Back Propagation Neural Network Model Research Articles

Multi-objective optimization study of airfoil fin printed circuit heat exchanger with tip gap based on machine learning

The printed circuit heat exchanger (PCHE), a compact and highly efficient device, is capable of operating effectively under demanding conditions, which makes it ideal for supercritical CO2 Brayton cycles. In this study, we present a novel airfoil fin PCHE with tip gap, and conduct multi-objective optimization on the tip gap and arrangement of airfoil fins based on the analysis of the supercritical CO2 flow and heat transfer characteristics. We conduct a parameter design using Design of Experiment to examine the impact of the dimensionless design parameters, including the horizontal number (ζh), staggered number (ζs), vertical number (ζv), and gap number (ζg). To forecast the flow and heat transfer performance, we utilize a neural network model called Particle Swarm Optimization-Back Propagation (PSO-BP). We employ the non-dominated sorting genetic algorithm II to obtain the Pareto optimal front by utilizing ζh, ζs, ζv, and ζg as variables for optimization, and the volumetric heat transfer coefficient (hv) and Fanning friction factor (f) as objectives for optimization. The results find that the utilization of tip gap can enhance heat transfer while reducing flow resistance. The PSO-BPNN model exhibits higher prediction accuracy and excellent generalization ability compared with traditional BPNN model. The VIKOR and TOPSIS methods identify compromise schemes with excellent thermal-hydraulic performance. The mechanism of heat transfer enhancement can be elucidated using the principle of field synergy.

Performance-based design of environmental parameters for offshore wind turbine foundations

Taking into account the coupling effects of wind, wave, and current on offshore wind turbines (OWTs), a combined approach utilizing non-linear finite element analysis (FEA), back propagation neural network (BPNN), and probabilistic statistical methodology (PSM) is proposed for accurate environmental parameter estimation in OWT foundation design. The structural bearing performances of OWT foundations are precisely predicted through a combination of FEA and BPNN models, and then serve as constraints for probabilistic statistical analysis. Furthermore, for PSM, a robust multivariate joint probability distribution is constructed leveraging copula functions, which not only maximizes marginal information but also effectively captures the intricate dependencies among ocean environmental parameters. Subsequently, the proposed models, along with traditional univariate analysis, are utilized to estimate the return values of diverse load combinations. The results indicate that, for a given return period, the conditional probability model yields lower wave heights and current velocities compared to univariate analysis, while the joint probability model results in lower values for all ocean parameters, making both approaches more suitable for OWT foundation design. This coupled approach presents a more rational and cost-effective solution compared to univariate analysis, potentially resulting in decreased investment costs.

A method for the simultaneous determination of the soil dry density and moisture content based on thermal effects

Abstract The thermodynamic parameters of soil are affected by both dry density and moisture content, leading to uncertainty in measuring moisture content using the heat source method. This study proposes a combined approach using a back propagation (BP) neural network and the point heat resource method to simultaneously determine soil dry density and moisture content. The segmented mean value extracted from the temperature time-history data during the cooling process of the heat source serves as the feature input, while measured values of dry density and moisture content serve as outputs. A calibrated BP neural network model is trained and utilized for simultaneous determination of both parameters. Numerical simulations and modeling tests demonstrate good agreement between inverse identification results and measurements, with root mean square errors of 1.65% for moisture content and 34.09 kg∙m-3 for dry density, along with coefficients of determination at 0.9482 and 0.9359 respectively. It is proved that the method combining soil thermal effect and BP neural network to measure soil dry density and moisture content is feasible.

Optimization of a radial inflow turbine rotor with splitter blades based on entropy production theory and artificial neural network

The radial inflow turbine is the core component of the organic Rankine cycle, and the application of splitter blades is an effective way to reduce its flow loss. It is helpful to study the energy loss characteristics of radial turbines with splitter blades and the optimal geometrical parameters of splitter blades for further improving their aerodynamic performance. In this paper, an optimization framework based on the combination of back-propagation neural network and non-dominated sorting genetic algorithm II is developed. Considering the turbine efficiency as the objective, the geometric parameters of splitter blades, including relative length of the blade, relative distance of the leading edge, circumferential offset, and outlet angle of the blade, are optimized. In addition, the entropy production method is used to locate and quantitatively evaluate the energy loss of the radial turbine with the optimal splitter blades. The results indicate that the genetic algorithm optimized back-propagation neural network model can accurately predict turbine efficiency with a maximum deviation of 2.47 %. In addition, the optimal geometrical parameters of the splitter blade are related to the turbine geometry. In this case, the optimal splitter blade has a relative blade length of 0.639, a relative leading edge distance of 0.3, a circumferential offset of 0.442, and an outlet angle of 38°. Compared with the original turbine, the efficiency of the radial turbine with optimal splitter blades increases by 4.65 %. The total entropy production of rotor and turbine is reduced by 33.3 % and 44.1 % respectively by the installation of optimal splitter blades. Compared with the case without splitter blades, the turbine with the optimal splitter blades can achieve higher efficiency at a flow rate 40 % higher than the design value, and it expands the efficient working range of the turbine by 60 %. This work can provide guidance for the optimization design of radial inflow turbines.

