Probabilistic Prediction Model Research Articles

Bubbles in biomass fluidized bed gasifiers: A phenomenological probabilistic predictive model

Bubble properties in fluidized beds including size, velocity, and path, vary randomly. This article proposes a new PPPM (Phenomenological Probabilistic Predictive Model) l, that allows one to establish a vrb-BAC prediction band, to describe bubble data, obtained in a lab scale biomass-sand fluidized bed, by using CREC Optiprobes. The PPPM, represents a significant improvement from an earlier Probabilistic Predictive Model (PPM) reported by Torres and de Lasa in 2021, with the new PPPM being based on a balance of forces including gravity, drag, buoyancy, and minimum fluidization excess velocity. The resulting PPPM bubble rise velocity (vrb) is obtained as a function of the bubble axial chord (BAC) and is specifically evaluated by studying bubbles in a high-density sand (above 2500 kg/m3) fluidized bed. The PPPM leads to a behavioral band showing a normal bubble frequency distribution that becomes increasingly skewed as the gas superficial velocity increases. The proposed PPPM is shown to be applicable to single bubbles in a sand bed at incipient fluidization and to multiple bubbles in a biomass-sand bubbling bed. The PPPM is validated in the present study, by comparing it against 46,000 bubbles, under different gas superficial velocities and biomass loadings. By using the PPPM, it is shown 91% of all bubbles studied fall within the PPPM prediction band. Given the demonstrated ability of the PPPM, it is anticipated that this model will significantly contribute to the development of large-scale multiphase biomass gasifier models.

Towards an explainable clinical decision support system for large-for-gestational-age births.

A myriad of maternal and neonatal complications can result from delivery of a large-for-gestational-age (LGA) infant. LGA birth rates have increased in many countries since the late 20th century, partially due to a rise in maternal body mass index, which is associated with LGA risk. The objective of the current study was to develop LGA prediction models for women with overweight and obesity for the purpose of clinical decision support in a clinical setting. Maternal characteristics, serum biomarkers and fetal anatomy scan measurements for 465 pregnant women with overweight and obesity before and at approximately 21 weeks gestation were obtained from the PEARS (Pregnancy Exercise and Nutrition with smart phone application support) study data. Random forest, support vector machine, adaptive boosting and extreme gradient boosting algorithms were applied with synthetic minority over-sampling technique to develop probabilistic prediction models. Two models were developed for use in different settings: a clinical setting for white women (AUC-ROC of 0.75); and a clinical setting for women of all ethnicity and regions (AUC-ROC of 0.57). Maternal age, mid upper arm circumference, white cell count at the first antenatal visit, fetal biometry and gestational age at fetal anatomy scan were found to be important predictors of LGA. Pobal HP deprivation index and fetal biometry centiles, which are population-specific, are also important. Moreover, we explained our models with Local Interpretable Model-agnostic Explanations (LIME) to improve explainability, which was proven effective by case studies. Our explainable models can effectively predict the probability of an LGA birth for women with overweight and obesity, and are anticipated to be useful to support clinical decision-making and for the development of early pregnancy intervention strategies to reduce pregnancy complications related to LGA.

Will the Order Be Canceled? Order Cancellation Probability Prediction Based on Deep Residual Model

Online ride-hailing is gradually becoming one of the indispensable travel modes for people. Online ride-hailing platforms receive tens of thousands of taxi travel orders every day. However, some orders are canceled by the user after the driver answers. User cancellation of orders will disrupt the regular operation of the taxi platform, reduce the efficiency and income of drivers on the online ride-hailing platform, and affect the ride experience of other users. To capture the relationship between the cancellation probability of online taxi orders and users, we first analyze the factors that affect users’ cancellation of online taxi orders. We find that the cancellation probability of online ride-hailing orders is highly correlated with the travel time of the online taxi, the distance between vehicle and passengers, road congestion conditions, and alternative transportation resources around the passengers. Second, a deep residual network-based ride-hailing order cancellation probability prediction model (DeepOCP) is designed to predict the probability of a user canceling an answered online-hailing order. Finally, the Didi Chengdu Express public data set is used to test the model’s effectiveness. This paper is the first study to use a deep-learning model to predict the cancellation probability of online ride-hailing orders.

