A commercial traffic site selection optimization method based on multi-source data fusion

Abstract

The application discloses a kind of commercial traffic site selection optimization method based on multi-source data fusion, it is related to big data processing and multi-source data fusion technical field, including the following steps: obtaining the multi-source data of the region to be measured, constructs multi-source feature, and through adaptive weight distribution mechanism multi-source feature cross-modal fusion, generate comprehensive feature representation;Based on comprehensive feature representation, through causal inference method, output candidate point in different scene passenger flow trend and causal contribution;Passenger flow trend and causal contribution are input, a multi-objective optimization model is established, and is solved, and the comprehensive score and ranking result of candidate point are output;According to comprehensive score and ranking result, combine causal contribution result, construct interpretive index matrix, and output site selection scheme through visual platform;The application is fused spatial geographic data, space-time flow data and consumer behavior data, solve the problem that data source is single in prior art, candidate point is not comprehensive.

Description

Technical Field

[0001] This invention relates to the field of big data processing and multi-source data fusion technology, and more specifically, to a commercial traffic location optimization method based on multi-source data fusion. Background Technology

[0002] In the development of modern urban commerce, the location of commercial outlets has always been a core issue of concern for industries such as retail, catering, and services. A reasonable location not only affects the operating efficiency of individual stores but also directly impacts the overall market expansion strategy and resource allocation efficiency of a company. However, traditional commercial site selection methods mostly rely on single data sources and empirical judgments, such as rough analyses based on population density, the size of the surrounding commercial area, or rental costs. While these methods had some reference value in the early stages of the business environment, their accuracy and foresight have been gradually challenged with the increasing complexity of urban structures and the diversification of consumer behavior, making it difficult to meet the refined decision-making needs of the modern business environment.

[0003] For example, the invention patent with publication number CN119849984B discloses a store location recommendation method integrating multi-dimensional features, including the following steps: S10, collecting multi-dimensional data of the area where candidate store locations are located, the multi-dimensional data including: pedestrian traffic, number of competitors, price, industry market size growth rate, regional planning trend, and consumer demographic change trend; S20, preprocessing the multi-dimensional data, including data cleaning and normalization; S30, inputting the preprocessed multi-dimensional data into the store location recommendation model to obtain a recommendation value for each candidate store location; S40, sorting the recommendation values ​​of each candidate store location in descending order to obtain a store location recommendation sequence. This multi-dimensional data-based store location recommendation fully considers various key factors affecting store operations, providing a more comprehensive depiction of the commercial environment of the candidate area, thus offering more accurate location recommendations and effectively improving the accuracy of location selection.

[0004] For example, the invention patent with publication number CN118278982A discloses a commercial site selection method and apparatus, which relates to the field of artificial intelligence technology. The method includes: acquiring spatiotemporal characteristic information of candidate locations; wherein the spatiotemporal characteristic information includes: time series information, spatial index characteristics at each time point in the time series, and the spatial index characteristics include spatial index characteristics affecting commercial customer flow; using a pre-trained pedestrian flow prediction model, predicting pedestrian flow information of the candidate locations based on the spatiotemporal characteristic information of the candidate locations; wherein the pedestrian flow information of the candidate locations includes pedestrian flow information of the candidate locations at each time point in the time series; the pedestrian flow prediction model is trained based on the spatiotemporal characteristic information and pedestrian flow information of historical locations; and determining suitable locations for commercial site selection based on the pedestrian flow information of each candidate location.

[0005] The above-disclosed technical solutions have at least the following technical problems:

[0006] Traditional commercial traffic location optimization methods often employ single-objective or weighted linear models, prioritizing the maximization of a single metric (such as foot traffic or rental return rate) while neglecting the balance between transportation convenience, consumer potential, and competitive pressure, easily leading to biased location selection results. To address these issues, this invention proposes a solution. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a commercial traffic location optimization method based on multi-source data fusion. By integrating spatial geographic data, spatiotemporal traffic data, and consumer behavior data, this method solves the problems of single data source and incomplete candidate point characterization in the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A commercial traffic location optimization method based on multi-source data fusion includes: acquiring multi-source data of the area to be tested, constructing multi-source features, and fusing the multi-source features across modalities through an adaptive weight allocation mechanism to generate a comprehensive feature representation; based on the comprehensive feature representation, outputting the passenger flow trend and causal contribution of candidate points under different scenarios through a causal inference method; using the passenger flow trend and causal contribution as inputs, establishing a multi-objective optimization model, solving it, and outputting the comprehensive score and ranking results of the candidate points; based on the comprehensive score and ranking results, combined with the causal contribution results, constructing an explanatory index matrix, and outputting the location scheme through a visualization platform.

