Method and system for collaborative optimization of text travel industry and social capital

Abstract

The invention relates to the technical field of document travel industry and social capital collaborative optimization, in particular to a document travel industry and social capital collaborative optimization method and system, and the method specifically comprises the steps: calculating the multi-dimensional interaction strength of document travel data and social capital data; using a cross-domain causal network to calculate the cross-domain causal strength of the reason variable to the result variable; determining posterior probability distribution of anti-fact result variables by using a Bayesian anti-fact intervention model; generating a plurality of policies for guiding document travel resource allocation and policy making based on an ethical constraint reinforcement learning model LARL; and finding an optimal strategy from a plurality of strategies for guiding document travel resource configuration and policy making by using quantum annealing. According to the method, the defects of a traditional method in the aspects of multi-scale coupling recognition, sudden passenger flow early warning, ethical risk control, privacy protection and the like are overcome, and effective technical support is provided for intelligent and sustainable development of the travel industry.

Description

Technical Field

[0001] The present invention relates to the technical field of collaborative optimization of cultural tourism industry and social capital, and in particular to a method and system for collaborative optimization of cultural tourism industry and social capital. Background Art

[0002] The "Cultural Tourism Industry" is short for the "Cultural Tourism Industry," a crucial component of the tourism industry. Chinese academia has a generalized understanding of the "Cultural Tourism Industry," with many considering the tourism industry as part of the cultural industry, including tourism transportation companies, tourist accommodations, and purely natural scenic areas.

[0003] With the vigorous development of the tourism industry and the continuous advancement of information technology, the contemporary cultural and tourism industry has also ushered in a wave of transformation towards digitalization, experience, and integration. That is, the diversification of tourist needs and the complexity of market dynamics have put forward higher requirements for the formulation of cultural and tourism policies. Although current cultural and tourism policy formulation has gradually introduced technologies such as big data and artificial intelligence, a series of flaws have been exposed in actual implementation. First, cultural and tourism data and social capital data differ in spatiotemporal scale and modality. Traditional regression models have difficulty processing high-dimensional, sparse, and nonlinear spatiotemporal data, resulting in poor fusion effects. Second, they often rely on expert experience to formulate fixed rules, such as the "investment amount-tourist volume" threshold matching, which cannot dynamically adapt to the nonlinear interaction between the cultural and tourism market and social capital, resulting in rigid analytical models. Third, existing optimization models often focus on single objectives, such as maximizing short-term returns or minimizing investment risks, ignoring the balance between the social benefits of cultural and tourism, such as ecological protection and cultural heritage, and the long-term appreciation of capital. Moreover, the closed-loop cycle from data collection to policy iteration is too long, resulting in low optimization efficiency. Fourth, privacy conflicts are prominent, and the boundaries between tourist behavior data collection and anonymization processing are blurred, which not only restricts the mining of data value but also easily triggers a crisis of public trust. Fifth, automatically generated cultural and tourism policies may ignore the carrying capacity of cultural heritage protection units, leading to overdevelopment and violation of cultural heritage protection regulations, thereby posing ethical risks.

[0004] Therefore, there is an urgent need for a method and system for deep collaborative optimization of the cultural tourism industry and social capital to solve the above technical problems. Summary of the Invention

[0005] In response to the deficiencies in the above-mentioned prior art, the present invention provides a method and system for collaborative optimization of the cultural tourism industry and social capital to solve the problems existing in the above-mentioned background technology.

[0006] The present invention provides a method for collaborative optimization of the cultural tourism industry and social capital, comprising: S1. Collect cultural travel data and social capital data; S2. Preprocessing the collected data, including data cleaning, data conversion and data normalization; S3. Based on the pre-processed data, calculate the multi-dimensional interaction intensity between cultural travel data and social capital data, where the multi-dimensional interaction intensity includes the interaction intensity at the micro level and the interaction intensity at the macro level; S4. Filter out causal variables, outcome variables, and confounding variables from the preprocessed data. Based on the multi-dimensional interaction strength between the cultural travel behavior data and the social capital data calculated in step S3, use a cross-domain causal network to calculate the cross-domain causal strength of the causal variables on the outcome variables, where confounding variables refer to variables that affect both the causal variables and the outcome variables. S5. Determine the intervention variable and the specific value of the intervention variable, and use the Bayesian counterfactual intervention model to determine the posterior probability distribution of the counterfactual outcome variable based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable, where the intervention variable is obtained by screening the cause variable; S6. Determine the optimization objectives and decision variables, and generate several strategies to guide cultural tourism resource allocation and policy formulation based on the ethical constraint reinforcement learning model LARL. The posterior probability distribution of the counterfactual outcome variable is used to design the reward function of the ethical constraint reinforcement learning model. S7. Based on the optimization goal determined in step S6 and the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies, quantum annealing is used to find the optimal strategy from the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies.

[0007] Preferably, the cultural travel data includes tourist data and scenic spot data, wherein the tourist data includes tourist number data, tourist GPS trajectory data, mobile device data, accommodation data, transportation data, payment data, shopping data, social media data and voice interview data; the scenic spot data includes scenic spot monitoring images, scenic spot ticket sales data and scenic spot evaluation data; Social capital data includes community data, policy data, and environmental data. Community data includes community population data, community relationship data, community organization and public service data, and community survey data. Policy data includes government-issued tourism development plans, community development policies, and cultural heritage protection regulations. Environmental data includes air quality, water quality, and noise levels. In addition, Laplace noise is used to protect the privacy of travel data. k - Anonymization and l -Diversity combination technology protects the privacy of social capital data, where Laplace noise satisfies ( ) - Laplace noise for differential privacy, with a noise scale of Δf / ϵ , Laplace noise is expressed as follows: ; Where, represents the data after applying the differential privacy mechanism, Represents the original data, Lap represents sampling from the Laplace distribution, represents global sensitivity; represents the privacy budget.

[0008] Preferably, before preprocessing, scenic area monitoring images and text data need to use a pre-trained deep learning model for feature extraction, where the scenic area monitoring images use a pre-trained ResNet model for feature extraction; and the text data use a pre-trained BERT model for feature extraction.

[0009] Preferably, data cleaning includes removing duplicate data, processing erroneous data and missing data; data conversion is to map cultural travel behavior data and social capital data into hyperbolic space through the Poincaré sphere model and convert them into a unified format, specifically: the Poincaré sphere model maps discrete events into hyperbolic space vectors , continuous space-time data is mapped into hyperbolic space vectors , its distance calculation satisfies: ;Wherein, the curvature of the Poincaré sphere is -1; In hyperbolic space, the point and the distance between them; represents the mapping of discrete events into hyperbolic space vectors, Indicates that continuous spatiotemporal data is mapped into hyperbolic space vectors; is the number of dimensions, express dimensional hyperbolic space, express dimensional hyperbolic space; Data normalization uses cross-modal contrastive learning to align the spatiotemporal and semantic features of the data and scale the data to the same range. The cross-modal contrastive learning loss function is shown below: ;; Where, Indicates the i text vectors and j The similarity between image vectors; N Indicates the number of samples, used as a normalization factor; Indicates the i text vectors; Indicates the j image vectors; ∈[0.2,0.4], is an adjustable temperature parameter used to control the sensitivity of contrastive learning; i, j, k Indicates the index of the sample.

[0010] Preferably, in step S3, the interaction intensity between travel behavior data and social capital data at the micro level is calculated based on the improved dynamic time warping algorithm, wherein the improved dynamic time warping algorithm is: introducing adaptive weights on the basis of the existing dynamic time warping algorithm It is used to suppress long-range alignment errors and improve the accuracy of time series matching. Its expression is as follows: ; Where, represents the optimized dynamic time warping distance; Represents two time series whose similarity needs to be calculated, u is a subset of time series U, and v is a subset of time series V; Represents all possible alignment paths; represents the adaptive weight in dynamic time warping; Representing time series The data points; Representing time series The data points; The interaction intensity between travel behavior data and social capital data at the macro level is calculated using a two-layer graph attention network. The macro-level interaction intensity indicator is the topological overlap rate. The expression for calculating the topological overlap rate using a two-layer graph attention network is as follows: Where, Indicates the Nodes in the layer For Node The attention weights between represents the weight vector in the attention mechanism, Represents the weight vector The transpose of Indicates the The weight matrix of the layer, Representation node The eigenvector of Representation node The eigenvector of Representation node The eigenvector of express activation function, Representation node neighbor nodes.

