Structure Learning of Deep Gaussian and Non-Gaussian Information Fusion Framework for Automated Predictive Data Analytics.

In the past years, latent variable models have played an important role in various industrial AI systems, among which quality prediction is one of the most representative applications. Inspired by the idea of deep learning, those basic latent variable models have been extended to deep forms, based on which the quality prediction performance has been significantly improved. However, different latent variable models have their own strengths and weaknesses, a model works well under one scenario might not provide satisfactory performance under another. The motivation of this article is based on the viewpoint of information fusion and ensemble learning for heterogeneous latent variable models. Particularly, a collaborative deep learning and model fusion framework is formulated for the purpose of industrial quality prediction. In the first stage of the framework, collaborative layer-by-layer feature extractions are implemented among different latent variable models, through which different patterns of latent variables are identified in different layers of the deep model. Then, in the second stage, an ensemble regression modeling strategy is proposed to fuse the quality prediction results from different latent variable models, which is based on a well-designed data description method. Two real industrial examples are used for performance evaluation of the proposed method, based on which we can observe that information fusions in terms of both collaborative layer-by-layer feature extraction and heterogeneous model ensemble have positive effects in improving prediction accuracy and stability.

To keep a long-term running for a deep learning based soft sensor in time-varying industrial processes, the main challenges lie in the frequent update of model structure and the heavy computation burden introduced by the deep learning algo-rithm. In this paper, a lightweight form of the deep learning model is introduced for recursive learning of a latent variable based soft sensor. Based on the fully automated deep latent vari-able network, both of the model structure and parameters can be automatically updated through the recursive learning process. As a result, the two main difficulties of developing a long-term run-ning soft sensor model can be well alleviated, while simultaneously the online prediction performance can be significantly improved by the frequently updated soft sensor in time-varying processes. Two real industrial examples are provided to validate the feasi-bility and effectiveness of the proposed method. Compared to the static form of the deep regression model, the soft sensing perfor-mance has been significantly improved during the recursive learning process.

Abstract. Deep-learning frameworks can effectively forecast the air pollution data for individual stations by decoding time series data. However, most of the existing time-series-based deep-learning models use offline spatial interpolation strategies and thus cannot reliably project the station-based forecast to the spatial region of interest. In this study, the station-based long short-term memory (LSTM) technique was extended for spatial air quality forecasting by combining a novel deep-learning layer, termed the broadcasting layer, which incorporates a learnable weight decay parameter designed for point-to-area extension. Unlike most existing deep-learning-based methods that isolate the interpolation from the model training process, the proposed end-to-end LSTM broadcasting framework can consider the temporal characteristics of the time series and spatial relationships among different stations. To validate the proposed deep-learning framework, PM2.5 and O3 forecasts for the next 48 h were obtained using 3D chemical transport model simulation results and ground observation data as the inputs. The root mean square error associated with the proposed framework was 40 % and 20 % lower than those of the Weather Research and Forecasting–Community Multiscale Air Quality model and an offline combination of the deep-learning and spatial interpolation methods, respectively. The novel LSTM broadcasting framework can be extended for air pollution forecasting in other regions of interest.

For the purpose of applying information technology to the creation of ink style painting, the algorithm of ink painting rendering based on the deep learning framework and convolutional neural network model is designed and improved. Firstly, the ink style rendering program is written in Python. Secondly, VGG under Caffe architecture and Illustration 2Vec models are transplanted to TensorFlow architecture, and the image is rendered in ink style based on deep learning framework and convolutional neural network model. Finally, based on Node.js, the server-side program for image ink style rendering is built. Among them, Express is adopted as the Web-side framework, and the front-end page effect is completed. The results show that the ink rendering logic program is applicable, and the expected purpose is achieved.