[0010] In a preferred embodiment, the process of acquiring multi-source data of the area to be tested, constructing multi-source features, and fusing the multi-source features across modalities through an adaptive weight allocation mechanism to generate a comprehensive feature representation is as follows: Multi-source data of the area to be tested is acquired and preprocessed. The multi-source data includes spatial geographic data, spatiotemporal traffic data, and consumer behavior data. Based on the spatiotemporal traffic data, the passenger flow intensity of candidate points at different time periods is calculated, and the time series is smoothed using a moving average method to form dynamic traffic features. Based on road network and public transportation station data in the spatial geographic data, the actual travel time matrix between candidate points and major population gathering areas is calculated, and travel time is converted into an accessibility index to form a transportation accessibility feature. Based on the consumer behavior data, the consumption potential index around the candidate points is calculated, and data of different dimensions are standardized to obtain consumer behavior features. The dynamic traffic features, transportation accessibility features, and consumer behavior features are fused using a weighted method to obtain weighted features, and the weights of each feature are dynamically adjusted based on historical prediction errors. The weighted features are input into a cross-modal embedding model, and different modal features are mapped to a unified vector space using a deep representation learning method to generate a comprehensive feature representation.

[0011] In a preferred embodiment, the actual travel time matrix between candidate points and major population gathering areas is calculated based on road network and public transportation station data from spatial geographic data. This travel time is then converted into an accessibility index, forming a transportation accessibility feature. Specifically, the process involves: acquiring urban road network data and constructing a road network model; collecting subway and bus route data, extracting station geographic locations and timetables, mapping stations to nodes in the road network model, and adding waiting and transfer time parameters to the paths between stations to form a public transportation network model; calculating the self-driving travel time from candidate points to major population gathering areas using the shortest path algorithm based on the road network model, and calculating the public transportation travel time from candidate points to major population gathering areas based on the public transportation network model; weighting and fusing the self-driving and public transportation travel times based on the obtained population travel mode proportions to obtain the comprehensive travel time from candidate points to each population gathering area; constructing a candidate point-population gathering area travel time matrix based on the comprehensive travel time, with the candidate point as the starting point and the major population gathering area as the ending point; and outputting the accessibility index of the candidate point, i.e., the transportation accessibility feature, based on the travel time matrix.

[0012] In a preferred embodiment, the step of calculating the consumption potential index around the candidate point based on consumption behavior data and standardizing data of different dimensions to obtain consumption behavior characteristics is as follows: Obtain census data and publicly available data from the National Bureau of Statistics; extract the number of permanent residents, age structure, family size, and income level within the neighborhood of the candidate point to form a population statistics set for the candidate point's neighborhood; obtain consumption preference data based on user surveys, and group the consumption preference data according to population attribute labels to obtain consumption preference parameter sets for different population categories; match the population statistics set of the candidate point's neighborhood with the consumption preference parameter sets, and match the corresponding consumption propensity coefficient for each type of population in the population data to obtain a weighted consumption vector for the candidate point's neighborhood; output the consumption potential index of the candidate point based on the weighted consumption vector, and normalize the consumption potential index of different candidate points using an interval scaling method to obtain consumption behavior characteristics.

[0013] In a preferred embodiment, the step of outputting the passenger flow trend and causal contribution of candidate points under different scenarios based on comprehensive feature representation and causal inference method is as follows: External event information affecting commercial passenger flow is acquired and transformed into structured data to obtain an event factor dataset; the event factor dataset is hierarchically encoded, establishing codes for sudden events and periodic events respectively, and multi-scale temporal features are constructed based on the occurrence time and duration of the events to form a hierarchical event temporal factor set; the comprehensive feature representation and the hierarchical event temporal factor set are time-aligned and spatially matched to generate a unified input tensor; a multi-task spatiotemporal prediction model containing a main prediction branch and an event influence branch is constructed based on the unified input tensor; based on the multi-task spatiotemporal prediction model, baseline scenario data and event scenario data are input respectively, and the passenger flow prediction results of candidate points under different event scenarios are output; based on the passenger flow prediction results, the causal effect of event factors is estimated using causal inference method to obtain the causal contribution results of passenger flow changes at candidate points; based on the passenger flow prediction results, the confidence interval of the passenger flow prediction results is calculated based on Bayesian approximate inference method to obtain the prediction uncertainty results of candidate points under different event scenarios.