[0011] Preferably, the cross-domain causal network function is as follows: ; Where, Represents the cross-domain causal strength from A to B, A represents the cause variable, B represents the result variable, Indicates the total time length, Represents the time index, Indicates the cause variable In time The value of Represents the weighted average of multi-dimensional causality; Z Represents a confounding variable, i.e., an influence and variables; For conditional mutual information, filter the confounding variable Z; Indicates the expected value, Indicates intervention, Indicates that the cause variable is artificially The value of is set to , Indicates that the cause variable is artificially The value of is set to 0; Represents the intervention cause variable A, and sets its value to Afterwards, the expected value of the outcome variable B; It represents the expected value of the outcome variable B after the intervention variable A is set to 0; is the weight coefficient; The Bayesian counterfactual intervention model is as follows: Where, represents the counterfactual outcome variable, represents the intervention variable, Represents the intervention variable Specific value of , D represents the historical intervention data of the intervention variable; Indicates that after observing historical intervention data D After that, intervention When the counterfactual outcome variable The posterior probability of Indicates that given the intervention variable and parameters Under the condition that probability; represents the model parameters, ; Indicates that given historical intervention data D Under the condition of The probability of; P represents the probability distribution, represents the differential, express A very small change.

[0012] Preferably, the steps for constructing the ethical constraint reinforcement learning model LARL are as follows: S61. Select a reinforcement learning algorithm; S62. Taking cultural heritage protection regulations as hard constraints, define an action shielding function to filter out actions that violate cultural heritage protection regulations in the state space; S63. Taking the posterior probability distribution of social moral norms and counterfactual outcome variables as soft constraints, design a reward function. The reward function expression is as follows: ; Where, Indicates that the status Take action Then, transfer to the state Rewards received when For economic benefit rewards, it means in the state Take action Transfer to state After that, the increase in tourism revenue; is the user satisfaction reward, expressed in the state Take action Transfer to state After that, the increment of tourist satisfaction score; is the ethical soft constraint penalty function, which means that in state Take action If the ethical soft constraint is violated, the function will penalize according to the median of the results of the counterfactual outcome variable; represents the median of the probability distribution of the Bayesian counterfactual intervention outcome variable, that is, Less than or equal to The probability is 50%; is an action shielding function, which is used to constrain cultural heritage protection. If it causes damage to cultural heritage, the function returns -∞, otherwise it returns 0; represents the weight coefficient; S64. Optimize the weights of the reward function using the experience of domain experts and the probability distribution of counterfactual outcome variables , guide the initialization and learning process of the strategy, and generate the initial ethical constraint reinforcement learning model; S65. Based on historical data, the initial ethical constraint reinforcement learning model is trained by reinforcement learning until the strategy converges or the maximum number of iterations is reached, and the ethical constraint reinforcement learning model is obtained. The historical data includes historical cultural travel behavior data and historical social capital data. The strategy optimization objective function of the ethical constraint reinforcement learning model LARL is shown as follows: Where, represents the objective function of the ethical constraint reinforcement learning model, s Indicates status, Indicates strategy, Indicates that in the strategy Induced state distribution Under expectations, Represents the discriminator output state The confidence level, represents the prior strategy distribution based on cultural heritage protection regulations; represents the adjustable constraint strength, , used to balance the relationship between strategy exploration and ethical constraints, ensuring that the generated strategy complies with social ethical constraints; KL express KL Divergence, a measure of the difference between two probability distributions.

[0013] Preferably, the Hamiltonian design in quantum annealing contains an illegal unconstraint term, which is expressed as: Where, represents the total Hamiltonian, represents the income weight coefficient, represents the state of the i-th spin, represents the state of the jth spin; Represents the constraint strength coefficient, which is used to adjust the penalty for illegal solutions; Indicates the illegal solution set The state of the spin, Indicates the illegal solution set The state of the spin, represents the set of illegal solutions; In addition, when the environment suddenly changes, several strategies with good performance are selected from the historical strategies as the initial solution, and the chaotic particle swarm optimization algorithm (CPSO) is used for optimization to obtain the best strategy and apply it to the sudden change of the environment. The particle velocity update formula of the chaotic particle swarm optimization algorithm (CPSO) is: ; , where Indicates the The particle in The speed of the iteration, represents the inertia weight, Indicates the The particle in The speed of the iteration, represents the acceleration factor, represents a random number, Indicates the The individual optimal solution found by particles so far, Indicates the The particle in t The position at the iteration, represents the global optimal solution, Indicates parameters, express Mapping the injected chaotic disturbance, Represents the variable used for chaotic perturbation.

[0014] The present invention also provides a collaborative optimization system for the cultural tourism industry and social capital, which is used to implement a collaborative optimization method for the cultural tourism industry and social capital, including: Data collection module, used to collect cultural travel behavior data and social capital data; The data processing module receives and preprocesses the data collected by the data acquisition module and transmits it to the coupling analysis module and the causal reasoning module. The preprocessing includes data cleaning, data conversion and data normalization. The coupling analysis module calculates the multi-dimensional interaction strength between cultural travel behavior data and social capital data based on pre-processed data and transmits it to the causal reasoning module. The multi-dimensional interaction strength includes the interaction strength at the micro level and the interaction strength at the macro level. The causal reasoning module includes a cross-domain causal submodule and a Bayesian counterfactual intervention submodule, where: The cross-domain causal submodule screens out causal variables, outcome variables, and confounding variables from the pre-processed data. Based on the multi-dimensional interaction strength between cultural travel behavior data and social capital data calculated in the coupling analysis module, the cross-domain causal network is used to calculate the cross-domain causal strength of the causal variable on the outcome variable. Confounding variables refer to variables that affect both the causal variable and the outcome variable. The Bayesian counterfactual intervention submodule determines the intervention variable and the specific value of the intervention variable. Based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable calculated by the cross-domain causal submodule, the Bayesian counterfactual intervention model is used to determine the posterior probability distribution of the counterfactual outcome variable, where the intervention variable is obtained by screening the cause variable. The strategy generation module determines the optimization objectives and decision variables, generates several strategies to guide the allocation of cultural and tourism resources and policy formulation based on the ethical constraint reinforcement learning model LARL, and then uses quantum annealing to find the optimal strategy from several strategies to guide the allocation of cultural and tourism resources and policy formulation. The posterior probability distribution of the counterfactual outcome variables calculated by the Bayesian counterfactual intervention submodule is used to design the reward function of the ethical constraint reinforcement learning model.

[0015] Preferably, it also includes a privacy protection module and an environment mutation module, wherein: The privacy protection module is used to protect the privacy of the collected travel behavior data and social capital data. Laplace noise is used to protect the privacy of travel behavior data. k - Anonymization and l -Diversity combination technology protects the privacy of social capital data; The environmental mutation module is used to select several well-performing strategies from historical strategies as initial solutions when the environment mutates. It uses the chaotic particle swarm optimization algorithm (CPSO) to optimize the strategy, thereby obtaining the best strategy and applying it to the environmental mutation.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention addresses the defect that existing technologies are difficult to effectively integrate cultural and tourism data and social capital data, resulting in poor fusion effect. By mapping cultural and tourism behavior data and social capital data into hyperbolic space through the Poincaré sphere model, it not only solves the problem that heterogeneous data are difficult to directly fuse, but also effectively improves the accuracy of subsequent analysis. In addition, the present invention also aligns the spatiotemporal and semantic features of the data through cross-modal comparative learning, scales the scenic area monitoring images and text data to the same range, and greatly improves the alignment accuracy compared with traditional methods, effectively ensuring the quality and effect of data fusion.

