Prior Parameters Research Articles

Image detection of insulator defects in complex environments

Aiming at the practical problems of low detection accuracy and low detection speed for the detection of power insulators and insulator defects in a complex environmental context, a method is proposed to improve the YOLOv4 algorithm for the detection of power insulator images and insulators with defects. By making a dataset of electric power insulators and insulators with defects, the K-means algorithm is used to cluster the electric power insulator image samples to obtain different sizes of prior box parameters; then a weighting factor is introduced by improving the balanced cross-entropy to increase the contribution degree of the loss function; finally, the depth of the network is deepened by adding convolution layers before and after the spatial pyramidal pooling structure. The experimental results show that the single detection time of the improved model is 3.27S, and the average detection accuracy for insulator defects is improved by 24.36% compared with the original YOLOv4 algorithm. Also the value of the mean average accuracy on the test set by the improved YOLOv4 algorithm is 84.05%, which is 17.83% better than the original YOLOv4 algorithm, fully demonstrating the ability to locate and identify the defects existing in power insulator images well.

Stock Assessment of Chub Mackerel (Scomber japonicus) in the Northwest Pacific Using a Multi-Model Approach

Chub mackerel (Scomber japonicus) is a major targeted species in the Northwest Pacific Ocean, fished by China, Japan, and Russia, and predominantly captured with purse seine fishing gear. A formal stock assessment of Chub mackerel in the region has yet to be implemented by the managing authority, that is, the North Pacific Fisheries Commission (NPFC). This study aims to provide a wider choice of potential models for the stock assessment of Chub mackerel in the Northwest Pacific using available data provided by members of the NPFC. The five models tested in the present study are CMSY, BSM, SPiCT, JABBA, and JABBA-Select. Furthermore, the influence of different data types and input parameters on the performance of the different models used was evaluated. These effects for each model are catch time series for CMSY, catch time series and prior of the relative biomass for BSM, prior information for SPiCT, and selectivity coefficients for JABBA-Select. Catch and CPUE (catch per unit effort) data used are derived from NPFC, while some life history information is referred from other references. The results indicate that Chub mackerel stock might be slightly overfished, as indicated by CMSY (B2020/BMSY = 0.98, F2020/FMSY = 1.12), BSM (B2020/BMSY = 0.97, F2020/FMSY = 1.21), and the base case run for the JABBA-Select (SB2020/SBMSY = 0.99, H2020/HMSY = 0.99) models. The results of the models SPiCT (B2020/BMSY = 2.30, F2020/FMSY = 0.31) and JABBA (B2020/BMSY = 1.40, F2020/FMSY = 0.62) showed that the state of this stock may be healthy. Changes in the catch time series did not affect CMSY results but did affect BSM. The present study confirms that prior information for BSM and SPiCT models is very important in order to obtain reliable results on the stock status. The results of JABBA-Select showed that different selectivity coefficients can affect the stock status of a species, as observed in the present study. Based on the optimistic stock status indicated by the best model, JABBA, a higher catch is allowable, but further projection is required for specific catch limit setting. Results suggested that, as a precautionary measure, management would be directed towards maintaining or slightly reducing the fishing effort for the sustainable harvest of this fish stock, while laying more emphasis on accurately estimating prior input parameters for use in assessment models.

Calibration method for sensor drifting bias in data center cooling system using Bayesian Inference coupling with Autoencoder