[0014] In a preferred embodiment, the step of using passenger flow trends and causal contributions as inputs to establish a multi-objective optimization model, solving it, and outputting the comprehensive score and ranking results of candidate points is as follows: A multi-objective optimization model is established based on passenger flow prediction results, causal contribution results, and prediction uncertainty results; the location scheme of candidate points is represented as a chromosome coding structure, and an initial population is randomly generated based on the chromosome coding structure of individuals; according to the multi-objective optimization model, the fitness value of each individual in the initial population is output, and individuals with fitness values ​​higher than a preset threshold are selected; the selected individuals exchange candidate point selection genes to generate new offspring individuals, and the fitness values ​​of the offspring individuals are re-output; based on the re-output fitness values, the evolved population is non-dominated and the crowding distance of individuals is calculated; based on the non-dominated ranking and crowding distance, individuals are selected to form a new generation population, which serves as the input for the next round of evolutionary iterations; if the number of iterations reaches a preset threshold or the convergence of the Pareto optimal solution set meets a preset standard, the final Pareto optimal solution set is output.

[0015] In a preferred embodiment, the step of constructing an explanatory index matrix based on the comprehensive score and ranking results, combined with the causal contribution results, and outputting the site selection scheme through a visualization platform, is as follows: An explanatory index matrix is ​​constructed from the comprehensive score, ranking results, and causal contribution results; a factor combination set is constructed for each candidate point in the explanatory index matrix; based on the factor combination set, a factor permutation sequence is generated using Monte Carlo sampling, and a single factor is added sequentially, calculating the marginal improvement value of the factor on the comprehensive score of the candidate point before and after its addition, thus obtaining the marginal contribution result of each factor in each permutation; a weighted average is applied to the marginal contribution results to obtain the Shapley value of the traffic-driven factor, traffic accessibility factor, and competitive pressure factor for each candidate point; the Shapley value results are correlated with the comprehensive score and ranking results of the candidate points to generate an explanatory quantitative result set.

[0016] In a preferred embodiment, the step of associating the Shapley value results with the comprehensive scores and ranking results of candidate points to generate an explanatory quantitative result set is as follows: The marginal contribution results of factors are mapped and associated with the comprehensive scores and ranking results corresponding to the candidate points to form a set of explanatory factors for candidate points; based on the explanatory factor set, an explanatory quantitative result set for candidate points is generated, and causal link tuples are constructed; the explanatory quantitative result set is stored in a distributed ledger system, and the hash value of the explanatory factor set for candidate points is recorded through a blockchain smart contract; the causal link tuples are mapped to a preset business location knowledge graph, and the query and visualization of causal links are realized based on the graph database; the explanatory quantitative result set and causal link inference results are output to the visualization display module through a preset data interface.

[0017] The technical effects and advantages of the commercial traffic location optimization method based on multi-source data fusion proposed in this invention are as follows:

[0018] 1. This invention establishes a multi-dimensional, cross-modal data support system by introducing spatial geographic data, spatiotemporal traffic data, and consumer behavior data. Compared with traditional methods that rely on single traffic or population density indicators, this method can more comprehensively characterize the true commercial potential of candidate sites. For example, dynamic traffic characteristics reflect the changing patterns of passenger flow at different times, transportation accessibility characteristics reflect the connectivity between population and commercial space, and consumer behavior characteristics reveal the potential purchasing power of consumer groups. These three types of features are fused through an adaptive weight allocation mechanism and dynamically adjusted based on historical prediction errors, resulting in a more accurate and robust comprehensive feature representation, thereby effectively improving the reliability of subsequent predictions and optimizations.

[0019] 2. This invention utilizes evolutionary mechanisms such as genetic algorithms, enabling the model to simultaneously consider multiple objective functions, such as maximizing passenger flow potential, improving transportation convenience, and reducing competitive pressure. The evolutionary process, through non-dominated ranking and congestion distance control, gradually approximates the Pareto optimal solution set, thereby outputting a comprehensive score and ranking result for candidate points. This process avoids the one-sidedness of traditional single-objective or linear weighted optimization, achieving global optimization under multi-objective trade-offs, and providing a more reasonable solution for commercial site selection. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a commercial traffic location optimization method based on multi-source data fusion according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted in advance that all information or data obtained and processed in this application are carried out in accordance with relevant national data protection laws and regulations and with the authorization granted by the owner of the relevant device.

