EVTOL travel selection prediction method based on potential category model

Abstract

The invention relates to an eVTOL travel selection prediction method based on a potential category model, and aims to optimize decision analysis of an urban air traffic system. According to the method, an expansion technology acceptance model is combined, psychological latent variables such as perception usefulness, perception trust, social influence and behavior intention are comprehensively considered, and decision analysis of an urban air traffic system is optimized. Firstly, traveler features are analyzed based on an extension technology acceptance model; constructing a time and price combination scene, and calculating a goodness-of-fit index; then, constructing a potential category Logit model, and checking a fitting degree; and finally, solving and analyzing in combination with a mixed influence factor model to realize accurate prediction of the eVTOL acceptance. According to the method, the acceptability of different traveler groups to the unmanned manned aircraft can be accurately predicted, so that data support is provided for traffic management and policy making, and the method is particularly suitable for selection behavior research of medium and short distance business trips.

Description

Technical Field

[0001] The present invention relates to the field of decision-making for urban air traffic systems, and particularly to a prediction method for eVTOL travel choices based on a latent class model. Background Art

[0002] In recent years, with the development of behavior choice theory, the research on travel mode choice behavior, especially for short- and medium-distance business travel, has gradually expanded from traditional observable variables to latent variables at the psychological level to better explain the deviation between the actual choice behavior of travelers and rational decision-making. In addition, the latent class model (LCM) has been widely used to capture the latent preferences of different groups, explain group heterogeneity, and play an important role in market segmentation and policy-making.

[0003] Although the current related technologies have made great progress, there is little research on short- and medium-distance business travel within a region, especially the lack of in-depth discussion on the choice behavior of public transportation modes such as unmanned manned aircraft. Secondly, traditional research mostly relies on linear models, which are difficult to handle complex non-linear data and lack a comprehensive integration of latent variables, resulting in limited model explanatory power and prediction accuracy. In addition, the application of the latent class model in travel mode choice mostly focuses on static analysis and lacks a comprehensive consideration of dynamic influencing factors. Summary of the Invention

[0004] The present invention provides a prediction method for eVTOL travel choices based on a latent class model, aiming to optimize the decision-making analysis of the urban air mobility (UAM) system. This method combines the extended technology acceptance model (TAM), comprehensively considers psychological latent variables such as perceived usefulness, perceived trust, social influence, and behavioral intention, and accurately predicts the acceptance of unmanned manned aircraft by different traveler groups by constructing a mixed choice model, providing data support for traffic management and policy-making, and is particularly applicable to the research on the choice behavior of short- and medium-distance business travel.

[0005] The present invention provides a prediction method for eVTOL travel choices based on a latent class model, including the following steps:

[0006] Step S1: Analyze the characteristics of travelers based on the technology acceptance model and conduct reliability and validity tests;

[0007] Step S2: Construct 16 combined scenarios based on time and price, and calculate the goodness-of-fit index;

[0008] Step S3: Construct a latent class Logit model, and calculate BIC and CALC to test the goodness of fit; the BIC index is the Bayesian information criterion, and the CALC index is the consistent Akaike information criterion, both of which are important indicators for evaluating the latent class Logit model.

[0009] Step S4: eVTOL acceptance analysis is performed by combining mixed-factor CFA model solution, traveler behavior analysis, and dynamic analysis. The CFA model is a confirmatory factor analysis model used to deeply analyze potential variables related to eVTOL travel willingness and acceptance.

[0010] The eVTOL is an electric vertical take-off and landing vehicle.

[0011] Furthermore, in step S1, the technology acceptance model includes four latent variables, namely perceived usefulness, perceived trust, social influence and behavioral intention. Each latent variable is represented by three observed variables. The question of each observed variable is set with five options of different degrees of recognition. The setting of the options is based on the Likert five-level scale method, with the selection range from "strongly disagree" to "strongly agree", and the assignment range corresponds to 1 to 5.

[0012] Furthermore, the reliability and validity tests performed in step S1 specifically include performing Cronbach's α coefficient test, KMO sample measurement and Bartlett's sphericity test on the recovered sample data to ensure that the reliability and validity of the questionnaire meet the requirements.

