Online Prediction Model Research Articles

A rigorous mathematical model for online prediction of tube skin temperature in an industrial top-fired steam methane reformer

During the steam reformer operation, monitoring of tube skin temperature is extremely important to any program concerned with maximizing furnace tube life and maintaining safe furnace operation. It can also be an important tool in troubleshooting, debottlenecking and optimizing the plant’s operations. As a result of having reliable tube skin temperature data, the heat flux profile can be adjusted to correct the unit performance for prevention of overheating, hot spots formation and thermal aging of tubes and pigtails before failure. The present study details a rigorous and reliable online model to predict the skin temperature of tubes in an industrial top-fired primary reformer. The model is designed to run online and has a solution time of less than 1.5min starting from a poor initial guess that can be determined automatically from plant data. The model predicts the most important variables such as heat flux, composition and conversion profiles along the tube length, process gas temperature, maximum tube-wall temperature and tube pressure drop based on considering mass transfer limitation in catalyst pellets using a one-dimensional heterogeneous model. The predictions of model were compared with two previously reported works and operating data from the top-fired steam reformer of an industrial ammonia plant under normal conditions which showed a well compatibility. The model results demonstrated a maximum skin temperature of 905°C at location 2.6m from top of tube length at plant rates greater than 90% of full capacity. Applying the model in the plant to adjust the tube skin temperature resulted in 605.4Nm3/h of natural gas fuel saving. The outcome from this work is not only valuable in terms of operation, but can also prevents the overheating and failure of tubes and pigtails which extends the life of these constitutes in the reformer. Before implementing modifications in reformer design or its configuration, it is important to evaluate the maximum temperature of reformer tube.

Fast auto-clean CNN model for online prediction of food materials

Online food image detection is a key issue for intelligent food materials receiving and food supply chain applications, how to efficiently, accurately and quickly detect the image of food materials is a challenging research topic. A fast auto-clean convolutional neural network (CNN) model for online prediction of food materials is proposed, which is aiming at problems of the complex characteristics of the food images such as the complexity of the food materials, the focus of the dislocation and the uniformity of illumination. Firstly, a new approach of the auto-clean CNN models is proposed for automatic image cleaning and classification, which starting from original images and ending with multi-class prediction of clean images. Given a vocabulary of K classes, and a Yes/No clean label, two CNN models will learn a class label and a clean label respectively. Secondly, after the forward pass of two CNN models, the joint features generated from the last convolutional layers will be fed into our two loss layers. Combined with multi-class classification method, it classifies and optimizes the image dataset intelligently. Finally, an online prediction algorithm is proposed to improve the image recognition efficiency. Experimental results show that the proposed model and algorithm have good efficiency and accuracy, and the results of this study have significance to optimize the efficiency of the food supply chain industry and food quality evaluation.

AEGEUS++: an energy-aware online partition skew mitigation algorithm for mapreduce in cloud

This paper investigates the partition skew problem at reduce phase in the mapreduce jobs. Our study summarize the skew problem in both offline and online manner. Offline is a heuristics based approach waits for the completion of map tasks and it involves computation overhead to estimate the partition size. In online approach, the overloaded tasks are distributed across other nodes that needs extra split and merge operations. These extra operations and ineffective utilization of resources in turn hamper the performance of the entire system. In this paper, we propose Aegeus++, to address the skew mitigation and adaptive data sampling problems for mapreduce jobs which enables to build an online prediction model with improved accuracy in minimal waiting time. In addition, we propose near linear skew detection and fine-grained Resource Allocation algorithms for identifying the skewed partition and allocating appropriate resources to reducers based on the partition size. Finally, our energy-aware opportunistic frequency tuning algorithm improves the performance of the reducer container on-fly, that can process the skewed data faster with minimal energy consumption. We evaluated Aegeus++ in the cloud setup by using benchmark datasets, compared its performance with native Hadoop and its other approaches. Based on our observation, Aegeus++ outperforms native Hadoop by 44% by maximizing its overall performance of the application and decreases the energy consumption by 37.67% when compared with existing approaches.

Predictive complex event processing based on evolving Bayesian networks

In the Big Data era, large volumes of data are continuously and rapidly generated from sensor networks, social network, the Internet, etc. Predicting from online event stream is an important task since users usually need to predict some future states and take some actions in advance. Many applications need online prediction models which can evolve automatically with data distribution drift and algorithms which can support single-pass processing of data, which are still faced with many challenges. In this paper, the authors propose a predictive complex event processing method based on evolving Bayesian networks. The Bayesian model is designed based on event type and time with inference method based on Gaussian mixture model and EM algorithm. When learning the structure of Bayesian network from event streams, this method supports calculating score metric incrementally when new data is arrived or edges in the network are changed. Evolving Bayesian network structure is supported based on hill-climbing method. The system can continuously monitor the Bayesian network model and modify it if it is found to be not appropriate for the new incoming data. The method of this paper is evaluated in road traffic domain with both real application data and data produced by a simulated transportation system. The total percentage error is 8.12% for real data and 7.78% for simulated data, while the best result for other methods is 11.79% for real data and 14.59% for simulated data. The experimental evaluations show that this method is effective for predictive complex event processing and it outperforms other popular methods when processing traffic prediction in intelligent transportation systems.

