Probabilistic Prediction Model Research Articles

Bayesian-Neural-Network-Based Approach for Probabilistic Prediction of Building-Energy Demands

Reliable prediction of building-level energy demand is crucial for the building managers to optimize and regulate energy consumption. Conventional prediction models omit the uncertainties associated with demand over time; hence, they are mostly inaccurate and unreliable. In this study, a Bayesian neural network (BNN)-based probabilistic prediction model is proposed to tackle this challenge. By quantifying the uncertainty, BNNs provide probabilistic predictions that capture the variations in the energy demand. The proposed model is trained and evaluated on a subset of the building operations dataset of Lawrence Berkeley National Laboratory (LBNL), Berkeley, California, which includes diverse attributes related to climate and key building-performance indicators. We have performed thorough hyperparameter tuning and used fixed-horizon validation to evaluate trained models on various test data to assess generalization ability. To validate the results, quantile random forest (QRF) was used as a benchmark. This study compared BNN with LSTM, showing that BNN outperformed LSTM in uncertainty quantification.

Parallel multiple-chain DRAM probabilistic model for ultrasonic pulse velocity of concrete performance in the pipe of arch bridge

The existing models for predicting the compaction state of concrete inside steel tubes based on ultrasonic wave velocity have a major limitation in that they cannot comprehensively consider concrete compressive strength, instar, and various sources of uncertainties, both subjective and objective. To overcome this limitation, the study introduced a probability model for ultrasonic wave velocity in the concrete performance inside arch bridge steel tubes, based on the parallel multiple-chain Delayed Rejection Adaptive Metropolis (DRAM) algorithm. Initially, the study considers the influence of concrete compressive strength and instar, establishing a simplified deterministic model for predicting concrete compactness inside the tubes. Building on this, the study incorporates material stochastic uncertainty and cognitive uncertainty, deriving analytical expressions for a probabilistic prediction model of concrete compactness. Furthermore, using parallel processing alongside the DRAM algorithm, the study explores how the number of parallel chains and the order of acceptance probabilities influence the probabilistic model parameters. The study selected and updated the posterior distribution information of the probabilistic model parameters. Finally, the effectiveness of the model was validated using experimental data. As shown in the analysis, with better prediction accuracy and lower dispersion, the model not only selects the optimal Markov chain iteration path in a short time, but also corrects the parameters affecting the prediction accuracy of the existing model, and also provides a probabilistic method for correcting the confidence intervals of the existing model.

A 5-Year Mortality Prediction Model for Prostate Cancer Patients Based on the Korean Nationwide Health Insurance Claims Database.

Prostate cancer is the fourth most common cancer and eighth leading cause of cancer-related mortality worldwide. Its incidence is increasing in South Korea. This study aimed to investigate a predictive model for the 5-year survival probability of prostate cancer patients in a Korean primary care setting. This retrospective study used data from the nationwide insurance claims database. The main outcome was survival probability 5 years after the initial diagnosis of prostate cancer. Potential confounding factors such as age, body mass index (BMI), blood pressure, laboratory results, lifestyle behaviors, household income, and comorbidity index were considered. These variables were available in the national health check-up information. A Cox proportional hazards regression model was used to develop the predictive model. The predictive performance was calculated based on the mean area under the receiver operating characteristic curve (AUC) after 10-fold cross-validation. The mean 5-year survival probability was 82.0%. Age, fasting glucose and gamma-glutamyl transferase levels, current smoking, and multiple comorbidities were positively associated with mortality, whereas BMI, alkaline phosphatase levels, total cholesterol levels, alcohol intake, physical activity, and household income were inversely associated with mortality. The mean AUC after 10-fold cross-validation was 0.71. The 5-year survival probability model showed a moderately good predictive performance. This may be useful in predicting the survival probability of prostate cancer patients in primary care settings. When interpreting these results, potential limitations, such as selection or healthy user biases, should be considered.

Probabilistic prediction model for critical chloride concentration of reinforcement corrosion based on improved Gaussian process regression

Probabilistic prediction model for critical chloride concentration of reinforcement corrosion based on improved Gaussian process regression

A machine learning-based probabilistic wind speed prediction model with multi-resolution data, quantile regression and bound estimation

Wind speed prediction can be important for management of long-span cross-sea bridges, and probabilistic prediction models are needed to take uncertainty of future wind speed into consideration. However, there are two shortcomings of existing models. One is the lack of supplementary input of detailed fluctuations of wind speed, and the other is that fixed quantiles are usually selected when the predicted probability distribution is converted into a confidence interval. To this end, we propose a novel machine learning-based probabilistic model with multi-resolution data, quantile regression and bound estimation. The proposed model is composed of the distribution prediction part and the interval prediction part. In the distribution prediction part, we employ multi-resolution data as input, which can complement details of fluctuations of wind speed to reduce uncertainty. The interval prediction part employs the lower upper bound estimation method to postprocess predicted quantiles and output a superior interval than selecting fixed quantiles. The proposed model is validated on the monitoring wind speed data of two long-span cross-sea bridges, where the superiority of the distribution part and the effectiveness of the interval part are demonstrated.