The optimal control strategy is considered a promising solution to reduce the energy consumption of cooling supply systems in data centers. The strategy is mainly based on sensor measurement data to adjust and control, thereby achieving energy-saving effects. However, the high thermal density characteristics of data centers can cause drifting bias of the sensor, resulting in failure to achieve the desired optimal control strategy. This study proposed a novel method based on Bayesian inference (BI) coupled with autoencoder to model the drifting bias correlation functions and calibrate the sensor drifting bias. Therein, Markov Chain Monte Carlo (MCMC) algorithm was used to generate the equivalent samples without complex integral solution process. Case studies of eight individual/multiple sensors drifting bias scenarios in the computer room air handler (CRAH), were conducted to comprehensively evaluate the calibration performance of the proposed method. Moreover, the effect of prior distribution parameters setting on the calibration performance was also discussed. The simulation results show that the proposed method performs well in terms of correction accuracy exceeding 92.60% and 96.34% for individual sensor and multiple sensors drifting bias scenarios, respectively. In addition, the updated strategy of the prior distribution parameter settings plays a crucial role in improving the correction accuracy, especially for multiple sensors drifting bias scenarios. This study filled the knowledge gap of sensor drifting bias correction for data center cooling supply systems and facilitates the application of advanced technologies in data centers.

A novel power prior approach for borrowing historical control data in clinical trials.

There has been an increased interest in borrowing information from historical control data to improve the statistical power for hypothesis testing, therefore reducing the required sample sizes in clinical trials. To account for the heterogeneity between the historical and current trials, power priors are often considered to discount the information borrowed from the historical data. However, it can be challenging to choose a fixed power prior parameter in the application. The modified power prior approach, which defines a random power parameter with initial prior to control the amount of historical information borrowed, may not directly account for heterogeneity between the trials. In this paper, we propose a novel approach to pick a power prior based on some direct measures of distributional differences between historical control data and current control data under normal assumptions. Simulations are conducted to investigate the performance of the proposed approach compared with current approaches (e.g. commensurate prior, meta-analytic-predictive, and modified power prior). The results show that the proposed power prior improves the study power while controlling the type I error within a tolerable limit when the distribution of the historical control data is similar to that of the current control data. The method is developed for both superiority and non-inferiority trials and is illustrated with an example from vaccine clinical trials.

Convergence properties of data augmentation algorithms for high-dimensional robit regression

The logistic and probit link functions are the most common choices for regression models with a binary response. However, these choices are not robust to the presence of outliers/unexpected observations. The robit link function, which is equal to the inverse CDF of the Student’s t-distribution, provides a robust alternative to the probit and logistic link functions. A multivariate normal prior for the regression coefficients is the standard choice for Bayesian inference in robit regression models. The resulting posterior density is intractable and a Data Augmentation (DA) Markov chain is used to generate approximate samples from the desired posterior distribution. Establishing geometric ergodicity for this DA Markov chain is important as it provides theoretical guarantees for asymptotic validity of MCMC standard errors for desired posterior expectations/quantiles. Previous work [16] established geometric ergodicity of this robit DA Markov chain assuming (i) the sample size n dominates the number of predictors p, and (ii) an additional constraint which requires the sample size to be bounded above by a fixed constant which depends on the design matrix X. In particular, modern high-dimensional settings where n<p are not considered. In this work, we show that the robit DA Markov chain is trace-class (i.e., the eigenvalues of the corresponding Markov operator are summable) for arbitrary choices of the sample size n, the number of predictors p, the design matrix X, and the prior mean and variance parameters. The trace-class property implies geometric ergodicity. Moreover, this property allows us to conclude that the sandwich robit chain (obtained by inserting an inexpensive extra step in between the two steps of the DA chain) is strictly better than the robit DA chain in an appropriate sense, and enables the use of recent methods to estimate the spectral gap of trace class DA Markov chains.