[0013] Furthermore, in step S2, the basic personal characteristics of the respondents are investigated, including gender, age, education level, occupation and monthly income. In addition, travel time and travel cost are selected as characteristic attributes, and the respondents choose between two travel modes: unmanned manned aircraft and traditional cars. A total of 16 scenarios are set up, and each respondent chooses options from 4 scenarios.

[0014] Furthermore, the construction of the latent class conditional logit model and the calculation of the BIC and CALC indicators in step S3 are specifically as follows:

[0015] Assume that there are N (n = 1, ..., N) respondents taking a survey, facing T (t = 1, ..., T) different choice scenarios, each with J (j = 1, ..., J) travel options to choose from. Also, assume that the N individuals can be divided into C (c = 1, ..., C) potential categories, each with a different parameter, i.e., β (c = 1, ..., C); if respondent n belongs to category c, then the probability of customer n choosing is P n (β c )for:

[0016]

[0017] Where c represents the coefficient of each category; y njt is the choice value, which is a binary variable, that is, when individual n chooses option j in scenario t, ynjt = 1, otherwise y njt = 0; x njt indicates that the explanatory variable varies with both individuals and the scenarios in different cases;

[0018] The classification of individual n is uncertain. Therefore, before calculation, the log-likelihood values of the overall sample under several classification cases need to be obtained first, and then the BIC value and CAIC value of these two index values are calculated using the log-likelihood value. Finally, the best classification case is determined according to the magnitude of the index values;

[0019] To obtain the log-likelihood value of the overall sample, the unconditional likelihood value of individual n under several classification cases needs to be obtained first. It is equal to the weighted average of the selection probabilities of each category obtained in the above formula. The specific formula is:

[0020]

[0021] where π cn (θ) is the weight of classification c, that is, the share of the number of individuals in this category in the overall sample quantity. The calculation formula is:

[0022]

[0023] where, θ = (θ1, θ2, …, θ c-1 ) is the classification quantity model parameter, and by default θ c = 0; the constant term z n is the coefficient that only varies with individuals;

[0024] Next, sum the log-unconditional likelihood values of each individual to obtain the log-likelihood value of the overall sample. The specific formula is:

[0025]

[0026] where, the parameters β and θ are obtained by the expectation maximization algorithm. The specific formula is:

[0027]

[0028] where, the superscript s or s + 1 represents the s-th or s + 1-th iteration estimate; η cn (β s , θ s ) is the posterior probability that individual n belongs to classification c in the s-th iteration estimate. The calculation formula is:

[0029]

[0030] Finally, calculate the BIC and CAIC index values for each classification case. The specific formula is:

[0031]

[0032] Among them, L is the maximum value of the overall logarithmic likelihood of the sample; m is the total number of parameters in the model. The smaller the BIC and CAIC index values, the higher the model fitting degree. When there are conflicts in the comparison of the index values, the comparison of the BIC index value shall prevail.

[0033] Furthermore, after using the latent class model to perform fitting analysis on the data, additional calculations can be performed on the latent class conditional Logit, which specifically includes the following four items:

[0034] Predict the unconditional choice probability of an individual, which is equal to the average value of the weighted category choice probabilities of the corresponding classification shares. The specific formula is:

[0035]

[0036] Predict the conditional choice probability of an individual in a specific category c;

[0037] Predict the category share and prior probability U of an individual in a specific category c p , and the calculation method corresponds to the category share predicted by the classification model parameters;

[0038] Predict the posterior probability C of an individual in a specific category c p , while considering the individual's choice sequence.

[0039] Furthermore, in the step S4, the specific content of the mixed influencing factor confirmatory factor analysis model solution, traveler behavior analysis, and dynamic analysis is as follows:

[0040] First, use the mixed influencing factor confirmatory factor analysis model to solve the latent variables, calculate the goodness-of-fit index of the latent variable model, and represent attitudes such as perceived usefulness, perceived trust, social influence, and behavioral intention that cannot be directly measured through exogenous measurement variables and endogenous manifest variables. The main goodness-of-fit indices for measuring the latent variable model include root mean square error of approximation, comparative fit index, Tucker-Lewis index, and root mean square of standardized residuals;

[0041] Determine the final sample category number through the CAIC and BIC values, record the influence of traveler characteristics on the category, and thus analyze the characteristics of various travelers;

[0042] Calculate the SP attribute, that is, the editing effect of the stated preference attribute, through the influence of time and price changes on the group preference, and at the same time combine population and socio-economic analysis to obtain more realistic analysis results.