Online Prediction with Variable Horizon for Vehicle's Future Driving-Cycle

Abstract With traditional driving cycle predictive model, the state point in vehicle-acceleration projection plane couldn’t cover the real driving state completely. And date-missing caused by this lead to interruption of the prediction process. So in this paper, a real-time prediction model with variable horizon is proposed to solve the problem. Real driving data is used to reconstruct the driving cycle and the accuracy of the real time prediction model could be estimated based on historical information. By using principal component analysis and cluster analysis, an online prediction model with variable horizon based on Marco Chain is established. The correctness of this method is verified by experiment of Hardware-in-loop simulation. And the result shows that the accuracy of variable time prediction model is 8.203km/h, which has been improved by 20% comparing with fixed time prediction model.

Vector Machines Regression Applied in Penicillin Fermentation Process Control

According to the characteristics of biochemical processes, this paper studied the small samples of the data process. Small amount of data can be quickly measured variable modeling. Introduces the support vector machine (SVM) modeling as small sample theory, the least squares support vector machine (LS-SVM) algorithm based on improved SVM applied to the typical fermented penicillin biochemical processes in the past. Online prediction model only off-line testing of important process variables based on the simulation results simulation platform show that the method is only by learning the few batches of sample data. Establish a penicillin product concentration, cell concentration and substrate concentration. This paper analyzes the characteristics of the penicillin fermentation process, and the use of SVM method for modeling, give potency relationship between the factors and its influence. Through experiments, the influence of SVM model parameters to adjust performance. Through a variety of models from the established field data can be found, SVM is better than ANN modeling.

Hybrid regression model for near real-time urban water demand forecasting

The most important factor in planning and operating water distribution systems is satisfying consumer demand. This means continuously providing users with quality water in adequate volumes at reasonable pressure, thus ensuring reliable water distribution. In recent years, the application of statistical, machine learning, and artificial intelligence methodologies has been fostered for water demand forecasting. However, there is still room for improvement; and new challenges regarding on-line predictive models for water demand have appeared. This work proposes applying support vector regression, as one of the currently better machine learning options for short-term water demand forecasting, to build a base prediction. On this model, a Fourier time series process is built to improve the base prediction. This addition produces a tool able to eliminate many of the errors and much of the bias inherent in a fixed regression structure when responding to new incoming time series data. The final hybrid process is validated using demand data from a water utility in Franca, Brazil. Our model, being a near real-time model for water demand, may be directly exploited in water management decision-making processes.

Remaining useful life estimation under degradation and shock damage

This article presents a prognostic approach to estimate remaining useful life for systems subjected to dependent competing failure processes. In the literature, shock damage is the damage to a soft failure process caused by a shock process. However, how the degradation process causes damage to a hard failure process has not been well studied. In this article, the degradation damage is modeled as the damage to a hard failure process from a degradation process. Degradation and shock processes, as “elemental processes,” result in failures via either a soft failure or a hard failure process, namely, “compound processes.” Instead of leading to a direct failure, elemental processes construct compound processes: the soft failure process consists of a degradation process and shock damage, and the hard failure process consists of a shock process and degradation damage. In this way, the damage in this article especially represents the effect of an elemental process on other compound processes. Furthermore, a particle filter is applied based on the established model for system statement estimation and on-line prediction of remaining useful life distribution with and without measurement noise in prognostics. Finally, a numerical example is presented with sensitivity analysis.

Non-invasive lactate- and pH-monitoring in porcine meat using Raman spectroscopy and chemometrics

The feasibility of using chemometrics and Raman spectroscopy as a fast and non-invasive method to monitor the early postmortem lactate accumulation and pH decline in pork meat has been investigated. For this application, an on-line monitoring methodology has not yet been established. Based on raw Raman spectra of porcine semimembranosus muscles, a range of spectral pre-processing and multivariate calibration techniques were investigated to develop and test on-line prediction models for the meat quality parameters. The influence of the pre-processing methods on the prediction speed, robustness and accuracy performance of the employed linear and non-linear algorithms was compared. Identification of the most effective chemometric evaluation procedure was performed using least square linear regression together with locally weighted regression and metaheuristic data optimization methods such as the genetic algorithm and the ant colony optimization. The herein presented analysis suggests that the locally weighted regression applied to the standard normal variate (SNV) normalized Raman spectra provides the most accurate and robust models with a cross-validated coefficient of determination (r2cv) of 0.97 for pH and lactate, a cross-validated root mean square error (RMSECV) of 4.5mmol/kg for the lactate prediction and 0.06pH-units for the pH prediction. These results demonstrate the great potential of combining chemometrics and Raman spectroscopy for on-line meat quality control applications.

