneticoptimization, related experimental design and validation work have been done to prove theaccuracy of the model. While, the randomness of original population in the GA results inneeding lots of genetic algebra and the algorithm costs a longer running time.（2）In order to solve the problem of time efficiency, complexity index is proposed todetermine the parameters’ approximate value, and reduce the blindness of initial population.Further, the rules between the complexity index and the parameters of regression model canbe found by experiments researching. By comparing the complexity index based regressionmodel and traditional regression model and regression algorithm with genetic optimization,simulation experiment results are given in the contrast test.（3）Multi-dimensional SVM theory has been studied and applied to the complexityindex based support vector regression model with genetic optimization. With the samplesconstructed by the data deducing from the multi-dimensional function mapping,experimental verification has been done to the multidimensional model.（4）Applying the new method proposed in this paper to the processing of seawaterdesalination by gas hydrate method, and designing related simulation experiments. Theresult proves the feasibility of the new complexity index based model using geneticoptimization. Further more, to facilitate the new model’s application, relevant softwaredevelopment work has been completed.