Data-driven Prediction Method Research Articles

Theory-guided deep neural network for boiler 3-D NOx concentration distribution prediction

Timely and accurate three-dimensional (3-D) NOx concentration distribution prediction is essential for achieving low-emission and efficient operation in power plants. This study proposed a theory-guided data-driven prediction method for the 3-D NOx concentration distribution prediction. Firstly, the method created a foundational dataset by fusing numerical simulation data from the computational fluid dynamics (CFD) with operational data from the distributed control system (DCS). Then, the data was classified into three load condition categories, and the center operating conditions for each category were computed separately. Subsequently, the K-means algorithm was employed to extract representative data to address the computational challenges associated with big data. Finally, a Theory-Guided Deep Neural Network model (TG-DNN) was established leveraging the principle of carbon element mass conservation and deep neural network. Experimental results demonstrate that the method effectively monitors the 3-D NOx concentration distribution, potentially facilitating efficient production processes.

Efficient data-driven event-triggered predictive control for cyber-physical systems

As a multi-dimensional complex system that integrates computing, network, and physical environments, cyber-physical system (CPS) involves a large amount of system data in its real-time control process. The efficiency of data processing and transmission in CPS plays an important role in its control performance, however, few results have addressed such problems. This paper proposes a novel data-driven event-trigger mechanism to alleviate the data processing and transmission load for the data-driven predictive control method in CPS. The design of the event-triggered law can be transformed into solving a linear matrix inequality (LMI) based on the measured system data, while ensuring the stability of the integrated system. The improved performances are illustrated by the explicit simulation results, which indicates its potential application in CPS.

Short-term cooling load prediction for central air conditioning systems with small sample based on permutation entropy and temporal convolutional network

In air-conditioning load predicting, the mainstream methods mainly include mechanism model based methods and data-driven prediction methods. The mechanism model can achieve accurate prediction, but are normally used to predict non-existent buildings and by default have to make several assumptions regarding post occupancy. This study proposes a novel data-driven short-term cooling load prediction method based on historical data of the existing buildings to improve the prediction performance when the size of the training sample datasets, including load data and air conditioning internal operation data, is small. Firstly, an improved version of variational mode decomposition (IVMD) technique is proposed to deconstruct complex historical cooling load signals. Then, the permutation entropy and Savitzky-Golay (PE-SG) smoothing method is developed to extract singles with distinct characteristics and high regularity, which comprises multiple single-component amplitude and frequency modulation (AFM) signals, thus mitigating the interference of noise in predictive accuracy. Subsequently, a data-driven model based on temporal convolutional network (TCN) is constructed, which effectively alleviates the risks of over-fitting and under-fitting. Finally, the efficacy of the proposed method is validated through two case studies, and the obtained results show that the issues confronted by small sample datasets are adeptly solved by this method. The precision of prediction is significantly enhanced according to the comparison to other traditional models, showcasing greater stability and generalization ability.

Short-term traffic flow prediction based on vehicle trip chain features

ABSTRACT Short-term traffic flow prediction can improve the efficiency of transportation operations. Historical data-driven prediction methods have been proved to perform well. However, saturated or oversaturated traffic operations cannot be accurately predicted based only on detector data from a single intersection. This study proposes a short-term traffic prediction method based on vehicle trip chain features. First, the video data is pre-processed and quality assessed. Then, vehicle trip chain features are mined to correlate upstream and downstream intersections.Convolutional neural networks and long-short-term-memory model are built next. The model is launched to train the predictor and output the traffic flow for all turns at each approach to the intersection. After cases we demonstrate that the prediction accuracy of CNNs-LSTM is usually better than other methods, especially during oversaturation. In addition, we demonstrate that vehicle trip chain features can improve prediction accuracy and shorten the time consumed by the model.