[0017] 2. The present invention addresses the rigidity of the cultural and tourism policy-making model. Existing technologies mostly rely on expert experience to formulate fixed rules and are unable to dynamically adapt to the nonlinear interaction between the cultural and tourism market and social capital, resulting in the defect of rigid analysis models. A cross-domain causal network is constructed, and the cross-domain causal strength of the cause variable to the result variable is calculated based on the multi-dimensional interaction intensity. This can effectively capture the nonlinear relationship between cultural and tourism behavior data and social capital data, and not only achieves a multi-scale coupling recognition F1-score of 0.93, which is much higher than the traditional method of 0.71, but the model can also dynamically adapt to changes in the cultural and tourism market and social capital, further improving the flexibility and adaptability of cultural and tourism policy-making.

[0018] 3. In response to the defects of existing optimization models that focus on a single goal, ignore the balance between the social benefits of cultural tourism and the long-term appreciation of capital, and have too long a closed-loop cycle and low optimization efficiency, the present invention introduces the ethical constraint reinforcement learning model LARL, which combines social moral norms and the posterior probability distribution of counterfactual outcome variables to design a reward function, thereby guiding the allocation of cultural tourism resources and policy formulation. While achieving economic benefits, it takes into account social equity and cultural heritage protection, greatly reduces the policy ethics violation rate, and significantly reduces the ethical risks that may arise in the implementation of cultural tourism policies. At the same time, the present invention also adopts a quantum annealing algorithm to accelerate the search process for the optimal strategy. Moreover, when the environment mutates, the present invention can also quickly search for new strategies through the chaotic particle swarm optimization algorithm CPSO, thereby obtaining the best strategy for application when the environment mutates, effectively solving the problem of low optimization efficiency.

[0019] 4. This invention aims to solve the problem of fuzzy boundary between the existing technology in the collection and anonymization of tourist behavior data, which not only restricts the mining of data value but also easily leads to a crisis of public trust. Laplace noise is used to protect the privacy of tourist behavior data. k - Anonymization and l -Diversity combination technology protects the privacy of social capital data, achieving a data reconstruction attack success rate of less than 5%. Compared with systems without privacy protection, the data reconstruction attack success rate is reduced from more than 80% to less than 5%, effectively balancing the relationship between data value mining and privacy protection, and enhancing the public's trust in the use of data in the cultural and tourism industry.

[0020] 5. To address the problem of high delay in sudden passenger flow warning in the existing technology, the present invention adopts a dynamic time warping algorithm to achieve a sudden passenger flow warning delay of <200ms, which is much smaller than the warning delay of >2s in the traditional method. It can achieve faster warning so that timely measures can be taken to deal with emergencies and ensure the safety and experience of tourists. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 This is a flow chart of a method for collaborative optimization of the cultural tourism industry and social capital in an embodiment of the present invention; Figure 2 This is a framework diagram of a collaborative optimization system between the cultural tourism industry and social capital in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the present invention provides a method for collaborative optimization of the cultural tourism industry and social capital, comprising the following steps: S1. Collect cultural travel data and social capital data.

[0025] In this application, travel data includes tourist data and scenic spot data, where tourist data includes but is not limited to tourist quantity data, tourist GPS trajectory data, mobile device data, accommodation data, traffic data, payment data, shopping data, social media data and voice interview data; scenic spot data includes but is not limited to scenic spot monitoring images, scenic spot ticket sales data and scenic spot evaluation data.

[0026] It should be noted that the voice interview data are tourists’ verbal descriptions of their travel experiences, which are used to understand tourists’ subjective feelings and service evaluations.

[0027] In this application, social capital data includes community data, policy data and environmental data, among which community data includes community population data, community relationship data, community organization and public service data and community survey data.

[0028] In the embodiment of the present application, community population data includes but is not limited to age, gender, education level, occupation, income level, employment status, place of residence, and length of residence; community relationship data includes but is not limited to the frequency and type of community members' participation in community activities, the degree of trust and willingness to cooperate among community members; community organization and public service data includes but is not limited to the number, type, and scale of community organizations, the number, quality, and distribution of community public facilities, the coverage, service quality, usage, and satisfaction of community public services; community survey data includes the opinions, suggestions, and needs of community residents collected through questionnaires and interviews.

[0029] In the embodiment of the present application, policy data includes but is not limited to tourism development plans, community development policies and cultural heritage protection regulations issued by the government.

[0030] In the embodiment of the present application, environmental data includes but is not limited to air quality, water quality and noise level.

[0031] In this application, Laplace noise is used to protect the privacy of travel data.k - Anonymization and l -Diversity combination technology protects the privacy of social capital data.

[0032] In the embodiment of this application, k =5, l =2.

[0033] In the embodiment of the present application, Laplace noise satisfies ( ) - Laplace noise for differential privacy, with a noise scale of Δf / ϵ , Laplace noise is expressed as follows: Where, represents the data after applying the differential privacy mechanism, Represents the original data, Lap represents sampling from the Laplace distribution, represents global sensitivity; represents the privacy budget.

[0034] It should be noted that tourist data is directly related to an individual's spatiotemporal behavior and contains extremely sensitive personal information. Compared with other types of cultural and tourism data, such as scenic spot introductions and travel guides, tourist data is more easily used to track, identify, and locate individuals. Once this information is leaked, it may pose serious privacy risks to individuals and violate laws and regulations. Therefore, a higher level of protection is required. Differential privacy technology is a very powerful privacy protection technology. Therefore, in this application, Laplace noise that satisfies differential privacy is added to cultural and tourism data including tourist data to effectively prevent data leakage. ( )-Differential privacy, which indicates the differential privacy protection scheme adopted in this application. Its privacy budget is 1.0, which is a relatively reasonable privacy protection parameter setting that can ensure both good privacy protection effect and data availability.

[0035] S2. Preprocess the collected data, which includes data cleaning, data conversion and data normalization.

[0036] Preferably, before preprocessing, scenic area monitoring images and text data need to use a pre-trained deep learning model for feature extraction, where the scenic area monitoring images use a pre-trained ResNet model for feature extraction; and the text data use a pre-trained BERT model for feature extraction.

[0037] In the embodiment of the present application, the text data includes text data converted from policy data and voice interview data.

[0038] It should be emphasized that before using the voice interview data in the embodiments of the present application, it must be converted into text data.

[0039] ResNet-50 is a 50-layer deep convolutional neural network that effectively extracts visual features from images. This application uses a ResNet-50 model pre-trained on the ImageNet dataset to extract visual features from scenic area surveillance images, and uses the output of the last convolutional layer as the feature representation of scenic area surveillance images.

[0040] BERT-Base is a BERT model with 12 layers of Transformer blocks that effectively captures the semantic information of text. This application uses a BERT-Base model pre-trained on the BooksCorpus and English Wikipedia datasets to capture the semantic information of text data and uses its [CLS] vector as the feature representation of the text data.

[0041] Because ResNet and BERT are already established models, they are not specifically described in this application. For specific methods, please refer to the references. For example, for ResNet-50, see He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep Residual Learning for Image Recognition. Proceedings of the IEEE conference on computer vision and pattern recognition, 770-778; for BERT-Base, see Devlin, J., Chang, MW, Lee, K., & Toutanova, K. (2018). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv preprint arXiv:1810.04805.

[0042] In the embodiment of the present application, data cleaning includes removing duplicate data, processing erroneous data and missing data; data conversion is converting data of different formats into a unified format; data normalization is scaling data of different scales to the same range to avoid certain data from having too great an impact on the analysis results.

[0043] In the embodiment of the present application, step S2 of data cleaning is specifically as follows: (1) Deduplication: A data deduplication method based on a hash algorithm is used to scan travel behavior data and social capital data, identify records with the same hash value, and delete duplicates. This method uses the SHA-256 algorithm to calculate the hash value to ensure the accuracy of deduplication. For text data, the SimHash algorithm is used to calculate text similarity. Texts with a similarity of more than 95% are considered duplicate data and deleted.

[0044] (2) Error data processing: For data whose data types do not meet the requirements, such as incorrect date formats or values ​​outside the reasonable range, data type conversion or outlier processing is performed. For GPS trajectory data, the DBSCAN clustering algorithm is used to identify abnormal trajectory points and correct or delete them. The parameters of the DBSCAN algorithm are adjusted according to the characteristics of the dataset, such as the neighborhood radius epsilon and the minimum number of neighborhood points minPts.