[0023] Example 1, Figure 1 This invention presents a commercial traffic location optimization method based on multi-source data fusion, comprising:

[0024] S1. Acquire multi-source data of the area to be tested, construct multi-source features, and fuse multi-source features across modalities through an adaptive weight allocation mechanism to generate a comprehensive feature representation. The multi-source features include: dynamic traffic features, traffic accessibility features, and consumer behavior features.

[0025] In this embodiment, multi-source data of the region to be tested is acquired, multi-source features are constructed, and the multi-source features are fused across modalities through an adaptive weight allocation mechanism to generate a comprehensive feature representation, as detailed below:

[0026] Acquire multi-source data for the area to be tested. The multi-source data includes: spatial geographic data (obtaining road network, points of interest, administrative divisions, and public transportation station data through the map open platform interface, and storing them under the same geographic coordinate system), spatiotemporal traffic data (obtaining anonymized mobile phone signaling data through telecommunications operators, obtaining taxi GPS trajectory and shared bicycle riding data through transportation departments, obtaining aggregated transaction records through third-party payment platforms, and obtaining population heat maps for different time periods through online map platforms), and consumer behavior data (obtaining population distribution and income level information through census and statistics bureau data, and collecting consumption preference information in conjunction with user surveys or membership system data).

[0027] The multi-source data is cleaned by removing duplicate, missing, and outlier values, and the spatiotemporal flow data is aligned to a time scale of hourly or daily using a unified timestamp standardization method to obtain the preprocessed multi-source dataset.

[0028] Based on the spatiotemporal flow data, the passenger flow intensity of the candidate points in different time periods (morning peak, noon, and evening) is calculated, and the time series is smoothed by the moving average method to form dynamic flow characteristics.

[0029] Based on road network and public transportation station data in spatial geographic data, the actual travel time matrix between candidate points and major population gathering areas is calculated, and the travel time is converted into an accessibility index to form a traffic accessibility feature.

[0030] Based on consumer behavior data, the consumption potential index around the candidate points is calculated, and the data of different dimensions are standardized to obtain consumer behavior characteristics.

[0031] The dynamic traffic flow characteristics, traffic accessibility characteristics, and consumer behavior characteristics are fused using a weighted method to obtain weighted features, and the weights of each feature are dynamically adjusted based on historical prediction errors.

[0032] Weighted features are input into a cross-modal embedding model, and deep representation learning methods are used to map features from different modalities to a unified vector space to generate a comprehensive feature representation.

[0033] In this embodiment, based on road network and public transportation station data in spatial geographic data, the actual travel time matrix between candidate points and major population gathering areas is calculated, and the travel time is converted into an accessibility index to form a traffic accessibility feature, as detailed below:

[0034] Acquire urban road network data, including arterial roads, secondary roads, branch roads and traffic control information, and divide the roads into directed edges with length and traffic speed attributes to construct a weighted graph model, i.e., a road network model, to represent the traffic relationship between different roads.

[0035] Collect subway and bus route data, extract station geographical locations and operating timetables, map the stations to the node locations of the road network model, and add waiting time and transfer time parameters to the paths between stations to form a public transportation network model;

[0036] Based on the road network model, the shortest path algorithm is used to calculate the driving time from the candidate point to each major population gathering area, and based on the public transportation network model, the public transportation travel time from the candidate point to each major population gathering area is calculated. The public transportation travel time includes walking time, waiting time, boarding time and transfer time.

[0037] The proportion of people's travel modes is obtained, and the travel time of self-driving and public transportation is weighted and integrated to obtain the comprehensive travel time from candidate points to various population gathering areas;

[0038] Starting from candidate points and ending at major population clusters, a travel time matrix between candidate points and population clusters is constructed based on comprehensive travel time.

[0039] Based on the travel time matrix, the accessibility index of the candidate points is output, i.e., the transportation accessibility feature.

[0040] The formula for calculating the accessibility index is as follows:

[0041]

[0042] In the formula: Represents the reachability index of candidate point i. This represents the population weight of population cluster j. This represents the total travel time from candidate point i to densely populated area j. This is a time decay function.