A two-step data-driven method for predicting the wear of train wheel treads using GA-BPNN and LSTM algorithms

ABSTRACT During train operations, the changing of wheel tread profile due to wheel-rail contact has a significant impact on the safe operation of the train and passenger comfort. Therefore, it is crucial to timely monitor the condition of the wheels and accurately predict wheel wear. In addition, wheel wear prediction can also guide wheel re-profiling, thereby extending the service life of wheels and minimizing operating costs. In this paper, a two-step data-driven wear prediction method was proposed to comprehensively predict wheel wear from two perspectives: the overall train and individual wheels. Firstly, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Radial Basis Function neural network (RBFNN) were used to preprocess historical data, remove outliers, and construct time series data. By analysing the wear trend of wheels, a two-step data-driven prediction method was proposed. The first step is to predict the average Wear of wheels in the future time, and the second step is to predict the deviation between each wheel Wear value and the mean in the future time. Different prediction models were utilized and compared to obtain the best predictive performance. Results show that back propagation neural network (BPNN) optimized by genetic algorithm (GA) and long short-term memory (LSTM) based model yield better performance for predicting the wear trend of the overall train and individual wheels respectively. In addition, comprehensive contour parameters were utilized in the prediction model to enhance the accuracy and robustness.

Data-model linkage prediction of tool remaining useful life based on deep feature fusion and Wiener process

Accurately predicting the tool remaining useful life (RUL) is critical for maximizing tool utilization and saving machining costs. Various physical model-based or data-driven prediction methods have been developed and successfully applied in different machining operations. However, many uncertain factors affect tool RUL during the cutting process, making it challenging to create a precise physical model to characterize the degradation of tool performance. The success of the purely data-driven technique depends on the amount and quality of the training samples, it does not consider the physical law of tool wear, and the interpretability of the prediction results is poor. This paper presents a data-model linkage approach for tool RUL prediction based on deep feature fusion and Wiener process to address the above limitations. A convolutional stacked bidirectional long short-term memory network with time-space attention mechanism (CSBLSTM-TSAM) is developed in the data-driven module to fuse the multi-sensor signals collected during the cutting process and then obtain the mapping relationship between signal features and tool wear values. In the physical modeling module, a three-stage tool RUL prediction model based on the nonlinear Wiener process is established by considering the evolution law of different wear stages and multi-layer uncertainty, and the corresponding probability density function is derived. The real-time estimated tool wear of the data-driven module is used as the observed value of the physical model, and the model parameters are dynamically updated by the weight-optimized particle filter (WOPF) algorithm under a Bayesian framework, thereby realizing the data-model linkage tool RUL prediction. Milling experiments demonstrate that the proposed method not only improves RUL prediction accuracy, but also has good generalization ability and robustness for prediction tasks under different working conditions.

Data-driven prediction method for shear capacity of corroded rectangular reinforced concrete shear walls under varied failure modes

The physical and mechanical properties of rectangular reinforced concrete shear walls (RRCSW) will deteriorate over time due to acidic erosion, causing changes in damage characteristics and a reduction in shear capacity under seismic loading. The aim of this paper is to develop a failure mode-based prediction method for the shear capacity of corroded RRCSW, as the existing formulas for calculating shear strength become invalid in such scenarios. Initially, a total of 2403 RRCSW samples with various corrosion levels and main design parameters were generated and simulated using the authors' previous numerical method to obtain a comprehensive database of shear capacity. Subsequently, an efficient damage mode identification method for RRCSW was established using the Random Forest algorithm, and its accuracy and applicability in identifying damage modes of corroded RRCSW were verified. Finally, the existing calculation formulas for RRCSW in flexure-shear and shear failure modes were selected and modified to consider the corrosion effect based on the shear capacity database. The comparative analysis revealed that the proposed formulas provide accurate predictions of the shear capacity of corroded RRCSW, making the results of this study invaluable for seismic design of new structures and seismic performance evaluation of existing structures.

