In order to improve the performance of fault isolation and diagnosis of Principal Component Analysis (PCA) based methods, this article proposes a novel fault detection and isolation approach using the Structured Joint Sparse PCA (SJSPCA). The objective function involves two regularization terms: the $l_{2,1}$ norm(sum of the Euclidean norms of the rows of a matrix) and the graph Laplacian. By imposing the $l_{2,1}$ norm, SJSPCA is able to achieve row-wise sparsity, and the graph Laplacian incorporates variable correlation information. The row-sparsity property of $l_{2,1}$ norm ensures that the score indices associated with normal variables approach zero and the graph Laplacian helps the isolation of correlated faulty variables. Once a fault is detected, a two stage fault isolation strategy is considered and a score index is calculated for each variable. It is proved that the proposed two stage strategy is capable of isolating faulty variables. The improved fault isolation performance of SJSPCA is illustrated by a simulation example and a gas flow fault observed in an industrial blast furnace.

It is of critical importance to keep a steady operation in the blast furnace to facilitate the production of high quality hot metal. In order to monitor the state of blast furnace, this article proposes a fault detection and identification method based on the multidimensional Gated Recurrent Unit (GRU) network, which is a kind of recurrent neural network and is highly effective in handling process dynamics. Comparing to conventional recurrent neural networks, GRU has a simpler structure and involves fewer parameters. In fault detection, a moving window approach is applied and a GRU model is constructed for each process variable to generate a series of residuals, which is further monitored using the support vector data description (SVDD) method. Once a fault is detected, fault identification is performed using the contribution analysis. Application to a real blast furnace fault shows that the proposed method is effective.

The background of the present study complies with silicon content prediction in hot metal in the blast furnace system. The blast furnace system is a highly complex industrial reactor in the conventional process. The system is subject to several problems (e.g., system automation, the thermal state of the blast furnace, and the life prediction of blast furnace) that should be addressed by professionals. To determine the prediction state of the heat in the blast furnace, the silicon content in the blast furnace molten iron commonly acts as a key indicator. Based on the assumption that the blast furnace system exhibits a stable state, the accuracy of hot metal silicon is analyzed by using a range of machine learning algorithms. In the present study, two derivative algorithms of gradient boosting decision tree are adopted to develop a strong boosting predictor based on the extreme gradient boosting (XGBoost) algorithm and the light gradient boosting machine (LightGBM) algorithm for prediction. Compared with the conventional algorithms (e.g., lasso, random forest, support vector machine and gradient boosting decision tree), the prediction by using the two boosting algorithms is capable of more effectively guiding and determining the state of the blast furnace. As revealed from experimentally simulated results, the mentioned two boosting algorithms exhibit better comprehensive prediction performance than the conventional algorithms on the datasets of two practical blast furnace systems, demonstrating that the R-square of the two blast furnaces in the training set is over 0.7. The mentioned two algorithms are of certain guiding significance for exploring blast furnace problems.

The proliferation of remote sensing imagery motivates a surge of research interest in image processing such as feature extraction and scene recognition, etc. Among them, scene recognition (classification) is a typical learning task that focuses on exploiting annotated images to infer the category of an unlabeled image. Existing scene classification algorithms predominantly focus on static data and are designed to learn discriminant information from clean data. They, however, suffer from two major shortcomings, i.e., the noisy label may negatively affect the learning procedure and learning from scratch may lead to a huge computational burden. Thus, they are not able to handle large-scale remote sensing images, in terms of both recognition accuracy and computational cost. To address this problem, in the paper, we propose a noise-resilient online classification algorithm, which is scalable and robust to noisy labels. Specifically, ramp loss is employed as loss function to alleviate the negative affect of noisy labels, and we iteratively optimize the decision function in Reproducing Kernel Hilbert Space under the framework of Online Gradient Descent (OGD). Experiments on both synthetic and real-world data sets demonstrate that the proposed noise-resilient online classification algorithm is more robust and sparser than state-of-the-art online classification algorithms.

Frequent trend change is an important feature of financial high-frequency data. In order to model the trend of financial high-frequency data, neural networks have been widely used. However, most research only considers the network performance under a single trend, hence the research results are not widely universal. In this regard, this article tests the performance Long-short Term Memory (LSTM) using three types of SSE 50 ETF datasets, corresponding to ascending, descending, and stationary trading scenarios. Considering the fact that the LSTM network has low computation efficiency due to the need for tuning a large number of hyperparameters, this paper proposed a fast sub-step Grid Search (SGS) method to optimize the model parameters, resulting in higher efficiency and smaller root mean square errors. It is found that comparing to Traditional LSTM networks, the SGS-LSTM network performs better in the trended datasets, but not as well in the stationary dataset, which indicates that multiple models are required to handle different trends in financial high-frequency time series.