Indeterministic Data Collection in UAV-Assisted Wide and Sparse Wireless Sensor Network

The widespread adoption of Internet of Things (IoT) applications has driven the demand for obtaining sensor data. Using unmanned aerial vehicles (UAVs) to collect sensor data is an effective means in scenarios with no ground communication facilities. In this paper, we innovatively consider an indeterministic data collection task in a UAV-assisted wide and sparse wireless sensor network, where the wireless sensor nodes (SNs) obtain effective data randomly, and the UAV has no pre-knowledge about which sensor has effective data. The UAV trajectories, SN serve scheduling and UAV-SN association are jointly optimized to maximize the amount of collected effective sensing data. We model the optimization problem and address the indeterministic effective indicator by introducing an effectiveness probability prediction model. The reformulated problem remains challenging to solve due to the number of constraints varying with the variable, i.e., the serve scheduling strategy. To tackle this issue, we propose a two-layer modified knapsack algorithm, within which a feasibility problem is resolved iteratively to find the optimal packing strategy. Numerical results demonstrate that the proposed scheme has remarkable advantages in the sum of effective data blocks, reducing the completion time for collecting the same ratio of effective data by nearly 30%.

Probabilistic model for high-level intention estimation and trajectory prediction in urban environments

To enable successful automated driving, precise behavior prediction of surrounding vehicles is indispensable in urban traffic scenarios. Furthermore, given that a vehicle’s behavior is influenced by the movements of other road users, it becomes crucial to estimate their intentions to anticipate precise future motion. However, the elevated complexity resulting from interdependencies among traffic participants and the uncertainty arising from the object recognition errors present additional challenges. Despite extensive research on inferring intentions, many studies have concentrated on estimating intentions from interactions, resulting in a lack of practicality in urban traffic environments due to low computational efficiency and low robustness against recognition failure of strongly interacting road users. In this paper, we introduce a practical stochastic model for intention estimation and trajectory prediction of surrounding vehicles in automated driving under urban traffic environments. The trajectory is forecasted based on hierarchically computed and probabilistically estimated intentions, which represent an interpretation of vehicle behavior, utilizing only the kinematic state of the focal vehicle and HD maps to ensure real-time performance and enhance robustness. The evaluated results demonstrate that the proposed model surpasses straightforward methods in terms of accuracy while maintaining computational efficiency and exhibits robustness against the recognition failure of traffic participants which strongly influence the focal vehicle.

Interpretable machine learning models for predicting probabilistic axial buckling strength of steel circular hollow section members considering discreteness of geometries and material

Steel circular hollow section (CHS) members are widely utilized as axial force-resisting structural members in civil engineering structures. The buckling strength under axial loads is one of the critical parameters to determine the performance of the steel CHS members, which is significantly affected by the discreteness introduced by geometries, material, and initial imperfections. However, the reduction factor employed in the modern design codes (i.e. Chinese codes and EC3) only accounts for the reduction caused by all kinds of discreteness and does not reflect the impacts of every single discreteness and imperfection. To fill the gap, this paper proposed an interpretable machine-learning method to provide the probabilistic axial buckling strength of steel CHS members prediction result in a distribution form with the consideration of detailed discreteness. The model to predict the nominal axial buckling strength of steel CHS members was first developed utilizing ten machine learning algorithms after sufficient numerical simulations, where the numerical model was verified using test results. The artificial neural network (ANN) was selected for developing the prediction model due to its highly reliable performance in testing. The developed ANN models were further interpreted utilizing Shapley Additive exPlanations (SHAP) to determine the interrelationship of different parameters. Then, the probabilistic axial bucking strength prediction model was established based on the developed ANN models, where the Latin hypercube sampling method was applied to address the discreteness of geometries, material, and initial imperfections. The generated probabilistic axial bucking strength prediction model’s effectiveness was verified by the evidence that the machine learning prediction results can highly match the numerical results' probability density function and the result from codes while significantly reducing the computation time. Finally, the design parameters’ impact on the axial buckling strength’s discreteness was evaluated using the global sensitivity analysis (GSA) method. The result shows that the discreteness of design parameters substantially influences the distribution of the axial buckling strength of the steel CHS members and the proposed prediction model can provide an accurate probabilistic distribution prediction.

