A user decision behavior analysis method and system based on mixed effect logistic regression in a digital twin environment, a terminal, and a storage medium

Abstract

The application relates to the technical field of digital twinning, and discloses a user decision behavior analysis method and system based on a mixed effect Logistic regression in a digital twinning environment, a terminal and a storage medium. The method comprises the following steps: collecting multi-source heterogeneous data of a subject in a VR experiment scene; generating a unified format data set according to continuous variables and discrete classification variables; generating multi-factor interaction items according to the unified format data set, constructing a generalized linear mixed model, and performing maximum likelihood method parameter estimation; generating a decision probability prediction value according to a parameter estimation result; verifying the effectiveness of the generalized linear mixed model; and outputting a user decision behavior analysis result. The application realizes effective integration of multi-source heterogeneous data and precise quantification of marginal utility of each influencing factor, and can provide scientific and precise quantitative basis for pedestrian behavior modeling and emergency management plan deduction in a digital twinning environment.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method, system, terminal, and computer-readable storage medium for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment. Background Technology

[0002] In the construction of digital twin models for pedestrian evacuation, the underlying supporting technology is a high-precision pedestrian simulation model. Its core task is to accurately simulate and predict the movement trajectory and decision-making patterns of pedestrians in complex or emergency environments through algorithms. Currently, Virtual Reality (VR) is widely used as the main technical means to collect multi-source heterogeneous data from subjects during the evacuation process by constructing immersive VR experimental systems.

[0003] Existing technical solutions heavily rely on classic significance testing tools such as t-tests and ANOVA. They typically treat a single factor as the independent variable and decision-making behavior as the dependent variable, grouping and comparing them to determine whether the single variable has a statistically significant impact on decision-making behavior. While this approach can initially establish a correlation between a single variable and decision-making behavior, pedestrian decision-making is essentially a dynamic process involving the coupling of multiple environmental, physiological, and psychological factors. Existing technical solutions, which overly rely on static independent sample significance tests, exhibit significant underlying logical limitations when dealing with multi-source, heterogeneous data with complex nested relationships in VR experiments. This makes it difficult to support the construction of high-precision pedestrian behavior models in digital twin cities.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide a method, system, terminal, and computer-readable storage medium for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment. This aims to solve the problem in the prior art that the over-reliance on significance analysis makes it impossible to effectively integrate and quantify multi-source heterogeneous data collected in VR experiments.

[0006] To achieve the above objectives, this invention provides a method for analyzing user decision-making behavior in a digital twin environment based on mixed-effects logistic regression. The method includes the following steps: Collect multi-source heterogeneous data of subjects in VR experimental scenarios, wherein the multi-source heterogeneous data includes continuous variables, discrete categorical variables and binary classification decision result variables; Generate a unified format dataset based on the continuous variables and the discrete categorical variables; Based on the unified format dataset, multi-factor interaction terms are generated. Based on the unified format dataset and the multi-factor interaction terms, a generalized linear mixed model is constructed. The parameters of the generalized linear mixed model are estimated by the maximum likelihood method to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. Based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept, a predicted decision probability value is generated. Based on the predicted decision probability values ​​and the binary classification decision result variables, the effectiveness of the generalized linear mixed model is verified to obtain the model effectiveness verification results. Based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results, the output shows the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value.

[0007] Furthermore, the collection of multi-source heterogeneous data from subjects in the VR experimental scenario specifically includes: Collect raw, multi-source, heterogeneous data from subjects in VR experimental scenarios; The original multi-source heterogeneous data is aligned according to timestamps to obtain structured multi-source heterogeneous data of subjects in VR experimental scenarios.

[0008] Furthermore, the step of generating a unified format dataset based on the continuous variable and the discrete categorical variable specifically includes: The continuous variable is subjected to Z-score standardization to obtain a standardized continuous variable, wherein the formula for calculating the Z-score standardization is as follows: ; in, Indicates the first The first subject was on the Standardized values ​​over continuous variables Indicates the first The first subject was on the The original observations on a continuous variable, Indicates the first The sample population mean of a continuous variable, Indicates the first Standard deviation of a continuous variable; The discrete categorical variables are encoded to obtain coded categorical variables; A unified format dataset is generated based on the standardized continuous variables and the encoded categorical variables.

[0009] Furthermore, the step of generating multi-factor interaction terms based on the unified format dataset, constructing a generalized linear mixture model based on the unified format dataset and the multi-factor interaction terms, and performing maximum likelihood estimation on the generalized linear mixture model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercepts specifically includes: Based on the standardized continuous variables and coded categorical variables in the unified format dataset, generate multi-factor interaction terms; Based on the standardized continuous variables, the coded categorical variables, and the multi-factor interaction terms, a generalized linear mixed model is constructed, and the parameters of the generalized linear mixed model are estimated by the maximum likelihood method to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. The mathematical expression for the generalized linear mixture model is: ; in, Indicates the first The first subject was on the The decision-making results in this experiment Indicates the first The first subject was on the The probability of making a specific evacuation decision in this experiment. Represents the natural logarithm function. The logarithmic form of the decision occurrence ratio. Indicates a fixed intercept term. This represents the total number of fixed-effects variables. Indicates the ordinal number of the fixed-effects variable. Indicates the first The regression coefficients of the fixed-effects variables, Indicates the first The first subject was on the In the experiment, the first The predictor values ​​corresponding to each fixed effect variable. Indicates the first Individual random intercept of each subject, This represents the residual term.