The execution of a human movement involves different muscles that are activated and coordinated by the brain at different temporal scales in a complex cognitive process. For this reason, studying human motion requires to properly model multiple temporal scales that fully describe its complexity. Current approaches are not able to address this requirement properly or are based on oversimplified models with obvious limitations. Data-driven models represent research frontiers able to provide new insights. In this work we will investigate different data-driven approaches. The first one is based on shallow models that, while achieving reasonably good recognition performance, require to handcraft features according to the domain knowledge The second one is based on deep models that can be extended to manage multiple temporal scales but they are hard to exploit as too many architecture configurations exist. For this reason, we will propose a new deep multiple temporal scale data-driven model, based on Temporal Convolutional Network, capable of learning features from the data at different temporal scales, of outperforming state of the art deep and shallow models, and of exploiting shallow models to tune the architecture configuration. We designed, collected data and tested our proposal in a specially devised experiment, to prove the validity of our approach. In particular, we collected motion capture data about dyad actions where two people exchange a ball. As the weight of the ball and the throwing intentions change, we will show how it is possible to automatically detect either the weight of the ball or the intention behind the throw just based on motion data. Data regarding our experiment and code of the methods proposed in this work are also made freely available to the research community. Results support both the proposal and the need for the use of deep multi scale models as a tool to better understand human movement and its multiple time scale nature.

An efficient deep learning-based workflow for real-time CO2 plume visualization in saline aquifer using distributed pressure and temperature measurements

There are many methods for dealing with spatial uncertainty. However, a separate discussion for attribute uncertainty of geographic information is still rare. With the rapid development of multiple sources of geographic data, the uncertainty of attributes and their fusion is important for accuracy assessment and decision making. Attribute uncertainty of geographic information is an important aspect of geographic information quality and will further affect geographic decision making. This paper has discussed the different types of attribute measurement and uncertainty of geo-data and information fusion schemes for multiple sources of geo-data. Two types of attributes, i.e. continuous variable attribute and categorical variable attribute, and their uncertainty, i.e. random and error, vagueness and ambiguity, are discussed. Five geographic information fusion schemes are given for multi-sources uncertainty attribute information, i.e. (a) univariate multi-sources data fusion, (b) multivariate continuous attributes fusion, (c) multivariate categorical attribute fusion, (d) multivariate normalized attribute fusion and (e) multivariate continuous-category attribute fusion. Methods for different fusion schemes are discussed, i.e. maximum likelihood estimation, multivariate regression, Artificial Neural Network, Dempster-Shafer evidence, multi-attribute decision making (MADA), and ordered weighted averaging (OWA). Finally an example of information fusion for precipitation estimation and flood susceptibility assessment are discussed.

Accurate aircraft engine fault detection and diagnosis is vitally important reducing operating costs and improving safety. Various data and knowledge could be collected from manufacture, test bed measurement systems, on-board measurement systems, maintenance history and experts' experience. Integrating and fusing these data and information to provide intelligent fault diagnosis and maintenance schedules are essentially to both civil and military engines. Information fusion strategies and architectures have been developed over the last several years for improving upon the accuracy, robustness and overall effectiveness of diagnostic and prognostic technologies. Fusion of relevant sensor data, maintenance database information, and various diagnostic and prognostic technologies has been proven effective in reducing false alarm rates, increasing confidence levels in early fault detection, and predicting time to failure or degraded condition requiring maintenance action. In this paper, we researched on the architectures of fusion systems. A four-level model was presented to fit the usage in aircraft engines fault detection and diagnosis. To generate the deviations for gas path fault detection, baseline values should be chosen firstly. We analyzed choosing baseline values and generating deviations in detail. Then, the fuzzy set was introduced to descript the degree of symptoms belonging to fault patterns. A method based on fuzzy set to isolate fault and quantify the deterioration of performance of components was presented. We also demonstrated the method with an example. For this method, it is not necessary to have more measurements than fault patterns. The fault diagnosis system based on it is very easy to construct and extend too

