Research On Red Tide Forecast Algorithm Based On Neural Network And SVM

With the development of industry in coastal areas,a large amount of untreated sewage flows into the ocean,resulting in frequent occurrence of red tide disasters.Red tide has become one of the most serious disasters that damage the marine ecological environment in China.Therefore,accurate prediction of red tide disasters is of great significance for the protection of marine ecological environment.The reasons for the formation of red tides are very complex,with sudden and nonlinear characteristics.Traditional red tide prediction methods are difficult to accurately predict red tide.The strong learning ability and nonlinear fitting ability of artificial neural network make it widely used in red tide prediction.In this paper,based on BP neural network,support vector machine and intelligent optimization algorithm BSO,a red tide prediction model based on BSO-BP neural network and BSOSVM are established to predict the density of Noctiluca scintillans.The specific research contents are as follows:There is a lot of redundant information and non-linear correlation between the variables of the original sample of red tide.In this paper,the dimensionality reduction of red tide samples is performed by KPCA.The original input variables include 6 characteristic parameters such as water temperature,salinity,total nitrogen,soluble inorganic phosphorus,and phytoplankton density.After KPCA dimensionality reduction processing,4 principal components with a cumulative contribution rate exceeding 90% are retained.Replace the input variables in the original sample with fewer principal components,and retain most of the information in the original sample.At the same time,the training time of the red tide prediction model is reduced,and the influence of the unrelated factors on the prediction accuracy of the network model is reduced.Aiming at the defect that BP neural network is easy to fall into local optimum,a red tide prediction model based on BSO-BP neural network is proposed.Firstly,KPCA is used to reduce the dimensionality of the input variables to speed up the convergence of the network.Secondly,use BSO to optimize the initial weights and thresholds of the BP neural network.In order to balance the global search ability and local search ability of the BSO algorithm,an inverted S-shaped function is introduced to adjust the inertia weight in the BSO algorithm.Therefore,the BSO algorithm performs a global search in the early stage to ensure the prediction accuracy of the model,and performs a local search in the later stage to accelerate the convergence rate of the model.Finally,the best weights and thresholds found by the BSO algorithm are substituted into the BP neural network.The simulation results show that the BP neural network model optimized by the intelligent optimization algorithm effectively reduces the average relative error.Compared with the BP neural network,PSO-BP,BASBP and other models,the BSO-BP model has advantages in prediction accuracy compared to other models,and has a better prediction effect on the density of Noctiluca scintillans.Aiming at the problem that BP neural network is prone to overfitting in the training of small sample data,a red tide prediction model based on BSO-SVM is proposed.Firstly,KPCA is used to reduce the original input variables of red tide.The dimensionality-reduced samples are used as the training samples and test samples of the BSO-SVM model.Then,through the BSO algorithm,the penalty parameter and the kernel width in the SVM are optimized.Finally,the best combination found by the BSO algorithm is substituted into the SVM to predict the density of Noctiluca scintillans.The simulation results show that,compared with the BSO-BP model,the BSO-SVM model has a better prediction effect on the density of Noctiluca scintillans and has better stability.This shows that the BSO-SVM model is more suitable for the prediction of red tides with small samples.