[0010] Furthermore, the step of generating decision probability prediction values ​​based on the unified format dataset, the estimated regression coefficients of each fixed-effects variable, and the estimated variance components of the individual random intercepts specifically includes: Based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept, the linear prediction results are calculated. Based on the linear prediction results, the inverse Logit function is used to map the linear prediction results to a probability interval to obtain the decision probability prediction value.

[0011] Furthermore, the step of validating the generalized linear mixed model based on the predicted decision probability value and the binary classification decision result variable to obtain the model validity validation result specifically includes: Based on the predicted decision probability value and the binary classification decision result variable, residual diagnosis is performed to obtain the residual diagnosis result; Based on the predicted decision probability value and the binary classification decision result variable, an ROC evaluation is performed to obtain the ROC evaluation result. Based on the residual diagnosis results and the ROC evaluation results, the effectiveness of the generalized linear mixed model is verified, and the model effectiveness verification results are generated.

[0012] Furthermore, the step of outputting the direction of influence and the magnitude of the specific utility value of each factor or variable on the user's decision based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results specifically includes: Based on the estimated regression coefficients of each fixed-effect variable and the estimated variance components of the individual random intercepts, the quantitative results of the influencing factors are generated. Based on the predicted decision probability, the model validity verification results, and the quantification results of the influencing factors, the direction of the effect of each factor or variable on the user's decision and the magnitude of its specific utility value are output.

[0013] Furthermore, to achieve the above objectives, the present invention also provides a user decision-making behavior analysis system based on mixed-effects logistic regression in a digital twin environment. This system is used to implement the user decision-making behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described above. The system includes: The data acquisition module is used to collect multi-source heterogeneous data of subjects in VR experimental scenarios to obtain multi-source heterogeneous datasets; wherein, the multi-source heterogeneous data includes continuous variables, discrete categorical variables and binary classification decision result variables; The data preprocessing module is used to generate a unified format dataset based on the continuous variables and discrete categorical variables in the multi-source heterogeneous dataset; The model building module is used to generate multi-factor interaction terms based on the unified format dataset, construct a generalized linear mixed model based on the unified format dataset and the multi-factor interaction terms, and perform maximum likelihood parameter estimation on the generalized linear mixed model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. The decision prediction module is used to generate decision probability prediction values ​​based on the unified format dataset, the regression coefficient estimates of each fixed effect variable, and the variance component estimates of the individual random intercept. The model validation module is used to validate the effectiveness of the generalized linear mixed model based on the predicted decision probability value and the binary classification decision result variable, and obtain the model effectiveness validation result. The results output module is used to output the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value, based on the estimated regression coefficients of each fixed effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results.

[0014] Furthermore, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment stored in the memory and executable on the processor. When the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment is executed by the processor, it implements the steps of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described above.

[0015] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment, and when the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment is executed by a processor, it implements the steps of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described above.

[0016] The beneficial effects of this invention are as follows: This invention collects multi-source heterogeneous data from subjects in VR experimental scenarios, enabling the acquisition of a complete data foundation covering continuous variables, discrete categorical variables, and binary decision outcome variables. A unified-format dataset is generated based on the continuous and discrete categorical variables, eliminating differences between different units of measurement and allowing each variable to participate in subsequent regression calculations and parameter estimation within the same framework. Based on the unified-format dataset, multi-factor interaction terms are generated and a generalized linear mixture model is constructed, simultaneously quantifying the main effects of each influencing factor and the interaction coupling effects between multiple factors. The individual random intercept introduced by the model effectively distinguishes between individual heterogeneity and the true effects of influencing factors, avoiding interference from inherent individual differences in parameter estimation. Decision probability prediction values ​​are generated based on the estimated model parameters, providing a quantified probabilistic expression for subjects' evacuation decisions under specific conditions, transforming decision-making behavior from a qualitative description into a calculable numerical indicator. The model's effectiveness is validated to ensure the statistically significant credibility of the output results. The system comprehensively outputs the regression coefficient estimates, random effect variance components, decision probability predictions, and model validity verification results of various influencing factors, providing directly accessible quantitative parameters and scientific basis for pedestrian behavior modeling, emergency management plan simulation, and building evacuation design optimization in digital twin environments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment according to the present invention; Figure 2 This is an overall flowchart of the user decision-making behavior analysis method based on mixed-effects logistic regression in the digital twin environment of the present invention; Figure 3 This is a diagnostic diagram of the model residuals; Figure 4 For predicting probability distribution; Figure 5 A visualization of the fixed effects coefficients of each influencing factor; Figure 6 This is a structural diagram of a preferred embodiment of the user decision-making behavior analysis system based on mixed-effects logistic regression in the digital twin environment of the present invention; Figure 7 This is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0018] This application provides a method, system, terminal, and storage medium for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment. To make the objectives, technical solutions, and effects of this application clearer and more explicit, the following detailed description, with reference to the accompanying drawings and embodiments, further illustrates this application. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0020] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0021] The preferred embodiment of the present invention describes a method for analyzing user decision-making behavior in a digital twin environment based on mixed-effects logistic regression, such as... Figure 1 As shown, the method for analyzing user decision-making behavior based on mixed-effects logistic regression in the digital twin environment includes the following steps: S10. Collect multi-source heterogeneous data of subjects in VR experimental scenarios, wherein the multi-source heterogeneous data includes continuous variables, discrete classification variables and binary classification decision result variables.