Multi-objective optimization for microbial electrolysis cell-assisted anaerobic digestion of swine manure

As a new technology, microbial electrolysis cell-assisted anaerobic digestion (MEC-AD) has been applied to the utilization of swine manure. Methane production and total energy efficiency are both important indicators in MEC-AD; thus, it is necessary to simultaneously optimize methane production and total energy efficiency. In this study, the Box–Behnken design (BBD) was used as the experimental design, the response surface methodology (RSM) and back propagation neural network (BP-NN) methods were used to construct multi-objective models, and the non-dominated sorting genetic algorithm II (NSGA II) was used for multi-objective optimization. The BP-NN model was superior to the RSM model in terms of goodness of fit. Additionally, the relative error between the predicted and experimental values of the Pareto front was determined to be <3%. The maximum methane production and total energy efficiency were 333.97 mL/g total solid and 61.38%, respectively. Therefore, multi-objective optimization of MEC-AD systems may be accomplished using BBD-(BP-NN)-NSGA II.

Spatio-temporal evolution mechanism and dynamic simulation of nitrogen and phosphorus pollution of the Yangtze River economic Belt in China

Excess nitrogen and phosphorus inputs are the main causes of aquatic environmental deterioration. Accurately quantifying and dynamically assessing the regional nitrogen and phosphorus pollution emission (NPPE) loads and influencing factors is crucial for local authorities to implement and formulate refined pollution reduction management strategies. In this study, we constructed a methodological framework for evaluating the spatio-temporal evolution mechanism and dynamic simulation of NPPE. We investigated the spatio-temporal evolution mechanism and influencing factors of NPPE in the Yangtze River Economic Belt (YREB) of China through the pollution load accounting model, spatial correlation analysis model, geographical detector model, back propagation neural network model, and trend analysis model. The results show that the NPPE inputs in the YREB exhibit a general trend of first rising and then falling, with uneven development among various cities in each province. Nonpoint sources are the largest source of land-based NPPE. Overall, positive spatial clustering of NPPE is observed in the cities of the YREB, and there is a certain enhancement in clustering. The GDP of the primary industry and cultivated area are important human activity factors affecting the spatial distribution of NPPE, with economic factors exerting the greatest influence on the NPPE. In the future, the change in NPPE in the YREB at the provincial level is slight, while the nitrogen pollution emissions at the municipal level will develop towards a polarization trend. Most cities in the middle and lower reaches of the YREB in 2035 will exhibit medium to high emissions. This study provides a scientific basis for the control of regional NPPE, and it is necessary to strengthen cooperation and coordination among cities in the future, jointly improve the nitrogen and phosphorus pollution tracing and control management system, and achieve regional sustainable development.

Predicting thermodynamic adhesion energies of membrane fouling in planktonic anammox MBR via backpropagation neural network model

Predicting thermodynamic adhesion energies was a critical strategy for mitigating membrane fouling. This study utilized a backpropagation (BP) neural network model to predict the thermodynamic adhesion energies associated with membrane fouling in a planktonic anammox MBR. Acid-base (ΔGAB), electrostatic double layer (ΔGEL), and Lifshitz–van der Waals (ΔGLW) energies were selected as output variables, the training dataset was collected by the advanced Derjaguin-Landau-Verwey-Overbeek (XDLVO) method. Optimization results identified “7-10-3″ as the optimal network structure for the BP model. The prediction results demonstrated a high degree of fit between the predicted and experimental values of thermodynamic adhesion energy (R2 ≥ 0.9278), indicating a robust predictive capability of the model in this study. Overall, the study presented a practical BP neural network model for predicting thermodynamic adhesion energies, significantly enhancing the prediction tool for adhesive fouling behavior in anammox MBRs.