[0045] (3) Missing data processing: For missing data, different filling strategies are adopted according to the missing ratio and data type. For numerical data with a low missing ratio, such as less than 5%, the mean or median is used for filling. For numerical data with a high missing ratio, the K-nearest neighbor algorithm (KNN) is used for filling. The K value of the KNN algorithm is adjusted according to the characteristics of the data set, and a value between 3 and 10 is usually selected. For categorical data, the mode is used for filling. For time series data, linear interpolation or seasonal decomposition interpolation (STL) is used for filling.

[0046] Note: The SHA-256 algorithm can be found in the National Institute of Standards and Technology (NIST). (2015). Secure Hash Standard (SHS). FIPS PUB 180-4, available for download at https: / / doi.org / 10.6028 / NIST.FIPS.180-4. The SimHash algorithm can be found in Charikar M S. Similarity estimation techniques from rounding algorithms[J]. Applied and Computational Harmonic Analysis, 2002: 380-388. The DBSCAN clustering algorithm can be found in Martin Ester, Hans-Peter Kriegel, Jörg Sander, and Xiaowei Xu. 1996. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD'96). AAAI Press, 226–231. K-nearest neighbor algorithm can be found in the literature: Thomas M Cover, Peter E Hart. Nearest neighbor pattern classification. IEEE Transactions on Information Theory, 1967. Linear interpolation or seasonal decomposition interpolation STL can be found in the literature: Cleveland, RB, Cleveland, WS, McRae, JE, & Terpenning, IJ (1990). STL: A seasonal-trend decomposition procedure based on loess. Journal of Official Statistics, 6(1), 3–33.

[0047] Preferably, the data conversion in step S2 is specifically as follows: mapping the cultural travel behavior data and social capital data to the hyperbolic space through the Poincaré sphere model, and converting them into a unified format; the normalization processing is specifically as follows: using cross-modal contrastive learning to align the spatiotemporal and semantic features of the data to scale the data to the same range, wherein the normalization processing is only for scenic area monitoring images and text data.

[0048] In this application, the Poincaré sphere model maps discrete events into hyperbolic space vectors , continuous space-time data is mapped into hyperbolic space vectors , its distance calculation satisfies: ;Wherein, the curvature of the Poincaré sphere is -1; In hyperbolic space, the point and the distance between them; represents the mapping of discrete events into hyperbolic space vectors, Indicates that continuous spatiotemporal data is mapped into hyperbolic space vectors; is the number of dimensions, express dimensional hyperbolic space, express dimensional hyperbolic space.

[0049] It should be noted that hyperbolic space has the characteristic of "exponential growth" and is more suitable for representing the hierarchical structure of social networks. The Poincaré sphere model is a representation of hyperbolic space that can better preserve the hierarchical relationship of data, reduce distance distortion, and convert data from different sources and types into a unified form suitable for subsequent processing. In this application, the data was mapped to hyperbolic space using the Poincaré sphere model. Compared with the case in Euclidean space, the distance error was reduced by 32%, effectively improving the accuracy of subsequent data analysis and utilization.

[0050] In this application, the cross-modal contrastive learning loss function is as follows: Where, Indicates the i text vectors and j The similarity between image vectors; N Indicates the number of samples, used as a normalization factor; Indicates the i text vectors; Indicates the j image vectors; ∈[0.2, 0.4], is an adjustable temperature parameter used to control the sensitivity of contrastive learning; i, j, k Indicates the index of the sample.

[0051] It should be noted that the highest accuracy of the traditional method is 72.3%, while the cross-modal alignment accuracy in this application is improved to 89.7%.

[0052] Cross-modal contrastive learning is a machine learning technique that normalizes shared representations across different modalities, such as text and images. For the same scene, the goal of cross-modal contrastive learning is to make the semantic representations of these modal data as close as possible after model processing. For example, it can align the semantics of a tourist's surveillance image and text data from a scenic spot.

[0053] S3. Based on the preprocessed data, calculate the multi-dimensional interaction intensity between cultural travel data and social capital data, where the multi-dimensional interaction intensity includes the interaction intensity at the micro level and the interaction intensity at the macro level.

[0054] In this application, based on the improved dynamic time warping algorithm, the interaction intensity between cultural travel behavior data and social capital data at the micro level is calculated. The improved dynamic time warping algorithm is: based on the existing dynamic time warping algorithm, the adaptive weight is introduced. It is used to suppress long-range alignment errors and improve the accuracy of time series matching. Its expression is as follows: Where, represents the optimized dynamic time warping distance; Represents two time series whose similarity needs to be calculated, u is a subset of time series U, and v is a subset of time series V; Represents all possible alignment paths; represents the adaptive weight in dynamic time warping; Representing time series The data points; Representing time series The data points.

[0055] It should be noted that the existing dynamic time warping (DTW) algorithm requires that the points of the two time series be matched one by one. If the two time series have a large time difference or a large time offset, then "long-range alignment" is required, that is, matching the points that are farther away. However, long-range alignment may introduce errors because the correlation between points that are farther away may be weaker. Therefore, this application introduces adaptive weighting. To solve the problem that long-range alignment may introduce errors in the DTW algorithm. In this application, the weight between point p and point q is The distance in time from them The farther the point is, the smaller its weight is; the closer the point is, the larger its weight is.

[0056] The smaller the distance, the stronger the association. This application adjusts weights based on social capital data. For example, if a community has high trust, the weight associated with visitor activity tracks is increased; if a community has an event, the weight associated with visitor volume and store traffic during that time period is increased.

[0057] The micro level focuses on individual or small-scale interactive behaviors, emphasizing details and precision. It focuses on analyzing specific, observable individual behaviors and how these behaviors directly affect another system. Specific content and data include: 1. Individual tourist behavior: such as the specific movement trajectory of tourists in the scenic area (GPS data), which stores they made purchases in, and which attractions they stayed at for a longer time. 2. Individual business operations: such as the customer flow, sales, and profit changes of a single store. 3. Specific community activities: such as cultural festivals and tourism promotion activities organized by a community organization. The data type is generally high-frequency, detailed time series data, such as: statistics on the number of tourists per minute, store sales records per hour, real-time traffic at specific locations (such as the entrance to the scenic area, near stores), etc.

[0058] In this application, a two-layer graph attention network is used to calculate the interaction intensity between cultural travel behavior data and social capital data at the macro level. The macro-level interaction intensity indicator is the topological overlap rate. The expression for calculating the topological overlap rate using the two-layer graph attention network is as follows: Where, Indicates the Nodes in the layer For Node The attention weights between represents the weight vector in the attention mechanism, Represents the weight vector The transpose of Indicates the The weight matrix of the layer, Representation node The eigenvector of Representation node The eigenvector of Representation node The eigenvector of express activation function, Representation node neighbor nodes.

[0059] In this application, the macro-level interaction intensity between the cultural tourism industry and social capital specifically refers to the overall degree of connectivity between the cultural tourism industry chain and the social capital network. For example, which tourism enterprises collaborate most closely with community organizations, and which government agencies contribute most to community development.

[0060] It is important to note that the macro-level focuses on overall or large-scale interactions, emphasizing globality and structure. Therefore, in this application, the macro-level interaction intensity between the cultural tourism industry and social capital focuses on analyzing the overall structure of the cultural tourism industry chain and the social capital network, as well as the relationship between them.

[0061] In the embodiment of the present application, the specific steps of using a two-layer graph attention network to calculate the interaction intensity between cultural travel behavior data and social capital data at the macro level are as follows: S31. Build a cultural tourism industry chain network and a social capital network. The cultural tourism industry chain network has tourism enterprises, suppliers, and distributors as nodes, and cooperative relationships and investment relationships as edges. Tourism enterprises include travel agencies, hotels, and scenic spots. The social capital network has community organizations, residents, and government agencies as nodes, and social relationships, trust relationships, and cooperative relationships as edges.

[0062] The data type is generally static data that describes the network structure, such as: equity relationship data between tourism companies, cooperation agreement data between community organizations, and government investment data in the cultural and tourism industry.

[0063] S32. Extract features for each node from the cultural tourism industry chain network / social capital network.

[0064] For example, enterprise scale, profitability, service quality, etc. are extracted from the cultural and tourism industry chain network; community population, economic level, organizational activity, etc. are extracted from the social capital network.

