Error Model Research Articles

Transient temperature and pressure coupling model of cementing injection stage in ultra deep wells considering segmental rheological model, casing eccentricity and U-tube effect

The geological environments of ultra deep wells such as high temperature and high pressure (HTHP), narrow safety density window pose significant challenges to cementing operation. Considering segmental rheological model and density model, casing eccentricity, U-tube effect and viscous dissipation, this paper established a transient temperature and pressure coupling model during cementing injection stage of ultra deep wells. The segmental rheological model can describe the rheological properties of cementing fluids under HTHP much better. Considering casing eccentricity and viscous dissipation can accurately model the variation of temperature and pressure profiles. Taking U-tube effect into the model, the formation process of vacuum section at top of the casing can be simulated. The fully implicit finite-difference approach was used to solve the coupling model, and the calculation results were compared with field measurement data and simulation data to verify its accuracy. The calculation errors of the developed model are respectively 10% and 5% compared to field measurement data and Wellplan’s simulation data. Numerical simulations were conducted to reveal the cementing fluid distribution, transient temperature distribution and pressure distribution during the cementing injection stage. The simulation results show that the variety of cementing fluids with different density, rheological properties and thermophysical parameters complicates the change mechanism of temperature and pressure. The temperature and pressure distributions both present unsteady characteristics during cementing injection stage, which puts forward higher requirements of pressure control for narrow safety density window formation. The influence of temperature and pressure on rheology of cementing fluids cannot be ignored, and segmental rheological model should be considered for accurate pressure prediction. Also the casing eccentricity cannot be ignored for cementing pressure calculation in deviated or horizontal wells. Managed pressure cementing (MPC) has a greater advantage in dealing with narrow density window formation. The synergistic control of cementing fluid density and back pressure can keep the equivalent circulating density (ECD) within the safe density window. Through the application of the transient temperature and pressure coupling model, we can simulate the variations of key dynamic parameters during cementing injection stage and optimize the cementing parameters.

Determination of outliers when building a multivariate regression model of prices in residential real estate market of the region

The paper is aimed at selecting the optimal method for identifying outliers in the initial data when building a multivariate regression model of prices in the regional residential real estate market. The study was based on offers for sale of apartments in prefab flat blocks located in Irkutsk. In this regard, a basic multiplicative multivariate regression model was built, describing the dependence of cost indicators on the pricing factors of real estate. The identified outliers were iteratively removed from the basic model. The methods for detecting outliers included calculation of standard deviation (z-score), calculation of the Mahalanobis distance, as well as a method developed in the study for bringing the prices of objects to the characteristics of the reference object. The optimal method for detecting outliers in the initial data was selected by comparing the characteristics of the final variable-based multivariate regression models obtained after removing outliers from them. The analysis of the results proved the method of bringing the prices of objects to the characteristics of the reference object to be the optimal method of identifying outliers when building a multivariate regression model of prices in the regional residential real estate market. This method significantly reduces the approximation errors of the basic multivariate regression model of the market, thereby increasing the adequacy of the results of the real estate valuation conducted on its basis.

An embedded differential evolution approach to adaptive covariance inflation with the ensemble Kalman filter

ABSTRACT The limited ensemble sizes and the neglected model errors in ensemble-based Data Assimilation (DA) usually lead to underestimation of the error variance, and covariance inflation ‘inflates’ the covariance of a forecast or analysis ensemble by some positive factor in each assimilation cycle. In practice, the value of the inflation factor is often set by trial and error. Thus, a variety of adaptive covariance inflation (ACI) algorithms have been proposed to improve assimilation performance by adjusting the optimal parameters. Most of these used traditional optimization algorithms, such as gradient methods and relaxation methods. However, Evolutionary Algorithms(EAs), which encompass a range of population-based stochastic search techniques widely used in various engineering optimization problems, are rarely seen in DA. In this study, we developed a new framework for adaptive inflation factors, introducing differential evolution (DE) algorithm into local ensemble Transform Kalman filter (LETKF) that can explore the most suitable/optimal inflation factors online. A ‘micro-DE’ based data assimilation method can obtain better assimilation results step by step without taking a much longer time. The feasibility and effectiveness of the algorithm is demonstrated in an idealized Lorenz-96 model with simulated observations. Under both perfect and imperfect-model scenarios, it is found that the ‘micro-DE’LETKF method is capable of outperforming the original LETKF with a considerable lower computational cost. The adaptive inflation factor is optimally updated at each assimilation cycle either around the upper bound or around the lower bound, which means that only an adaptive factor helps mitigate the ensemble collapse due to a lack of spread. However, the further study will consider the application of the methodology to more complex atmospheric models.

