Improvement Of Heavy Metal LSTM Prediction Based On Multivariate Chaotic Sequence Environment

With the improvement of the economy and people's living standards,heavy metal pollution in life has become more serious,and heavy metals have increased toxicity,which will cause people's production and life to hinder the harm.For the sustainable development of the river basin,if the heavy metals in the river basin exceed the standard,it will have a conductive impact on the natural environment and residents.Therefore,it is of practical significance to accurately predict the medium and short-term heavy metal content.Then,taking the heavy metal content of the soil in the Daxia River Basin as the research object,the following related work was carried out:First of all,because the heavy metal content is affected by various factors such as temperature,daily runoff,PH value,the traditional traditional single-factor prediction model cannot achieve the ideal prediction effect.Therefore,combined with temperature and daily runoff,a more reduced multi-phase chaotic phase space for heavy metal content is established.This phase space covers all possible states of heavy metal content after being affected by other factors,which will provide subsequent predictions.Comprehensive forecast support.In fact,after establishing an appropriate multivariate chaotic phase space,it is necessary to select an appropriate prediction method.At present,the emergence of neural networks has been widely concerned.Therefore,the article uses radial basis neural network(RBF)as a comparative experiment of heavy metal content,and A PS-LSTM prediction model is proposed based on multi-chaotic phase space reconstruction and long-short-term memory(LSTM)recurrent neural network with increased peephole connections.Configure multiple chaotic phase spaces to restore the true state of the input data.the improved LSTM neural network can obtain more of the previous unit and the current input data in the "memory cell",and improve the prediction effect of the model.The proposed PS-LSTM model has an RMSE of 0.0927 and a MAE of 0.2102,stepwise Volterra series one-step prediction,RBF neural network prediction and a support vector regression prediction model that is more suitable for small amounts of data.Finally,this paper also uses PS-LSTM to predict cotton yield in a cotton test field in the Mississippi River Basin,and demonstrates the universality of the model.By calculating the RMSE and MAE,it is 9.8% and 2.78% higher than the traditional LSTM model with the best prediction accuracy.therefore,the model has better prediction effect for complex nonlinear system prediction.