Modeling Uncertainty Quantification of NDVI of Agricultural Fields through Bayesian Linear Regression in Time Series Prediction

The current research discusses the applications of Bayesian linear regression to predict the uncertainty of remote sensing data. To predict the uncertainty, the study considered the SENTINEL-2 satellite data of agricultural fields of Uttar Pradesh state of India. Using the stratified sampling method in Google Earth Engine, the random points generated are mapped to agricultural fields. Data was collected in the form of maximum Normalized Difference Vegetation Index (NDVI) values of each agricultural field. The dynamics of the time series predictions were explored with Bayesian linear regression, a probabilistic deep learning method. The model uncertainty defined as epistemic uncertainty is evaluated with the prior and posterior probability parameters of Bayesian statistics in linear regression. The number of regression lines predicted for the same data shows evidence of uncertainty. The Bayesian linear regression models show evidence of high uncertainty for the predicted NDVI values. The variation in model uncertainty is measured by dividing the dataset into samples and it is observed that with increase in data the uncertainty is reduced. Also, with the increase in data, the posterior density becomes sharper which corresponds to a decrease in variance. Further, the study extended the concept of regression analysis with Gaussian basis functions to determine the effect of model uncertainty with an increase in data. The analysis has shown the same result in knowing the effect of uncertainty with the increase in data. Further, a nonlinear polynomial regression model with a Gaussian distribution as a basis function was developed to evaluate the marginal probabilities of the evidence function in capturing the uncertainty with varying degrees of freedom. The polynomial regression with a Gaussian distribution using Bayesian statistics has captured the uncertainty and confirmed that the uncertainty is captured at lower degrees of freedom.

Joint Activity Detection and Channel Estimation for Massive IoT Access Based on Millimeter-Wave/Terahertz Multi-Panel Massive MIMO

The multi-panel array, as a state-of-the-art antenna-in-package technology, is very suitable for millimeter-wave (mmWave)/ terahertz (THz) systems, due to its low-cost deployment and scalable configuration. But in the context of non-uniform array structures it leads to intractable signal processing. Based on such an array structure at the base station, this paper investigates a joint active user detection (AUD) and channel estimation (CE) scheme based on compressive sensing (CS) for application to the massive Internet of Things (IoT). Specifically, by exploiting the structured sparsity of mmWave/THz massive IoT access channels, we firstly formulate the multi-panel massive multiple-input multiple-output (mMIMO)-based joint AUD and CE problem as a multiple measurement vector (MMV)-CS problem. Then, we harness the expectation maximization (EM) algorithm to learn the prior parameters (i.e., the noise variance and the sparsity ratio) and an orthogonal approximate message passing (OAMP)-EM-MMV algorithm is developed to solve this problem. Our simulation results verify the improved AUD and CE performance of the proposed scheme compared to conventional CS-based algorithms.

Weighing patient attributes in antibiotic prescribing for upper respiratory tract infections: A discrete choice experiment on primary care physicians in Hubei Province, China.

This study aimed to determine how primary care physicians weigh intervenable patient attributes in their decisions of antibiotic prescribing for upper respiratory tract infections (URTIs). A discrete choice experiment (DCE) was conducted on 386 primary care physicians selected through a stratified cluster sampling strategy in Hubei province, China. The patient attributes tested in the DCE were identified through semi-structured interviews with 13 primary care physicians, while the choice scenarios were determined by a D-efficient design with a zero prior parameter value. Conditional logit models (CL) and mixed logit models (MXL) were established to determine the preference of the study participants in antibiotic prescribing for URTI patients with various attributes. Relative importance (RI) was calculated to reflect the influence of each attribute. In addition to age and duration of symptoms, the interventionable patient attributes were also considered by the primary care physicians in their antibiotic prescribing decisions. They preferred to prescribe antibiotics for URTI patients with difficulties to schedule a follow-up appointment (p < 0.001) and for those without a clear indication of refusal to antibiotics (p < 0.001). Patient request for antibiotics had an RI ranging from 15.2 to 16.3%, compared with 5.1-5.4% for easiness of follow-up appointment. The influence of these two interventionable patient attributes was most profound in the antibiotic prescribing decisions for patients aged between 60 and 75 years as indicated by their interaction effects with age (β = 0.69 for request for antibiotics, p < 0.01; β = -1.2 for easiness of follow-up, p < 0.001). Reducing patient pressure and improving accessibility and continuity of care may help primary care physicians make rational antibiotic prescribing decisions for URTIs.