Noise-Aware Active Learning to Develop High-Temperature Shape Memory Alloys with Large Latent Heat.

Shape memory alloys (SMAs) with large latent heat absorbed/released during phase transformation at elevated temperatures benefit their potential application on thermal energy storage (TES) in high temperature environment like power plants, etc. The desired alloys can be designed quickly by searching the vast component space of doped NiTi-based SMAs via data-driven method, while be challenging with the noisy experimental data. A noise-aware active learning strategy is proposed to accelerate the design of SMAs with large latent heat at elevated phase transformation temperatures based on noisy data. The optimal noise level is estimated by minimizing the model error with incorporation of a range of noise levels as noise hyper-parameters into the noise-aware Kriging model. The employment of this strategy leads to the discovery of the alloy with latent heat of -36.08 J g-1, 9.2% larger than the best value (-33.04 J g-1) in the original training dataset within another four experiments. Additionally, the alloy represents high austenite finish temperature (481.71°C) and relatively small hysteresis. This promotes the latent heat TES application of SMAs in high temperature circumstance. It is expected that the noise-aware approach can be convenient for the accelerated materials design via the data-driven method with noisydata.

Spatio-temporal effects and influence mechanism of digital technology on tourism efficiency in Chinese provinces

Digital technology serves as a new industrial driving force, playing a crucial role in promoting the efficiency of regional tourism. To explore the inherent logic between digital technology and tourism development, the study employed the DEA-BCC model and the entropy-weighted TOPSIS method to measure the tourism efficiency and digital technology development level of 31 provinces (municipalities and districts) in China from 2011 to 2022. The evolutionary characteristics of the relationship between digital technology and tourism efficiency were analyzed from the spatio-temporal dimension, and empirical tests were carried out using the spatial error model (SEM) and the spatial lag model (SLM) to explore the mechanism of the impact of digital technology on tourism efficiency. The results show that: during the study period, the overall trend of digital technology development and tourism efficiency is upward, but there is a certain degree of spatial mismatch between the two; digital technology is not only conducive to the improvement of tourism efficiency in the province, but also acts on neighboring provinces through spatial spillover effects, mechanism of action tests show that digital technology can positively moderate the effects of tourism economic growth and tourism industry structure on tourism efficiency; in western and northeast China, the positive effect of digital technology on tourism efficiency is more obvious. The conclusions provide a new perspective for understanding and analyzing the development of provincial tourism in China, as well as a reference for the rational use of digital technology to promote tourism development.

A method for temperature sensor and model selection for machine tool thermal error modelling using ANFIS and ANN

AbstractThermal errors account for a significant part of the dimensional errors of components produced by precision machine tools. These errors are commonly compensated using predictions from temperature-based empirical models. The accuracy and robustness of these predictions are affected by the locations of temperature sensors used to obtain the model’s input data. Methods for sensor selection found in literature are often difficult to replicate and automate because they require tuning of several hyperparameters. This work presents a sensor and model selection approach that uses proper orthogonal decomposition (POD) and QR pivoting to select a subset of sensors that have been preinstalled in the machine tool as possible model inputs. These sensors are then sorted according to their correlation with the thermal error being modelled. The final set of inputs and thermal error model structure is chosen using Bayesian Information Criterion (BIC) to limit model overfitting. The approach was tested by modelling the Y-axis thermal error measured from air-cutting experiments performed under different spindle speeds and feed rates. This enabled determination of the structure and the choice inputs for Adaptive Neuro Fuzzy Inference System (ANFIS) and Artificial Neural Network (ANN) thermal error models. The accuracy of these models was compared to that of models trained using inputs selected by conventional approaches: the least absolute shrinkage and selection operator (LASSO), fuzzy c-means clustering (FCM), and principal component regression (PCR). The presented approach had better or comparable results to the conventional approaches while using fewer inputs. The presented approach is also well suited for automation compared to conventional approaches that require expert input.