[0022] Furthermore, the collection of multi-source heterogeneous data from subjects in the VR experimental scenario specifically includes: S11. Collect raw, multi-source, heterogeneous data of subjects in the VR experimental scenario; S12. Align the original multi-source heterogeneous data according to the timestamp to obtain the multi-source heterogeneous data of the subjects in the structured VR experimental scene.

[0023] In this embodiment, data acquisition relies on the synchronous acquisition capabilities of the VR experimental platform. The VR device is equipped with a positioning tracking system with a sampling frequency of 90Hz, used to record the spatial coordinates and motion trajectory data of the subject in the virtual scene in real time. The synchronously acquired physiological monitoring devices include a heart rate sensor (sampling frequency of 60Hz) and an eye tracker (sampling frequency of 120Hz), used to record the subject's heart rate variability and fixation point distribution, respectively. During the experiment, the data collected by each sensor carries a unified timestamp information to ensure time alignment between behavioral data, physiological data, and environmental parameters.

[0024] The data collection encompassed three core categories: continuous variables, including General Self-Efficacy Scale (GSES) scores, Interpersonal Trust Scale (ITS) scores, age, scene familiarity, conformity tendency, self-determination tendency, follower rate, and social hierarchy; these indicators reflect individual attributes and psychological states. Discrete categorical variables included gender, 3D vision activation status, augmented reality device type, leader type, and subway riding frequency; these indicators described scene parameters and individual characteristics. Binary decision-making outcome variables represented path selection results (defined as "0 = path A selected, 1 = path B selected"), directly characterizing pedestrian evacuation decision-making behavior. During the data collection process, sensors on the VR device synchronously recorded data, ensuring the timestamp alignment of behavioral data, physiological data, and environmental parameters, forming a structured raw dataset that provides a complete data foundation for subsequent processing.

[0025] In a specific example of this scheme, the system collected and constructed a multidimensional structured raw dataset containing 315 valid decision data points covering 67 subjects. Multiple decision data points from the same subject shared the same subject ID, thus providing a data foundation for subsequent processing of nested relationships.

[0026] It should be noted that the VR experimental scenario described in step S10 is a pre-constructed virtual subway station evacuation scenario. The scenario includes multiple selectable evacuation routes, and throughout the evacuation, at each decision point, the subject will face a binary decision choice (e.g., left or right). Subjects will need to make route selection decisions when encountering emergencies (such as a fire alarm) in the virtual environment. The specific types of continuous and discrete categorical variables are not limited to those listed in this embodiment and can be expanded or replaced according to actual research needs, as long as they belong to the same data type.

[0027] S20. Generate a unified format dataset based on the continuous variables and the discrete categorical variables.

[0028] The purpose of this step is to convert the original heterogeneous data into an input format suitable for the mixed-effects logistic regression model, eliminate the influence of different units on the model convergence, and ensure that the regression coefficients of each variable are comparable.

[0029] Furthermore, generating a unified format dataset based on the continuous variable and the discrete categorical variable specifically includes: S21. Perform Z-score standardization on the continuous variable to obtain a standardized continuous variable; S22. Encode the discrete categorical variables to obtain coded categorical variables; S23. Generate a unified format dataset based on the standardized continuous variables and the coded categorical variables.

[0030] In this embodiment, Z-score normalization is used to transform continuous variables. The calculation formula for Z-score normalization is as follows: ; in, Indicates the first The first subject was on the Standardized values ​​over continuous variables Indicates the first The first subject was on the The original observations on a continuous variable, Indicates the first The sample population mean of a continuous variable, Indicates the first The standard deviation of each continuous variable. This formula transforms all continuous variables into a standard normal distribution with a mean of 0 and a standard deviation of 1, eliminating the influence of different units on model convergence and ensuring the stability of parameter estimation.

[0031] For discrete categorical variables, one-hot encoding is used for numerical transformation. Taking the leader type variable as an example, which includes leaderless, security leader, station leader, humanoid robot leader, and ordinary pedestrian leader, one-hot encoding converts them into binary variables (0 or 1), enabling the categorical variable to participate in regression calculations. For binary variables (such as gender, 3D vision activation status), 0 or 1 binary encoding is directly applied.

[0032] After preprocessing, the differences in the scales of the aforementioned continuous variables (such as age, scale scores, etc.) are eliminated, and they are uniformly transformed into standard normal distribution data with a mean of 0 and a standard deviation of 1. This ultimately generates a unified format dataset of standardized continuous variables, coded categorical variables, and binary classification decision results, ensuring absolute comparability of the regression coefficients of each variable during subsequent model parameter estimation.