The backward problem of a stochastic PDE with bi-harmonic operator driven by fractional Brownian motion

ABSTRACT This paper is concerned with a backward problem of a stochastic partial differential equation with bi-harmonic operator. The source term is driven by fractional Brownian motion. Based on the Gevrey-type space, the regularity of the mild solution is studied. However, this problem is ill-posed since it is unstable. The instability is discussed in the sense of expectation and variance. Moreover, a regularization method is proposed. The error estimation between the regularization solution and the mild solution is given using an a prior parameter choice rule.

Inverse Estimation of Soil Hydraulic Parameters in a Landslide Deposit Based on a DE-MC Approach

Extreme rainfall is a common triggering factor of landslide disasters, for infiltration and pore water pressure propagation can reduce suction stress and shear strength at the slip surface. The subsurface hydrological model is an essential component in the early-warning system of rainfall-triggered landslides, whereas soil moisture and pore water pressure simulated by the Darcy–Richards equation could be significantly affected by uncertainties in soil hydraulic parameters. This study conducted an inverse analysis of in situ measured soil moisture in an earthquake-induced landslide deposit, and the soil hydraulic parameters were optimized with the Differential Evolution Markov chain Monte Carlo method (DE-MC). The DE-MC approach was initially validated with a synthetic numerical experiment to demonstrate its effectiveness in finding the true soil hydraulic parameters. Besides, the soil water characteristic curve (SWCC) and hydraulic conductivity function (HCF) described with optimized soil hydraulic parameter sets had similar shapes despite the fact that soil hydraulic parameters may be different. Such equifinality phenomenon in inversely estimated soil hydraulic parameters, however, did not affect the performance of simulated soil moisture dynamics in the synthetic numerical experiment. The application of DE-MC to a real case study of a landslide deposit also indicated satisfying model performance in terms of accurate match between the in situ measured soil moisture content and ensemble of simulations. In conclusion, based on the satisfying performance of simulated soil moisture and the posterior probability density function (PDF) of parameter sets, the DE-MC approach can significantly reduce uncertainties in specified prior soil hydraulic parameters. This study suggested the integration of the DE-MC approach with the Darcy–Richards equation for an accurate quantification of unsaturated soil hydrology, which can be an essential modeling strategy to support the early-warning of rainfall-triggered landslides.

Corrosion fatigue crack growth prediction of bridge suspender wires using Bayesian gaussian process

This paper proposes a Bayesian gaussian process-based corrosion fatigue damage assessment framework for bridge suspender wires. Prior statistical parameters in the fatigue life prediction model are obtained by corrosion fatigue crack growth test under different stress ratios. A numerical simulation method of corrosion fatigue crack growth is developed combined with ABAQUS and FRANC3D. Following this, a Bayesian network-based framework is established by integrating fatigue crack growth model, stress concentration factor and observed fatigue crack length information to reduce the uncertainties. The uncertainties exist in stress concentration factor are quantified by gaussian process regression-based surrogate model, and the pit geometry of length, width and depth are selected as three-dimensional input. Markov Chain Monte Carlo simulation is used to obtain the posterior distribution of parameters in the Bayesian network. The results show that the fatigue crack growth predictions agree well with experimental observations. The proposed method can simplify the complicated modeling process of stress concentration factor under the influence of corrosion pit, and can effectively update the model parameters to accurately predict the corrosion fatigue crack of bridge suspender wires.

A simple and robust model for enrollment projection in clinical trials

Enrollment projection in clinical trials is a topic gaining attention in the statistics literature in recent years. A number of methods have been proposed in this area. Some approaches are sophisticated but complicated to implement. We aim to implement a simple and robust empiric Bayes Poisson Gamma model (PGM) that is suitable for practical use. We assume a constant and site-specific underlying enrollment rate over time, which comes from a common Gamma distribution. Choice of prior parameters is data driven. We tested the proposed PGM in a simulation study as well as a number of oncology trials with various enrollment patterns, which yield satisfactory results. Compared to a flexible nonparametric model (Zhang and Long, 2010), the PGM is associated with a narrower credible interval as a result of parametric assumptions. However, the model prediction may be off when the assumptions are substantially violated.