Heat and mass transfer performance of multi-solute electroplating wastewater in spray evaporation separation process

Electroplating wastewater is complex in composition and contains many harmful substances. Based on the single droplet heat and mass transfer method, a one-dimensional mathematical model for the evaporation separation heat and mass transfer process of multi-solute electroplating wastewater is established. In order to validate the model, the experimental system of evaporation separation tower is established, and Nusselt numbers corresponding to metal ion solutions are fitted, reliabilities of models are also verified. The maximum relative errors of mathematical models are within 10.0 %. In addition, effects of the type and concentration of metal ions and air inlet temperature on heat and mass transfer characteristics of spray separation tower are investigated. Experimental results show that when the wastewater inlet concentration is the same, the improvement effects of increasing air inlet temperature on evaporation and separation performance from high to low are ZnCl2, CuCl2, and NiCl2 solutions. When the air inlet temperature increases from 90 °C to 130 °C, the evaporation speed of ZnCl2, CuCl2 and NiCl2 solutions increase by 90.3 %, 84.3 % and 45.7 %, respectively. When the air inlet temperature is the same, with the increase of wastewater inlet concentration, the evaporation separation performance of electroplating wastewater in descending order is ZnCl2, CuCl2, and NiCl2 solutions.

Safe robust multi-agent reinforcement learning with neural control barrier functions and safety attention mechanism

In this paper, a novel safe robust multi-agent reinforcement learning method integrated with decentralized robust neural control barrier functions (CBFs) and a safety attention mechanism (SAM) is proposed for the safety-critical multi-agent system (MAS). Safety is fundamental in the safety-critical MAS but can be affected by factors such as modeling errors, external unknown disturbances, and time-varying observable agents. Several appropriate measures are implemented to address these issues. First, modeling errors and external disturbances are regarded as an adversary for each agent. The agent learns a policy that is robust to disturbances created by the adversary. Accordingly, decentralized robust neural CBFs are introduced to maintain the safety of the MAS, particularly when the general handcrafted CBFs are difficult to construct. The SAM, in combination with the robust neural CBFs, provides a control policy with the capacity to handle time-varying observable agents and increases its attention to dangerous events. The online fine-tuning procedure further enhances the safety. Finally, experiments demonstrate the safety and effectiveness of the proposed method.

Research on compressor cascade flow field modeling method based on finite volume flux-informed neural network

For compressor cascade flow field modeling, there exists strong velocity shear in the leading edge separation flow, boundary layer, and wake, which leads to increased modeling errors. To improve the accuracy of the flow field modeling method, this paper introduces the concept of numerical flux from the finite volume method into the loss function to implement Euler equation physics-informed learning, and a finite volume flux-informed neural network (FVFI-net) is constructed. Selecting a high-load, large-turning-angle compressor cascade as the study object, a comparative analysis is conducted on the advantages and disadvantages of purely data-driven, weak physical constraint, and finite volume flux-informed methods in compressor cascade flow field modeling. The study found that compared to purely data-driven and weak physical constraint methods, FVFI-net can reduce the average error of aerodynamic parameters in the flow field by approximately 45.6% and 29.5%, respectively, at a 0° angle of attack. For the flow separation problem occurring at the suction side leading edge and the blade wake area caused by a 5° angle of attack, FVFI-net can effectively reduce modeling errors near the leading edge, in the wake region, and near the periodic boundaries, thus reducing the average error of the aerodynamic parameters of the flow field by about 49.2%and 31.3%, respectively, compared to pure data-driven and weak physical constraint methods.

A New Coupled Biogeochemical Modeling Approach Provides Accurate Predictions of Methane and Carbon Dioxide Fluxes Across Diverse Tidal Wetlands

AbstractTidal wetlands provide valuable ecosystem services, including storing large amounts of carbon. However, the net exchanges of carbon dioxide (CO2) and methane (CH4) in tidal wetlands are highly uncertain. While several biogeochemical models can operate in tidal wetlands, they have yet to be parameterized and validated against high‐frequency, ecosystem‐scale CO2 and CH4 flux measurements across diverse sites. We paired the Cohort Marsh Equilibrium Model (CMEM) with a version of the PEPRMT model called PEPRMT‐Tidal, which considers the effects of water table height, sulfate, and nitrate availability on CO2 and CH4 emissions. Using a model‐data fusion approach, we parameterized the model with three sites and validated it with two independent sites, with representation from the three marine coasts of North America. Gross primary productivity (GPP) and ecosystem respiration (Reco) modules explained, on average, 73% of the variation in CO2 exchange with low model error (normalized root mean square error (nRMSE) <1). The CH4 module also explained the majority of variance in CH4 emissions in validation sites (R2 = 0.54; nRMSE = 1.15). The PEPRMT‐Tidal‐CMEM model coupling is a key advance toward constraining estimates of greenhouse gas emissions across diverse North American tidal wetlands. Further analyses of model error and case studies during changing salinity conditions guide future modeling efforts regarding four main processes: (a) the influence of salinity and nitrate on GPP, (b) the influence of laterally transported dissolved inorganic C on Reco, (c) heterogeneous sulfate availability and methylotrophic methanogenesis impacts on surface CH4 emissions, and (d) CH4 responses to non‐periodic changes in salinity.

