Hierarchical Quality-Relevant Feature Representation for Soft Sensor Modeling: A Novel Deep Learning Strategy

Yuan, Xiaofeng; Zhou, Jiao; Huang, Biao; Wang, Yalin; Yang, Chunhua; Gui, Weihua

doi:10.1109/tii.2019.2938890

Cited by 193 publications

(65 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Artificial intelligent learning methods benefit from the rapid growth of computing power and exploit artificial intelligence algorithm to learn a mapping relationship between input and output, mainly focusing on nonlinear mapping models [27][28][29]. Artificial intelligence methods have been widely used in various domains, ranging from abnormal detection [30], power grids [31,32], energy consumption [33], pattern recognition [34][35][36], and have become an excellent tool for PV power generation prediction [37,38].…”

Section: Introductionmentioning

confidence: 99%

Photovoltaic Power Forecasting With a Hybrid Deep Learning Approach

Xie

Wang

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Solar energy is the key to clean energy, which can generate large amounts of electricity for the future smart grid. Unfortunately, the randomness and intermittency of solar energy resources bring difficulties to the stable operation and management of the power systems. To reduce the negative impact of photovoltaic (PV) plants accessing on the power systems, it is great significant to predict PV power accurately. In light of this, we propose a hybrid deep learning approach based on convolutional neural network (CNN) and long-short term memory recurrent neural network (LSTM) for the PV output power forecasting. The CNN model is leveraged to discover the nonlinear features and invariant structures exhibited in the previous output power data, thereby facilitating the prediction of PV power. The LSTM is used to model the temporal changes in the latest PV data, and predict the PV power of next time step. Then, the prediction results in the two models are comprehensively considered to obtain the expected output power. The proposed approach is extensively evaluated on real PV data in Limberg, Belgium, and numerical results demonstrate that the proposed approach can provide good prediction performance in PV systems. INDEX TERMS Solar energy, deep learning, photovoltaic (PV) power forecasting, power systems.

show abstract

Section: Introductionmentioning

confidence: 99%

Photovoltaic Power Forecasting With a Hybrid Deep Learning Approach

Xie

Wang

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…In recent years, deep‐learning‐based architectures have been applied in data modeling of soft sensor . Based on quality prediction and process monitoring methods in process industries, deep quality‐related feature extraction with hybrid Variable‐Wise weighted stack auto encoder (VW‐SAE), hierarchical quality‐relevant feature representation and extended deep belief network have been developed for soft sensing modeling …”

Section: Introductionmentioning

confidence: 99%

“…36,37 Based on quality prediction and process monitoring methods in process industries, deep qualityrelated feature extraction with hybrid Variable-Wise weighted stack auto encoder (VW-SAE), hierarchical quality-relevant feature representation and extended deep belief network have been developed for soft sensing modeling. [38][39][40] Although the model-driven soft sensor is more effective for specific plants, in the previous research of acetylene hydrogenation reactor, a simple soft sensor calibration scheme based on output correction is proposed. 41 In addition, dynamic modeling of soft sensor is a significant problem that should be addressed.…”

Section: Introductionmentioning

confidence: 99%

Online estimation for catalyst activity of acetylene hydrogenation reactor

Xie

Zhang

et al. 2020

Asia-Pacific J Chem Eng

View full text Add to dashboard Cite

In the ethylene industry, the high purity of the ethylene product depends on hydrogenation in acetylene hydrogenation reactor. Because the catalyst deactivation leads to the moving of the operating point, the operation scheme must be adjusted continually according to the catalyst activity. It is necessary to estimate the catalyst activity online. Based on the discrete dynamic model of the acetylene hydrogenation reactor, the extended Kalman filter (EKF) is used to build the soft sensor for catalyst activity. Considering that EKF involves the large computation costs, we propose a method that estimates the parameters of the time‐varying deactivation kinetics model for the tradeoff of accuracy and complexity. The method is effective to reduce computation complexity of estimation, and simultaneously, the accuracy satisfies the process requirement.

show abstract

“…As a result, latent variable data modeling methods have become very popular in process data analytics. 10 For example, principal component analysis (PCA), 11 partial least squares, 12 independent component regression, 13 and recent deep learning techniques [14][15][16] are widely used to extract key information from data while simultaneously improve the efficiency of data analytic procedures. Moreover, to solve the problem that PCA is susceptible to noise interference, Tipping et al introduced the probability form into PCA and proposed probabilistic principal component analysis (PPCA).…”

Section: Introductionmentioning

confidence: 99%

An ensemble just‐in‐time learning soft‐sensor model for residual lithium concentration prediction of ternary cathode materials

Chen

Gui

Dai

et al. 2020

Journal of Chemometrics

Self Cite

View full text Add to dashboard Cite

The surface-free lithium content, which is a critical index that reflecting the quality of ternary cathode materials, usually cannot be monitored in real time due to technical restriction. To this end, soft sensor technique is applied to predict the surface-free lithium content by readily available process variables, making timely control on operation parameters. In this paper, a modeling method based on ensemble just-in-time learning (JITL) soft-sensor is developed. In the model, data feature indices are firstly designed to properly extract the real-time and short-time features between the batching and the loading procedure. Accordingly, ensemble JITL soft-sensor model is constructed with the semisupervised local weighted probability principal component regression.Moreover, considering varying working conditions, an adaptive moving window technique is adopted to improve the adaptability of the model. The validation and the flexibility of the developed modeling method are testified with the practical manufacturing data. K E Y W O R D Sjust-in-time learning, moving window, prediction of residual lithium content, semisupervised learning | INTRODUCTIONWith the growing demand for electric energy reliability in modern society, as well as the widespread use of intermittent new energy sources such as wind power and photovoltaics, electric energy storage technology has received extensive attention from all over the world. 1 Lithium-ion batteries, which have many merits such as portability, recyclability, high energy density, and long cycle life, are nowadays recognized as a key power storage material. In the process of manufacturing lithium-ion batteries, surface-free lithium content, namely, the content of lithium-ion on the surface of cathode materials, plays a decisive role in determining the characteristic of this material. Thus, high requirements are put forward for the real-time monitoring of surface-free lithium content. However, surface-free lithium content cannot be directly measured since the mensuration is inefficient, hysteresis, and costly. Therefore, it is a good choice to predict surface-free lithium content by using the soft-sensor model. 2,3 There are two most commonly used methods for building soft-sensor models, namely, mechanism modeling and data-driven modeling. The mechanism modeling method starts with the physical, chemical, and biological knowledge of the object process. Then, mathematical equations based on conservation laws such as mass conservation are formed to find the relationship between process variables and the key variables. The mechanism models have more application

show abstract

Hierarchical Quality-Relevant Feature Representation for Soft Sensor Modeling: A Novel Deep Learning Strategy

Cited by 193 publications

References 41 publications

Photovoltaic Power Forecasting With a Hybrid Deep Learning Approach

Photovoltaic Power Forecasting With a Hybrid Deep Learning Approach

Online estimation for catalyst activity of acetylene hydrogenation reactor

An ensemble just‐in‐time learning soft‐sensor model for residual lithium concentration prediction of ternary cathode materials

Contact Info

Product

Resources

About