Research And Implementation Of Ice Hockey Prediction System

With the successful bidding of Beijing Winter Olympic Games in 2022,fan's attention to ice hockey game has been significantly increased,and more and more methods of predicting and analyzing hockey games have been developed.With the rapid development of information technology,volume and dimension of ice hockey statistics continued to grow explosively,and the increase in data scale and characteristic dimension brings new challenges to the study of hockey match situation prediction.At present,the traditional prediction algorithm has been unable to meet the need of ice hockey tournament prediction.At this time,it is found that a support vector machine(SVM)provides a good solution to the classification and prediction problem,and has obvious advantages in dealing with the classification of non-linear and high-dimensional data.However,the existing Support Vector Machine(SVM)methods cannot meet the requirement of hockey tournament prediction for timeliness in the classification and prediction of large-scale high-dimensional data.Therefore,this thesis focuses on the feature selection of game data and the parameter optimization method of support vector machine,and realizes the prototype of an ice hockey game situation prediction system according to the actual demand.The main work of this thesis are as follows:(1)In order to solve the problem of high computational complexity of Support Vector Machine in high-dimensional space caused by the large-scale and dimension of hockey data,an effective solution is to select a small number of feature subsets from many candidate features of high-dimensional data,and remove irrelevant and redundant features.In this thesis,we adopt sparse score as well as the correlation between categories and features to study feature selection(ISSFS)algorithm based on sparse score and correlation analysis.The dimension reduction preprocessing is carried out on the input features of the prediction model.By calculating the sparse representation ability of each feature of the data establish and the degree of correlation between the category,the optimal feature combination is obtained through comprehensive analysis,so as to achieve the purpose of dimension reduction of high-dimensional data.Simulation results show that ISSFS algorithm can reduce the data dimension and reach an ideal classification effect.(2)For the problem that the penalty factor c and the kernel functioned parameter g in SVM has great influence on the generalization ability of SVM and the parameters are difficult to adjust,this thesis studies the parallel mechanism algorithm that iterates from the random solution to find the current optimal solution in order to find the global optimal solution,and realizes the purpose of SVM parameter optimization.Based on the good global optimization performance of genetic algorithm(GA)and the advantages of fast search speed and high efficiency of particle swarm optimization(PSO),we study the parameter optimization algorithm of hybrid genetic particle swarms to support vector machine(HGAPSO-SVM)by improving genetic algorithm in the iterative update step which will automatically optimize SVM parameters.Simulation results demonstrate that the accuracy of HGAPSO-SVM method is 3.6% higher than that of traditional GA-SVM and 7.4% higher than that of PSO-SVM.(3)Combining with the requirement,organization structure and main function flow of the hockey match situation prediction system,this thesis applies the hockey match situation prediction model to the hockey match situation prediction system,and gives the detailed realization of the key technology and typical operation interface of the hockey match situation prediction system...