Fluvial Flood Losses in the Contiguous United States Under Climate Change

AbstractFlooding is one of the most devastating natural disasters causing significant economic losses. One of the dominant drivers of flood losses is heavy precipitation, with other contributing factors such as built environments and socio‐economic conditions superimposed to it. To better understand the risk profile associated with this hazard, we develop probabilistic models to quantify the future likelihood of fluvial flood‐related property damage exceeding a critical threshold (i.e., high property damage) at the state level across the conterminous United States. The model is conditioned on indicators representing heavy precipitation amount and frequency derived from observed and downscaled precipitation. The likelihood of high property damage is estimated from the conditional probability distribution of annual total property damage, which is derived from the joint probability of the property damage and heavy precipitation indicators. Our results indicate an increase in the probability of high property damage (i.e., exceedance of 70th percentile of observed annual property damage for each state) in the future. Higher probability of high property damage is projected to be clustered in the states across the western and south‐western United States, and parts of the U.S. Northwest and the northern Rockies and Plains. Depending on the state, the mean annual probability of high property damage in these regions could range from 38% to 80% and from 46% to 95% at the end of the century (2090s) under RCP4.5 and RCP8.5 scenarios, respectively. This is equivalent to 20%–40% increase in the probability compared to the historical period 1996–2005. Results show that uncertainty in the projected probability of high property damage ranges from 14% to 35% across the states. The spatio‐temporal variability of the uncertainty across the states and three future decades (i.e., 2050s, 2070s, and 2090s) exhibits nonstationarity, which is driven by the uncertainty associated with the probabilistic prediction models and climate change scenarios.

Probabilistic Hotspot Prediction Model Based on Bayesian Inference Using Precipitation, Relative Dry Spells, ENSO and IOD

Increasing global warming can potentially increase the intensity of ENSO and IOD extreme phenomena in the future, which could increase the potential for wildfires. This study aims to develop a hotspot prediction model in the Kalimantan region using climate indicators such as precipitation and its derivatives, ENSO and IOD. The hotspot prediction model was developed using Principal Model Analysis (PMA) as the initial model basis. The overall model performance is evaluated using the concept of Cross-Validation. Furthermore, the model’s performance will be improved using the Bayesian Inference principle so that the average performance increases from 28.6% to 61.1% based on the model’s coefficient of determination (R2). The character of each year in the model development process is also evaluated using the concept of cross validation. Since the climate indicator we used was integrated with the ENSO and IOD index, model performance is strongly influenced by the ENSO and IOD phenomena. To obtain better performance when estimating future forest fires (related to El Niño and positive IOD), years with a high number of hotspots and coinciding with the occurrence of El Niño and IOD are better used as early model years (PMA). However, the model tends to overestimate the hotspot value, especially with a lower strength El Niño and positive IOD. Therefore, years with a low number of hotspots, as in normal years and La Niña, are better used in the model performance improvement stage (Bayesian Inference) to correct the overestimation.

Study on the Limit of Moisture Content of Smoldering Humus during Sub-Surface Fires in the Boreal Forests of China