[0065] S33 and two-layer GAT learn the node representations of the cultural and tourism industry chain network and the social capital network respectively. At the same time, each node updates its own representation based on the characteristics and attention weights of its neighboring nodes.

[0066] S34. After obtaining the node representations of the two networks, the topological overlap rate between them is calculated to measure the overall connectivity of the two networks.

[0067] It should be noted that the higher the topological overlap rate, the closer the connection between the cultural and tourism industry chain network and the social capital network, and the greater the interaction intensity.

[0068] S4. Filter out causal variables, outcome variables, and confounding variables from the preprocessed data. Based on the multi-dimensional interaction intensity between the cultural travel behavior data and the social capital data calculated in step S3, use a cross-domain causal network to calculate the cross-domain causal intensity of the causal variable on the outcome variable, where the confounding variable refers to a variable that affects both the causal variable and the outcome variable.

[0069] In this application, the cross-domain causal network function is shown as follows: ; Where, Represents the cross-domain causal strength from A to B, A represents the cause variable, B represents the result variable, Indicates the total time length, Represents the time index, Indicates the cause variable In time The value of Represents the weighted average of multi-dimensional causality; Z Represents a confounding variable, i.e., an influence and variables; For conditional mutual information, filter the confounding variable Z; Indicates the expected value, Indicates intervention, Indicates that the cause variable is artificially The value of is set to , It means that the value of the cause variable A is artificially set to 0; Represents the intervention cause variable A, and sets its value to Afterwards, the expected value of the outcome variable B; It represents the expected value of the outcome variable B after the intervention variable A is set to 0; is the weight coefficient.

[0070] In this application, the specific steps for constructing a cross-domain causal network are: (1) Determine nodes: Determine the nodes in the network, each node represents a variable. These variables include factors related to the cultural tourism industry and factors related to social capital.

[0071] For example, the cultural and tourism industry: number of tourists, tourism revenue, scenic spot ticket prices, hotel occupancy rates, investment amounts of tourism enterprises, etc.; social capital: community residents’ income, employment rate, education level, trust, community organization activity, etc.

[0072] Determine edges: Determine the causal relationship between nodes, represented by directed edges. That is, if A is the cause of B, then establish a directed edge from A to B.

[0073] In the embodiment of the present application, the causal relationship between expert knowledge and data-driven nodes is combined, specifically: first, in-depth interviews are conducted with field experts to sort out the potential causal relationship between the cultural tourism industry and social capital, and construct a preliminary causal relationship network; second, historical data is used to adopt a causal discovery algorithm to learn the causal relationship between variables from historical data, thereby verifying and correcting the initial causal relationship network; finally, the expert knowledge and data-driven results are comprehensively analyzed to solve the problem of inconsistency between expert knowledge and data-driven results. If the expert knowledge and data-driven results are consistent, they are directly adopted; if the expert knowledge and data-driven results are inconsistent, it is necessary to further analyze the reasons, such as data quality issues, expert knowledge bias, etc., and make adjustments.

[0074] The causal discovery algorithm used in the present embodiment is the Directed Cycle Discovery (CCD) algorithm. By allowing loops in the graph, the CCD algorithm can more accurately reflect the complex relationship between the cultural tourism industry and social capital. It automatically discovers potential causal relationships from the data.

[0075] It should be noted that the CCD algorithm for directed cyclic discovery can be found in the literature Richardson, TS (1996). A discovery algorithm for directed cyclic graphs. Sixth International Workshop on Artificial Intelligence and Statistics, 347-364.

[0076] It should be emphasized that when combining expert knowledge and the causal relationship between data-driven nodes, the causal relationship judgment criteria must be followed, including: ① temporal order, that is, the cause must occur before the result; ② statistical dependence, that is, there must be a statistical correlation between the cause and the result; ③ intervention effect, that is, by intervening in the cause, a significant impact on the result can be observed.

[0077] In the embodiment of the present application, the time sequence of the cause and the result is determined by the timestamp or the order in which the events occur.

[0078] (3) Connect the nodes related to the cultural tourism industry with the nodes related to social capital to form a cross-domain causal network.

[0079] S5. Determine the intervention variable and the specific value of the intervention variable, and use the Bayesian counterfactual intervention model to determine the posterior probability distribution of the counterfactual outcome variable based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable, where the intervention variable is obtained by screening the cause variable.

[0080] It should be noted that since the factors affecting the development of the cultural tourism industry vary from region to region, when using this method, researchers need to screen out intervention variables that are consistent with the development of the cultural tourism industry in their region from the causal variables determined in step S4 based on the actual situation of the region. In addition, the specific value of the intervention variable refers to the actual value of the intervention variable, which can be single or multiple.

[0081] In this application, the Bayesian counterfactual intervention model is obtained by combining the Bayesian network with counterfactual reasoning, wherein the Bayesian counterfactual intervention model is shown as follows: Where, represents the counterfactual outcome variable, represents the intervention variable, Represents the intervention variable Specific value of , D represents the historical intervention data of the intervention variable; Indicates that after observing historical intervention data D After that, intervention When the counterfactual outcome variable The posterior probability of Indicates that given the intervention variable and parameters Under the condition that probability; represents the model parameters, ; Indicates that given historical intervention data D Under the condition of The probability of; P represents the probability distribution, represents the differential, express A very small change.

[0082] In this application, combined with historical intervention data D , which reduced the mean square error (MSE) of policy effect prediction by 47%, effectively improving the prediction accuracy.

[0083] For example, the selected intervention variable is ticket price, and the specific value of ticket price is 120. The ticket price, the specific value of ticket price, and the historical intervention data of ticket price are input into the Bayesian counterfactual intervention model. According to the output results of the Bayesian counterfactual intervention model, it can be known that when the specific value of ticket price is 120, what is the probability of income increasing, what is the probability of income remaining unchanged, and what is the probability of income decreasing.

[0084] S6. Determine the optimization objectives and decision variables, and generate several strategies to guide the allocation of cultural and tourism resources and policy formulation based on the ethical constraint reinforcement learning model LARL. The posterior probability distribution of the counterfactual outcome variable is used to design the reward function of the ethical constraint reinforcement learning model.

[0085] The optimization objective can be a single objective, such as maximizing tourism revenue; or it can be multi-objective, such as maximizing tourism revenue, improving residents' quality of life and protecting cultural heritage at the same time.

[0086] In this application, the optimization goal is to maximize total tourism revenue and improve tourist satisfaction under the premise of protecting cultural heritage and safeguarding community interests.

[0087] In this application, the steps for constructing the ethical constraint reinforcement learning model LARL are as follows: S61. Select reinforcement learning algorithm This application adopts the proximal policy optimization PPO algorithm as the core algorithm of ethical constraint reinforcement learning.

[0088] PPO is an Actor-Critic method and a type of policy gradient algorithm. This algorithm ensures the stability of training by limiting the magnitude of each policy update to avoid excessive policy updates that lead to performance degradation. There are two main variants of the PPO algorithm: PPO-Clip and PPO-Adaptive KL. The PPO-Clip algorithm is used in this application. The PPO algorithm uses the Actor-Critic architecture to update the policy network by maximizing the clipped surrogate objective function, and uses the mean square error loss function to update the value network. The PPO algorithm has the advantages of stable training and insensitivity to hyperparameters. It can effectively solve the problems of high-dimensional state space and complex action space in cultural and tourism resource allocation and policy formulation.

[0089] Use PPO algorithm to train policy network and value network , by updating the policy network parameters using the clippedsurrogate objective function To update the policy network, update the value network parameters by minimizing the mean square error To update the value network. Indicates input status , the output is in state Take each action The probability of Represents the parameters of the policy network; Indicates input status , output status The value of , that is, the expected cumulative return, Represents the parameters of the value network.

[0090] The policy network generates policies, while the value network evaluates them and provides guidance. The policy network attempts to select actions that lead to high-value states. The two networks collaborate to learn and optimize policies.