Abstract 14455: Impact of Cardiac Rehabilitation on Peak Vo 2 in the Renal Transplant Candidate

Introduction: Cardiopulmonary fitness measured by peak VO 2 is effective in cardiovascular screening prior to renal transplantation and even predicts survival. Hypothesis: Will cardiac rehabilitation impact functional status on renal transplant candidates when compared to other factors that affect peak VO 2 Methods: A peak VO 2 test is performed with the initial transplant evaluation and repeated annually for patients on the waiting list. Peak V0 2 , heart rate max (HRmax), respiratory exchange ratio (RER), pulmonary or cardiac output limitation, deconditioning, obesity were evaluated. If any of the prior parameters were noted to be abnormal, they were acted upon. Results: A total of 107 patients (38 females and 69 males) received initial and repeat CPET testing. The average time interval between repeat tests was 342 days (11 months). The data was adjusted for various interventions to determine the magnitude to which they affected increases in peak VO 2 . The patients were able to complete an average of 36 exercise sessions (phase III) and demonstrated an average increase in peak VO 2 of 5.02 mL·min -1 ·kg -1 which was the largest difference among interventions observed. Patients that improved HRmax or RER between tests demonstrated similar changes in VO 2 (1.98, and 0.42 mL·min -1 ·kg -1 respectively). Among patients that lost weight, an average decrease of 13.1 lbs was noted with their respective increase of peak VO 2 by 0.68 mL·min -1 ·kg -1 . The initial test identified 83 patients with cardiac output limitation and 10 patients received cardiac interventions between their testing interval, resulting in an average VO 2 increase of 0.74 mL·min -1 ·kg -1 . Only 32 patients were identified to have pulmonary abnormalities from their initial test, 3 had pulmonary interventions and the average VO 2 increase was 1.53 mL·min -1 ·kg -1 Conclusion: Cardiac rehabilitation program has the highest impact on peak VO 2 that may translate into a better and healthier renal transplant candidate.

PENet: Prior evidence deep neural network for bladder cancer staging.

Bladder cancer is a heterogeneous, complicated, and widespread illness with high rates of morbidity, death, and expense if not treated adequately. The accurate and exact stage of bladder cancer is fundamental for treatment choices and prognostic forecasts, as indicated by convincing evidence from randomized trials. The extraordinary capability of Deep Convolutional Neural Networks (DCNNs) to extract features is one of the primary advantages offered by these types of networks. DCNNs work well in numerous real clinical medical applications as it demands costly large-scale data annotation. However, a lack of background information hinders its effectiveness and interpretability. Clinicians identify the stage of a tumor by evaluating whether the tumor is muscle-invasive, as shown in images by the tumor's infiltration of the bladder wall. Incorporating this clinical knowledge in DCNN has the ability to enhance the performance of bladder cancer staging and bring the prediction into accordance with medical principles. Therefore, we introduce PENet, an innovative prior evidence deep neural network, for classifying MR images of bladder cancer staging in line with clinical knowledge. To do this, first, the degree to which the tumor has penetrated the bladder wall is measured to get prior distribution parameters of class probability called prior evidence. Second, we formulate the posterior distribution of class probability according to Bayesian Theorem. Last, we modify the loss function based on posterior distribution of class probability which parameters include both prior evidence and prediction evidence in the learning procedure. Our investigation reveals that the prediction error and the variance of PENet may be reduced by giving the network prior evidence that is consistent with the ground truth. Using MR image datasets, experiments show that PENet performs better than image-based DCNN algorithms for bladder cancer staging.

