Water quality parameter prediction method based on online sequential extreme learning machine

Abstract

The invention provides a water quality parameter prediction method based on an online sequential extreme learning machine. The water quality parameter prediction method comprises the following steps of 1 acquiring water quality parameter historical data from an automatic water quality monitoring station and performing data pre-processing; 2 using front 2 / 3 data in the pro-processed water quality parameter historical data as a learning sample set and using rear 1 / 3 data as a testing sample set; 3 taking a part of data from the learning sample set to perform OSELM initialization, and then utilizing the remaining data in the learning sample set to enable the OSELM to perform learning; 4 using the testing sample set to inspect an OSELM model after learning so as to obtain a water quality parameter prediction model based on the OSELM; 5 using the OSELM model to predict new water quality parameters. The water quality parameter prediction method is good in prediction effect, high in prediction accuracy and short in learning time, and can meet the requirement for quick water quality parameter prediction in the future development trend.

Description

technical field [0001] The invention relates to the field of water quality monitoring, in particular to a water quality parameter prediction method based on an Online Sequential Extreme Learning Machine (OSELM for short). Background technique [0002] Establishing an effective water quality parameter prediction model can accurately point out the degree of pollution of the water body and its development trend in the future, thereby providing a reliable basis for formulating policies and specific measures for water environmental protection, and can change people's traditional concepts of water environment application and governance , Transform the previous post-event governance into pre-event prevention. Although people can find a certain degree of change law from the long-term continuous observation data, because the water quality parameters are affected by many factors, the change shows a high degree of nonlinearity, time delay and uncertainty, and the change law is complex....

AI Technical Summary

Problems solved by technology

Although people can find a certain degree of change law from the long-term continuous observation data, because the water quality parameters are affected by many factors, the change shows a high degree of nonlinearity, time delay and uncertainty, and the change law is complex. It is difficult to express with a mechanism model, and traditional time series prediction methods such as autoregressive (AR), autoregressive moving average (ARMA) and other limited parameter linear models cannot well characterize its variation characteristics

The emergence of intelligent models such as artificial neural networks provides a new method for water quality parameter prediction, but because artificial neural networks use gradient-based learning algorithms and adjust the parameters in the network iteratively, there are the following defects: (1) Learning The speed is slow, which cannot meet some occasions that require online learning; (2) the learning process tends to converge to local minimum points, and it is difficult to guarantee the learning accuracy; (3) the network generalization ability is poor, and it cannot ensure good prediction results for unknown samples

The existence of these defects is not conducive to the rapid and accurate prediction of water quality parameters by artificial neural network, which restricts the popularization and application of artificial neural network in water quality parameter prediction.

Extreme Learning Machine (Extreme Learning Machine, ELM) is a new learning algorithm proposed by Huang et al. in 2006. In this algorithm, the input weight and hidden layer threshold are randomly selected, while the output weight is determined by the least square method It is directly calculated, and the whole process is completed at one time without iteration, so the learning time is short, and it also effectively avoids the defect that artificial neural network learning is easy to fall into local minima, but in practical applications, all data may not be added at one time Yes, when new data is added, the ELM algorithm will relearn new data and old data together, so that the learning process takes too much time

Images