A sub-surface forest fire is a kind of fire that spreads slowly with no flames and lower temperatures, and threatens the ecosystem and human life. The moisture content of humus is considered to be an important factor in determining fire occurrence and sustaining. The humus of the Larix gmelinii in the Daxing’an Mountains was selected for the experiment, the limit moisture content condition of sub-surface forest fires was determined by an experiment simulating smoldering, and the prediction model of the probability of sub-surface forest fire occurrence was established. The results will be of great significance for the prevention, monitoring, and fighting of sub-surface forest fires in the boreal forest. The results showed that when the moisture content of humus in the upper layer was low, the smoldering process could be self-sustaining at 20%. For deeper layers of a depth of 18 cm, this increased to 30% moisture content of the humus and was the critical depth for sub-surface fires. The moisture content of 40% was a limit to burning where smoldering can only last for a short duration and is then extinguished. When the moisture content of the humus was 20%, the smoldering temperature was higher and the rate of spread was faster, with smoldering being maintained for longer periods at 30% moisture content. The regression prediction model of the highest temperature and vertical rate of spread in a column of humus was correlated to moisture content and depth, and the model significance was good at p < 0.01. Based on moisture content and depth, the occurrence probability prediction model of sub-surface fires has a good correlation (R2 = 0.93) and high prediction accuracy (AUC = 0.995). The effect of moisture content (Or = 4.008) on the occurrence probability of sub-surface fires is higher than that of depth (Or = 2.948). The results point out that it is necessary to prevent and monitor the occurrence of sub-surface fires when the humus moisture content is less than 40%. In order to reduce the risk of sub-surface fires, the monitoring time of the fire field should be extended after the fire is extinguished due to the slow-burning process of the sub-surface fire. Increasing the moisture content of the humus is an important method to reduce the probability and restrain the spread of sub-surface fires.

Reliability-based assessment of scour at pier in gravel beds

ABSTRACT Estimation of local scour depth at a bridge pier must be accurate and precise for the safety assessment of pier foundation. Many deterministic scour depth prediction models for bridge piers in sands and gravels have already been reported in the literature. However, the various types of uncertainties may be involved in the deterministic scour depth prediction models and thus affect the accuracy of the scour depth. To account for such uncertainties, a reliability analysis method is generally adopted. A sizable amount of laboratory and field data on scour has been procured in the present study to assess the performance of existing scour prediction models for gravel beds and develop a probabilistic scour prediction model using a First-Order Reliability Method (FORM) based on the object-oriented constraints optimization spreadsheet algorithm for the most appropriate selected deterministic scour prediction model. A new relationship of the silt factor has also been developed to estimate the scour depth in gravel bed as per Indian Practice. A safety factor based on the target reliability index has also been proposed in this study.

Strong aftershocks traffic light system: A case study of the 8 January 2022 MS6.9 Menyuan earthquake, Qinghai Province, China

Strong aftershocks, especially the disaster-causing M≥5.0 kind, are a key concern for mitigation of seismic risks because they often lead to superimposed earthquake damage. However, the real-time forecasting results of the traditional probability prediction models based on statistics are usually far from accurate and therefore unsatisfactory. Borrowing an idea from the foreshock traffic light system (FTLS), which is based on observations of decreasing b-values or increasing differential stress just before a strong aftershock, we constructed a strong aftershock traffic light system (SATLS) that uses data-driven technology to improve the reliability of time sequence b-value calculations, and analyzed the b-value variations of strong aftershocks in the China continent. We applied this system to the MS6.9 Menyuan earthquake occurred on 8 January 2022. The earthquake occurrence rates before the largest aftershock (MS5.2) forecast by the Omi-R-J model were too low, although the model could accurately forecast aftershock rates for each magnitude interval in most time-periods. However, reliable b-values can be calculated using the time-sequence b-value data-driven (TbDD) method, and the results showed that the b-values continued declining from 1.3 days before the MS5.2 aftershock and gradually recovered afterward. This would suggest that the stress evolution in the focal area can provide data for deciding when to post risk alerts of strong aftershocks. In the process of building the SATLS, we studied thirty-four M≥6.0 intraplate earthquake sequences in the China continent and concluded that the differences between the b-values of the aftershock sequences and of the background events, △b = bafter - bbg = ±0.1, could be used as thresholds to determine whether M≥5.0 aftershocks would occur. The △b value obtained using the events before the MS5.2 aftershock of the MS6.9 Menyuan sequence was about -0.04, which would have caused the SATLS to declare a yellow alert, but there would have been some gap expected before a red alert was triggered by the b-value difference derived from the events associated with this strong aftershock. To accurately forecast a strong aftershock of M≥5.0, a deeper understanding of the true b-value and a detailed description of the stress evolution state in the source area is necessary.

