Research On Cyanobacteria Bloom Prediction And Decision-making Method Based On Improved Bi-LSTM-ARIMA Model

In recent years,with the continuous strengthening of environmental protection awareness,water environment issues have become a research hotspot,and the prediction and control of cyanobacteria blooms is one of the water environment issues that have plagued people for many years.However,because the growth mechanism of cyanobacteria blooms is highly nonlinear,time-varying and multivariate,its prediction and control have always been the focus and difficulty.In response to the mentioned problems,this paper mainly focuses on the cyanobacteria bloom prediction and management decision-making.The main work includes:(1)The basic principles of data background and related theories are introduced.First,the background and source of the experimental data required for the research are explained,and the data is preprocessed;Secondly,the theoretical introduction and principle analysis of the main algorithms used in the research are carried out to provide a theoretical basis for modeling.(2)Aiming at the problem of inaccurate prediction by single algorithm,a Bi-LSTM-ARIMA hybrid prediction algorithm based on wavelet decomposition is proposed.Wavelet decomposition is firstly used to decompose the original data of cyanobacteria density into low-frequency and high-frequency data.Then,Bi-LSTM algorithm is used to predict the trend of low-frequency data,and the ARIMA model is used to predict the high-frequency data in detail.Finally,two prediction results are integrated to obtain a mixed prediction result.(3)Aiming at the problem of ignoring the influence of environmental variables on cyanobacteria,a residual regression compensation model based on QPSO(Particle Swarm Optimization)is proposed.This model is to input the cyanobacteria time series training data into the trained Bi-LSTM model to obtain the training result,and make the difference with the real cyanobacteria training value to obtain the residual value of the time series training.Then,the obtained residual value is used as output signal,the explanatory variable is used as input signal to build a neural network model,QPSO is used to optimize the parameters of the model,and the built model is used to predict the residual error of the cyanobacteria density.Finally,the predicted residual is compensated to the predicted result in(1).(4)The established prediction model is used to build a cyanobacteria bloom management decision-making model,and the decision-making model is evaluated using AHP,and then,specific decision-making measures are obtained.The built predictive model is used to establish the model between the cyanobacteria density at the last moment and the cyanobacteria density at the current moment as well as independent variables.On this basis,the lowest cyanobacteria density value is predicted through the permutation and combination of different environmental factors,and therefore the best permutation and combination method is obtained.On the theoretical basis of cyanobacteria management decision-making strategies,three specific measures for cyanobacteria management are proposed,and the AHP is used to establish a comprehensive evaluation model of cyanobacteria treatment plan,and then,the optimal plan for the cyanobacteria treatment is obtained.(5)Cyanobacteria prediction and decision model simulation results are compared and analyzed.First,the experimental environment is introduced,and the selection of evaluation indicators are explained;then compare and analyze the prediction results of different algorithms.The final results show that the hybrid algorithm can solve the problem of poor adaptability compared to single time series forecasting algorithm.The compensation model can reasonably combine the time series model and the regression model,and improve the prediction accuracy;then,the impact of different training set ratios on the prediction results is compared.The larger the training set ratio,the more accurate the prediction is;finally,simulation results of the decision model are given.