[0043] Compared with the prior art, the present invention adopts the above technical solutions and has the following beneficial effects:

[0044] (1) Precise Market Segmentation and Decision-Making: By constructing a latent class conditional Logit model and a mixed factor CFA model, different groups of travelers can be accurately identified, such as groups with a preference for unmanned aerial vehicles for passenger transportation and neutral transportation mode preference groups. Based on the segmentation results, traffic management departments can formulate differentiated policies for different groups. For example, increase the promotion efforts for preference groups and strengthen the publicity and guidance for neutral groups, improving the scientificity and pertinence of decision-making and promoting the rational allocation of urban air traffic resources.

[0045] (2) Efficiently Alleviate Urban Traffic Congestion: Deeply analyze travelers' acceptance of eVTOL, and clarify the preference changes of different groups under time and price fluctuations. For short and medium-distance travel demands such as business trips, guide more travelers to choose eVTOL, effectively diverting ground traffic pressure, alleviating urban congestion, improving the operating efficiency of urban traffic, and providing strong support for the sustainable development of the city.

[0046] (3) Promote the Development of Urban Air Traffic Industry: Accurately predict travelers' willingness to accept eVTOL, providing a reliable basis for the planning and construction of urban air traffic systems. Attract more investment into this field, promote the research and development of eVTOL technology, infrastructure construction, and the improvement of related supporting services, accelerating the development process of the UAM industry and enhancing the level of urban modern transportation. Description of the Drawings

[0047] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is a step block diagram of an eVTOL travel choice prediction method based on a latent class model provided by an embodiment of the present invention;

[0049] Figure 2 It is a detailed step flow chart of an eVTOL travel choice prediction method based on a latent class model provided by an embodiment of the present invention;

[0050] Figure 3 It is an expanded schematic diagram of an extended TAM2 technology acceptance model provided by an embodiment of the present invention;

[0051] Figure 4 It is a graph of the acceptance change of all interviewees provided by an embodiment of the present invention;

[0052] Figure 5The acceptance change graph of the highly educated group provided by the embodiments of the present invention;

[0053] Figure 6 The acceptance change graph of the male group provided by the embodiments of the present invention. Detailed implementation manners

[0054] Next, certain details of the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] The above content of the present invention will be further described in detail below in the form of embodiments, but it should not be understood that the scope of the above subject matter of the present invention is limited to the following embodiments. All technologies implemented based on the above content of the present invention fall within the scope of the present invention.

[0056] The present invention provides an eVTOL travel choice prediction method based on a latent class model, aiming to optimize the decision-making analysis of the urban air mobility (UAM) system. This method combines the extended technology acceptance model (TAM), comprehensively considers psychological latent variables such as perceived usefulness, perceived trust, social influence, and behavioral intention, and accurately predicts the acceptance of unmanned passenger aircraft by different traveler groups by constructing a mixed choice model, providing data support for traffic management and policy making, and is particularly applicable to the research on the choice behavior of medium and short-distance business trips.

[0057] The present invention provides an eVTOL travel choice prediction method based on a latent class model, and the flowchart is as Figure 1 shown, including the following steps:

[0058] Step S1: Analyze the characteristics of travelers based on the technology acceptance model (TAM), and conduct reliability and validity tests.

[0059] Step 1.1: Based on the extended technology acceptance model framework as Figure 2 shown, consider 4 latent variables of perceived usefulness, perceived trust, social influence, and behavioral intention, as well as the influence and role of socio-economic attributes such as gender, age, and income on the selection behavior of manned eVTOL.

[0060] Step 1.2: The Technology Acceptance Model (TAM) includes four latent variables, namely Perceived Usefulness (PU), Perceived Trust (PT), Subjective Norms (SN), and Behavioral Intention (BI). Each latent variable is characterized by three observed variables. For each observed variable, five options with different degrees of agreement are set in the question, and the options are set according to the Likert five-point scale method. The selection range is from "strongly disagree" to "strongly agree", and the corresponding value range is 1 - 5. The framework of the latent variables and the corresponding manifest variables is shown in the following table:

[0061] Table 1 Manifest Variables Table Representing Psychological Latent Variables

[0062]

[0063]

[0064] Step S1.3: For the reliability and validity tests, first conduct Cronbach's Alpha, KMO sample measure, and Bartlett spherical test on the collected sample data to ensure that the reliability and validity of the questionnaire meet the requirements. The KMO value of each latent variable in the questionnaire sample is greater than 0.7, the factor loadings are greater than 0.8, and Cronbach's Alpha is greater than 0.8, indicating that the reliability and validity of the questionnaire meet the requirements. The specific reliability and validity tests are shown in the following table:

[0065] Table 2 Reliability and Validity Test Results Table of Sample Data

[0066]

[0067] Step S2: Construct 16 combined scenarios based on Time (minutes) and Price (yuan), and calculate the goodness-of-fit index.

