ELM-Based AFL–SLFN Modeling and Multiscale Model-Modification Strategy for Online Prediction

Abstract

Online prediction of key parameters (e.g., process indices) is essential in many industrial processes because online measurement is not available. Data-based modeling is widely used for parameter prediction. However, model mismatch usually occurs owing to the variation of the feed properties, which changes the process dynamics. The current neural network online prediction models usually use fixed activation functions, and it is not easy to perform dynamic modification. Therefore, a few methods are proposed here. Firstly, an extreme learning machine (ELM)-based single-layer feedforward neural network with activation-function learning (AFL-SLFN) is proposed. The activation functions of the ELM are adjusted to enhance the ELM network structure and accuracy. Then, a hybrid model with adaptive weights is established by using the AFL-SLFN as a sub-model, which improves the prediction accuracy. To track the process dynamics and maintain the generalization ability of the model, a multiscale model-modification strategy is proposed. Here, small-, medium-, and large-scale modification is performed in accordance with the degree and the causes of the decrease in model accuracy. In the small-scale modification, an improved just-in-time local modeling method is used to update the parameters of the hybrid model. In the medium-scale modification, an improved elementary effect (EE)-based Morris pruning method is proposed for optimizing the sub-model structure. Remodeling is adopted in the large-scale modification. Finally, a simulation using industrial process data for tailings grade prediction in a flotation process reveals that the proposed method has better performance than some state-of-the-art methods. The proposed method can achieve rapid online training and allows optimization of the model parameters and structure for improving the model accuracy.Processes 2019, 7, 893 2 of 23 Two kinds of methods are typically used to solve model-mismatch problems [5,6]. The first involves retraining the model using the most recent samples. For example, Feng et al. [6] used the object-reidentification method to deal with object changes. The second kind involves adjusting the structure and parameters of the model according to the original model [7][8][9]. Reidentification of the model parameters cannot always maintain good model performance, because the model structure is important. Thus, updating the model at different scales is necessary. For online updating, the computation time of the algorithm is also significant.As for retraining and updating the model parameters, in a common method, the most recent samples are used to re-estimate the model parameters offline or online. For example, Feng et al. [6] used an object-reidentification method to deal with changes in the object model and object-mismatch problems. Wang et al.[10] used a four-step recursive algorithm to estimate the weights of a neural network. Here, online spatiotemporal measurements were performed to obtain time-varying model parameters.In recent years, for selecting relate...

Development of hybrid SVM-FA, DT-FA and MLR-FA models to predict the flexural strength (FS) of recycled concrete

“…• However, the above methods used to evaluate the flexural strength of recycled aggregate concrete has disadvantages as follow: • Laboratory testing involves casting, curing and testing samples, which requires a large amount of cost, substantial effort and time (Jamei et al, 2021), especially considering the influence of multiple factors on the flexural strength of recycled aggregate concrete, the amount of work will increase exponentially. • Some researchers have realized that the prediction effect of hybrid machine learning models is better than that of single machine learning models (Guo and Wang, 2017;Wang et al, 2019;Zhu et al, 2021;Hasanipanah et al, 2022), and studied the evaluation effect of hybrid machine learning models, but it is necessary to compare the evaluation effect of different hybrid machine learning models and select the model with higher prediction effect. • In order to compare the prediction effect of different hybrid machine learning models on the FS of recycled concrete, and select the model with high prediction accuracy for engineers as an environmentally friendly tool to evaluate the FS of recycled concrete, this study proposed to use the SVM-FA, DT-FA and MLR-FA to predict the FS of recycled concrete.…”