Classification of coal bursting liability of some chinese coals using machine learning methods

The classification of coal bursting liability (CBL) is essential for the mitigation and management of coal bursts in mining operations. This study establishes an index system for CBL classification, incorporating dynamic fracture duration (DT), elastic strain energy index (WET), bursting energy index (KE), and uniaxial compressive strength (RC). Utilizing a dataset comprising 127 CBL measurement groups, the impacts of various optimization algorithms were assessed, and two prominent machine learning techniques, namely the back propagation neural network (BPNN) and the support vector machine (SVM), were employed to develop twelve distinct models. The models’ efficacy was evaluated based on accuracy, F1-score, Kappa coefficient, and sensitivity analysis. Among these, the Levenberg–Marquardt back propagation neural network (LM-BPNN) model was identified as superior, achieving an accuracy of 96.85%, F1-score of 0.9113, and Kappa coefficient of 0.9417. Further validation in Wudong Coal Mine and Yvwu Coal Industry confirmed the model, achieving 100% accuracy. These findings underscore the LM-BPNN model’s potential as a viable tool for enhancing coal burst prevention strategies in coal mining sectors.

Machine learning-based modeling of interface creep behavior of grouted soil anchors with varying soil moistures

The interface creep behavior of the grouted soil anchor subject to varying soil moisture was investigated using the combined incorporation of experimental and data-driven modeling methods to establish an efficient and robust forecasting framework. This study carried out the rapid and creep pullout tests of element anchor specimens at various saturations and then utilized machine learning methods to predict the development of interface creep displacement. The stepwise loading strategy and nonlinear superposition method were combined to generate the interface shear creep curves of the element anchor specimens. A total of 936 data groups of the interface shear displacement were collected with changing soil moisture contents, interface shear time, and interface shearing stress. Next, this study explored the Back Propagation Neural Network (BPNN) and four other machine learning algorithms in predicting the interface creep behavior of the grouted soil anchor under various moisture conditions. As for the hyperparameters, the beetle antennae search (BAS) approach was employed to optimize the BPNN and random forest (RF) models. Finally, the boxplot and Taylor diagrams proved the BAS-BPNN demonstrated a better performance than BAS-RF in predicting the interface creep behavior. The consequent correlation coefficients ranged from 0.9613 to 0.9805 for BPNN, indicating the accuracy and reliability of the interface creep prediction. A partial dependence plot (PDP) was also introduced to visualize the established machine learning model. The threshold of moisture content near 28.7 % is found to switch the interface shear stress-displacement response from strain-stabilizing to strain-softening behavior and to result in the main moisture-increase-induced interface strength degradation. The soil moisture fluctuation leads to the development of interface shear displacement mainly observed in the early phase of 20 h after the onset of moisture change. The uncovered coupled impact of soil moisture condition and interface shear stress state can provide insights into the evaluation of the time-dependent in-service performance of grouted soil anchors embedded in clayey soils.

Multi-objective optimization design of a solar-powered integrated multi-generation system based on combined SCO2 Brayton cycle and ORC using machine learning approach

In order to reduce the use of fossil fuels and meet the needs of different energy products, this paper proposes an integrated multi-generation system that can produce power, cooling, and freshwater. A mathematical model is established to analyze the system performance from the perspectives of energy, exergy and economics. Different machine learning algorithms are used to develop prediction models of the system performance. By comparing the predictive performance of different machine learning models, the optimal model is obtained to replace the thermodynamic model of the proposed system to accelerate the optimization process. The results show that the backpropagation neural network (BPNN) model demonstrates the highest predictive accuracy and the lowest relative error fluctuation. When comparing the multi-objective optimization results of the BPNN model with those of the thermodynamic model, it can be observed that the results achieved through both models are very close. Furthermore, the optimization time using the BPNN model is significantly shorter than the thermodynamic model. Finally, under the conditions of solar radiation intensity of 950 W/m2, heliostat field area of 1000 m2, and heat source temperature of 838.15 K, the total product output, thermal efficiency, exergy efficiency, and levelized cost of energy at the final optimal point obtained through BPNN model-based optimization are 2.486 MW, 26.16 %, 23.64 %, and 0.105 $/kWh, respectively. The net power output, cooling capacity, and freshwater flow rate are 2.006 MW, 131.06 kW, and 8.27 m3/h, respectively.