[0043] In this embodiment, based on consumer behavior data, the consumption potential index around the candidate point is calculated, and the data of different dimensions are standardized to obtain consumer behavior characteristics, as follows:

[0044] Obtain census data and publicly available data from the statistics bureau, extract the number of permanent residents, age structure, family size and income level within the neighborhood of candidate points, and form a population statistics set for the neighborhood of candidate points;

[0045] Based on user surveys or membership system data, we extract the consumption frequency and amount of different groups in categories such as catering, retail and entertainment to obtain consumption preference data. We then group the consumption preference data according to demographic attribute tags (age, income level, etc.) to obtain consumption preference parameter sets for different groups.

[0046] The population statistics set of the candidate point neighborhood is matched with the consumption preference parameter set. For the number of various groups in the population data, the corresponding consumption tendency coefficient is matched to obtain the weighted consumption vector of the candidate point neighborhood.

[0047] Based on the weighted consumption vector, the consumption potential index of the candidate points is output, and the consumption potential index of different candidate points is normalized by the interval scaling method, and the consumption potential index is mapped to the [0,1] interval to obtain the standardized consumption potential index, i.e. consumption behavior characteristics.

[0048] The formula for calculating the accessibility index is as follows:

[0049]

[0050] In the formula: This represents the consumption potential index of candidate point i. This represents the number of people of the k-th class within the neighborhood of candidate point i. This represents the consumption propensity coefficient of the k-th group.

[0051] S2, based on comprehensive feature representation, outputs the passenger flow trend and causal contribution of candidate points in different scenarios through causal inference methods;

[0052] In this embodiment, based on comprehensive feature representation, a causal inference method is used to output the passenger flow trend and causal contribution of candidate points in different scenarios, as detailed below:

[0053] Obtain information on external events that affect commercial customer traffic, including both sudden and periodic events;

[0054] External event information is transformed into structured data with labels for occurrence time, duration, and event type, resulting in an event factor dataset;

[0055] The event factor dataset is hierarchically encoded, and codes for sudden events and periodic events are established separately. Based on the occurrence time and duration of the events, multi-scale time series features are constructed, including three time dimensions: short-term, medium-term and long-term, thus forming a hierarchical event time series factor set.

[0056] The comprehensive feature representation is temporally aligned and spatially matched with the hierarchical event time series factor set to generate a unified input tensor containing spatial features, temporal features and event factors;

[0057] Based on a unified input tensor, a multi-task spatiotemporal prediction model is constructed, which includes a main prediction branch and an event influence branch. The main prediction branch is used to jointly model the data using a graph neural network and a time series model, and output the passenger flow prediction value of candidate points. The event influence branch is used to quantify the contribution of the hierarchical event factors at different time scales through a causal attention mechanism.

[0058] Based on the multi-task spatiotemporal prediction model, the baseline scenario data (excluding event factors) and the event scenario data (including specific event factors) are input respectively, and the passenger flow prediction results of the candidate points under different event scenarios are output.

[0059] Based on the passenger flow forecast results, the causal effect of the event factors is estimated by the causal inference method to obtain the causal contribution results of the passenger flow changes at the candidate points. The causal inference method includes a controlled experimental framework or a causal forest model, which is used to remove the influence of codependent variables.

[0060] Based on the passenger flow forecast results, the confidence interval of the passenger flow forecast results is calculated using the Bayesian approximate inference method to obtain the prediction uncertainty results of candidate points under different event scenarios.

[0061] S3 takes passenger flow trends and causal contributions as inputs, establishes a multi-objective optimization model, solves it, and outputs the comprehensive score and ranking results of candidate points;

[0062] In this embodiment, passenger flow trends and causal contributions are used as inputs to establish a multi-objective optimization model, which is then solved to output the comprehensive score and ranking results of the candidate points, as detailed below:

[0063] Based on passenger flow forecast results, causal contribution results, and forecast uncertainty results, a multi-objective optimization model is established. The multi-objective optimization model includes: passenger flow maximization objective, cost minimization objective, competition conflict avoidance objective, and risk minimization objective.

[0064] The candidate point location scheme is represented as a chromosome coding structure, where the chromosome contains binary decision variables for the candidate points, where a value of 1 indicates that the candidate point is selected and a value of 0 indicates that the candidate point is not selected.

[0065] Based on the individual's chromosome coding structure, an initial population is randomly generated, and a weighted sampling strategy based on causal contribution results is used during the generation process to increase the probability of high-potential candidate points being sampled and included in the initial population.

[0066] Based on the multi-objective optimization model, the fitness value of each individual in the initial population is output. The fitness value includes the objective functions of maximizing passenger flow, minimizing cost, avoiding competition, and minimizing risk.