Calibration strategies for long-term strain forecasting in high-rise building columns

Concrete exhibits long-term time-dependent behavior due to creep and shrinkage that impacts the safety and serviceability of high-rise buildings. These rheological effects are difficult to predict due to their random nature, dependence on environmental conditions, and loading history. Current methods include the use of creep and shrinkage tests of structure-specific concrete samples to update compliance and shrinkage, and sophisticated numerical models for prediction of the long-term structural behavior. However, creep and shrinkage tests are time-consuming, and simulation requires reliable numerical models and often proprietary solvers that are not available to the structural health monitoring (SHM) practitioner. Further, uncertainty propagation in complex numerical models is rarely seen in the relevant literature. In contrast, data-driven prediction methods using SHM data and simplified analytical models have shown to be successful for prestressed concrete bridges. In this work, we investigate calibration strategies of creep and shrinkage models using SHM data toward data-driven forecasting of long-term time-dependent behavior of high-rise buildings. A calibration strategy is identified that enables significant and consistent improvement of forecasting of long-term time-dependent behavior. It is also shown that continuous calibration can provide good predictions at least 30 days ahead. First-order analytical uncertainty propagation formulas are also provided. The calibration strategies are evaluated on data from two residential high-rise buildings in Singapore. Recommendations to the SHM practitioners are also given.

A domain adversarial graph convolutional network for intelligent monitoring of tool wear in machine tools

The prediction of tool wear on CNC machine tools holds great significance for enhancing both the safety of tool machining and the quality of product machining. The current data-driven prediction methods for tool wear in machining have yielded very promising results. However, tool wear prediction models for the same machine tool under different machining conditions lack universality since the data feature space distribution under varying work conditions is different. To bridge the gap between source and target domains, data structure information is all too commonly ignored beyond domain and class labels. To attack these problems, a transfer learning method for acquiring domain-invariant features of tool wear based on a dynamic domain adversarial graph convolutional network (DDAGCN) is proposed in this paper. The network integrates three effective alignment mechanisms - domain, data structure, and class centroid alignments - and evaluates the relative importance of the marginal and conditional distributions quantitatively. In addition, a dynamic factor is introduced to quantitatively assess the relative significance of marginal and conditional distributions, which effectively facilitate the learning of domain-invariant features and decrease differences between source and target domains. The superiority of the proposed method was validated through experiments under two distinct conditions.

Associations and predictive power of dietary patterns on metabolic syndrome and its components

Background and aimsMetabolic syndrome (MetS) defines important risk factors in the development of cardiovascular diseases and other serious health conditions. This study aims to investigate the influence of different dietary patterns on MetS and its components, examining both associations and predictive performance. Methods and resultsThe study sample included 10,750 participants from the seventh survey of the cross-sectional, population-based Tromsø Study in Norway. Diet intake scores were used as covariates in logistic regression models, controlling for age, educational level and other lifestyle variables, with MetS and its components as response variables. A diet high in meat and sweets was positively associated with increased odds of MetS and elevated waist circumference, while a plant-based diet was associated with decreased odds of hypertension in women and elevated levels of triglycerides in men. The predictive power of dietary patterns derived by different dimensionality reduction techniques was investigated by randomly partitioning the study sample into training and test sets. On average, the diet score variables demonstrated the highest predictive power in predicting MetS and elevated waist circumference. The predictive power was robust to the dimensionality reduction technique used and comparable to using a data-driven prediction method on individual food variables. ConclusionsThe strongest associations and highest predictive power of dietary patterns were observed for MetS and its single component, elevated waist circumference.

Robust production forecast and uncertainty quantification for waterflooding reservoir using hybrid recurrent auto-encoder and long short-term memory neural network

History-matching-based production forecast and uncertainty quantification are essential to achieve reliable risk assessment for waterflooding reservoir in the community of petroleum engineering, however, the conventional model-based history matching procedure presents intensive computation-cost due to numerous high-fidelity reservoir simulations. The use of direct forecast approach, e.g., data-space inversion (DSI) and its variants, is urgent to achieve direct production forecast without explicitly performing history matching. This work presents a generic data-driven post-history production forecasting framework using a novel deep recurrent neural network. We construct a hybrid recurrent neural network proxy through combining recurrent autoencoder with long short-term memory neural network, e.g., referred to as RAE-LSTM. The proposed RAE-LSTM takes history prediction and post-history well controls as input, while the post-history production responses (oil and water production, water injection rate) as output. Training samples are constructed by simulating high-fidelity models parameterized with various post-history well-controls and geomodels. Once the deep recurrent neural network proxies are trained offline, the online post-history production forecast and uncertainty quantification can be efficiently achieved by input user-specific well control sequences and history measurement. The proposed proxy model is demonstrated on two examples with varying complexity, e.g., a synthetic 2D Gaussian model and a 3D channelized EGG model. The use of our proposed RAE-LSTM neural network can be regarded as a nonlinear alternative to the DSI-type of linear statistical interpolation method. The comparisons between DSI and RAE-LSTM proxy confirm that the application of our proposed data-driven prediction method can effectively obtain more robust predictions and thus support rapid and reliable decision-making during the process of real-time waterflooding production management and optimization.