The occurrence probability prediction model of 2D and 3D freak waves generated by wave superposition

The wave superposition is a common method to generate freak waves by assigning energy of wave spectrum to a focusing wave train and a random wave train. Traditionally, the freak wave generation by wave superposition requires repeatedly try for many times before obtaining satisfactory results. In this process, freak wave may not occur when wave component number or energy portion of focusing wave train is small and freak wave may break up when that is too large. It may also contribute to the discrepancy of significant wave height between target value and generated result when energy portion of focusing wave train is large. To improve simulation efficiency, the paper proposes a model to quantitatively predict the occurrence probability of freak waves in superposition cases of two and three wave trains. The error of significant wave height could be artificially controlled. The occurrence probability prediction model of freak wave is also extended to three-dimensional wave fields. The accuracy is well validated by statistical data. According to prediction results, two-dimensional and three-dimensional freak waves are successfully generated.

Progression of Bio-Modified Adobes with Derivatives of Cassava Periderm and Clay Brick Waste by Experimental and Probabilistic Prediction Models

Progression of Bio-Modified Adobes with Derivatives of Cassava Periderm and Clay Brick Waste by Experimental and Probabilistic Prediction Models

Fundamental study on probabilistic generative modeling of earthquake ground motion time histories using generative adversarial networks

AbstractThis study proposes a probabilistic model for earthquake ground motion prediction, named ground motion generation model, which can generate ground motion time history data directly. The ground motion generation model is based on a data‐driven technique called generative adversarial networks, allowing generation of ground motion time history data without making assumptions about physical or statistical models. A method to quantitatively and qualitatively evaluate the performance of constructed model is also proposed and the ground motion generation model is optimized for high performance from earthquake engineering and deep learning perspectives. Numerical experiments show that our proposed model is probabilistic, approximating the probabilistic distribution of the dataset of observed records and generating realistic ground motion time histories with various characteristics in the time and frequency domains.

Data-driven ensemble model for probabilistic prediction of debris-flow volume using Bayesian model averaging

Data-driven ensemble model for probabilistic prediction of debris-flow volume using Bayesian model averaging

Development of physiologically-based gut absorption model for probabilistic prediction of environmental chemical bioavailability_suppl

Development of physiologically-based gut absorption model for probabilistic prediction of environmental chemical bioavailability_suppl

Development of physiologically-based gut absorption model for probabilistic prediction of environmental chemical bioavailability.

Absorption in the gastrointestinal tract is a key factor determining the bioavailability of chemicals after oral exposure but is frequently assumed to have a conservative value of 100% for environmental chemicals, particularly in the context of high-throughput toxicokinetics for in vitro-to-in vivo extrapolation (IVIVE). For pharmaceutical compounds, the physiologically based advanced compartmental absorption and transit (ACAT) model has been used extensively to predict gut absorption but has not generally been applied to environmental chemicals. Here we develop a probabilistic environmental compart­mental absorption and transit (PECAT) model, adapting the ACAT model to environmental chemicals. We calibrated the model parameters to human in vivo, ex vivo, and in vitro datasets of drug permeability and fractional absorption by con­sidering two key factors: (1) differences between permeability in Caco-2 cells and in vivo permeability in the jejunum, and (2) differences in in vivo permeability across different gut segments. Incorporating these factors probabilistically, we found that given Caco-2 permeability measurements, predictions of the PECAT model are consistent with the (limited) available gut absorption data for environmental chemicals. However, the substantial chemical-to-chemical variability observed in the cal­ibration data often led to wide probabilistic confidence bounds in the predicted fraction absorbed and resulting steady state blood concentration. Thus, while the PECAT model provides a statistically rigorous, physiologically based approach for incor­porating in vitro data on gut absorption into toxicokinetic modeling and IVIVE, it also highlights the need for more accurate in vitro models and data for measuring gut segment-specific in vivo permeability of environmental chemicals.