Water and Thermal Management in PEM Fuel Cells Using Feasible Humidity Plots and Model Predictive Controllers

Water and thermal management are critical for the performance, efficiency and longevity of PEM fuel cells (PEMFCs). Effective water and thermal management require the design of control systems that can maintain the water balance and temperature at stable and optimal levels. In this paper, we consider a stack of PEM fuel cells integrated with a water recovery and cooling system. A mechanistic dynamic model is developed to be able to predict the water content and temperature in response to the fuel cell inputs. Water management uses a cascade arrangement of a supervisory Model Predictive Controller (MPC) and local anode and cathode PID humidity controllers to balance the membrane water content. Thermal management consists of a separate MPC controller to regulate the fuel stack temperature. One novelty of this work lies in identifying and utilizing the feasible region for the relative humidities of the anode and cathode when controlling the membrane water content. We introduce the feasible humidity plots (FHP) which define the feasible values for the anode and cathode relative humidities for a given fuel cell design and its operating conditions. This useful information helps to assign the set-point values to the local PID humidity controllers of the water management system. It is shown by simulations that the water and thermal management MPC controllers work in tandem and successfully track the desired set-point changes in humidity and temperature while rejecting external disturbances such as load changes. In addition, the control system is robust against modeling errors and possible model-plant mismatch introduced by fuel cell aging.

The association between hunger-coping economic tradeoffs and food insecurity among female recipients of charitable food assistance

Food insecurity is an indicator of well-being in the United States. A high proportion of recipients of charitable food assistance (CFA) are women and are often in charge of specific household managerial responsibilities (e.g., childcare, transportation). Consequently, they frequently face choices between paying for food and paying for other basic need(s). This study aims to examine which hunger-coping economic tradeoffs place females with at least one dependent child in the house and females without a dependent child in the house at risk for experiencing food insecurity. Data was collected at 10 Houston-area and 10-Atlanta-area food pantries in 2022 (N=883). Using USDA cutoff criteria, households were considered food insecure based on ≥3 affirmative responses to the 18-item Food Security Scale Module. Hunger-coping economic tradeoff experiences were based on affirmative responses to whether anyone in the household ever had to choose between food and six basic needs (i.e. childcare, medicine/medical care, utilities, rent/mortgage, transportation, education). Covariate-adjusted logistic regression models were conducted to understand the relationship between six hunger-coping economic tradeoffs and food insecurity for the entire analytic sample and stratified by whether the female participant had a child in house. Standard errors in all regression models were corrected to account for multiple observations within a pantry. Adults, on average, were 55 years old (58% food insecure; 47% Hispanic; 42% black). Four hunger-coping economic tradeoffs were related to experiencing food insecurity. Economic tradeoffs between food and a) medicine/medical care and b) transportation elevated the likelihood of food insecurity, regardless of child status. Tradeoffs between food and childcare increased the risk for experiencing food insecurity among females with a dependent child. Deciding to pay between food and utilities was related to food insecurity experiences among females without a dependent child. Increases in Supplemental Nutrition Assistance Program (SNAP) benefits and eligibility along with programs to enhance resources related to medical care, transportation, childcare and utilities could help reduce food insecurity, especially among CFA recipients.

Towards a Spectral Library of Medicinal and Aromatic Plant species (MAPs): Plant Discrimination and Wavelength Selection

The recognition and identification of Medicinal and Aromatic Plants species (MAPs) is a challenge for researchers and professionals in the field. To address this issue, this study aimed to examine the potential of using hyperspectral remote sensing, chemometrics and Machine Learning in discriminating between MAP species. Specifically, we tested the applicability of UV-Near-infrared Spectroscopy (UV-NIR), PCA and Partial Least-Squares Discriminant Analysis (PLS-DA) as tools for discrimination. The spectral signature data were obtained by UV-near infrared spectroscopy, covering a range of 325–1075 nm. Principal component analysis (PCA) was applied to reduce data dimensionality in order to minimize errors in PLS-DA model. This model was used to build a discriminant model based on the wavelengths obtained from PCA. The results show that it is effective to combine UV-NIR spectral features and PLS-DA to establish a discriminant model for accurately classifying the MAP species. The key wavelengths most involved in the discrimination were found in the NIR region and especially in ranges 516–548 and 628–1070 nm.