Online NIR Analysis and Prediction Model for Synthesis Process of Ethyl 2-Chloropropionate.

Online near-infrared spectroscopy was used as a process analysis technique in the synthesis of 2-chloropropionate for the first time. Then, the partial least squares regression (PLSR) quantitative model of the product solution concentration was established and optimized. Correlation coefficient (R 2) of partial least squares regression (PLSR) calibration model was 0.9944, and the root mean square error of correction (RMSEC) was 0.018105 mol/L. These values of PLSR and RMSEC could prove that the quantitative calibration model had good performance. Moreover, the root mean square error of prediction (RMSEP) of validation set was 0.036429 mol/L. The results were very similar to those of offline gas chromatographic analysis, which could prove the method was valid.

Data-based adaptive online prediction model for plant-wide production indices

A data-based adaptive online prediction model is proposed for plant-wide production indices based on support vector regression, a general method which we customized specifically to model very large data sets that are generated dynamically and periodically. The proposed model can update its parameters online according to the statistical properties of the training samples. Further, in order to improve the prediction precision, each sample is weighted with a dynamic penalty factor that considers the effect of each sample on the prediction model accuracy. Moreover, a customized procedure is introduced to handle large training sets. After having been convincingly evaluated on benchmark data, effectiveness and performance of our approach for plant-wide production indices is demonstrated using industrial data from an operating ore dressing plant over a range of scale in training data set size. The higher accuracy and shorter computation times than existing methods suggest that it may prove advantageous in actual application to dynamic production processes.

External Validation of Online Predictive Models for Prediction of Cancer-specific Mortality and All-cause Mortality in Patients with Urothelial Carcinoma of the Urinary Bladder

The objective of the study was to validate the previously reported lookup Table and Bladder Cancer Research Consortium (BCRC) nomogram in predicting cancer-specific mortality (CSM) and all-cause mortality (ACM) after radical cystectomy using an external cohort from South Korea. The study comprised 409 patients. Discrimination was quantified with the concordance index. The relationship between the model-derived and actual CSM and ACM was graphically explored within calibration plots. Clinical net benefit was evaluated by decision curve analysis. Of the 409 patients, 147 (35.9 %) had died from various causes. One hundred two deaths were attributable to bladder cancer. For CSM at 5 years, the bootstrap-corrected concordance indices of the American Joint Committee on Cancer (AJCC) staging system, lookup Table, and BCRC nomogram were 71.8 % (95 % confidence interval [CI] 66.9-76.5), 73.0 % (95 % CI 67.9-78.0), and 76.2 % (95 % CI 71.6-80.9), respectively. For ACM at the same time point, the discrimination accuracies of these models were 70.7 % (95 % CI 66.7-74.6), 72.8 % (95 % CI 68.5-76.9), and 76.2 % (95 % CI 72.3-80.2), respectively. The calibration plots tended to exaggerate both survival outcomes in all models. When compared to the lookup Table as well as the AJCC staging system, the BCRC nomogram performed well across a wide range of threshold probabilities using decision curve analysis. The BCRC nomogram was characterized by higher accuracy and larger potential clinical benefit compared to the lookup Table. However, there is a great need for additional models that consider outcomes of patients for whom the existing models do not apply.

Meat quality prediction using Raman spectroscopy and chemometrics

The feasibility of using Raman spectroscopy as a fast and non-invasive method to monitor the quality parameters in pork meat has been investigated. For this application an online prediction methodology has not been established yet. Based on raw Raman spectra of 10 pork semimembranosus muscles a range of data pre-processing and multivariate calibration methodology have been used to develop online predictive models for the meat quality parameters: the lactate and pH. The linear and nonlinear algorithms studied were comparatively analysed for speed, robustness and accuracy. Identifying the best “efficiency” evaluation procedure represented the final milestone of the present study. Thus with a cross-validated r2 value for both pH and lactate of 0.97, a RMSECV of 4.5 mmol/l for the lactate prediction and 0.06 units for the pH prediction, locally weighted regression provided the most accurate and robust model. This prove the feasibility of using Raman spectroscopy for online meat quality control applications.