Prediction of dislocation - grain boundary interactions in FCC aluminum bicrystals using a modified continuum criterion and machine learning methods

Mechanical properties of metals such as strength and toughness are strongly correlated to complex interactions between various defects in the crystalline structure. While elementary interactions between these defects have been investigated using recent micro- and nano-characterization techniques, understanding of the detailed interaction mechanisms has hardly been obtained. To understand defect-driven plasticity at various time and length scales, it is necessary to formulate a general guideline to predict both the interaction type (transmission or reflection) and the dislocation's subsequent slip system after the interaction. Many criteria based on the geometric alignment of the defects have been developed to predict this phenomenon, but these have yet to be found to be accurate when applied to general data sets of grain boundaries (GBs). With this motivation, we conduct a systematic study using molecular dynamics (MD) models of bicrystals to analyze defect interaction process between a prismatic dislocation loop and eleven different grain boundaries of the following character: three tilt, three twist, and five mixed. Based on the MD observations, two new prediction methods are developed: the first is a new data-driven parametric score function based on the classical geometric criteria, and the second is by applying Gaussian process machine learning methods to find the probability distribution of a hidden function. The proposed data-driven prediction methods could pave a new way to predict the unit interaction of dislocations with various GBs, which could show much higher accuracy compared to pre-existing geometric criteria.

Development of hybrid neural network and current forecasting model based dead reckoning method for accurate prediction of underwater glider position

Buoyancy-driven underwater gliders are considered to be advanced platforms for large-scale ocean exploration. However, it is greatly affected by ocean currents, and traditional dead reckoning navigation methods are inadequate in accurately predicting its own position, causing great difficulties in underwater target detection. To solve this problem, a navigation estimation method based on the current prediction model and a convolutional neural network-long short-term memory hybrid neural network is proposed in this paper to predict the position of underwater gliders. Analysis of the experimental results shows that the hybrid neural network model trained with underwater glider sea trial data and simulated motion data can predict the glider speed more accurately than the current-assisted dead reckoning navigation, and can predict the relative position of the glider more accurately with the updated feedback from the current prediction model. The test results show that the data-driven prediction method can greatly help to predict the position of underwater gliders in the absence of other underwater positioning and navigation equipment.

Data-Driven Method for Porosity Measurement of Thermal Barrier Coatings Using Terahertz Time-Domain Spectroscopy

Accurate measurement of porosity is crucial for comprehensive performance evaluation of thermal barrier coatings (TBCs) on aero-engine blades. In this study, a novel data-driven predictive method based on terahertz time-domain spectroscopy (THz-TDS) was proposed. By processing and extracting features from terahertz signals, multivariate parameters were composed to characterize the porosity. Principal component analysis, which enabled effective representation of the complex signal information, was introduced to downscale the dimensionality of the time-domain data. Additionally, the average power spectral density of the frequency spectrum and the extreme points of the first-order derivative of the phase spectrum were extracted. These extracted parameters collectively form a comprehensive set of multivariate parameters that accurately characterize porosity. Subsequently, the multivariate parameters were used as inputs to construct an extreme learning machine (ELM) model optimized by the sparrow search algorithm (SSA) for predicting porosity. Based on the experimental results, it was evident that the predictive accuracy of SSA-ELM was significantly higher than the basic ELM. Furthermore, the robustness of the model was evaluated through K-fold cross-validation and the final model regression coefficient was 0.92, which indicates excellent predictive performance of the data-driven model. By introducing the use of THz-TDS and employing advanced signal processing techniques, the data-driven model provided a novel and effective solution for the rapid and accurate detection of porosity in TBCs. The findings of this study offer valuable references for researchers and practitioners in the field of TBCs inspection, opening up new avenues for improving the overall assessment and performance evaluation of these coatings.