[0033] It should be noted that the encoding process described in step S22 is not limited to one-hot encoding. For ordered categorical variables, label encoding can also be used for conversion, as long as the categorical variable can be converted into a numerical input. The Z-score standardization described in step S21 is calculated based on the population mean and standard deviation. In practical applications, if new data is added subsequently, the mean and standard deviation estimated from the training set can be used to standardize the new data.

[0034] S30. Based on the unified format dataset, generate multi-factor interaction terms. Based on the unified format dataset and the multi-factor interaction terms, construct a generalized linear mixed model. Perform maximum likelihood estimation on the generalized linear mixed model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept.

[0035] The purpose of this step is to achieve a quantitative analysis of the decision-making influence mechanism by constructing a generalized linear mixture model, integrating fixed effects and random effects, quantifying the main effects and interaction effects of each influencing factor on the decision, and capturing individual heterogeneity to avoid averaging bias.

[0036] Further, the step of generating multi-factor interaction terms based on the unified format dataset, constructing a generalized linear mixture model based on the unified format dataset and the multi-factor interaction terms, and performing maximum likelihood parameter estimation on the generalized linear mixture model to obtain estimated regression coefficients of each fixed effect variable and estimated variance components of the individual random intercepts specifically includes: S31. Generate multi-factor interaction terms based on the standardized continuous variables and coded categorical variables in the unified format dataset; S32. Based on the standardized continuous variables, the coded categorical variables, and the multi-factor interaction terms, construct a generalized linear mixed model, and perform maximum likelihood estimation on the generalized linear mixed model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept.

[0037] In this embodiment, a binomial distribution family is used as the basic distribution, coupled with a Logit link function, to map the decision probability to a linear prediction space. The model structure integrates fixed effects and random effects to form a dual-effect architecture: the fixed effects term includes standardized continuous variables, encoded categorical variables, and multi-factor interaction terms, used to quantify the main effects and interaction effects of each influencing factor on the decision. The multi-factor interaction terms are generated by combining the standardized continuous variables and encoded categorical variables pairwise to generate second-order interaction terms (such as "follower rate × leader type", "hierarchy × conformity tendency", etc.), used to capture the interaction coupling effects between multiple factors.

[0038] The random effects term introduces the individual random intercept of the subjects. And assume that it follows a normal distribution. ,in, This indicates that the mean is 0 and the variance is 0. The normal distribution; Let be the mean of this normal distribution. This represents the variance component of the individual random intercept. It is used to capture individual heterogeneity (such as inherent differences in risk preference and decision-making style) in hierarchically nested data generated from multiple repeated experiments on the same subject, addressing the averaging bias problem caused by traditional models neglecting individual differences. The mathematical expression of the generalized linear mixture model is: ; in, Indicates the first The first subject was on the The decision-making results in this experiment Indicates the first The first subject was on the The probability of making a specific evacuation decision in this experiment. Represents the natural logarithm function. The logarithmic form of the decision occurrence ratio (i.e., the Logit transformation). Indicates a fixed intercept term. This represents the total number of fixed-effects variables. Indicates the ordinal number of the fixed-effects variable. Indicates the first The regression coefficients of the fixed-effects variables, Indicates the first The first subject was on the In the experiment, the first The predictor values ​​corresponding to each fixed effect variable. Indicates the first Individual random intercept of each subject, This represents the residual term.

[0039] Through maximum likelihood method fitting and calculation, this scheme successfully outputs the specific quantization parameter matrix: a. Random effects quantification: The model accurately extracted the variance component among individual subjects as 0.245 (standard deviation 0.495), effectively eliminating the averaging bias caused by individual risk preferences.

[0040] b. Fixed Effects Quantification: Taking the independent variable "level of location" as an example, the model calculates its regression coefficients. β =0.322( P =0.029<0.05). Within the Logistic framework, this is transformed into an occurrence ratio. .

[0041] c. Quantification Mechanism Output: The above results are directly translated into a practically operable numerical mechanism. After eliminating the interference of individual differences and keeping other conditions constant, for every standard deviation increase in the pedestrian's level, the probability of choosing the target path increases by 37.9%. Similarly, for every unit increase in the following rate, this probability increases by 34.5%. These precise marginal utility parameters completely overcome the qualitative limitations of traditional significance analysis.

[0042] It should be noted that the specific generation method of the multi-factor interaction terms in step S31 is not limited to pairwise product combinations. Third-order or higher-order interaction terms can also be generated according to research needs. Moreover, the multi-factor interaction terms are generated in the data preprocessing stage and participate in model construction as independent predictive variables.

[0043] S40. Based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept, generate a predicted decision probability value.

[0044] The purpose of this step is to generate predicted decision probabilities for individuals under specific conditions based on the estimated model parameters, thereby providing quantitative instructions for driving the virtual pedestrian behavior of the digital twin system.

[0045] Further, the step of generating decision probability prediction values ​​based on the unified format dataset, the estimated regression coefficients of each fixed-effects variable, and the estimated variance components of the individual random intercepts specifically includes: S41. Calculate the linear prediction result based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept; S42. Based on the linear prediction result, the inverse Logit function is used to map the linear prediction result to the probability interval to obtain the decision probability prediction value.