[0091] The PPO algorithm is a mature reinforcement learning algorithm, so its principles and process are not detailed in this application. For detailed principles, formulas, and experimental results, please refer to: Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal Policy Optimization Algorithms. arXiv preprint arXiv:1707.06347. S62. Taking cultural heritage protection regulations as hard constraints, define an action shielding function to filter out actions that violate cultural heritage protection regulations in the state space; S63. Taking the posterior probability distribution of social moral norms and counterfactual outcome variables as soft constraints, design a reward function. The reward function expression is as follows: ; Where, Indicates that the status Take action Then, transfer to the state Rewards received when For economic benefit rewards, it means in the state Take action Transfer to state After that, the increase in tourism revenue; is the user satisfaction reward, expressed in the state Take action Transfer to state After that, the increment of tourist satisfaction score; is the ethical soft constraint penalty function, which means that in state Take action If the ethical soft constraint is violated, the function will penalize according to the median of the results of the counterfactual outcome variable; represents the median of the probability distribution of the Bayesian counterfactual intervention outcome variable, that is, Less than or equal to The probability is 50%; is an action shielding function, which is used to constrain cultural heritage protection. If it causes damage to cultural heritage, the function returns -∞, otherwise it returns 0; Represents the weight coefficient.

[0092] In the embodiments of this application, the social ethics in the soft constraints specifically include: ① Community Interest Protection: Ensuring that the development of the cultural tourism industry does not harm the interests of community residents, such as through demolition and environmental pollution. ② Fairness and Justice: Ensuring fairness and justice in the allocation of cultural tourism resources and policy-making processes, avoiding discrimination or unfair treatment. ③ Sustainable Development: Ensuring that the development of the cultural tourism industry complies with the principles of sustainable development, focusing on ecological environmental protection and resource conservation and utilization.

[0093] S64. Optimize the weights of the reward function using the experience of domain experts and the probability distribution of counterfactual outcome variables , guiding the initialization and learning process of the strategy and generating the initial ethical constraint reinforcement learning model.

[0094] It should be noted that in step S64, the domain experts manually optimize the weights of the reward function based on their experience and the probability calculation of the counterfactual outcome variables. .

[0095] S65. Based on historical data, the initial ethical constraint reinforcement learning model is subjected to reinforcement learning training until the strategy converges or the maximum number of iterations is reached, thereby obtaining an ethical constraint reinforcement learning model, wherein the historical data includes historical cultural travel behavior data and historical social capital data.

[0096] In this application, the strategy optimization objective function of the ethical constraint reinforcement learning model LARL is as follows:

[0097] Where, represents the objective function of the ethical constraint reinforcement learning model, s Indicates status, Indicates strategy, Indicates that in the strategy Induced state distribution Under expectations, Represents the discriminator output state The confidence level, represents the prior strategy distribution based on cultural heritage protection regulations; represents the adjustable constraint strength, , used to balance the relationship between strategy exploration and ethical constraints, ensuring that the generated strategy complies with social ethical constraints; KL express KL Divergence, a measure of the difference between two probability distributions.

[0098] It should be noted that Domain experts use rule bases or expert systems to transform the requirements of cultural heritage protection regulations into operational rules or constraints, and generate an initial policy distribution for each state. That is, for actions that violate regulations, the probability is set to 0 or a very low probability; for actions that comply with regulations, domain experts assign an initial probability value based on experience.

[0099] Reinforcement learning is a machine learning method that aims to enable an intelligent agent to learn how to make decisions in an environment to maximize accumulated rewards. Reinforcement learning can solve decision-making problems in dynamic environments. Its core idea is to learn through the interaction between the intelligent agent and the environment, rather than training the model with pre-collected data.

[0100] In the embodiment of the present application, social ethics include but are not limited to fairness, justice, and sustainable development.

[0101] By maximizing this objective function, the LARL model can learn an optimal strategy that achieves economic benefits while also balancing social equity and cultural heritage preservation. This strategy, within given constraints, finds the optimal balance and achieves the coordinated development of multiple objectives.

[0102] S7. Based on the optimization goal determined in step S6 and the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies, quantum annealing is used to find the optimal strategy from the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies.

[0103] In this application, several strategies for guiding cultural and tourism resource allocation and policy formulation are manually encoded into a variable S in quantum annealing, where S represents a spin state, and each spin represents a strategy. For example, S1 represents strategy 1, S2 represents strategy 2, and so on. Quantum annealing finds the optimal spin state, which is the optimal strategy.

[0104] In this application, the Hamiltonian design in quantum annealing contains illegal unconstrained terms, expressed as: Where, represents the total Hamiltonian, represents the income weight coefficient, represents the state of the i-th spin, represents the state of the jth spin; Represents the constraint strength coefficient, which is used to adjust the penalty for illegal solutions; Indicates the illegal solution set The state of the spin, Indicates the illegal solution set The state of the spin, Represents the set of illegal solutions.

[0105] It should be noted that a single-dimensional consideration, such as focusing solely on the number of tourists, may overlook the actual effectiveness of the strategy. Therefore, to more accurately evaluate the strategy, this application introduces micro- and macro-interaction intensity to more comprehensively assess the strategy's combined impact on the cultural tourism industry and social capital.

[0106] In quantum annealing, this application introduces the micro- and macro-interaction intensity of cultural tourism behavior data and social capital data. The first purpose is to improve the effectiveness of the strategy, that is, by more accurately evaluating the strategy, screening out the strategy that can truly promote the coordinated development of the cultural tourism industry and social capital, and avoid blind investment and ineffective policies; the second purpose is to improve the efficiency of resource allocation, that is, to guide the quantum annealing algorithm to search in a more favorable direction, find the optimal strategy more quickly, shorten the decision-making cycle, and improve the efficiency of resource allocation; the third purpose is to better promote the sustainable development of the cultural tourism industry, that is, by comprehensively considering multiple dimensions, the development strategy formulated is also more in line with actual needs.

[0107] In the embodiment of the present application, the constraint strength coefficient , which penalizes solutions that violate constraints. In a 1000-dimensional coupled matrix optimization, the probability of finding the global optimal solution increased to 93%, compared to only 65% ​​for traditional gradient methods.

[0108] Preferably, when the environment suddenly changes, several strategies with good performance are selected from historical strategies as initial solutions, and the chaotic particle swarm optimization algorithm CPSO is used for optimization to obtain the best strategy and apply it to the sudden change of the environment.

[0109] In this application, sudden environmental changes include but are not limited to a surge or a sharp drop in the number of tourists.

[0110] It's important to note that the CPSO used in this application doesn't directly monitor real-world mutations. Instead, it uses manual monitoring indicators to define normal ranges or abnormal situations, such as a sudden surge or drop in visitor numbers, or a slow increase in visitors exceeding the scenic area's carrying capacity. Once a mutation is detected, staff manually select several high-performing strategies from historical data as initial solutions. They then use the chaotic particle swarm optimization (CPSO) algorithm to optimize the solution, rapidly searching for new strategies and ultimately obtaining the optimal strategy for the sudden change.

[0111] The carrying capacity of a scenic area refers to the capacity for tourist activities within a given spatial scope of tourism resources under certain time conditions. This capacity refers to the number of visitors that can be accommodated within the area's tourism resources and spatial scale, while meeting minimum visitor requirements, including physical space and psychological atmosphere, and environmental standards consistent with the protection of tourism resources. Each scenic area determines its maximum carrying capacity based on the method outlined in the National Tourism Administration's "Guidelines for Determining the Maximum Carrying Capacity of Scenic Areas." This data remains unchanged for the long term, unless the scenic area's infrastructure or service facilities undergo major upgrades or changes, in which case the maximum carrying capacity needs to be recalculated.

[0112] It should be noted that since the initial solution in CPSO is selected from historical strategies, and these historical strategies are generated by the ethical constraint reinforcement learning model LARL, which already meets ethical standards, the CPSO optimization process no longer pre-sets constraints that comply with ethical standards. Instead, it only sets the required constraints based on its actual situation, thereby searching for the optimal strategy more efficiently.

[0113] In this application, the particle velocity update formula of the chaotic particle swarm optimization algorithm CPSO is: ; Where, Indicates the The particle in The speed of the iteration, represents the inertia weight, Indicates the The particle in The speed of the iteration, represents the acceleration factor, represents a random number, Indicates the The individual optimal solution found by particles so far, Indicates the The particle in t The position at the iteration, represents the global optimal solution, Indicates parameters, express Mapping the injected chaotic disturbance, Represents the variable used for chaotic perturbation.

