Dissolved oxygen prediction on time series pond data
Answers
Answered by
0
Highlights
•
We proposed CSELM model to predict dissolved oxygen change from time series data.
•
CSELM prediction model was verified by experiments with two counterpart models.
•
We used clustering method to capture characteristics of data in similar time slots.
•
CSELM achieved satisfied efficiency and accuracy in dissolved oxygen prediction.
Abstract
Accurate and efficient prediction of dissolved oxygen from time series data is critical for aquaculture that needs intelligent management and control. However, data streams of dissolved oxygen that are nonlinear and continuously generated challenge the existing prediction methods. This paper provides a novel Clustering-based Softplus Extreme Learning Machine method (CSELM) to accurately and efficiently predict dissolved oxygen change from time series data. The CSELM adopts k-medoids clustering to group the dataset into different clusters based on Dynamic Time Warping (DTW) distance, and uses a new Softplus ELM algorithm to discover a common trend in a cluster of time series pieces (within the same period) and then predict the future trend. The Softplus ELM improves ELM using a new activation function, Softplus, to solve the nonlinear and continuous problems of time series data streams and adopting partial least squares (PLS) to avoid the instability of output weight coefficients. The DTW based clustering in CSELM improves the efficiency while tolerating some data loss and uncertain outliers of sensor time series. Softplus based on PLS optimizes the performance and increases the accuracy of ELM. We have demonstrated that CSELM achieves better prediction results than PLS-ELM and ELM models in terms of accuracy and efficiency in a real-world dissolved oxygen content prediction.
•
We proposed CSELM model to predict dissolved oxygen change from time series data.
•
CSELM prediction model was verified by experiments with two counterpart models.
•
We used clustering method to capture characteristics of data in similar time slots.
•
CSELM achieved satisfied efficiency and accuracy in dissolved oxygen prediction.
Abstract
Accurate and efficient prediction of dissolved oxygen from time series data is critical for aquaculture that needs intelligent management and control. However, data streams of dissolved oxygen that are nonlinear and continuously generated challenge the existing prediction methods. This paper provides a novel Clustering-based Softplus Extreme Learning Machine method (CSELM) to accurately and efficiently predict dissolved oxygen change from time series data. The CSELM adopts k-medoids clustering to group the dataset into different clusters based on Dynamic Time Warping (DTW) distance, and uses a new Softplus ELM algorithm to discover a common trend in a cluster of time series pieces (within the same period) and then predict the future trend. The Softplus ELM improves ELM using a new activation function, Softplus, to solve the nonlinear and continuous problems of time series data streams and adopting partial least squares (PLS) to avoid the instability of output weight coefficients. The DTW based clustering in CSELM improves the efficiency while tolerating some data loss and uncertain outliers of sensor time series. Softplus based on PLS optimizes the performance and increases the accuracy of ELM. We have demonstrated that CSELM achieves better prediction results than PLS-ELM and ELM models in terms of accuracy and efficiency in a real-world dissolved oxygen content prediction.
Similar questions