Research on particle accelerator control network deformation prediction algorithm with machine learning

Regular measurement of particle accelerators is imperative to ensure their long-term, stable operation and to meet future construction needs. The aim of this study is to minimise unplanned downtime and to provide data analysis and predictive insights for the remeasurement of similar particle accelerators. The focus is on the three-dimensional coordinate prediction and performance evaluation of a control network for a particle accelerator. The study uses observational data from the laser tracker in the Hefei Light Source storage ring control network, gathered from 2013 to 2023. Innovative use of the linear regression model, hyperbolic model, grey prediction model, and backpropagation neural network model has been made in the alignment of particle accelerators. A comparative analysis with measured data reveals that various deformation prediction methods can achieve a prediction accuracy of 0.25 mm. We recommend using the linear regression model for predicting the plumb direction, and the grey prediction model for predicting the horizontal direction. In addition, reducing the number of backward iteration steps can further improve the prediction accuracy. This study offers guidance for the deformation prediction of particle accelerators, and provides scientific data support and a basis for decision making on their long-term, stable operation.

Identification of historical maximum load on concrete beams based on crack and deformation response using knowledge-embedded machine learning algorithms

This study introduces an innovative method for identifying the historical maximum load on concrete beams, utilizing a multi-layer back propagation (BP) neural network model with physically interpretable features as inputs. The method establishes a mapping relationship among four-dimensional features (boundary constraints, system parameters, cracks, and deformations) and the maximum load in the load history, using incremental load action and structural response. Moreover, a hybrid optimization algorithm, combining Hunter Prey Optimization (HPO) and Particle Swarm Optimization (PSO), is introduced to optimize the model's parameters. The method is compared with commonly used machine learning models, demonstrating its accuracy and reliability. Additionally, the study performs a sensitivity analysis of the input features using the proposed intelligent identification model. The results reveal that the proposed intelligent identification model exhibits small prediction errors on both the training and testing sets, with average absolute percentage errors of 16.46 % and 19.56 %, and determination coefficients of 0.922 and 0.8185, respectively, showcasing its strong robustness and generalization capability. Compared to the non-optimized multi-layer BP neural network model, the proposed model achieves a reduction in average absolute percentage errors of 39.59 % and 48.33 % on the training and testing sets, respectively.

Multiparameter Mechanical Phenotyping for Accurate Cell Identification Using High-Throughput Microfluidic Deformability Cytometry.

Mechanical phenotyping has been widely employed for single-cell analysis over recent years. However, most previous works on characterizing the cellular mechanical properties measured only a single parameter from one image. In this paper, the quasi-real-time multiparameter analysis of cell mechanical properties was realized using high-throughput adjustable deformability cytometry. We first extracted 12 deformability parameters from the cell contours. Then, the machine learning for cell identification was performed to preliminarily verify the rationality of multiparameter mechanical phenotyping. The experiments on characterizing cells after cytoskeletal modification verified that multiple parameters extracted from the cell contours contributed to an identification accuracy of over 80%. Through continuous frame analysis of the cell deformation process, we found that temporal variation and an average level of parameters were correlated with cell type. To achieve quasi-real-time and high-precision multiplex-type cell detection, we constructed a back propagation (BP) neural network model to complete the fast identification of four cell lines. The multiparameter detection method based on time series achieved cell detection with an accuracy of over 90%. To solve the challenges of cell rarity and data lacking for clinical samples, based on the developed BP neural network model, the transfer learning method was used for the identification of three different clinical samples, and finally, a high identification accuracy of approximately 95% was achieved.

Accurate measurement techniques and prediction approaches for the in-situ rock stress

The precise calculation and evaluation of the in-situ rock stress tensor is a crucial factor in addressing the major challenges related to subsurface engineering applications and earth science research. To improve the accuracy of in-situ stress measurement and prediction, an improved overcoring technique involving a measurement circuit, temperature compensation, and calculation method is presented for accurately measuring the in-situ rock stress tensor. Furthermore, an embedded grey BP neural network (GM–BPNN) model is established for predicting in-situ rock stress values. The results indicate that the improved overcoring technique has significantly improved the stress measurement accuracy, and a large number of valuable stress data obtained from many mines have proved the testing performance of this technique. Moreover, the mean relative errors of the prediction results of GM(0, 1) for the three principal stresses all reach 6–30%, and the accuracy of the model fails to meet the requirements. The average relative errors of the prediction results of the BPNN model are all less than 10%, and the model accuracy meets the requirements and has sufficient credibility. Compared with the GM and BPNN models, the embedded GM–BPNN model produces the best results, with mean relative errors of 0.0001–4.8338%. The embedded GM–BPNN model fully utilizes the characteristics of grey theory and BP neural network, which require a small sample size, weaken the randomness of the original data, and gradually approach the accuracy of the model, making it particularly suitable for situations with limited stress data.