Data-driven models for predicting tensile load capacity and failure mode of grouted splice sleeve connection

Grouted splice sleeve connection (GSSC) is an important connection technology in prefabricated structures, and the tensile load capacity and failure models of GSSC are very important to joint safety. In this study, two data-driven prediction methods (i.e., the machine learning (ML) method, and the threshold method) are proposed to predict the tensile load capacity and failure mode of GSSC. To this end, a database containing 418 existing GSSC experimental data is built. The database is used for eleven ML algorithms (i.e., linear prediction (LP), artificial neural network (ANN), support vector machine (SVM), k-nearest neighbors (KNN), decision tree (DT), random forest (RF), extremely randomized trees (ET), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost)) to establish ML models interpreted by shapley additive explanations (SHAP) and partial dependence plot (PDP). Further, the database is applied to the genetic programming (GP) algorithm to generate a simplified equation for the bond strength between rebar and grouting materials, which is the key mechanical parameter for the threshold method. The results show that both of these methods can effectively predict the tensile load capacity and failure mode of GSSC with various common construction defects, and the predictive performance of ML is slightly greater than that of the threshold method.

Artificial Intelligent Power Forecasting for Wind Farm Based on Multi-Source Data Fusion

Wind power forecasting is a typical high-dimensional and multi-step time series prediction problem. Data-driven prediction methods using machine learning show advantages over traditional physical or statistical methods, especially for wind farms with complex meteorological conditions. Thus, effective use of different data sources and data types will help improve power forecasting accuracy. In this paper, a multi-source data fusion method is proposed, which integrates the static information of the wind turbine with observational and forecasting meteorological information together to further improve the power forecasting accuracy. Firstly, the characteristics of each time step are re-characterized by using the self-attention mechanism to integrate the global information of multi-source data, and the Res-CNN network is used to fuse multi-source data to improve the prediction ability of input variables. Secondly, static variable encoding and feature selection are carried out, and the time-varying variables are combined with static variables for collaborative feature selection, so as to effectively eliminate redundant information. A forecasting model based on the Encoder–Decoder framework is constructed with LSTM as the basic unit, and the Add&Norm mechanism is introduced to further enhance the input variable information. In addition, the self-attention mechanism is used to integrate the global time information of the decoded results, and the Time Distributed mechanism is used to carry out multi-step prediction. Our training and testing data are obtained from an operating wind farm in northwestern China. Results show that the proposed method outperforms a classic AI forecasting method such as that using the Seq2Seq+attention model in terms of prediction accuracy, thus providing an effective solution for multi-step forecasting of wind power in wind farms.

Improving building energy consumption prediction using occupant-building interaction inputs and improved swarm intelligent algorithms

Building energy consumption prediction is important for sustainable building and city construction. However, traditional data-driven prediction methods weakly consider real-time interactions among humans, energy systems, and the environment, which reduces forecasting accuracies. In addition, compared to traditional machine learning algorithms, more suitable algorithms may be required to improve forecasting accuracy and accelerate convergence speed. In this study, an improved energy-consumption prediction system is proposed that includes input system modification and algorithm improvement. For inputs, the probability of sockets, lighting, and air-conditioning related behaviors are included as inputs. Compared with previous studies, the duration of behaviors instead of the number of behavior occurrences is used in this study to solve the behavior probability, which can more accurately reflect occupant-building interactions during forecasting. For algorithms, swarm intelligence algorithms with circle mapping and self-adaptive weight adjustment are used, retaining the diversity of the population and optimizing the weight of the neural network, thereby resulting in better algorithm performance. Results showed that the case using occupant-building interaction inputs and improved swarm intelligence algorithms had the best performance. The coefficients of determination are 0.9588–0.9901, and the mean absolute percentage errors are 4.44%–11.60%. When using the same algorithms and optimization, the modified inputs can improve the prediction performance by 9.15%–21.59%. Also, when using the same inputs, the improved algorithms can improve the precision performance by 5.46%–22.37%. Also, this study discusses the priority of selecting a combination of inputs and algorithms, which will help users to identify different inputs and algorithms when there are different data collection conditions and accuracy requirements. All cases in this study had coefficients of determination above 0.95, indicating good fits. In addition, models of occupant behavior probability are discussed and may be used to help develop occupant behavior research, particularly for random behavior.