[0046] In this embodiment, the prediction calculation uses the inverse Logit function (i.e., the Sigmoid function), which maps the linear prediction value output by the model back to the probability interval [0,1]. The formula for calculating the inverse Logit function is as follows: ; in, Indicates the first The first subject was on the The predicted probability of making a specific evacuation decision in this experiment. This represents an exponential function with the natural constant e as its base. The maximum likelihood estimate or restricted maximum likelihood estimate of the fixed intercept term represents the average decision tendency of all subjects when all influencing factors are zero. Indicates the first The maximum likelihood estimate or restricted maximum likelihood estimate of the regression coefficient of the nth fixed effect variable is used to quantify the nth fixed effect variable regression coefficient. The degree of influence of each influencing factor on the log odds of a decision; Indicates the first The maximum likelihood estimate or restricted maximum likelihood estimate of the random intercept for each individual subject is used to adjust for the baseline decision bias bias caused by the subject's inherent individual traits. This formula combines the various quantified values ​​obtained above. β The coefficient directly measures the absolute probability of a pedestrian's decision under a specific combination, thus providing direct quantitative instructions for driving the behavior of the virtual pedestrian (Non-Player Character, NPC) in the digital twin system.

[0047] It should be noted that the inverse Logit function and the Sigmoid function mentioned in step S42 are two different names for the same mathematical function.

[0048] S50. Based on the predicted decision probability value and the binary classification decision result variable, the effectiveness of the generalized linear mixed model is verified to obtain the model effectiveness verification result.

[0049] The purpose of this step is to objectively evaluate the effectiveness and prediction accuracy of the model through a dual-index verification system, ensuring the reliability of the model's output parameters.

[0050] Further, the step of validating the generalized linear mixed model based on the predicted decision probability value and the binary classification decision result variable to obtain the model validity verification result specifically includes: S51. Based on the predicted decision probability value and the binary classification decision result variable, perform residual diagnosis to obtain the residual diagnosis result; S52. Based on the predicted decision probability value and the binary classification decision result variable, perform ROC (Receiver Operating Characteristic) evaluation to obtain the ROC evaluation result; S53. Based on the residual diagnosis results and the ROC evaluation results, the effectiveness of the generalized linear mixed model is verified, and the model effectiveness verification results are generated.

[0051] In this embodiment, to demonstrate the scientific validity of the output parameters of this scheme, a dual-indicator verification system is adopted, namely: Residual Diagnostic Assessment: The residuals between the actual decision results and the predicted probabilities are extracted and plotted as a scatter plot. The residuals are extracted by calculating the difference between the actual decision result and the predicted probability for each sample; that is, the residual equals the actual decision result minus the predicted probability. A scatter plot of the residuals is plotted with the residuals on the vertical axis and the predicted probability on the horizontal axis. The residuals are uniformly and randomly distributed on both sides of the zero baseline, showing no significant systematic bias (no heteroscedasticity), proving the high reliability of the mathematical fit of this model.

[0052] ROC Curve and AUC (Area Under the Curve) Test: The predicted probabilities of the model are compared with the actual VR decision-making results, and the receiver operating characteristic (ROC) curve is plotted. The ROC curve is plotted as follows: predicted probabilities are sorted from high to low, and the corresponding True Positive Rate (TPR) and False Positive Rate (FPR) are calculated using different thresholds. The curve is plotted with the False Positive Rate on the horizontal axis and the True Positive Rate on the vertical axis. Based on the actual calculation results of this invention, the area under the ROC curve (AUC) of this model reaches 0.791. The measured AUC value close to 0.8 objectively demonstrates that the technical architecture of this invention not only deeply analyzes the decision-making mechanism but also has extremely high predictive accuracy and explanatory power for real pedestrian evacuation decision-making behavior in complex scenarios.

[0053] It should be noted that the AUC value in the ROC evaluation described in step S52 typically ranges from 0.5 to 1.0. A value closer to 1.0 indicates better classification prediction performance. Generally, an AUC value between 0.7 and 0.8 indicates moderate prediction accuracy, while an AUC value above 0.8 indicates high prediction accuracy. In actual testing, this scheme achieved an AUC value of 0.791, demonstrating good predictive ability. The "residual diagnosis" described in step S51 is not limited to the residual scatter plot method; other residual diagnosis methods such as standardized residuals and deviation residuals can also be used, as long as they can assess the model's goodness of fit.

[0054] S60. Based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results, output the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value.

[0055] The purpose of this step is to comprehensively output the quantitative results of influencing factors, the prediction results of decision probability, and the model verification results, so as to provide a complete quantitative basis for the modeling of digital twin city evacuation systems, smart city emergency management, and building evacuation design optimization.

[0056] Furthermore, the step of outputting the direction of influence and the magnitude of the specific utility value of each factor or variable on the user's decision based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results specifically includes: S61. Based on the estimated regression coefficients of each fixed-effect variable and the estimated variance components of the individual random intercept, generate the quantitative results of the influencing factors. S62. Based on the predicted decision probability value, the model validity verification result, and the quantification result of the influencing factors, output the direction of the effect of each factor or variable on the user's decision and the magnitude of its specific utility value.