Two-stage fusion framework driven by domain knowledge for penetration prediction of laser welding

To solve the problem of industrial application difficulties caused by insufficient depth exploration and poor interpretability of the pure data drive neural network. A two-stage fusion framework driven by domain knowledge was proposed and focused on industrial applications: laser welding penetration monitoring. First, a multi-sensor acquisition system based on plasma and molten pool signals is developed, extracting multiple features from plasma and molten pool images. Two Back Propagation (BP) neural network information fusion models are established in the first fusion stage, leveraging the characteristics of the plasma plume and molten pool front. In the second fusion stage, the Dempster-Shafer (DS) evidence theory is employed to fuse the information from the two BP neural network models. The proposed method, achieving a high accuracy of 98.7 %, is validated by comparing its average accuracy and anti-interference capability with classical Convolutional Neural Network (CNN) and BP neural network models. The physical relationship between input and output is elucidated through an established physical model, enhancing the interpretability and anti-interference capability of the model, and thereby improving the stability and accuracy of the system.

Early screening and staging of melanoma using blood based on laser-induced breakdown spectroscopy

For melanoma, early screening could increase the cure rate, while staging helps to make treatment strategies. Blood sampling has advantages of little damage, convenient operation and low cost, which combined with laser-induced breakdown spectroscopy (LIBS) has been utilized for tumor diagnoses. Here, we proposed to accurately attain early screening and staging of melanoma blood using LIBS. The serum was collected from 25 melanoma mice and 10 healthy controls on the 7th, 14th, 21st and 28th days. Compared with k nearest neighbor (kNN), support vector machine (SVM) and back propagation neural network (BPNN) models, the adaptive boosting of BPNN (BP_AdaBoost) models had the best accuracies of 83.37 % for early screening and 96.18 % for staging, respectively. Using mutual information (MI) method to select features, the accuracies of BP_AdaBoost models were improved to 86.11 % for early screening and 96.91 % for staging, respectively. Besides, the difference significance of elements and molecular bands in the serum was examined by the Kruskal-Wallis (K-W) test. The test results showed that obvious differences of Ca and Na existed in both early screening and staging, while K and Mg made significant differences in staging, consistent with roles of Ca and Na in the whole process of tumor development and roles of K and Mg in tumor proliferation and metastasis. Overall, all results demonstrated that early screening and staging of melanoma could be accurately realized using blood based on LIBS.

Wearable glove gesture recognition based on fiber Bragg grating sensing using genetic algorithm-back propagation neural network

The recognition of gestures through intelligent wearable devices has a broad array of applications, including low-visibility industrial sites and robotics. However, accurately and swiftly recognizing gestures poses a significant challenge. This study addresses the complex interplay between multimodal information sensing and signal processing with flexible sensors in wearable data gloves by conducting theoretical and experimental investigations into gesture recognition using fiber Bragg grating (FBG) flexible sensors. First, the degree of bending and bending angle of the FBG sensor were calibrated, and the impact of temperature was investigated. Subsequently, a human hand wearable data glove gesture recognition system based on fiber grating sensing was constructed, and gesture recognition experiments were conducted. Finally, an enhanced error backpropagation (BP) neural network model for gesture recognition based on genetic algorithms (GA) was established and tested and the results were compared with those achieved using a conventional error BP neural network. The comparison revealed that under the same network structure, the recognition accuracy of the GA-BP neural network of approximately 100% was higher than the conventional BP neural network’s accuracy of 95.77%. Thus, the development of gesture recognition using FBG sensors provides a theoretical foundation and technical support for multiple applications of wearable data gloves, driving their adoption in areas such as human–computer interactions, intelligent robotics, and complex high-risk environments.

High-Temperature Characteristics of Polyphosphoric Acid-Modified Asphalt and High-Temperature Performance Prediction Analysis of Its Mixtures

