Lake TN prediction method based on VMD-CSSA-LSTM-MLR combined model

Abstract

The invention discloses a lake TN prediction method based on a VMD-CSSA-LSTM-MLR combined model. The method comprises the following steps: firstly, performing decomposition to obtain K intrinsic mode functions through VMD; then processing and predicting high-frequency signals by adopting an LSTM neural network, and optimizing hyper-parameters of the LSTM neural network by adopting a CSSA; processing and predicting low-frequency signals by adopting MLR; and superposing prediction values of all the mode functions to obtain an actual prediction result. According to the method, the problems of mode mixing, end effects and the like existing in other common signal decomposition methods are effectively solved, and the method is high in operation speed and stable in decomposition results; and meanwhile, the operation efficiency of a algorithm and the prediction precision of a model are improved, the problem that the hyper-parameters of the LSTM neural network are difficult to manually determine is solved, and the efficiency and precision of a prediction model are improved.

Description

technical field [0001] The invention belongs to the technical field of water quality prediction, and in particular relates to a lake TN prediction method based on a VMD-CSSA-LSTM-MLR combination model. Background technique [0002] At present, there are mainly four types of methods for water quality prediction: the first is the water quality simulation model prediction method, the second is the prediction method based on the traditional mathematical statistical model, the third is the data-driven prediction method based on machine learning, and the fourth is the prediction method based on the traditional mathematical statistics model. is the method of combining forecasting models. [0003] Since the 1960s, people began to study the uncertainty in the water environment system, and established various types of random water pollutant simulation models, among which the stochastic analysis techniques that have been proposed mainly include random walk, Markov chain, Kalman filter,...

AI Technical Summary

Problems solved by technology

[0003] Since the 1960s, people began to study the uncertainty in the water environment system, and established various types of random water pollutant simulation models, among which the stochastic analysis techniques that have been proposed mainly include random walk, Markov chain, Kalman filter, a Although the water quality prediction model can be used to predict the change trend of pollutants in the water body, it is only suitable for simulation prediction on a small time scale.

Forecasting methods based on traditional mathematical statistical models, such as exponential smoothing method, time series analysis method, etc., are simple in model calculation and have a fast prediction speed, but the prediction ability for complex nonlinear, non-stationary or strong random data is relatively weak. Difference

With the rapid development of machine learning, especially deep learning, data-driven machine learning prediction methods have attracted extensive attention from experts and scholars in the field. Compared with traditional methods, they have higher prediction accuracy in nonlinear data prediction. However, machine learning algorithms generally have the problem that hyperparameters are difficult to determine, and different machine learning models have large differences in actual predictions

The method of combining several models for water quality prediction can combine the advantages of each model to make up for the shortcomings of a single model to make water quality prediction more accurate. One method of combining prediction models is to obtain the final combined prediction by weighting the prediction results of different models As a result, another method is to preprocess the original sequence, decompose it into multiple components of different time scales by using wavelet decomposition, empirical mode decomposition (EMD) and ensemble empirical mode decomposition (EEMD), and then analyze each The prediction models of each component are respectively established, and the prediction results of each component are superimposed to obtain the final prediction value, but the selection of different wavelet bases for wavelet decomposition has a great influence on the decomposition results, and EMD and EEMD will have endpoint effects, mode mixing and noise. remaining issues

Images