State-of-Health Prediction of Lithium-Ion Batteries Based on Diffusion Model with Transfer Learning

An accurate state-of-health (SOH) prediction of lithium-ion batteries (LIBs) is crucial to their safe and reliable. Although recently the data-driven methods have drawn great attention, owe to its efficient deep learning, it is worthwhile to continue devoting many efforts to prediction performance. In practice, fast charging mode has been widely applied in battery replenishing, which poses challenges for SOH prediction due to the diversity of charging conditions and electrochemical properties of LIBs; although, the process is stable and detectable. Furthermore, most previous data-driven prediction methods based discriminative model cannot describe the whole picture of the problem though sample data, affecting robustness of model in real-life applications. In this study, it is presented a SOH prediction model based on diffusion model, as an efficient new family of deep generative model, with time series information tackled through Bi-LSTM and the features derived from the voltage profiles in multi-stage charging process, which can identify distribution characteristics of training data accurately. The model is further refined by means of transfer learning, by adding a featured transformation from the base model for SOH prediction of different type LIBs. Two different types of LIBs datasets are used to evaluate the proposed model and the verified results revealed its better performance than those of other methods, reducing efforts required to collect data cycles of new battery types with the generality and robustness.

A Comparative Study of the Kalman Filter and the LSTM Network for the Remaining Useful Life Prediction of SOFC

The solid oxide fuel cell (SOFC) system is complicated because the characteristics of gas, heat, and electricity are intricately coupled. During the operation of the system, problems such as frequent failures and a decrease in the stack’s performance have caused the SOFC system to work less well and greatly shortened the SOFC’s practical life. As such, it is essential to accurately forecast its remaining useful life (RUL) to make the system last longer and cut down on economic losses. In this study, both model-based and data-driven prediction methods are used to make predictions about the RUL of SOFC. First, the linear degradation model of the SOFC system is established by introducing degradation resistance as the index of health status. Using the Kalman filtering (KF) method, the health status of SOFC is evaluated online. The results of the health state estimation indicated that the KF algorithm is accurate enough to provide a good basis for the model-based RUL prediction. Then, a long short-term memory (LSTM) network-recursive (data-driven) method is presented for RUL prognostics. The multi-step-ahead recursive strategy of updating the network state with actual test data improves the prediction accuracy. Finally, a comparison is made between the LSTM network prediction approach suggested and the model-based KF prognostics. The results of the experiments indicate that the LSTM network is more suitable for RUL prediction than the KF algorithm.

Data-driven depth-averaged current prediction methods for underwater gliders with sailing parameters

The study of depth-averaged currents is of great significance for the application of underwater gliders. In order to solve the problem of low prediction accuracy of the time series-based depth-averaged current prediction method, the factors affecting the prediction of depth-averaged currents are analyzed and a data-driven prediction method for depth-averaged currents of an underwater glider with sailing parameters is proposed in this paper. First, depth-averaged currents of the underwater glider’s historical profile period and navigation parameters of the underwater glider are taken as inputs to construct multi-input and double-output characteristics. Then, based on the two sets of the real sea trial data and two groups of the generic set of evaluation criteria, five different data-driven methods are used to predict depth-averaged currents. Experimental results show that the prediction result of depth-averaged currents of an underwater glider driven by data with sailing parameters is better than that based on time series, and the prediction accuracy of depth-averaged currents of a future profile period is improved.