[0068] The personal basic characteristic information of the respondents, including gender, age, education level, occupation, and monthly income. In addition, two factors such as Travel time (tt) and Travel cost (tc) are selected as characteristic attributes, and the respondents make choices between two travel modes: unmanned aerial vehicle for passenger transportation and traditional car. A total of 16 scenarios are set, and each respondent makes choices for the options in 4 scenarios. The specific statistical results are shown in the following table:

[0069] Table 3 Sample Descriptive Statistics Table

[0070]

[0071]

[0072] Step S3: Construct a latent class Logit model and calculate the BIC and CALC metrics to test the goodness of fit.

[0073] Step S3.1: The specific steps for constructing the latent class conditional Logit model are as follows:

[0074] Suppose there are N (n = 1, …, N) respondents being surveyed, facing T (t = 1, …, T) different choice scenarios, and each scenario has J (j = 1, …, J) travel options available. At the same time, assume that the N individuals can be divided into C (c = 1, …, C) latent classes, and the parameters of each class are different, namely β (c = 1, …, C). If respondent n belongs to class c, then the choice probability P n (β c ) is:

[0075]

[0076] where c represents the coefficient of each class; y njt is the choice value, which is a binary variable, that is, when individual n selects option j in scenario t, y njt = 1, otherwise y njt = 0; x njt represents the explanatory variable that varies with both the individual and the option in different scenarios.

[0077] The classification of individual n is uncertain. Therefore, before calculation, the log-likelihood value of the overall sample under several classification situations needs to be obtained first, and then the BIC value (Bayesian Information Criterion) and CAIC value (Consistent Akaike Information Criterion) of these two metrics are calculated using this log-likelihood value. Finally, the best classification situation is determined according to the magnitudes of the metric values.

[0078] Step S3.2: The specific content of calculating the BIC and CALC metrics is as follows:

[0079] To obtain the log-likelihood value of the overall sample, the unconditional likelihood value of individual n under several classification situations needs to be obtained first. It is equal to the weighted average of the choice probabilities obtained for each class in the above formula. The specific formula is:

[0080]

[0081] where π cn (θ) is the weight of class c, that is, the share of the number of individuals in this class in the total sample size. The calculation formula is:

[0082]

[0083] Among them, θ = (θ1, θ2, …, θ c-1 ) is the parameter of the classification quantity model, and by default θ c = 0; the constant term z n is a coefficient that only varies with individuals.

[0084] By summing the log-unconditional likelihood values of each individual, the log-likelihood value of the overall sample can be obtained. The specific formula is:

[0085]

[0086] Among them, the parameters β and θ are obtained by the EM algorithm (Expectation-Maximization Algorithm). The specific formula is:

[0087]

[0088] Among them, the superscript s or s + 1 represents the s-th or (s + 1)-th iteration estimate; η cn (β s , θ s ) is the posterior probability that individual n belongs to class c in the s-th iteration estimate. The calculation formula is:

[0089]

[0090] Finally, calculate the BIC and CAIC index values for each classification situation. The specific formula is:

[0091]

[0092] Among them, L is the maximum value of the log-likelihood of the overall sample; m is the total number of parameters in the model. The smaller the BIC and CAIC index values, the higher the model fitting degree. When there are conflicts in the comparison of the index values, the comparison situation of the BIC index value shall prevail.

[0093] Step S3.3: After using LCM to perform fitting analysis on the data, additional calculations can also be performed for the latent class conditional Logit, which specifically includes the following four items:

[0094] Under the selected scenario, the unconditional choice probability of an individual can be predicted, which is equal to the average value of the weighted class choice probabilities of the corresponding classification shares. The specific formula is:

[0095]

[0096] Predict the conditional selection probability of an individual in a specific category c; predict the category share of an individual in a specific category c and the prior probability U p , the calculation method corresponds to the classification share predicted by the classification model parameters; predict the posterior probability C of an individual in a specific category c p , while considering the individual's selection sequence.