High-resolution inverse modelling of European CH4 emissions using the novel FLEXPART-COSMO TM5 4DVAR inverse modelling system

Abstract. We present a novel high-resolution inverse modelling system (“FLEXVAR”) based on FLEXPART-COSMO back trajectories driven by COSMO meteorological fields at 7 km×7 km resolution over the European COSMO-7 domain and the four-dimensional variational (4DVAR) data assimilation technique. FLEXVAR is coupled offline with the global inverse modelling system TM5-4DVAR to provide background mole fractions (“baselines”) consistent with the global observations assimilated in TM5-4DVAR. We have applied the FLEXVAR system for the inverse modelling of European CH4 emissions in 2018 using 24 stations with in situ measurements, complemented with data from five stations with discrete air sampling (and additional stations outside the European COSMO-7 domain used for the global TM5-4DVAR inversions). The sensitivity of the FLEXVAR inversions to different approaches to calculate the baselines, different parameterizations of the model representation error, different settings of the prior error covariance parameters, different prior inventories, and different observation data sets are investigated in detail. Furthermore, the FLEXVAR inversions are compared to inversions with the FLEXPART extended Kalman filter (“FLExKF”) system and with TM5-4DVAR inversions at 1∘×1∘ resolution over Europe. The three inverse modelling systems show overall good consistency of the major spatial patterns of the derived inversion increments and in general only relatively small differences in the derived annual total emissions of larger country regions. At the same time, the FLEXVAR inversions at 7 km×7 km resolution allow the observations to be better reproduced than the TM5-4DVAR simulations at 1∘×1∘. The three inverse models derive higher annual total CH4 emissions in 2018 for Germany, France, and BENELUX compared to the sum of anthropogenic emissions reported to UNFCCC and natural emissions estimated from the Global Carbon Project CH4 inventory, but the uncertainty ranges of top-down and bottom-up total emission estimates overlap for all three country regions. In contrast, the top-down estimates for the sum of emissions from the UK and Ireland agree relatively well with the total of anthropogenic and natural bottom-up inventories.

Status Forecasting Based on the Baseline Information Using Logistic Regression

In the status forecasting problem, classification models such as logistic regression with input variables such as physiological, diagnostic, and treatment variables are typical ways of modeling. However, the parameter value and model performance differ among individuals with different baseline information. To cope with these difficulties, a subgroup analysis is conducted, in which models’ ANOVA and rpart are proposed to explore the influence of baseline information on the parameters and model performance. The results show that the logistic regression model achieves satisfactory performance, which is generally higher than 0.95 in AUC and around 0.9 in F1 and balanced accuracy. The subgroup analysis presents the prior parameter values for monitoring variables including SpO, milrinone, non-opioid analgesics and dobutamine. The proposed method can be used to explore variables that are and are not medically related to the baseline variables.

Beneficial effect of SPE-ZnO nanoparticles against hepato-renal toxicity and oxidative stress induced by fluoride in wistar rats

Our study aimed to investigate the effect of ZnNPs formulated by spinach extract (SPE-ZnNPs) fluoride induced hepato-renal toxicity in rats. 18 male rats were divided into three groups (n=6); Control, NaF and NaF + SPE-ZnNPs. various biological parameters and histopathological analysis were estimated. Phytochemical of spinach and characterizations of ZnNPs were analyzed using standard technics. Results show that the plant is rich in active substances and ZnNPs were formulated with a size of up to 30 nanometers, which is proven by the Uv-VIS, FTIR and SEM results. Results show that NaF exposed induce an alteration in hemato-biochemical parameters, and hepato-renal oxidative stress and histopathological alteration. However, SPEZnNPs’ administration to NaF rats resulted in a partial improvement and correction of the prior parameters. Use of SPE-ZnNPs appears to be the main preventative measure against fluoride poisoning and the oxidative stress it causes.