[0114] like Figure 2 As shown, the present invention also provides a collaborative optimization system for the cultural tourism industry and social capital, which is used to execute a collaborative optimization method for the cultural tourism industry and social capital, including: Data collection module, used to collect cultural travel behavior data and social capital data; The data processing module receives and preprocesses the data collected by the data acquisition module and transmits it to the coupling analysis module and the causal reasoning module. The preprocessing includes data cleaning, data conversion and data normalization. The coupling analysis module calculates the multi-dimensional interaction strength between cultural travel behavior data and social capital data based on pre-processed data and transmits it to the causal reasoning module. The multi-dimensional interaction strength includes the interaction strength at the micro level and the interaction strength at the macro level. The causal reasoning module includes a cross-domain causal submodule and a Bayesian counterfactual intervention submodule, where: The cross-domain causal submodule screens out causal variables, outcome variables, and confounding variables from the pre-processed data. Based on the multi-dimensional interaction strength between cultural travel behavior data and social capital data calculated in the coupling analysis module, the cross-domain causal network is used to calculate the cross-domain causal strength of the causal variable on the outcome variable. Confounding variables refer to variables that affect both the causal variable and the outcome variable. The Bayesian counterfactual intervention submodule determines the intervention variable and the specific value of the intervention variable. Based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable calculated by the cross-domain causal submodule, the Bayesian counterfactual intervention model is used to determine the posterior probability distribution of the counterfactual outcome variable, where the intervention variable is obtained by screening the cause variable. The strategy generation module determines the optimization objectives and decision variables, generates several strategies to guide the allocation of cultural and tourism resources and policy formulation based on the ethical constraint reinforcement learning model LARL, and then uses quantum annealing to find the optimal strategy from several strategies to guide the allocation of cultural and tourism resources and policy formulation. The posterior probability distribution of the counterfactual outcome variables calculated by the Bayesian counterfactual intervention submodule is used to design the reward function of the ethical constraint reinforcement learning model.

[0115] In this application, the system also includes a privacy protection module for protecting the privacy of the collected travel behavior data and social capital data, wherein Laplace noise is used to protect the privacy of travel behavior data, and k - Anonymization and l -Diversity combination technology protects the privacy of social capital data.

[0116] In this application, the system also includes an environment mutation module, which is used to select several well-performing strategies from historical strategies as initial solutions when the environment mutates, and use the chaotic particle swarm optimization algorithm CPSO for optimization to obtain the best strategy and apply it to the environment mutation.

[0117] The present invention verifies the method and system provided by this application through actual measurement on the home culture and tourism big data platform, and uses indicators such as F1-score, latency, violation rate, and reconstruction attack success rate for evaluation. The evaluation results are as follows: (1) The F1-score of the traditional method for multi-scale coupling recognition is 0.71, while the F1-score of the multi-scale coupling recognition of this method reaches 0.93; (2) The sudden passenger flow warning delay of the traditional method is greater than 2s, while the sudden passenger flow warning delay of this method is less than 200ms; (3) This method reduces the policy ethics violation rate from 12.7% to 0.3%; (4) For systems without privacy protection, the data reconstruction attack success rate is >80%, while the data reconstruction attack success rate of this system is <5%.

[0118] In summary, this application demonstrates significant superiority in multi-scale coupling identification, sudden passenger flow warning, ethical risk control, and privacy protection, providing effective technical support for the intelligent and sustainable development of the cultural and tourism industry.

[0119] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A collaborative optimization method for the cultural tourism industry and social capital, characterized in that: include: S1. Collect cultural travel data and social capital data; S2. Preprocessing the collected data, including data cleaning, data conversion and data normalization; S3. Based on the pre-processed data, calculate the multi-dimensional interaction intensity between cultural travel data and social capital data, where the multi-dimensional interaction intensity includes the interaction intensity at the micro level and the interaction intensity at the macro level; S4. Filter out causal variables, outcome variables, and confounding variables from the preprocessed data. Based on the multi-dimensional interaction strength between the cultural travel behavior data and the social capital data calculated in step S3, use a cross-domain causal network to calculate the cross-domain causal strength of the causal variables on the outcome variables, where confounding variables refer to variables that affect both the causal variables and the outcome variables. S5. Determine the intervention variable and the specific value of the intervention variable, and use the Bayesian counterfactual intervention model to determine the posterior probability distribution of the counterfactual outcome variable based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable, where the intervention variable is obtained by screening the cause variable; S6. Determine the optimization objectives and decision variables, and generate several strategies to guide cultural tourism resource allocation and policy formulation based on the ethical constraint reinforcement learning model LARL. The posterior probability distribution of the counterfactual outcome variable is used to design the reward function of the ethical constraint reinforcement learning model. S7. Based on the optimization goal determined in step S6 and the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies, quantum annealing is used to find the optimal strategy from the several strategies for guiding the allocation of cultural and tourism resources and the formulation of policies.

2. A collaborative optimization method for cultural tourism industry and social capital according to claim 1, characterized in that: Cultural travel data includes tourist data and scenic spot data. Tourist data includes tourist number data, tourist GPS trajectory data, mobile device data, accommodation data, transportation data, payment data, shopping data, social media data, and voice interview data; scenic spot data includes scenic spot monitoring images, scenic spot ticket sales data, and scenic spot evaluation data. Social capital data includes community data, policy data, and environmental data. Community data includes community population data, community relationship data, community organization and public service data, and community survey data. Policy data includes government-issued tourism development plans, community development policies, and cultural heritage protection regulations. Environmental data includes air quality, water quality, and noise levels. In addition, Laplace noise is used to protect the privacy of travel data. k - Anonymization and l -Diversity combination technology protects the privacy of social capital data, where Laplace noise satisfies ( )-Laplace noise of differential privacy, the noise scale is Δf / ϵ , Laplace noise is expressed as follows: , where represents the data after applying the differential privacy mechanism, Represents the original data, Lap represents sampling from the Laplace distribution, represents global sensitivity; represents the privacy budget.

3. A collaborative optimization method for cultural tourism industry and social capital according to claim 2, characterized in that: Before preprocessing, scenic area monitoring images and text data need to be extracted using a pre-trained deep learning model. The pre-trained ResNet model is used for feature extraction of scenic area monitoring images, and the pre-trained BERT model is used for feature extraction of text data.

4. A collaborative optimization method for cultural tourism industry and social capital according to claim 2, characterized in that: Data cleaning includes removing duplicate data, processing erroneous data and missing data; data conversion uses the Poincaré sphere model to map cultural travel behavior data and social capital data to hyperbolic space and convert them into a unified format. Specifically, the Poincaré sphere model maps discrete events to hyperbolic space vectors. , continuous space-time data is mapped into hyperbolic space vectors , its distance calculation satisfies: ;Where the curvature of the Poincaré sphere is -1; In hyperbolic space, the point and the distance between them; Represents the mapping of discrete events to hyperbolic space vectors, Indicates that continuous spatiotemporal data is mapped into hyperbolic space vectors; is the number of dimensions, express dimensional hyperbolic space, express dimensional hyperbolic space; Data normalization uses cross-modal contrastive learning to align the spatiotemporal and semantic features of the data and scale the data to the same range. The cross-modal contrastive learning loss function is shown below: ; Where, Indicates the i text vectors and j The similarity between image vectors; N Indicates the number of samples, used as a normalization factor; Indicates the i text vectors; Indicates the j image vectors; ∈[0.2,0.4], is an adjustable temperature parameter used to control the sensitivity of contrastive learning; i, j, k Indicates the index of the sample.