[0097] Step S3.4: Actual calculation of model CAIC and BIC:

[0098] For the latent class conditional logit model, it is necessary to determine the number of categories C in the sample population. Assuming that the sample can be divided into 2 to 6 subsets, the CAIC and BIC values are calculated for each case. The results are shown in the following table:

[0099] Table 4 Comparison of model CAIC and BIC

[0100] Category CAIC BIC 2 397619.975960 397467.589653 3 356065.921299 355837.438392 4 345953.848755 345649.333993 5 353097.625869 352717.144283 6 350068.101669 349611.718574

[0101] The latent class conditional logit model was used to determine the optimal number of classes for the sample population. The CAIC and BIC values were calculated for classes ranging from 2 to 6, as shown in Table 4. The optimal BIC was 345953.848755, corresponding to 4 latent classes, and the optimal CAIC was 345649.333993, corresponding to 4 latent classes. Therefore, the sample can ultimately be divided into 4 classes.

[0102] Based on the above CAIC and BIC values, travelers are divided into four categories. The following analyzes the characteristics of each type of travelers:

[0103] Class 1: This group has a significant positive sensitivity to perceived usefulness, indicating a strong focus on the practical functionality and value of UAMs. Gender, education, and income significantly influence their choices, with high-income, highly educated men being the majority. However, occupation negatively influences their behavior, potentially conflicting with their desire for stable employment. This group highly appreciates technological innovation and new technologies, is willing to try new things, and exhibits a high risk tolerance. They also prefer fast and efficient commuting options, particularly during peak hours or for medium- and long-distance trips.

[0104] Class 2: This group is highly sensitive to behavioral intention and perceived usefulness, but negatively responds to perceived trust (PT) and social influence. This suggests their choices are driven more by intrinsic motivations than external social influences. This group has strong behavioral intentions but may be averse to excessive social pressure. They are neutral toward various modes of transportation, with a relatively scattered preference and lacking a clear preference. They maintain a wait-and-see attitude toward new technologies such as unmanned passenger vehicles (eVTOLs), not actively trying them but also expressing aversion to them.

[0105] Class 3: This group of people shows strong sensitivity to perceived usefulness, while their behavioral intention has marginal significance, indicating that gender, education, and income significantly affect their choices during the observation and trial stages), and occupation and age have no significant effects. They value practical benefits and have a relatively high acceptance of the new technological functions of unmanned aerial vehicles for passenger transportation, but they do not fully and firmly support it.

[0106] Class 4: This group of people has a positive sensitivity to perceived trust and perceived usefulness, indicating that trust building is crucial for their acceptance of UAM, but their behavioral intention (BI) is not significant. Age has marginal significance, and occupation has a negative impact, indicating that this group still has safety concerns about unmanned aerial vehicles for passenger transportation, they have more trust in traditional services, habitually rely on existing transportation modes, have a low acceptance of new things, and are willing to pay a premium for familiar services.

[0107] In summary, define Class 1 as the group with a preference for unmanned aerial vehicles for passenger transportation, Class 2 as the group with a neutral preference for transportation modes, Class 3 as the group with a preference for trial transportation modes, and Class 4 as the group with a preference for traditional taxis.

[0108] Step S4: Conduct eVTOL acceptance analysis by combining the solution of the mixed influencing factor CFA model, traveler behavior analysis, and dynamic analysis

[0109] Step S4.1: Use the mixed influencing factor CFA model to solve for latent variables, calculate the goodness-of-fit indices of the latent variable model, and represent attitudes such as perceived usefulness, perceived trust, social influence, and behavioral intention that cannot be directly measured through exogenous measured variables and endogenous manifest variables. The main goodness-of-fit indices for measuring the latent variable model are RMSEA, CFI, TLI, and SRMR. The CFA model has a relatively high goodness of fit and basically meets the requirements. The model fit indices and CFA results are shown in the following table:

[0110] Table 5 Results of Model Test Indices

[0111]

[0112]

[0113] Note: ***, **, * represent the significance levels of 1%, 5%, and 10% respectively

[0114] Table 6 CFA Results Table

[0115]

[0116] Step S4.2: Calculate the editing effect of the SP attribute based on the impact of time and price changes on group preferences, and combine demographic and socioeconomic analyses to obtain more realistic analysis results.