Using sequential conditioning to explore uncertainties in geostatistical characterization and in groundwater transport predictions

Rapid transmission of contaminants in groundwater can occur in alluvial gravel aquifers that are permeated by highly conductive small-scale open framework gravels (OFGs). This open framework gravel structure and the associated distribution of hydraulic properties is complex, and so assessments of contamination risks in these aquifers are highly uncertain. Geostatistical models, based on lithological data, can be used to quantitatively characterize this structure. These models can then be used to support analyses of the risks of contamination in groundwater systems. However, these geostatistical models are themselves accompanied by significant uncertainty. This is seldom considered when assessing risks to groundwater systems. Geostatistical model uncertainty can be reduced by assimilating information from hydraulic system response data, but this process can be computationally challenging. We developed a sequential conditioning method designed to address these challenges. This method is demonstrated on a transition probability based geostatistical simulation model (TP), which has been shown to be superior for representing the connectivity of high permeability pathways, such as OFGs. The results demonstrate that the common modelling practice of adopting a single geostatistical model may result in realistic predictions being overlooked, and significantly underestimate the uncertainties of groundwater transport predictions. This has important repercussions for uncertainty quantification in general. It also has repercussions if using ensemble-based methods for history matching, since it also relies on geostatistical models to generate prior parameter distributions. This work highlights the need to explore the uncertainty of geostatistical models in the context of the predictions being made.

A Bayesian stable isotope mixing model for coping with multiple isotopes, multiple trophic steps and small sample sizes

Abstract We introduce a Bayesian stable isotope mixing model for estimating the relative contributions of different dietary components to the tissues of consumers within food webs. The model is implemented with the probabilistic programming language Stan. The model incorporates isotopes of multiple elements (e.g. C, N, H) for two trophic levels, when the structure of the food web is known. In addition, the model allows inclusion of latent trophic levels (i.e. for which no empirical data are available) intermediate between sources and measured consumers. Running the model in simulations driven by a real dataset from Finnish lakes, we tested the sensitivity of the posterior distributions by altering critical prior parameters and assumptions in the data‐generating process. Importantly, we found that the model estimations were particularly sensitive to the assigned prior value for ω (the fraction of H in aquatic consumer tissue that is derived from environmental water rather than diet) so that reliable empirical data for this parameter are required. When reliable information is not available for ω, we suggest that an uninformative prior should be used. The proposed model and inferences are suitable for studies where resources for collecting new data are limited, but useful prior information for each specific trophic level is available from earlier studies.

Characteristics of Chemical Accidents and Risk Assessment Method for Petrochemical Enterprises Based on Improved FBN

Refining and chemical integration is the major trend in the development of the world petrochemical industry, showing intensive and large-scale development. The accident risks caused by this integration are complex and diverse, and pose new challenges to petrochemical industry safety. In order to clarify the characteristics of the accident and the risk root contained in the production process of the enterprise, avoid the risk reasonably and improve the overall safety level of the petrochemical industry, in this paper, 159 accident cases of dangerous chemicals in China from 2017–2021 were statistically analyzed. A Bayesian network (BN)-based risk analysis model was proposed to clarify the characteristics and root causes of accident risks in large refining enterprises. The prior probability parameter in the Bayesian network was replaced by the comprehensive weight, which combined subjective and objective weights. A hybrid method of fuzzy set theory and a noisy-OR gate model was employed to eliminate the problem of the conditional probability parameters being difficult to obtain and the evaluation results not being accurate in traditional BN networks. Finally, the feasibility of the methods was verified by a case study of a petrochemical enterprise in Zhoushan. The results indicated that leakage, fire and explosion were the main types of accidents in petrochemical enterprises. The human factor was the main influencing factors of the top six most critical risk root causes in the enterprise. The coupling risk has a relatively large impact on enterprise security. The research results are in line with reality and can provide a reference for the safety risk management and control of petrochemical enterprises.