The super-resolution algorithms based on deep learning can effectively increase optical remote sensing image (ORSI) details for further analysis tasks. Deep unfolding methods have been studied in recent years to bridge the gap between optimization-based and learning-based methods, which can take good advantage of the prior knowledge. However, these unfolding methods usually ignore the utilization of intermediate network features between different iteration stages, thereby limiting the performance of super-resolution results. We propose a multi-source information fusion network (MSFNet) for ORSI super-resolution to address this problem. We mainly consider three strategies to enhance the image super-resolution performance, including feature extraction strategy, information fusion strategy, and the structure of the unfolding network. Firstly, image information of various scales is helpful for mining potential features of images for image super-resolution. Therefore, we introduce multi-scale implicit constraints to the objective function. Secondly, we unfold the optimization process into a neural network by alternating direction method of multipliers (ADMM). The network framework follows the ADMM method's iteration process. This unfolding strategy can effectively utilize the prior information for image reconstruction. Thirdly, we propose a row-column decoupling Transformer module based on this unfolding framework for feature fusion. Specifically, the row Transformer block completes the feature fusion of various scales, and the column Transformer block completes the feature fusion of various channels. The fused features are transmitted to the next iteration stage for feature enhancement. And this fusion strategy enables the network to focus on the global information of the features. We perform experiments on three remote sensing image datasets to fully demonstrate the algorithm's effectiveness. Experiment results show that the proposed algorithm can achieve better image reconstruction performance.

Emotion recognition has become an important research area for advanced human-robot interaction. Through recognizing facial expressions, a robot can interact with a person with a more friendly manner. In this paper, we proposed a bimodal emotion recognition system by combining image and speech information. A novel information fusion strategy is proposed to set proper weights to two feature modalities based on their recognition reliability. The fusion weights are determined by the distance between test data and the classification hyperplane and the standard deviation of training samples. After normalization using the mean distance between training samples and the hyperplane, the fusion weight is set to represent the classification reliability of individual modality. In the latter bimodal SVM classification, the recognition result with higher weight is selected. The complete procedure has been implemented in a DSP-based system to recognize five facial expressions on-line in real time. The experimental results show that a recognition rate of 86.9% is achieved, an improvement of 5% compared with using only image information.

An integrated framework of input determination for ensemble forecasts of monthly estuarine saltwater intrusion

Expressway lane change strategy of autonomous driving based on prior knowledge and data-driven

Electrolysis systems are critical to several societal applications, particularly energy storage and conversion. Developing these systems requires a detailed knowledge of the chemistry and thermodynamics of the materials used in the electrolysis cell. This work focuses on using embedded scientific machine learning as an efficient way to build an interpretable model for the reaction and transport kinetics in the LSCF electrode, whose performance directly influences the electrolysis system’s performance. The models developed in this study are trained using the publicly available machine learning package, FoKL-GP. This package incorporates a robust Gibbs sampler that employs a forward variable selection process to construct Karhunen–Lo`eve decomposed Gaussian Process (GP) functions in an automated manner. The model training process uses experimental data from open literature and scientifically derived models to obtain GP functions that accurately represent vital parameters such as the equilibrium constant, rate constant, and diffusivity in LSCF. The bulk defect equilibrium model is derived by interpreting the half-reaction at the LSCF electrode and computed using thermogravimetric measurements. Once quantified, a model for the equilibrium constant is built as a function of the site fraction of oxygen vacancies in LSCF and the temperature. Using data from electrical conductivity relaxation (ECR) experiments, GP functions for the rate constant and diffusivity are built as a function of the site fraction of oxygen vacancies in LSCF and the temperature. The data-driven models for the LSCF electrode can be paired with models for the other components in the overall cell to optimize the electrolysis system. Embedded scientific machine learning is an advancement in the design of chemical processes, offering precise and predictive insights crucial for optimizing performance and efficiency. By harnessing datasets from both experimental and theoretical sources, these models provide a solid foundation for informed decision-making. This approach deepens our understanding of complex chemical interactions and fosters innovation by identifying optimal conditions, reducing trial-and-error experimentation, and paving the way for more sustainable and cost-effective solutions.