[0117] Through questionnaire analysis, when the time increases by 10%, the preferences of Class 1, Class 2, and Class 3 all increase, by 6.5%, 3.5%, and 4.3% respectively, indicating that these groups have a certain acceptance of time extension. In particular, the preference of Class 1 for technology makes it the least sensitive to time changes, while the preference of Class 4 decreases slightly (-1.6%), showing its relatively high adaptability to time changes. However, when the price increases by 10%, the preferences of all groups decline. In particular, Class 4 (-5.5%) is the most sensitive to price, and Class 2 and Class 3 also decline by 4.3% and 3.3% respectively, while Class 1 is the least affected by price changes, only increasing slightly by 0.5%. The acceptance changes of all respondents are as Figure 4 shown, and the marginal effect table of the SP attribute is as follows:

[0118] Table 7 Marginal Effect Table of SP Attribute

[0119] Class1 Class2 Class3 Class4 The time increases by 10% 6.5％ 3.5％ 4.3％ -1.6％ The price increases by 10% 0.5％ -4.3％ -3.3％ -5.5％

[0120] Through demographic and socioeconomic analyses, men and those with higher education (above undergraduate level) have a higher likelihood. However, women have a lower likelihood, and this finding is consistent with previous studies. As mentioned above, the impact of personality traits and situational characteristics on the acceptance of unmanned aerial vehicles for manned use can be altered by personal characteristics such as gender and education level. This indicates the need for targeted strategies to increase acceptance. Figure 5 、 Figure 6 respectively describe the tendency changes in the acceptance of unmanned aerial vehicles for manned use among the highly educated population and the tendency changes among men.

[0121] And it can be clearly found that among male respondents, minor changes in price and time do not change the acceptance of this group for unmanned aerial vehicles for manned use. This indicates that this group has a low price sensitivity and is more inclined to choose travel modes based on high value or long-term benefits (such as time savings and efficiency improvement) rather than focusing on short-term economic cost changes. It may also have a firm attitude towards this new type of transportation, unmanned aerial vehicles. The analysis results are as Figure 5 shown.

[0122] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A prediction method for eVTOL travel choices based on a latent class model, characterized in that It includes the following steps: Step S1: Analyze traveler characteristics based on the Technology Acceptance Model and conduct reliability and validity tests. Step S2: Construct 16 combined scenarios based on time and price, and calculate the goodness-of-fit index. Step S3: Construct a latent class Logit model, and calculate BIC and CALC to test the goodness of fit; the BIC index is the Bayesian Information Criterion, and the CALC index is the Consistent Akaike Information Criterion, both of which are important indicators for evaluating the latent class Logit model. Step S4: Conduct eVTOL acceptance analysis by combining the solution of the mixed influence factor CFA model, traveler behavior analysis, and dynamic analysis; the CFA model is the Confirmatory Factor Analysis model, which is used to deeply analyze the latent variables related to the acceptance of eVTOL travel willingness. The eVTOL is an electric vertical takeoff and landing aircraft.

2. The eVTOL travel choice prediction method based on the latent class model according to claim 1, characterized in that, In the step S1, the Technology Acceptance Model includes four latent variables, namely perceived usefulness, perceived trust, social influence, and behavioral intention. Each latent variable is characterized by three observed variables, and each observed variable is set with five options of different degrees of agreement. The options are set according to the Likert five-point scale method, and the selection range is from "strongly disagree" to "strongly agree", and the corresponding value range is 1-5.

3. A method for predicting eVTOL travel choices based on a latent class model according to claim 1, characterized in that, The reliability and validity tests conducted in the step S1 are specifically to conduct Cronbach's α coefficient test, KMO sample measure, and Bartlett spherical test on the recovered sample data to ensure that the reliability and validity of the questionnaire meet the requirements.

4. A method for predicting eVTOL travel choices based on a latent class model according to claim 1, characterized in that, In the step S2, the personal basic characteristic information of the respondents is investigated, including gender, age, education level, occupation, and monthly income; in addition, two factors of travel time and travel cost are selected as characteristic attributes, and the respondents choose between two travel modes of unmanned aerial vehicle for manned flight and traditional car; a total of 16 scenarios are set, and each respondent makes a choice for the options in 4 scenarios.