A novel real-time non-linear wavelet-based model predictive controller for a coupled tank system

This article presents the design, simulation and real-time implementation of a constrained non-linear model predictive controller for a coupled tank system. A novel wavelet-based function neural network model and a genetic algorithm online non-linear real-time optimisation approach were used in the non-linear model predictive controller strategy. A coupled tank system, which resembles operations in many chemical processes, is complex and has inherent non-linearity, and hence, controlling such system is a challenging task. Particularly important is low-level control where often instability and oscillatory responses are observed. This article designs a wavelet neural network with high predicting precision and time–frequency localisation characteristics for an online prediction model in the non-linear model predictive controller to show the effectiveness of this approach in controlling the liquid at low level. To speed up the training process, a fast global search stochastic non-linear conjugate wavelet gradient algorithm is initially used to train the wavelet neural network structure before the genetic algorithm optimisation technique is utilised to tune adaptively the wavelet neural network parameters. The non-linear model predictive controller algorithm is tested for both approaches: first, in a simulation using identified models, and second, in a real-time practical application to a single-input single-output system coupled tank system. The results show an excellent control performance with respect to mean square error and average control energy values obtained.

Coiling Temperature Prediction and Application Based on Genetic-Neural Network on Hot Strip Mill

Coiling temperature control (CTC) is very important to the quality of the strip steel in Hot Strip Rolling Mill. In the paper, genetic algorithm and neural network method to predict coiling temperature on hot strip mill were put forward. The genetic-neural network was trained and checked with actual production data. The result indicates that the method can real-time predict the strip coiling temperature. The on-line prediction model and step track method has been put into use. The result shows that the method can settle lag influence in feedback control and the CTC control precision is improved greatly.

LS_SVM Based on DTW and Sliding Window Pruning Algorithm and its Application

Compared with the traditional SVM, LS_SVM improves the speed of the large sample study,however ,LS_SVM loses its sparsity, which is an advantage of the the traditional SVM . For LS_SVM,almost all of the samples are support vectors .When a new sample added all samples are re-involved in the calculation, and that greatly increases the amount of the computation. In response to these issues,this paper presents an algorithm called the LSSVM sliding window pruning algorithm,which is based on DTW. With the combination of local modeling ,the algorithm selects the most similar samples with the current sample point in the process of the establishment of the on-line model of the batch process as the training samples .It can remove those samples that are relatively ineffective,which can restore LSSVM 's sparsity.Taking the cell concentration in the fermentation process of E. Coli for example, this paper proves that the method can establish a more accurate on-line prediction model in the case of a smaller number of samples . Compared with the off-line LSSVM ,the on-line model has a higher accuracy and a better dynamic adaptability.

On-line prediction of ship roll motion during maneuvering using sequential learning RBF neuralnetworks

The on-line prediction of ship roll motion during maneuvering plays an important role in navigation safety and ship control applications. This paper presents an on-line prediction model of ship roll motion via a variable structure radial basis function neural network (RBFNN), whose structure and parameters are tuned in real time based on a sliding data window observer. The RBFNN is sequentially constructed by adding the new sample in the hidden layer and pruning the obsolete hidden units at each epoch, with the connecting parameters adjusted simultaneously. Gaussian functions with multi-scale kernel width are adopted to provide more flexible representations of model input terms and to achieve better generalization capability. Simulation study of ship roll motion prediction is conducted with measurement data of turning circle test and zigzag test in full-scale sea trial. Results demonstrate that the proposed neural network predictive model can on-line predict the roll angle with high accuracy. The predictive model is also featured with its compact network structure and fast computationalspeed.

Erratum to: Architecture for development of adaptive on-line prediction models

B. Gabrys e-mail: BGabrys@bournemouth.ac.uk which emphasises that the weightsw(m),i must add up to 1 and preserves the interpretations of L(m), j as a PDF. 2. In order for Eq. (19) to map the error e(m) into a performance index u(m), so that the best performing local expert receives a weight (score) equal to 1 and the weights of the remaining local experts decay exponentially with the increasing error, whereas the median error is mapped to the value 0.5, Eq. (19) should take the following form:

Online prediction model based on the SVD–KPCA method

This paper proposes a new method for online identification of a nonlinear system modelled on Reproducing Kernel Hilbert Space (RKHS). The proposed SVD–KPCA method uses the Singular Value Decomposition (SVD) technique to update the principal components. Then we use the Reduced Kernel Principal Component Analysis (RKPCA) to approach the principal components which represent the observations selected by the KPCA method.

An Online Prediction Model for BFG Output in Steel Industry

The output prediction of blast furnace gas (BFG), influenced by many complex production factors, is a very important and difficult problem concerning the byproduct gas balance in steel industry. A new online least squares support vector machine (LSSVM) prediction model is proposed in this paper, in which the training data is filtered by an improved empirical mode decomposition threshold filtering (IEMDTF). The model is solved and optimized by an online learning algorithm and an online bayesian parameters optimization, respectively. The experimental results using practical BFG output data from BaoSteel Co. Ltd., China show the proposed model is effective and enable to offer reasonable gas balance scheduling for operators.