Improvement Of Twin Support Vector Machine Model And Its Application In Runoff Forecasting

Accurate runoff forecasting plays an important role in water resources management.However,the runoff sequence is characterized by high factors such as meteorological,geographical conditions,and human activities,such as high signal-to-noise ratio,nonlinearity,and randomness.It is difficult to improve the accuracy of runoff prediction by a single model.To this end,this paper studies the combined forecasting method based on wavelet analysis theory combined with Twin Squares Support Vector Machines(TSVM)and the use of artificial fish swarm algorithm to optimize TSVM parameters.The forecasting model was established by MATLAB language programming,and the monthly runoff forecast was realized by using the twin kernel support vector machine based on mixed kernel function.The research content of this paper mainly includes the following aspects:(1)Discusses the knowledge and concepts related to forecasting.Firstly,the definition of wavelet is introduced.The discrete wavelet transform and Mallat fast algorithm are discussed.The theoretical reference basis for dealing with high SNR runoff data is introduced.The VC dimension theory,structural risk minimization and support vector machine are introduced comprehensively.The theory is proposed.(2)The traditional twin sub-vector model is studied.Firstly,the theory and basic kernel function of Gemini support vector machine are discussed and analyzed,and the shortcomings of commonly used radial basis functions and polynomial functions are proved by simulation experiments.In this paper,we construct a mixed kernel function by linearly superimposing these two functions,and set the weights to different values ??for performance comparison.The results show that the improved hybrid kernel function has stronger advantages in prediction than the commonly used kernel function,and can effectively improve the learning ability and fitting ability of the Gemini support vector machine.(3)Because the selection of parameters in the Gemini support vector machine method has a crucial impact on the prediction performance,this paper proposes to improve the artificial fish swarm algorithm to optimize the kernel parameters.Firstly,the basic idea of artificial fish swarm algorithm is introduced.The chaotic mechanism is added to the initialization of fish school and its visual field and step size are adaptively improved.The test function optimization experiment was designed and compared with the basic artificial fish swarm algorithm and particle swarm optimization(PSO).The improved precision and convergence of the improved artificial fish swarm algorithm were effectively improved.Finally,with the minimum mean square error as the optimal objective function of the model,the improved model of the twin support vector machine is established.(4)Analyze the interannual and intra-annual trends of the Shangjiang River Basin andperform wavelet preprocessing on the runoff data.Coupled with the Gemini support vector machine model,the selection of input variables for the improved artificial fish swarm algorithm for the artificial fish swarm algorithm(IAFSA-TSVM)model is discussed,which helps us to better understand the observation.The difference between data and forecast data.In order to test the advantages of the IAFSA-TSVM model,the runoff is predicted by the TSVM model and the PSO-TSVM model without parameter optimization.When the runoff hysteresis variable is 4 and the rainfall data is added,the prediction effects of the three models are relatively stable.The IAFSA-TSVM model has the best forecast performance.In addition,in order to further analyze the prediction performance of the IASSA algorithm for TSVM,this paper optimizes the parameters of BP neural network(BPNN),radial basis neural network(RBNN)and SVM to verify the effectiveness of the IAFSA algorithm..Finally,a box plot is used to evaluate the performance of all models.The case study shows that the simulation results of the historical monthly runoff of Shangyoujiang Hydropower Station show that the IAFSA-TSVM model has higher precision than other models.Therefore,the IAFSA-TSVM model proposed in this paper provides an effective method for monthly runoff forecasting.