5. A method for predicting eVTOL travel choices based on a latent class model according to claim 1, characterized in that, The specific content of the construction of the latent class conditional Logit model and the calculation of BIC and CALC indicators in the step S3 is as follows: Suppose there are N (n = 1, …, N) interviewees being surveyed, facing T (t = 1, …, T) different choice scenarios, and there are J (j = 1, …, J) travel options available for each scenario. At the same time, assume that the N individuals can be divided into C (c = 1, …, C) potential categories, and the parameters of each category are different, namely β (c = 1, …, C); if interviewee n belongs to category c, then the choice probability P n (β c ) is as follows: where c represents the coefficient for each category; y njt is the choice value, which is a binary variable, that is, when individual n selects alternative j in scenario t, y njt = 1, otherwise y njt = 0; x njt is the explanatory variable, which is an observed attribute that varies with individuals, choice scenarios, or alternative options in the latent class conditional Logit model and is used to analyze its impact on the individual's choice behavior y njt ; the explanatory variable x njt varies both with individuals and with the alternatives in different scenarios; The classification of individual n is uncertain. Therefore, before calculation, the log-likelihood value of the overall sample under several classification situations needs to be obtained first, and then these two index values of BIC value and CAIC value are calculated using this log-likelihood value. Finally, the best classification situation is determined according to the size of the index values. To obtain the log-likelihood value of the overall sample, the unconditional likelihood value of individual n under several classification situations needs to be obtained first, which is equal to the weighted average of the selection probabilities of each category obtained in the above formula. The specific formula is: where π cn (θ) is the weight of class c, that is, the share of the number of individuals in this class in the overall number of samples, and the calculation formula is: Among them, θ = (θ1, θ2, …, θ c-1 ) is the classification quantity model parameter, and by default θ c = 0; the constant term z n is a coefficient that only varies with individuals; Then, sum the log-unconditional likelihood values of each individual to obtain the log-likelihood value of the overall sample. The specific formula is: Among them, the parameters β and θ are obtained by the Expectation-Maximization algorithm. The specific formula is: where the superscript s or s + 1 represents the s-th or (s + 1)-th iteration estimate respectively; η cn (β s , θ s ) is the posterior probability that individual n belongs to class c in the s-th iteration estimate, and the calculation formula is: Finally, calculate the BIC and CAIC index values of each classification situation. The specific formula is: Among them, L is the maximum value of the overall sample log-likelihood; m is the total number of parameters in the model; the smaller the BIC and CAIC index values, the higher the model goodness of fit; when there is a conflict in the comparison of the index values, the comparison of the BIC index value shall prevail.

6. The method according to claim 5, characterized in that, After fitting and analyzing the data using the latent class conditional Logit model, additional calculations can be performed on the latent class conditional Logit, specifically including the following four items: Predict the unconditional choice probability of an individual, which is equal to the average of the weighted class choice probabilities of the corresponding classification shares. The specific formula is: Predict the conditional choice probability of an individual in a specific class c; Predicting the class share of an individual in a specific class c and the prior probability U p , where the calculation method corresponds to the classification share predicted using the classification model parameters; Predict the posterior probability C of an individual in a specific class c p , while taking into account the individual's selection sequence.

7. The method according to claim 1, characterized in that, In step S4, the specific content of the mixed influencing factor confirmatory factor analysis model solution, traveler behavior analysis, and dynamic analysis is as follows: First, use the mixed influencing factor confirmatory factor analysis model to solve the latent variables, calculate the goodness-of-fit index of the latent variable model, and represent attitudes such as perceived usefulness, perceived trust, social influence, and behavioral intention that cannot be directly measured through exogenous measurement variables and endogenous manifest variables; The main goodness-of-fit indicators for measuring the latent variable model are the root mean square error of approximation, comparative fit index, Tucker-Lewis index, and standardized root mean square residual; Determine the final number of sample classes through the CAIC and BIC values, record the influence of traveler characteristics on the classes, and thus analyze the characteristics of various types of travelers; Calculate the SP attributes through the influence of time and price changes on group preferences, that is, the editing effect of the stated preference attributes, and at the same time combine demographic and socioeconomic analyses to obtain more realistic analysis results.

Images