High precision aerodynamic heat prediction method based on data augmentation and transfer learning

Data-driven modeling methods have become one of the main technologies for predicting aerodynamic heat in hypersonic conditions. However, due to the limitations of wind tunnel experimental conditions, the spatial distribution of aerothermal wind tunnel experimental data is often sparse, and the sample size is relatively small. Furthermore, there is a lack of direct correlation in the aerodynamic heat distribution data among different shapes of vehicles, which poses challenges for constructing high-performance data-driven aerodynamic heat prediction models. To address these issues, this paper proposes a high-precision aerodynamic heat modeling and prediction method based on data augmentation and transfer learning. First, integrating the concept of data fusion, we propose to enhance the sparse aerothermal wind tunnel experimental data by using deep neural networks and introducing low-precision numerical computation results. Next, based on the close physical correlation between boundary layer outer edge information and wall surface aerodynamic heat, we construct the aerodynamic heat prediction model ED-ResNet using a double-series residual neural network. Finally, by fine-tuning the ED-ResNet model for transfer learning, high-precision predictions of aerothermal wind tunnel experimental results for different shaped vehicles are achieved under small sample conditions. Verification using hypersonic double-ellipsoid, blunt cone, and blunt bicone shows that after data augmentation, the prediction error of the aerodynamic heat prediction model is significantly reduced to 1/3 of that when data augmentation is not used. Moreover, through transfer learning, the model effectively leverages existing hypersonic double-ellipsoid aerothermal wind tunnel experimental data to achieve high-precision predictions of aerodynamic heat distribution for blunt cone and blunt double cone under different incoming flow conditions, with normalized root mean square error(NRMSE) maintained below 10%.

A precise and efficient K-means-ELM model to improve ultra-short-term solar irradiance forecasting

To address the problems of intermittent and uncontrollable solar irradiance faced by grid-connected photovoltaic power plants, this paper proposes an ultra-short-term solar irradiance prediction model that combines K-means clustering with the Extreme Learning Machine (ELM). The K-means algorithm is first applied to cluster solar irradiance data from different seasons in Shanghai based on spatial similarity. Subsequently, the ELM algorithm is employed to train the model, significantly improving both prediction accuracy and training speed. Compared with the prediction accuracy of the model before clustering, the root-mean-square error (RMSE) of the clustered model is significantly reduced by 28.75 % in spring, and 0.30 %, 6.70 % and 5.92 % in summer, Autumn and winter, respectively. In addition, the model demonstrates stable predictive performance at different time resolutions (5, 10 and 15 minutes) with R2 values close to 1, confirming its accuracy and stability under small sample conditions. In terms of training speed, ELM’s training time is more than 100 times faster than Support Vector Regression (SVR) and significantly shorter than traditional models such as PSO-BP and Random Forest (RF), showing its great advantage in application scenarios that require fast response. Overall, the K-means-ELM model is able to accurately capture the overall trend and sudden changes in solar irradiance, which is of great application value for enhancing the efficiency and stability of solar power generation systems.

A highly accurate and robust prediction framework for drilling rate of penetration based on machine learning ensemble algorithm

The rate of penetration (ROP) is a key indicator of drilling efficiency. Many researchers have explored the application of machine learning in ROP prediction. However, few studies have focused on the robustness of the constructed models, and developing a ROP prediction model that can achieve both high accuracy and strong robustness remains a challenge. This paper introduces a novel machine learning approach to constructing a ROP prediction model through ensemble learning algorithms. The model is based on field data from oilfields, incorporating 16 input parameters that influence ROP. First, the feasibility of the collected dataset is verified using correlation analysis. Then, ROP prediction models are developed based on various machine learning algorithms, including Decision Tree Regression (DTR), Random Forest (RF), eXtreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGBM), Support Vector Regression (SVR), and Backpropagation Neural Network (BPNN). By comparing the performance of these models under noise levels of 0%, 1.7%, and 5.1%, RF, LGBM, XGB, and SVR are selected as base learners. These base learners are then combined to construct multiple ensemble models, and the performance of the optimal ensemble model is evaluated under varying noise levels. The results show that the prediction error of the optimal model remains within 10%, and R2 is greater than 0.96. Finally, the Shapley Additive Explanations (SHAP) method is applied to perform interpretability analysis on the optimal ROP prediction model, examining the impact of different input factors on the model's predictive performance. Compared to single models and other ensemble models, the proposed ensemble model not only achieves higher accuracy but also demonstrates strong robustness and generalization capability.