A parallel and multi-scale probabilistic temporal convolutional neural networks for forecasting the key monitoring parameters of gas turbine

As a crucial equipment in the power industry, gas turbines need effective condition monitoring techniques to maintain safe and reliable operations. Fast yet accurate long-term forecasting of the key monitoring parameters of gas turbine is vital for achieving the overall effective condition monitoring. This however poses challenges in terms of both efficacy and efficiency for conventional time series models like recurrent neural networks (RNN), since they run in a sequential manner. To this end, this paper introduces a novel parallel probabilistic time series prediction model. Particularly, the proposed model leverages the parallelism of temporal convolutional neural network (TCN) and exploits the power of latent space in the Bayesian framework. Besides, a multi-scale feature extraction strategy based on dense connections and a lagged observation strategy are presented to improve the model performance. In the comparative study against state-of-the art competitors on four classical system identification benchmarks, the proposed model significantly reduces the training time, while consistently achieving highly accurate predictions and providing appropriate uncertainty quantification. Thereafter, the proposed model is adopted to build a systematic workflow for the fast yet accurate long-term forecasting of the temperature of blade channel of gas turbine. The comprehensive results again highlight the superiority of the proposed model in terms of both the prediction quality and the training efficiency.

Comparison of the quality of logistic regression models and a classification tree in predicting hospital mortality in elderly patients with non-ST-elevation myocardial infarction

Aim. Using the CHAID (Chi Squared Automatic Interaction Detection) method to develop a classification tree for predicting hospital mortality in patients with non-ST-elevation myocardial infarction (non-STEMI) aged 75 years and older and compare the quality of the constructed model with the logistic regression model.Material and methods. A retrospective analysis of the case histories of 119 sequentially enrolled patients aged ≥75 years who were hospitalized in a cardiology department due to non-STEMI in 2020-2021 was carried out. The construction of a predictive model of probability of dying was carried out using the logistic regression method. To assess the impact of various predictors affecting the probability of dying during the of hospitalization period in patients with non-STEMI, a classification tree was developed using the CHAID method. To compare the quality of logistic regression models and the classification tree, the areas under the ROC curve and confidence intervals were estimated.Results. Based on the construction of a binary logistic regression, it was found that the factors increasing hospital mortality were cardiogenic shock (CS): odds ratio (OR) 47.55; 4.00-589.16; p=0.002; new-onset atrial fibrillation: OR 6.45; 1.39-30.42; p=0.018; and the number of points on the GRACE scale: for each increase by 1 point: OR 1.03; 1,00-1,05; p=0.046. Similar data were obtained when analyzing the classification tree: in patients with CS, the predicted mortality was 91.7%. The probability of an unfavorable outcome based on the constructed classification tree was higher than the average in the analyzed sample in persons without CS, 2-3 degree atrioventricular blocks, and pulmonary edema, but with right bundle branch block on the electrocardiogram (25.0%) and in persons without CS and atrioventricular blockages of 2-3 degrees, but with pulmonary edema and a Q wave on the electrocardiogram (50.0%). Both methods of predicting hospital mortality are applicable. There were no statistically significant differences in the quality of both constructed models; the difference in the areas under the ROC curves was 0.043±0.268 with a 95% confidence interval of -0.055-0.141, p=0.387.Conclusion. Both developed methods can be used to determine the probability of dying in a hospital. Currently, the recruitment of patients into a prospective study of a similar design has begun and is continuing, during which validation of the constructed forecasting models is planned.

A probabilistic prediction model based on stacking ensemble learning for completion time in flexible shop-floor

A probabilistic prediction model based on stacking ensemble learning for completion time in flexible shop-floor

Prediction method of loess landslides based on faster R-CNN and WACM

Due to the complexity of the environment and geological conditions in which the loess slope is located, there are many challenges in the accuracy and prediction of loess landslide detection. Therefore, this study introduces a fast convolutional neural network model to solve the problems of traditional detection methods in terms of technology, cost, and detection accuracy, and to achieve real-time detection of the morphology of loess landslides. A weight absorption coupling model is constructed to address the uniform moisture content in loess with hidden dangers. Combined with instability probability, the probability of shallow loess landslides is predicted. The results showed that the mAP value of the Faster R-CNN algorithm using the ResNet125 network exceeded 90%, which was 46.23% and 32.01% higher than the algorithm models using ResNet50 and VGG16, respectively. The proposed model performed fractal analysis on four different loess particle samples, with correlation coefficients R2 above 0.9. The difference between the predicted and actual moisture content of upper and surface loess was within 11%. Compared with existing methods, the research and construction of a loess landslide detection and probability prediction model has greatly improved reliability and accuracy, which is of great significance for predicting the probability of different loess landslides.