Online Fault Prediction Based on Collaborative Filtering in Smart Grid

Smart grid, responsible for upgrading traditional power networks by integrating with cutting‐edge information and communication networks, forms coupled networks but also pose potential hazards in the face of fault cascade. In coupled networks, fault prediction is of significance because tight interaction between power nodes and communication nodes makes the smart grid more vulnerable. Unfortunately, most existing works of fault prediction are specific to a single network and do not consider the correlation of coupled elements. To address these limitations, in this paper, we highlight the interdependence of networks and define fault correlation. Further, we propose a probabilistic prediction model using collaborative filtering in machine learning. We finally present an online prediction algorithm. We conduct experiments to illustrate the effectiveness of our prediction algorithm with different parameters and give some observations that may give more insight into interdependent networks.

A unified probabilistic fatigue life prediction model for natural rubber components considering strain ratio effect

AbstractThe fatigue life of natural rubber (NR) components under a constant‐amplitude load is dispersed. The dispersion can be reflected by establishing a probabilistic fatigue life prediction model (also called the P–S–N curve). However, the effect of strain ratio on fatigue life is typically ignored in conventional P–S–N curves. Aiming at improving the prediction accuracy of conventional models, a unified P–S–N curve for NR components is proposed in the present study, in which the strain ratio effect is considered in the calculations. Using the dumbbell‐shaped rubber specimen fatigue test data, a polynomial function is proposed to describe the correlation between the equivalent strain amplitude, strain amplitude, and strain ratio. Then a calculation model of equivalent strain amplitude is established. Furthermore, a unified P–S–N curve for NR is established with equivalent strain amplitude as the damage parameter. By comparing the measured and predicted fatigue life of NR under different reliabilities, the accuracy of the established unified P–S–N curve is verified.

Probabilistic prediction of joint shear strength using Gaussian process regression with anisotropic compound kernel

Shear strength of reinforced concrete (RC) beam-column joints is inevitably affected by various uncertainties existing in complex shear behaviors. However, traditional mechanics-based and data-driven methods fail to reasonably quantify these uncertainties. In order to overcome this limitation, a probabilistic prediction model for shear strength of RC beam-column joints based on the Gaussian process regression (GPR) with anisotropic compound kernel was proposed. Firstly, the GPR algorithm was employed to establish the probabilistic shear strength model of RC beam-column joints. Then the anisotropic compound kernel capturing complex variation characteristics of shear strength of RC beam-column joints was designed based on the sum of single kernels and the automatic relevance determination function. Meanwhile, the characteristic length-scales of the designed kernel were used to analyze the feature importance of shear strength of RC beam-column joints. Moreover, the anisotropic compound kernel was integrated to the GPR algorithm and the hyper-parameter optimization of the developed GPR was performed based on the maximum likelihood estimation. Finally, the developed GPR was verified by comparing with traditional models and 237 sets of experimental data. Analysis results show that the proposed kernel can effectively improve the prediction accuracy and generalization performance of the GPR. Comparing with traditional data-driven methods, the developed GPR not only has satisfactory estimation and generalization ability, but also describes probabilistic characteristics of shear strength of RC beam-column joints. Mechanics-based models were further calibrated based on the confidence interval of the developed GPR.

Simultaneous Astronaut Accompanying and Visual Navigation in Semi-Structured and Dynamic Intravehicular Environment

The application of intravehicular robotic assistants (IRA) can save valuable working hours for astronauts in space stations. There are various types of IRA, such as an accompanying drone working in microgravity and a dexterous humanoid robot for collaborative operations. In either case, the ability to navigate and work along with human astronauts lays the foundation for their deployment. To address this problem, this paper proposes the framework of simultaneous astronaut accompanying and visual navigation. The framework contains a customized astronaut detector, an intravehicular navigation system, and a probabilistic model for astronaut visual tracking and motion prediction. The customized detector is designed to be lightweight and has achieved superior performance (AP@0.5 of 99.36%) for astronaut detection in diverse postures and orientations during intravehicular activities. A map-based visual navigation method is proposed for accurate and 6DoF localization (1~2 cm, 0.5°) in semi-structured environments. To ensure the robustness of navigation in dynamic scenes, feature points within the detected bounding boxes are filtered out. The probabilistic model is formulated based on the map-based navigation system and the customized astronaut detector. Both trajectory correlation and geometric similarity clues are incorporated into the model for stable visual tracking and trajectory estimation of the astronaut. The overall framework enables the robotic assistant to track and distinguish the served astronaut efficiently during intravehicular activities and to provide foresighted service while in locomotion. The overall performance and superiority of the proposed framework are verified through extensive ground experiments in a space-station mockup.