Kalman filter-driven state observer for thermal error compensation in machine tool digital twins

Sustainable reduction of thermal errors during production is the key challenge in modern high-precision manufacturing. Numerical compensation models provide an energy-efficient solution, but in the case of data-driven models, high-quality experimental data must be time-consuming and expensive to produce, negatively impacting overall productivity. Furthermore, robustness concerns arise in the case of new operating conditions, which were not contained in the training data. This paper presents a novel use of a Kalman filter together with model order reduced finite element models to observe the entire thermal state, which allows the subsequent solution of the mechanical model and computation of the thermal errors in real-time without requiring any training data but instead purely based on the physical system model. The effectiveness of this approach is evaluated using experiments on a thermal test bench with 16 out of 40 temperature sensors employed for observation and demonstrated on a 5-axis machine tool (MT) with 13 out of 25 temperature sensors used. Due to the combination of the reduced order model and Kalman filter these 13 temperature sensors are sufficient to represent a MT mesh of more than 350’000 elements. The entire temperature profile of the thermal test bench is reconstructed to achieve a root mean square error (RMSE) of the unobserved temperature sensors of only 2.7 °C, which accounts for more than 83% of all temperature variations and 1.3 °C for the 5-axis MT. For the thermal error of the thermal test bench, the RMSE could be reduced from 67.4μm to 33.3μm, corresponding to a reduction of 52.7 %. This could be achieved without the need for experimental data for model calibration, in a real-time capable physics-based model.

Modeling of Fabrication Errors Caused by Operator Habits in Substrate Alignment Process

Modeling of Fabrication Errors Caused by Operator Habits in Substrate Alignment Process

Modal test and finite element updating of sprayer boom truss

In addressing the finite element model and actual structural error of the sprayer boom truss, this study aims to achieve high-precision dynamic characteristics, enhance simulation credibility, make informed optimization decisions, and reduce testing costs. The research investigates the dynamic behavior of the sprayer boom truss through modal experiments and finite element simulations. Initially, modal parameters of the sprayer boom are obtained through experimental testing, validating their reasonableness and reliability. Subsequently, Ansys Workbench18.0 simulation software was employed to analyze the finite element model of the sprayer boom, revealing a maximum relative error of 11.93% compared to experimental results. To improve accuracy, a kriging-based response surface model was constructed, and multi-objective parameter adjustments using the MOGA algorithm reduce the maximum relative error to 4.6%. Sensitivity analysis further refines the model by optimizing target parameters, resulting in a maximum relative error of 4.96%. These findings demonstrate the effective enhancement of the corrected finite element model’s precision, with the response surface method outperforming sensitivity analysis the maximum relative error between the updated finite element model and experimental results was within the engineering allowable range, confirming the effectiveness of the updated model.

Machine Learning-Driven precise design of stable OLED Materials: Predicting and enhancing Multi-State C-N bond dissociation energies

The intrinsic stability of organic light-emitting diode (OLED) materials critically hinges on the bond dissociation energies (BDEs) of fragile bonds, particularly C-N bonds. Despite the necessity of considering BDEs in multiple electronic states due to the complex working conditions of organic materials in OLED devices, comprehensive exploration remains challenging. Herein, we constructed a high-quality dataset, CNBDE-24, comprising nearly 1500 data points for BDEs across multiple electronic states, obtained through an active learning-guided density functional theory (DFT) calculation strategy. Machine learning models trained on the CNBDE-24 dataset exhibit excellent accuracy in predicting BDEs for OLED materials in neutral, triplet excited (T1), and anionic states, achieving an adjusted coefficient of determination exceeding 0.90 on the test set and bypassing the need for high-cost calculation. Moreover, utilizing neutral state BDEs as pre-training data significantly reduces model error and provides deeper insights into the relationships between different BDEs. We find that BDEs in neutral are influenced by the electronic properties of the nitrogen atom, as indicated by descriptors such as Gasteiger charges. Incorporating C-N bonds into aromatic systems and slightly dispersing N atom charges can inherently increase multi-state BDEs without altering luminescent properties. Additionally, anionic BDEs are determined by the extent of charge transfer during bond cleavage and can be regulated by modifying the C-side fragment. High-throughput virtual screening reveals a new stable purple emitter with a dominant 0–0 transition at 360 nm, which was validated through DFT and excited-state property evaluations. Strategies for enhancing BDEs in multiple resonance thermally activated delayed fluorescence materials without altering excited-state properties are also demonstrated.