Automated Interstitial Lung Abnormality Probability Prediction at CT: A Stepwise Machine Learning Approach in the Boston Lung Cancer Study.

Background It is increasingly recognized that interstitial lung abnormalities (ILAs) detected at CT have potential clinical implications, but automated identification of ILAs has not yet been fully established. Purpose To develop and test automated ILA probability prediction models using machine learning techniques on CT images. Materials and Methods This secondary analysis of a retrospective study included CT scans from patients in the Boston Lung Cancer Study collected between February 2004 and June 2017. Visual assessment of ILAs by two radiologists and a pulmonologist served as the ground truth. Automated ILA probability prediction models were developed that used a stepwise approach involving section inference and case inference models. The section inference model produced an ILA probability for each CT section, and the case inference model integrated these probabilities to generate the case-level ILA probability. For indeterminate sections and cases, both two- and three-label methods were evaluated. For the case inference model, we tested three machine learning classifiers (support vector machine [SVM], random forest [RF], and convolutional neural network [CNN]). Receiver operating characteristic analysis was performed to calculate the area under the receiver operating characteristic curve (AUC). Results A total of 1382 CT scans (mean patient age, 67 years ± 11 [SD]; 759 women) were included. Of the 1382 CT scans, 104 (8%) were assessed as having ILA, 492 (36%) as indeterminate for ILA, and 786 (57%) as without ILA according to ground-truth labeling. The cohort was divided into a training set (n = 96; ILA, n = 48), a validation set (n = 24; ILA, n = 12), and a test set (n = 1262; ILA, n = 44). Among the models evaluated (two- and three-label section inference models; two- and three-label SVM, RF, and CNN case inference models), the model using the three-label method in the section inference model and the two-label method and RF in the case inference model achieved the highest AUC, at 0.87. Conclusion The model demonstrated substantial performance in estimating ILA probability, indicating its potential utility in clinical settings. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Zagurovskaya in this issue.

Supervised probabilistic dynamic-controlled latent-variable model for quality pattern prediction and optimisation

A supervised probabilistic dynamic-controlled latent-variable (SPDCLV) model is proposed for online prediction, as well as real-time optimisation of process quality indicators. Compared to existing probabilistic latent-variable models, the key advantage of the proposed method lies in explicitly modelling the dynamic causality from the manipulated inputs to the quality pattern. This is achieved using a well-designed, dynamic-controlled Bayesian network. Furthermore, the algorithms for expectation-maximisation, forward filtering, and backward smoothing are designed for learning the SPDCLV model. For engineering applications, a framework for pattern-based quality prediction and optimisation is proposed, under which the pattern-filtering and pattern-based soft sensor are explored for online quality prediction. Furthermore, quality optimisation can be realised by directly controlling the pattern to the desired condition. Finally, case studies on both an industrial primary milling circuit and a numerical example illustrate the benefits of the SPDCLV method in that it can fully model the process dynamics, effectively predict and optimise the quality indicators, and monitor the process.

A Probabilistic Runoff Prediction Model Based on Improved Long Short-Term Memory and Interval Correction

A Probabilistic Runoff Prediction Model Based on Improved Long Short-Term Memory and Interval Correction

Nomograms for prognosis prediction in esophageal adenocarcinoma: realities and challenges.

Prognostic assessment is of great significance for individualized treatment and care of cancer patients. Although the TNM staging system is widely used as the primary prognostic classifier for solid tumors in clinical practice, the complexity of tumor occurrence and development requires more personalized probability prediction models than an ordered staging system. By integrating clinical, pathological, and molecular factors into digital models through LASSO and Cox regression, a nomogram could provide more accurate personalized survival estimates, helping clinicians and patients develop more appropriate treatment and care plans. Esophageal adenocarcinoma (EAC) is a common pathological subtype of esophageal cancer with poor prognosis. Here, we screened and comprehensively reviewed the studies on EAC nomograms for prognostic prediction, focusing on performance evaluation and potential prognostic factors affecting survival. By analyzing the strengths and limitations of the existing nomograms, this study aims to provide assistance in constructing high-quality prognostic models for EAC patients.

