Open AccessPrediction of heavy metal Cd content in basin soil with time series input
Author(s) -
Shengwei Wang,
Yulin Zhan,
Hao Ji,
Ping Li
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/804/4/042079
Subject(s) - artificial neural network , recurrent neural network , content (measure theory) , mean absolute percentage error , mean squared error , series (stratigraphy) , time series , computer science , dimension (graph theory) , artificial intelligence , data mining , machine learning , mathematics , statistics , geology , mathematical analysis , paleontology , pure mathematics
The content of heavy metal Cd in basin soil is important to human health and environmental management. Traditional pollution prediction models have problems such as a single input dimension and the inability to retain historical legacy information of time series data. This paper uses high-dimensional time series data as model inputs. A Backward Propagation Network (BP) and two variants of Recurrent Neural Network (RNN) named Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks were employed to develop Cd content prediction models of basin soil heavy metal. According to the experimental results, we explored the performance of different models about the prediction of the heavy metal Cd content. Experimental results show that the BP neural network converges faster but has a greater error. Compared with the BP neural network, the LSTM neural network error is decreased by 0.0895, 0.3124, 0.0159, 0.1533 and the GRU neural network error is decreased by 0.0743, 0.2985, 0.0259, 0.1441 in term of MAE, MAPE, MSE, RMSE. The experimental results showed the GRU is more efficient in time and space compared to the LSTM neural network. It is more suitable for high-precision content prediction of the heavy metal Cd.