The digital revolution in the cultural and creative industries is the factor, which has radically changed the way the audience communicated with the cultural content and organizations. Digital platforms are becoming a more and more significant instrument which cultural organizations utilize in order to connect with more people, become more participatory, such as websites, social media, mobile applications, cultural archives on the Internet. It has been observed that data analytics are a huge source, in this case, to understand the audience behavior, increase the engagement strategy, and evidence-based decision-making. The paper research analyzes how data analytics is used in the engagement and cultural participation of the audience and further offers a comprehensive outline of data analytics that should be used by culture institutions. The first analysis of the paper is the currently existing data-driven models of engagement, including audience analytics frameworks, smart cultural analytics systems, and engagement modeled on the basis of personalization. The elementary limitations and gaps in research will be conducted through comparison analysis to identify gaps in research in the current systems. Based on such findings, a hierarchical data analytics framework can be proposed, which includes many components, including cultural data sources, data collection and data integration module, scalable data storage platform, analytics and machine learning, visualization dashboards, and decision-making operations. The proposed framework can enable cultural institutions to accumulate, process, and analyze the audience data on different online platforms to generate meaningful insights into the audience preferences and activity patterns. Evaluation and key performance indicators measurements are also provided to show whether the given framework was successful in improving the outcomes of cultural participation and engagement. The results prove that data-driven cultural analytics has the potential to enhance audience engagement, strategy-oriented approaches to cultural decisions significantly, and the sustainability of cultural organizations. The other implementation challenges are also discussed in the study and the future research directions are outlined as applying artificial intelligence, predictive analytics, immersive technologies, and smart cultural ecosystems

Data-Driven subsurface modeling technology has been proven, for the past few years, to yield technical and commercial success in several oil fields worldwide. A data-driven model is constructed for the first time for an oil field onshore Abu Dhabi, and used for evaluation of a reservoir with substantial reserves and comprehensive development plan; for the purpose of predicting production rates, dynamic reservoir pressure and water saturation, improving reservoir understanding, supporting field development optimization and identifying optimum infill well locations. The objective is to provide the asset with a decision-support tool to make better field development planning and management. The subject reservoir is a low permeability carbonate reservoir and characterized by lateral and vertical variations in its reservoir rocks and fluid properties. More than 8 years of Phase-I development and production/injection data and extensive amount of well tests and log data (SCAL, PVT, MDT) from more than 37 wells were used to construct the Data Driven Model for this asset. This new modeling technology, (TDM), integrates reservoir engineering analytical techniques with Artificial Intelligence, Machine Learning & Data Mining in order to formulate an empirical and spatiotemporally calibrated full field model. In this work, it is leveraged with other conventional reservoir modeling and management tools such as streamline modeling, isobaric maps and flooding conformance. Several analyses were performed using the full field data-driven model; complementing the existing conventional numerical model. The accomplishments of the data-driven reservoir model for this project included, but not limited to, comprehensive history matching (including blind validation) and then forecast of Oil rate, GOR, WC, reservoir pressure and water saturation, injection optimization, and choke size optimization. The results generated by the data-driven model proved to be quite eye-opening for the asset management; as the model was able to identify potential areas of improving field efficiency and cost reduction. When combined with numerical techniques, the calibrated data-driven model assist to obtain a reliable short term forecast in a shorter time and help make quick decisions on day-to-day operational optimization aspects. The use of facts (all field measurements) instead of human biases, pre-conceived notions, and gross approximations distinguishes data-driven modeling from other existing modeling technologies. Its innovative combination of Artificial Intelligence and Machine Learning (the technologies that are transforming all industries in the 21st century) with reservoir engineering, reservoir modeling and reservoir management clearly demonstrates the potentials that these pattern recognition technologies offer to the upstream oil and gas industry for its realistic digital transformation.