5. The method for collaborative optimization of cultural tourism industry and social capital according to claim 1, characterized in that: In step S3, the interaction intensity between cultural travel behavior data and social capital data at the micro level is calculated based on the improved dynamic time warping algorithm, wherein the improved dynamic time warping algorithm is: based on the existing dynamic time warping algorithm, the adaptive weight is introduced. It is used to suppress long-range alignment errors and improve the accuracy of time series matching. Its expression is as follows: ; Where, Represents the optimized dynamic time warping distance; U and V represent the two time series whose similarity needs to be calculated, u is a subset of time series U, and v is a subset of time series V; Represents all possible alignment paths; represents the adaptive weight in dynamic time warping; Representing time series The data points; Representing time series The data points; The interaction intensity between travel behavior data and social capital data at the macro level is calculated using a two-layer graph attention network. The macro-level interaction intensity indicator is the topological overlap rate. The expression for calculating the topological overlap rate using a two-layer graph attention network is as follows: ; Where, Indicates the Nodes in the layer For Node The attention weights between represents the weight vector in the attention mechanism, represents the transpose of the weight vector b, Indicates the The weight matrix of the layer, Representation node The eigenvector of Representation node The eigenvector of Representation node The eigenvector of express activation function, Representation node neighbor nodes.

6. A collaborative optimization method for cultural tourism industry and social capital according to claim 1, characterized in that: The cross-domain causal network function is shown as follows: ; ; Where, Represents the cross-domain causal strength from A to B, A represents the cause variable, B represents the result variable, Indicates the total time length, Represents the time index, Indicates that the cause variable A is at time The value of Represents the weighted average of multi-dimensional causality; Z represents a confounding variable, i.e., a variable that affects both A and B; For conditional mutual information, filter the confounding variable Z; Indicates the expected value, Indicates intervention, It means artificially setting the value of the cause variable A to , It means that the value of the cause variable A is artificially set to 0; Represents the intervention cause variable A, and sets its value to Afterwards, the expected value of the outcome variable B; It represents the expected value of the outcome variable B after the intervention variable A is set to 0; is the weight coefficient; The Bayesian counterfactual intervention model is as follows: ; Where, represents the counterfactual outcome variable, represents the intervention variable, represents the specific value of the intervention variable, and D represents the historical intervention data of the intervention variable; Indicates that after observing historical intervention data D After that, intervention When the counterfactual outcome variable The posterior probability of Indicates that given the intervention variable and parameters Under the condition that probability; represents the model parameters, ; Indicates that given historical intervention data D Under the condition of The probability of; P represents the probability distribution, represents the differential, express A very small change.

7. A collaborative optimization method for cultural tourism industry and social capital according to claim 1, characterized in that: The steps for building the ethical constraint reinforcement learning model LARL are as follows: S61. Select a reinforcement learning algorithm; S62. Taking cultural heritage protection regulations as hard constraints, define an action shielding function to filter out actions that violate cultural heritage protection regulations in the state space; S63. Taking the posterior probability distribution of social moral norms and counterfactual outcome variables as soft constraints, design a reward function. The reward function expression is as follows: ; ; Where, Indicates that the status Take action Then, transfer to Rewards obtained when in status; is the economic benefit reward, indicating that s takes action in state Transfer to state After that, the increase in tourism revenue; is the user satisfaction reward, expressed in the state Take action Transfer to state After that, the increment of tourist satisfaction score; is the ethical soft constraint penalty function, expressed as Taking action on a state If the ethical soft constraint is violated, the function will penalize according to the median of the results of the counterfactual outcome variable; represents the median of the probability distribution of the Bayesian counterfactual intervention outcome variable, that is, Less than or equal to The probability is 50%; is an action shielding function, which is used to constrain cultural heritage protection. If it causes damage to cultural heritage, the function returns -∞, otherwise it returns 0; represents the weight coefficient; S64. Optimize the weights of the reward function using the experience of domain experts and the probability distribution of counterfactual outcome variables , guide the initialization and learning process of the strategy, and generate the initial ethical constraint reinforcement learning model; S65. Based on historical data, the initial ethical constraint reinforcement learning model is trained by reinforcement learning until the strategy converges or the maximum number of iterations is reached, and the ethical constraint reinforcement learning model is obtained. The historical data includes historical cultural travel behavior data and historical social capital data. The strategy optimization objective function of the ethical constraint reinforcement learning model LARL is shown as follows: Where, represents the objective function of the ethical constraint reinforcement learning model, Indicates status, Indicates strategy, Indicates that in the strategy Induced state distribution Under expectations, Represents the discriminator output state The confidence level, represents the prior strategy distribution based on cultural heritage protection regulations; represents the adjustable constraint strength, , used to balance the relationship between strategy exploration and ethical constraints, ensuring that the generated strategy complies with social ethical constraints; KL express KL Divergence, a measure of the difference between two probability distributions.

8. The method for collaborative optimization of cultural tourism industry and social capital according to claim 1, characterized in that: The Hamiltonian design in quantum annealing contains illegal unconstrained terms, expressed as: Where, represents the total Hamiltonian, represents the income weight coefficient, represents the state of the i-th spin, represents the state of the jth spin; Represents the constraint strength coefficient, which is used to adjust the penalty for illegal solutions; Indicates the illegal solution set The state of the spin, represents the state of the qth spin in the illegal solution set, represents the set of illegal solutions; In addition, when the environment suddenly changes, several strategies with good performance are selected from the historical strategies as the initial solution, and the chaotic particle swarm optimization algorithm (CPSO) is used for optimization to obtain the best strategy and apply it to the sudden change of the environment. The particle velocity update formula of the chaotic particle swarm optimization algorithm (CPSO) is: ; ; Where, Indicates the The particle in The speed of the iteration, represents the inertia weight, Indicates the The particle in The speed of the iteration, represents the acceleration factor, represents a random number, Indicates the The individual optimal solution found by particles so far, Indicates the The particle in t The position at the iteration, represents the global optimal solution, Indicates parameters, express Mapping injected chaotic disturbance, Represents the variable used for chaotic perturbation.

9. A collaborative optimization system for cultural tourism industry and social capital, used to execute a collaborative optimization method for cultural tourism industry and social capital according to any one of claims 1 to 8, characterized in that: include: Data collection module, used to collect cultural travel behavior data and social capital data; The data processing module receives and preprocesses the data collected by the data acquisition module and transmits it to the coupling analysis module and the causal reasoning module. The preprocessing includes data cleaning, data conversion and data normalization. The coupling analysis module calculates the multi-dimensional interaction strength between cultural travel behavior data and social capital data based on pre-processed data and transmits it to the causal reasoning module. The multi-dimensional interaction strength includes the interaction strength at the micro level and the interaction strength at the macro level. The causal reasoning module includes a cross-domain causal submodule and a Bayesian counterfactual intervention submodule, where: The cross-domain causal submodule screens out causal variables, outcome variables, and confounding variables from the pre-processed data. Based on the multi-dimensional interaction strength between cultural travel behavior data and social capital data calculated in the coupling analysis module, the cross-domain causal network is used to calculate the cross-domain causal strength of the causal variable on the outcome variable. Confounding variables refer to variables that affect both the causal variable and the outcome variable. The Bayesian counterfactual intervention submodule determines the intervention variable and the specific value of the intervention variable. Based on the historical intervention data of the intervention variable and the cross-domain causal strength of the cause variable on the outcome variable calculated by the cross-domain causal submodule, the Bayesian counterfactual intervention model is used to determine the posterior probability distribution of the counterfactual outcome variable, where the intervention variable is obtained by screening the cause variable. The strategy generation module determines the optimization objectives and decision variables, generates several strategies to guide the allocation of cultural and tourism resources and policy formulation based on the ethical constraint reinforcement learning model LARL, and then uses quantum annealing to find the optimal strategy from several strategies to guide the allocation of cultural and tourism resources and policy formulation. The posterior probability distribution of the counterfactual outcome variables calculated by the Bayesian counterfactual intervention submodule is used to design the reward function of the ethical constraint reinforcement learning model.

10. A collaborative optimization system for cultural tourism industry and social capital according to claim 9, characterized in that: It also includes a privacy protection module and an environment mutation module, where: The privacy protection module is used to protect the privacy of the collected travel behavior data and social capital data. Laplace noise is used to protect the privacy of travel behavior data. k - Anonymization and l -Diversity combination technology protects the privacy of social capital data; The environmental mutation module is used to select several well-performing strategies from historical strategies as initial solutions when the environment mutates. It uses the chaotic particle swarm optimization algorithm (CPSO) to optimize the strategy, thereby obtaining the best strategy and applying it to the environmental mutation.

Images