Stochastic optimisation of district integrated energy systems based on a hybrid probability forecasting model

Day-ahead scheduling optimisation plays a significant role in solving stable operation problems of district integrated energy systems (DIESs) by carrying out scheduling optimisation of key equipment of the system based on forecasting data. However, the stable operation of DIESs is affected by multiple uncertainties from source-load. Day-ahead scheduling optimisation method is designed in this study for DIESs based on “prediction-optimisation” to reduce the impact of source-load uncertainty on the stable operation of system. Considering multidimensional source-load data, hybrid probability forecasting model based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN)-Long Short-Term Memory (LSTM)-Gaussian Process Quantile Regression (GPQR) is introduced to quantify the uncertainty using prediction interval. K-means method is used to generate typical scenarios in optimisation step and these scenarios are fed into stochastic optimisation to obtain day-ahead scheduling optimisation results. The experimental results in case study show that hybrid probability prediction model exhibits good prediction performance, while the day-ahead scheduling optimisation based on stochastic optimisation for system's key equipment has the smallest error in typical summer and winter days compared with the real situation, which is 0.018 and 0.0176, respectively. The “prediction-optimisation” approach proposed in this paper shows significant advantages for improving the performance of day-ahead scheduling optimisation of DIESs.

Clustered remote sensing target distribution detection aided by density-based spatial analysis

Small target detection in remote sensing is integral to a range of applications, including smart city systems and emergency rescue operations. However, the challenges posed by weak features and complex backgrounds in remote sensing images have hindered the efficacy of detection. Current models tend to focus on identifying individual targets, resulting in algorithms with larger parameters, slower detection efficiency, and difficulty in striking a balance between false positives and negatives. Given that many tasks do not require precise target location, a more efficient approach involves swiftly predicting target areas with models involving fewer parameters. This paper introduces the concept of group target distribution detection, gathering targets with similar distances and semantic similarities for clustered detection. A Gaussian probability map, formed from target density, is used to train a probability prediction model. We propose a new metric for evaluating this innovative group target distribution detection paradigm and provide a comparative assessment with traditional single-object detection models. In experimental evaluation, our proposed DenseUGE network — employing ResNet34 and ResNet50 as its backbone — surpasses the best baseline method by 3.37% on the AI-TOD dataset using our metrics. Additionally, visualizations demonstrate the ability of our proposed methodology to effectively identify the concentrated distribution of small target groups.

Physics-based natural gradient boosting probabilistic prediction model for seismic failure modes of reinforced concrete columns

A novel physics-based natural gradient boosting (PNGB) probabilistic prediction model for flexure failure (FF), flexure-shear failure (FSF) and shear failure (SF) of reinforced concrete (RC) columns under strong earthquakes has been developed to address the limitations of empirical and machine learning (ML) models, which often fail to learn physical laws, insufficiently consider uncertainties, and exhibit poor generalization performance. Firstly, an improved natural gradient boosting (NGB) probabilistic model of seismic failure modes was created by using natural gradient boosting and gradient boosting regressor algorithms. Then a new hyperparameter optimization method for the improved NGB probabilistic model was proposed based on the ML algorithms and the physical laws. Finally, the predictive performance and engineering practicability of the PNGB probabilistic model were verified compared to empirical and ML models. According to the analysis results, the PNGB model can effectively improve the predictive recall of FF, FSF and SF by about 19–65 %, 20–67 % and 20–56 % respectively compared with empirical models, and by about 1–7 %, 7–43 % and 6–32 % respectively compared with ML models. Moreover, the PNGB model exhibits satisfactory generalization performance and minimal dispersion, which can be used to calibrate the ML models based on the confidence intervals. Furthermore, the result predicted by the PNGB model satisfies the physical laws relating to the seismic failure modes and design parameters, which can further represent the competitive relationships of various seismic failure modes.

Probabilistic fatigue life prediction of notched specimens based on modified stress field intensity method under multiaxial loading

In practical engineering components, due to the existence of non-uniform stress and strain field near the notch, it brings severe challenges to fatigue life prediction when evaluating the integrity of notched components. In this study, a probabilistic fatigue life prediction model for notched specimens was established by coupling the stress field intensity (SFI) method and Weibull distribution. Firstly, the position of the dangerous point is determined by finite element calculation, and the maximum strain energy density plane through the dangerous point is defined as the critical plane. Secondly, from the perspective of 2D features, the traditional SFI method is modified based on the stress distribution on the critical plane, and a new concept of effective stress is proposed to predict the fatigue life of notched specimens by the experimental data of smooth specimens. Finally, a new non-proportional additional hardening factor is established to characterize the influence of material properties and loading path on fatigue life. The experimental data of Q345 low alloy steel and GH4169 nickel base alloy are used to compare and analyze the proposed model. The results show that the predicted life of the proposed model is in good agreement with the experimental life.