[0057] In this embodiment, the quantification results of influencing factors in step S61 are numerical expressions of the intensity and direction of the influence of each influencing factor on evacuation decisions, including but not limited to: the regression coefficient values ​​and confidence intervals of each fixed-effect variable, the odds ratio (OR) values ​​and their confidence intervals, and the variance components and standard deviations of random effects. Taking "level of location" as an example, the fixed-effect regression coefficient is 0.322, and the odds ratio is 1.379. That is, under the condition that individual differences are excluded and other conditions remain unchanged, for every standard deviation unit increase in the level of a pedestrian, the probability of choosing the target path increases by 37.9%. After the quantification results of each influencing factor are generated, step S62 summarizes and formats all output results based on the decision probability prediction values ​​obtained in step S40 and the model validity verification results obtained in step S50. Application scenarios in digital twin systems include: inputting the marginal utility parameters of various influencing factors into the behavior decision-making module of the pedestrian simulation model as driving parameters for virtual pedestrian path selection; and inputting the predicted decision probability values ​​into the real-time simulation module of the digital twin model for dynamic simulation and optimization evaluation of evacuation plans in urban high-density crowd gathering scenarios. Step S62 simultaneously outputs the residual diagnosis results and the AUC value from the ROC evaluation results as objective evidence of model effectiveness.

[0058] right Figures 2 to 5 The explanation is as follows: Figure 2 This is a flowchart illustrating the overall process of the user decision-making behavior analysis method based on mixed-effects logistic regression in the digital twin environment of this invention. Figure 2 As shown, this invention first collects multi-source heterogeneous data from subjects using a virtual reality experimental platform. This data includes continuous variables, discrete categorical variables, and binary decision outcome variables. Then, the continuous variables are standardized, and the discrete categorical variables are coded to generate a unified format dataset. Next, a generalized linear mixed model is constructed based on the preprocessed data, and model fitting and parameter estimation are performed, outputting fixed-effects coefficients, random-effects variances, and residuals. Then, significant factors are screened based on the parameter estimation results to identify influencing factors that significantly affect decision-making behavior. Subsequently, the model parameters are substituted into the inverse Logit function to calculate the predicted decision probability values ​​of the subjects under specific conditions. A dual validation system is then employed: the classification accuracy of the model is evaluated using the subject operating characteristic curve and the area under the curve; residual diagnosis is used to analyze the relationship between residuals and predicted probabilities to verify whether the model has systematic biases. Finally, the fixed-effects coefficients, random-effects variances, predicted decision probability values, and model validation indicators are visualized. These steps are executed sequentially, forming a complete processing chain from digital twin experimental data to influencing factor parameters, individual difference parameters, and decision probabilities.

[0059] Figure 3 This is a diagnostic diagram of the model residuals. For example... Figure 3 As shown, this is used to verify the systematic bias of the model fit. The horizontal axis represents the predicted decision probability output by the model, and the vertical axis represents the residual between the actual decision result and the predicted probability. The figure shows that the residuals are uniformly and randomly distributed on both sides of the zero baseline, without exhibiting a clear funnel-shaped or curved trend. This indicates that the model residuals do not have heteroscedasticity or nonlinear characteristics, and the mathematical fit of this model has high reliability, satisfying the basic assumptions of the generalized linear mixture model regarding error distribution.

[0060] Figure 4 This is a probability distribution plot used to demonstrate the model's ability to distinguish the predicted probabilities of two actual decision outcomes (actual outcome = 0 and actual outcome = 1). For example... Figure 4 As shown in the figure, the horizontal axis represents the predicted decision probability value output by the model, and the vertical axis represents the sample frequency within the corresponding probability interval. The figure shows that samples with an actual result of 0 are concentrated in the low probability interval (0.50 to 0.65), while samples with an actual result of 1 are concentrated in the high probability interval (0.65 to 0.75). This indicates that the model has good discriminatory power in predicting the probabilities of the two types of decision outcomes, further validating the model's predictive effectiveness for pedestrian evacuation decisions.

[0061] Figure 5This is a visualization of the fixed effects coefficients of each influencing factor. The horizontal axis represents the estimated regression coefficients of the fixed effects variables, and the vertical axis represents the names of each influencing factor. Figure 5 The coefficient estimates of each variable are represented by dots, and the confidence intervals are represented by horizontal lines. The coefficient estimates of each influencing factor in the figure show a differentiated distribution. The confidence intervals of some variables (such as intercept, subway frequency (3), subway frequency (2), interpersonal trust, and self-efficacy) are far from the zero baseline, indicating that the above factors have a significant impact on pedestrian evacuation decisions. The coefficient estimates of some variables (such as age, following rate, and autonomous decision-making tendency) are small and their confidence intervals cross the zero baseline, indicating that the above factors have a relatively limited impact on decision-making behavior. The figure intuitively reflects the direction and intensity of the influence of each influencing factor on pedestrian evacuation decisions. Positive coefficients indicate that the probability of choosing the target path increases when the value of the factor increases, and negative coefficients indicate that the probability of choosing the target path decreases when the value of the factor increases.

