Search

Abstract Soft sensor plays a decisive role in making control strategies and production plans. However, the difficulty in establishing accurate and robust soft sensors using an individual model is continuously increasing due to the increasing scale and complexity in modeling data. To handle this problem, an effective ensemble model using partial least squares regression (PLSR) integrated with extreme learning machine (ELM) with multiple activation functions (PLSR-MAFELM) is proposed in this paper. The proposed PLSR-MAFELM is simple in construction: firstly, train several ELM models assigned with different activation functions using the least squares solution; secondly, combine ELM models for enhancing accuracy and stability performance; finally, obtain the optimal ensemble outputs by aggregating the outputs of individual ELM models using PLSR. To test the performance of the proposed PLSR-MAFELM model, a UCI benchmark dataset and two real-world applications are selected to carry out simulation case studies. Simulation results show that PLSR-MAFELM can achieve good stability and accuracy performance, which indicates that the generalization capability of soft sensors can be improved through combining some single models. Highlights • A multi-activation functions based extreme learning machine ensemble is proposed. • The ensemble model utilizes different and effective nonlinear activation functions. • Partial least squares regression is used to aggregate the outputs of single models. • The ensemble model can be accurate, stable and efficient in intelligent measurement. • Simulation results confirm the effectiveness of the proposed ensemble framework. [ABSTRACT FROM AUTHOR]

Most of existing modeling methods are based on point prediction. However, the accuracy of point prediction cannot meet the actual demand due to existence of high noise, volatility, complexity and irregularity inherent in the chemical process data. In order to solve this problem, a hybrid high-quality prediction intervals (PIs) method integrating complementary ensemble empirical mode decomposition (CEEMD), sample entropy (SE), and improved multi-kernel relevant vector machine (RVM) is proposed in the paper. The proposed PIs method mainly consists of three aspects: Firstly, CEEMD is adopted to decompose the original data into several independent intrinsic mode functions (IMFS), and then SE is used to analyze the complexity of the extracted IMFs to obtain recombinant components; Secondly, an improved multi-kernel RVM (MRVM) is presented to predict recombinant components independently, in which the linear kernel and the Gaussian kernel are combined; Thirdly, the predicted components are aggregated to obtain an ensemble result using another MRVM for constructing the high-quality PIs. To verify the performance of the proposed PIs method, a purified Terephthalic acid (PTA) solvent system is selected. Comparative simulation results demonstrate that the proposed PIs method greatly outperforms on coverage probability and sharpness in all the step predictions. [ABSTRACT FROM AUTHOR]

Some key variables in the complex chemical processes are very difficult to measure due to the nonlinearity, the disturbances, and the technological limitations. In order to accurately predict the difficult-to-measure variables, soft sensor based on a novel robust bagging nonlinear model integrating improved extreme learning machine with partial least square (RB-PLSIELM) is developed. Motivated by the ensemble ideas, the proposed RB-PLSIELM model is based on the bagging ensemble scheme to combine some individual nonlinear models integrating improved extreme learning machine with partial least square (PLSIELM). The sub-data for building the individual PLSIELM model are re-sampled from the original training data using the bagging tool. The problem of over-training in the PLSELM model can be avoided by using the bagging re-sampling technology. The proposed RB-PLSIELM model was demonstrated by applying it to predicting the key variables of the Tennessee Eastman Process (TEP) and the Purified Terephthalic Acid Process (PTAP). The simulation results obtained by RB-PLSIELM are compared with those obtained by the individual PLSELM model, the ELM model, and the partial least square regression (PLSR) model. Compared with the other models, the RB-PLSIELM can achieve higher prediction accuracy and stability. [ABSTRACT FROM AUTHOR]

With the increasing complexity of large-scale industrial production processes, the number of variable factors is increasing. As a result, it is demanding to predict process key variables accurately. Currently, most of soft sensor models using support vector regression and artificial neural networks are based on point prediction. The soft measurement models using the technique of point prediction can only track or fit set values. It is difficult to deal with the problem of system uncertainty and to make reliability analysis using the point prediction based soft sensors. To address this problem, this paper proposes a development method of soft sensor using the technique of prediction intervals. Under this condition, the prediction intervals instead of the point prediction of the stable operation of the industrial process system are used. The interval boundaries of the trend change can be utilized to quantify and estimate the associated uncertainty. The proposed prediction intervals based soft sensor is based on fuzzy information granularity and improved recurrent extreme learning machine. First, the fuzzy information granularity is adopted to get the lower bound, trend and upper bound of the interval. Secondly, an improved recurrent extreme learning machine is built to further enhance the ability of prediction intervals. In the improved extreme learning machine model, a feedback layer is adopted to store the hidden layer output, calculate the data trend change and dynamically update the outputs of the feedback layer. Third, the comprehensive interval evaluation function is used to evaluate the rationality of the interval results. Through case studies using a University of California Irvine dataset and the purified Terephthalic acid solvent system, the provided prediction intervals method can directly generate the upper and lower bounds for process key variables with high accuracy. • Novel soft sensor development based on prediction intervals is presented. • PIs using fuzzy information granularity and IRELM is proposed to predict variables. • Simulation results confirm the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]