To promote the application of economical and sustainable polyphosphoric acid (PPA)-modified asphalt in road engineering, styrene-butadiene block copolymer (SBS), styrene-butadiene rubber (SBR), and PPA were used to prepare PPA/SBS and PPA/SBR composite-modified asphalts, which were tested and the data analyzed. Fourier transform infrared spectroscopy (FTIR) tests and thermogravimetric analysis (TG) tests were carried out to study the modification mechanisms of the composite-modified asphalts, and the high-temperature performance of the PPA-modified asphalt and asphalt mixtures was analyzed by dynamic shear rheology (DSR) tests and wheel tracking tests. A gray correlation analysis and a back-propagation (BP) neural network were utilized to construct a prediction model of the high-temperature performance of the asphalt and asphalt mixtures. The test results indicate that PPA chemically interacts with the base asphalt and physically integrates with SBS and SBR. The PPA-modified asphalt has a higher decomposition temperature than the base asphalt, indicating superior thermal stability. As the PPA dosage increases, the G*/sinδ value of the PPA-modified asphalt also increases. In particular, when 0.6% PPA is combined with 2% SBS/SBR, it surpasses the high-temperature performance achieved with 4% SBS/SBR, suggesting that PPA may be a good alternative for polymer modifiers. In addition, the creep recovery of PPA-modified asphalt is influenced by the stress level, and as the stress increases, the R-value decreases, resulting in reduced elastic deformation. Furthermore, the BP neural network model achieved a fit of 0.991 in predicting dynamic stability, with a mean percentage of relative error (MAPE) of 6.15% between measured and predicted values. This underscores the feasibility of using BP neural networks in predictive dynamic stability models.

Evaluation and Screening of Technological Innovation and Entrepreneurship Based on Improved BPNN Model

Aiming at the low success rate of incubation investment of China’s technology-based start-ups, how to scientifically evaluate technology-based start-ups has become an important issue that needs to be faced. Considering the expert consultation characteristics of the Delphi method is extremely suitable for decision-making problems in the field of uncertainty, as well as the strong nonlinear mapping ability and learning ability of the backpropagation neural network (BPNN). According to the characteristics of the evaluation object and following the principle of index selection, the study uses the Delphi method to determine the evaluation index system suitable for technology-based entrepreneurial enterprises in the current environment and obtain the scores of each index. Based on the established evaluation index system, the BPNN evaluation model is further constructed, and its parameters are optimised to improve its performance. Aiming at the problem that it is easy to fall into a local optimal, a genetic algorithm (GA) is used to optimise it, and a GA-BPNN model is constructed. The comprehensive capability of GA-BPNN is evaluated by using the excellent nonlinear characteristic analysis ability of GA-BPNN to provide a reference for important decisions such as investment. Using BPNN simulation, it was concluded that the correct rate of evaluation of qualified enterprises was between 23.32% and 89.99%, with an average correct rate of 58.32%. The average correct rate was 80.99%. The evaluation accuracy rate was unstable and the average accuracy rate was low. The optimised GA-BPNN model had an average evaluation accuracy rate of 80.32% for qualified enterprises and 93.66% for unqualified enterprises, and the average evaluation accuracy rate increased by 21.99% and 12.66%, respectively. The effectiveness of the model and algorithm was verified. It shows that the GA-BPNN model can be used as an effective tool for the evaluation and screening of technology-based entrepreneurial enterprises. The evaluation system of technology-based entrepreneurial enterprises established by research is scientific and can be applied in practice.

Exploring microstructure evolution and machine-learning methods based on SCAT-CIWOA-BP-DMM theory during hot deformation of 56Ni–32Ti–12Hf alloy

Hot compression simulation tests were carried out on a 56Ni–32Ti–12Hf alloy at temperatures ranging from 800 to 1000 °C and strain rates ranging from 0.001 to 1 s−1. The study aimed to explore the hot deformation behavior and microstructure evolution mechanism of the alloy. Two flow stress prediction models were developed: a strain-compensated Arrhenius constitutive (SCAT) model based on phenomenology and a chaotic mapping adaptive inertia weight whale optimization algorithm-optimized back propagation neural network (CIWOA-BPNN) model based on machine learning. In comparison to the SCAT model, the CIWOA-BPNN model demonstrates higher prediction accuracy, improved error correlation coefficient, decreased average relative error, and lower root mean square error and mean absolute error. The hot processing map of the 56Ni–32Ti–12Hf alloy was developed using dynamic material modeling (DMM) theory and flow stress data predicted by the CIWOA-BPNN model. The optimal hot processing range was found to be between temperatures of 875–975 °C and strain rates of 0.001–1 s−1. The study found that within the optimal processing interval, the softening mechanisms observed were discontinuous dynamic recrystallization (DDRX) and continuous dynamic recrystallization (CDRX). An increase in power dissipation efficiency (η) led to an increase in the dynamic recrystallization (DRX) fraction from 27.1 % to 41.9 %, while the substructure fraction decreased from 34.2 % to 19.8 %. The decrease in the number of dislocations around the DDRX grains further validated the accuracy of the identified optimum machining interval in this research.