[0062] The beneficial effects of this invention are as follows: This invention collects multi-source heterogeneous data from subjects in VR experimental scenarios, enabling the acquisition of a complete data foundation covering continuous variables, discrete categorical variables, and binary decision outcome variables. A unified-format dataset is generated based on the continuous and discrete categorical variables, eliminating differences between different units of measurement and allowing each variable to participate in subsequent regression calculations and parameter estimation within the same framework. Based on the unified-format dataset, multi-factor interaction terms are generated and a generalized linear mixture model is constructed, simultaneously quantifying the main effects of each influencing factor and the interaction coupling effects between multiple factors. The individual random intercept introduced by the model effectively distinguishes between individual heterogeneity and the true effects of influencing factors, avoiding interference from inherent individual differences in parameter estimation. Decision probability prediction values ​​are generated based on the estimated model parameters, providing a quantified probabilistic expression for subjects' evacuation decisions under specific conditions, transforming decision-making behavior from a qualitative description into a calculable numerical indicator. The model's effectiveness is validated to ensure the statistically significant credibility of the output results. The system comprehensively outputs the regression coefficient estimates, random effect variance components, decision probability predictions, and model validity verification results of various influencing factors, providing directly accessible quantitative parameters and scientific basis for pedestrian behavior modeling, emergency management plan simulation, and building evacuation design optimization in digital twin environments.

[0063] Furthermore, such as Figure 6 As shown, based on the above-mentioned method for analyzing user decision-making behavior in a digital twin environment using mixed-effects logistic regression, this invention also provides a system for analyzing user decision-making behavior in a digital twin environment using mixed-effects logistic regression. The system includes: The data acquisition module 51 is used to acquire multi-source heterogeneous data of subjects in VR experimental scenarios to obtain multi-source heterogeneous datasets; wherein, the multi-source heterogeneous data includes continuous variables, discrete classification variables and binary classification decision result variables; Data preprocessing module 52 is used to generate a unified format dataset based on the continuous variables and discrete categorical variables in the multi-source heterogeneous dataset; The model building module 53 is used to generate multi-factor interaction terms based on the unified format dataset, construct a generalized linear mixed model based on the unified format dataset and the multi-factor interaction terms, and perform maximum likelihood parameter estimation on the generalized linear mixed model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. The decision prediction module 54 is used to generate a decision probability prediction value based on the unified format dataset, the regression coefficient estimates of each fixed effect variable, and the variance component estimates of the individual random intercept. The model validation module 55 is used to validate the effectiveness of the generalized linear mixed model based on the predicted decision probability value and the binary classification decision result variable, and obtain the model effectiveness validation result. The result output module 56 is used to output the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value based on the estimated regression coefficients of each fixed effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results.

[0064] Furthermore, such as Figure 7 As shown, based on the above-mentioned method and system for analyzing user decision-making behavior in a digital twin environment using mixed-effects logistic regression, the present invention also provides a terminal, which includes a processor 10, a memory 20, and a display 30. Figure 7 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0065] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a user decision behavior analysis program 40 based on mixed-effects logistic regression in a digital twin environment. This user decision behavior analysis program 40 based on mixed-effects logistic regression in a digital twin environment can be executed by the processor 10, thereby implementing the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described in this application.

[0066] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing a user decision behavior analysis method based on mixed-effects logistic regression in the digital twin environment.

[0067] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0068] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment, and when the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment is executed by a processor, the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment implements the steps of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described above.

[0069] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0070] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for analyzing user decision-making behavior in a digital twin environment based on mixed-effects logistic regression, characterized in that, The method for analyzing user decision-making behavior based on mixed-effects logistic regression in the digital twin environment includes the following steps: Collect multi-source heterogeneous data of subjects in VR experimental scenarios, wherein the multi-source heterogeneous data includes continuous variables, discrete categorical variables and binary classification decision result variables; Generate a unified format dataset based on the continuous variables and the discrete categorical variables; Based on the unified format dataset, multi-factor interaction terms are generated. Based on the unified format dataset and the multi-factor interaction terms, a generalized linear mixed model is constructed. The parameters of the generalized linear mixed model are estimated by the maximum likelihood method to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. Based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept, a predicted decision probability value is generated. Based on the predicted decision probability values ​​and the binary classification decision result variables, the effectiveness of the generalized linear mixed model is verified to obtain the model effectiveness verification results. Based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results, the output shows the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value.

2. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The collection of multi-source heterogeneous data from subjects in VR experimental scenarios specifically includes: Collect raw, multi-source, heterogeneous data from subjects in VR experimental scenarios; The original multi-source heterogeneous data is aligned according to timestamps to obtain structured multi-source heterogeneous data of subjects in VR experimental scenarios.

3. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The step of generating a unified format dataset based on the continuous variable and the discrete categorical variable specifically includes: The continuous variable is subjected to Z-score standardization to obtain a standardized continuous variable, wherein the formula for calculating the Z-score standardization is as follows: ； in, Indicates the first The first subject was on the Standardized values ​​over continuous variables Indicates the first The first subject was on the The original observations on a continuous variable, Indicates the first The sample population mean of a continuous variable, Indicates the first Standard deviation of a continuous variable; The discrete categorical variables are encoded to obtain coded categorical variables; A unified format dataset is generated based on the standardized continuous variables and the encoded categorical variables.

4. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The process involves generating multi-factor interaction terms based on the unified format dataset, constructing a generalized linear mixture model based on the unified format dataset and the multi-factor interaction terms, and performing maximum likelihood estimation on the generalized linear mixture model to obtain estimated regression coefficients for each fixed effect variable and estimated variance components of the individual random intercepts. Specifically, this includes: Based on the standardized continuous variables and coded categorical variables in the unified format dataset, generate multi-factor interaction terms; Based on the standardized continuous variables, the coded categorical variables, and the multi-factor interaction terms, a generalized linear mixed model is constructed, and the parameters of the generalized linear mixed model are estimated by the maximum likelihood method to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. The mathematical expression for the generalized linear mixture model is: ； in, Indicates the first The first subject was on the The decision-making results in this experiment Indicates the first The first subject was on the The probability of making a specific evacuation decision in this experiment. Represents the natural logarithm function. The logarithmic form of the decision occurrence ratio. Indicates a fixed intercept term. This represents the total number of fixed-effects variables. Indicates the ordinal number of the fixed-effects variable. Indicates the first The regression coefficients of the fixed-effects variables, Indicates the first The first subject was on the In the experiment, the first The predictor values ​​corresponding to each fixed effect variable. Indicates the first Individual random intercept of each subject, This represents the residual term.

5. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The step of generating decision probability predictions based on the unified format dataset, the estimated regression coefficients of each fixed-effects variable, and the estimated variance components of the individual random intercepts specifically includes: Based on the unified format dataset, the estimated regression coefficients of each fixed effect variable, and the estimated variance components of the individual random intercept, the linear prediction results are calculated. Based on the linear prediction results, the inverse Logit function is used to map the linear prediction results to a probability interval to obtain the decision probability prediction value.

6. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The step of validating the generalized linear mixture model based on the predicted decision probability value and the binary classification decision result variable, and obtaining the model validity validation result, specifically includes: Based on the predicted decision probability value and the binary classification decision result variable, residual diagnosis is performed to obtain the residual diagnosis result; Based on the predicted decision probability value and the binary classification decision result variable, ROC evaluation is performed to obtain the ROC evaluation result; Based on the residual diagnosis results and the ROC evaluation results, the effectiveness of the generalized linear mixed model is verified, and the model effectiveness verification results are generated.

7. The method for analyzing user decision-making behavior based on mixed-effects logistic regression in a digital twin environment according to claim 1, characterized in that, The step involves outputting the direction of influence and the magnitude of the specific utility value of each factor or variable on the user's decision-making process, based on the estimated regression coefficients of each fixed-effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity validation results. Specifically, this includes: Based on the estimated regression coefficients of each fixed-effect variable and the estimated variance components of the individual random intercepts, the quantitative results of the influencing factors are generated. Based on the predicted decision probability, the model validity verification results, and the quantification results of the influencing factors, the direction of the effect of each factor or variable on the user's decision and the magnitude of its specific utility value are output.

8. A user decision-making behavior analysis system based on mixed-effects logistic regression in a digital twin environment, characterized in that, The user decision behavior analysis system based on mixed-effects logistic regression in the digital twin environment is used to implement the user decision behavior analysis method based on mixed-effects logistic regression in the digital twin environment as described in any one of claims 1-7. The user decision behavior analysis system based on mixed-effects logistic regression in the digital twin environment includes: The data acquisition module is used to collect multi-source heterogeneous data of subjects in VR experimental scenarios to obtain multi-source heterogeneous datasets; wherein, the multi-source heterogeneous data includes continuous variables, discrete categorical variables and binary classification decision result variables; The data preprocessing module is used to generate a unified format dataset based on the continuous variables and discrete categorical variables in the multi-source heterogeneous dataset; The model building module is used to generate multi-factor interaction terms based on the unified format dataset, construct a generalized linear mixed model based on the unified format dataset and the multi-factor interaction terms, and perform maximum likelihood parameter estimation on the generalized linear mixed model to obtain the regression coefficient estimates of each fixed effect variable and the variance component estimates of the individual random intercept. The decision prediction module is used to generate decision probability prediction values ​​based on the unified format dataset, the regression coefficient estimates of each fixed effect variable, and the variance component estimates of the individual random intercept. The model validation module is used to validate the effectiveness of the generalized linear mixed model based on the predicted decision probability value and the binary classification decision result variable, and obtain the model effectiveness validation result. The results output module is used to output the direction of influence of each factor or variable on the user's decision and the magnitude of its specific utility value, based on the estimated regression coefficients of each fixed effect variable, the estimated variance components of the individual random intercept, the predicted decision probability, and the model validity verification results.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment, stored in the memory and executable on the processor. When the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment is executed by the processor, it implements the steps of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment. When the user decision behavior analysis program based on mixed-effects logistic regression in a digital twin environment is executed by a processor, it implements the steps of the user decision behavior analysis method based on mixed-effects logistic regression in a digital twin environment as described in any one of claims 1-7.

Images