A probabilistic model predicting host specificity and host range expansion in mites parasitic on mammals

The emergence of new mammalian diseases through the switching of parasitic organisms to novel hosts has significant wildlife, livestock, and human health impacts, however, new host switches are notoriously difficult to predict. The factors influencing the host switches are not fully understood because complete unbiased large-scale datasets of host-parasite relationships are lacking. Building a large-scale model to identify these factors and predict potential host switching (host range expansion) could hold substantial benefits from theoretical and applied perspectives (e.g., disease emergence prediction). Here we analyzed a large, curated database of host-parasite relationships of 1906 species of acariform mites forming permanent associations with 1235 species of mammals, including humans, and built a probabilistic predictive model of host-specificity for these mites. There was a total of 3,125 unique host-parasite records.

A model established for predicting natural pregnancy possibility based on the imaging characteristics of 4-dimensional hysterosalpingo-contrast sonography.

The incidence of infertility is increasing, more than 30% of them having related abnormal tubal patency. Four-dimensional (4D) hysterosalpingo-contrast sonography (HyCoSy) overcomes the shortcomings of 3D HyCoSy in the diagnosis of tubal patency, showing high specificity and accuracy. In addition, 4D HyCoSy discards iodine allergy and X-ray radiation and possesses easy-operating, contributing to good acceptance in clinical practice. However, there is no research to explore the imaging standards related to the possibility of natural pregnancy after 4D HyCoSy. If a predictive model of postoperative natural pregnancy was established using the analysis of clinical data combined with imaging characteristics of 4D HyCoSy of patients with tubal factor infertility, clinical decision-making can be wisely guided in the future. This study aims to establish a predictive model of natural pregnancy after 4D HyCoSy based on clinical data and imaging characteristics of patients with tubal factor in fertility. A retrospective study was conducted for patients who were diagnosed with tubal factor infertility in Hunan Guangxiu Hospital from February 2017 to May 2018. The patients ought to possess complete 4D HyCoSy imaging data and at least one-side-unobstructed fallopian tube. General clinical data and imaging data were collected. Pregnancy outcome was followed up till 3 months after 4D HyCoSy. According to pregnancy outcome, patients were divided into a pregnancy group and a non-pregnancy group. Binary logistic regression was used to analyze the correlation between various variables and natural pregnancy after 4D HyCoSy. The variables with significant difference (P<0.05) in single-factor logistic regression were included in the natural pregnancy probability prediction model. The classification accuracy was further verified with 10-fold cross-validation. A total of 1 085 patients with clinically suspected tubal factor infertility who met the requirements and followed the doctors' prescription were collected. Clinical characteristics (age and duration of infertility) and 4D HyCoSy imaging characteristics (thickness of endometrium from the 3rd to the 7th day after the end of menstruation, visualization of the left fallopian tube, the diffusion of contrast agent around the left ovary, and the diffusion of contrast agent around the right ovary) were independent predictors for natural pregnancy 3 months after 4D HyCoSy. A natural pregnancy probability prediction model was established with the area under the curve (AUC) verified by the 10-fold cross-validation all greater than 0.75, and the best AUC was 0.868. The Q value obtained by the prediction model was the probability of natural pregnancy, and the cutoff value was 0.5. When the Q value was greater than 0.5, it was recommended to attempt natural pregnancy for 3 months, and when the Q value was less than 0.5, in vitro fertilization was adviced. A predictive model for the evaluating probability of natural pregnancy in women with tubal factor infertility after 4D HyCoSy is successfully established based on the analysis for clinical data and imaging characteristics. This model shows a great potential in assisting clinical decision making.