[0067] Based on the fitness value, individuals with fitness values ​​higher than a preset threshold are swapped to select genes for candidate points, generating new offspring individuals, and the fitness values ​​of the offspring individuals are re-output.

[0068] Based on the re-output fitness values, the evolved population is subjected to non-dominated sorting, and the crowding distance of individuals is calculated. The non-dominated sorting is used to determine the dominance relationship of different individuals in the Pareto optimal solution set, and the crowding distance is used to maintain the diversity of the solution set.

[0069] Based on non-dominated ranking and crowding distance, individuals with better fitness and higher diversity are selected to form a new generation of population, which serves as the input for the next round of evolutionary iteration.

[0070] If the number of iterations reaches a preset threshold or the convergence of the Pareto optimal solution set meets a preset standard, then the final Pareto optimal solution set is output.

[0071] S4. Based on the comprehensive scoring and ranking results, combined with the causal contribution results, an explanatory index matrix is ​​constructed, and the site selection scheme is output through a visualization platform.

[0072] In this embodiment, based on the comprehensive scoring and ranking results, combined with the causal contribution results, an explanatory index matrix is ​​constructed, and the site selection scheme is output through a visualization platform, as follows:

[0073] An explanatory index matrix is ​​constructed by combining the comprehensive score, ranking results, and causal contribution results. The explanatory index matrix includes traffic driving factors, traffic accessibility factors, and competitive pressure factors, and corresponds to the comprehensive score and ranking position of the candidate points.

[0074] For each candidate point in the explanatory index matrix, a factor combination set is constructed. The factor combination set is a set containing any subset of factors. The subset is formed by combining and selecting from traffic-driven factors, traffic accessibility factors, and competitive pressure factors.

[0075] Based on the factor combination set, a factor permutation sequence is generated by Monte Carlo sampling, and a single factor is added one by one. The marginal improvement value of the factor on the comprehensive score of the candidate point before and after the addition is calculated to obtain the marginal contribution result of each factor in each permutation.

[0076] The marginal contribution results are weighted and averaged to obtain the Shapley values ​​of the flow-driving factor, traffic accessibility factor, and competition pressure factor for each candidate point. The weights are determined based on the probability of the factors appearing in different permutations.

[0077] The Shapley value results are correlated with the comprehensive scores and ranking results of the candidate points to generate an interpretive quantitative result set. This interpretive quantitative result set is then passed as input to the visualization step to generate an interpretive graphical display of the candidate points and an output of the site selection optimization scheme.

[0078] In this embodiment, the Shapley value results are correlated with the comprehensive scores and ranking results of the candidate points to generate an explanatory quantitative result set, as follows:

[0079] The marginal contribution results of factors are mapped and associated with the comprehensive scores and ranking results corresponding to the candidate points to form a set of explanatory factors for candidate points. The mapping rule is based on the candidate point identifier code and timestamp for unique indexing to ensure the traceability of results from different batches.

[0080] Based on the explanatory factor set, an explanatory quantitative result set of candidate points is generated. The result set includes candidate point identifier, comprehensive score, ranking position and Shapley value of each factor. Furthermore, a causal link tuple is constructed, which consists of "factor → Shapley value → comprehensive score contribution → candidate point ranking position".

[0081] The explanatory quantification result set is stored in a distributed ledger system, and the hash value of the candidate point explanatory factor set is recorded through a blockchain smart contract. The blockchain node is maintained by multiple participating entities to ensure the immutability of the explanatory results and the trusted sharing among multiple parties.

[0082] The causal link tuples are mapped to a preset business site selection knowledge graph. The query and visualization of causal links are realized based on the graph database. The knowledge graph nodes are used to represent candidate points or explanatory factors, and the edges are used to represent causal contribution relationships, thereby realizing traceable reasoning for the explanatory nature of candidate points.

[0083] The explanatory quantification result set and causal link reasoning results are output to the visualization display module through a preset data interface.

[0084] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0085] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0086] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0087] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0089] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A commercial traffic location optimization method based on multi-source data fusion, characterized in that, include: Acquire multi-source data of the region to be tested, construct multi-source features, and fuse multi-source features across modalities through an adaptive weight allocation mechanism to generate a comprehensive feature representation; Based on comprehensive feature representation, the causal inference method is used to output the passenger flow trend and causal contribution of candidate points in different scenarios; By taking passenger flow trends and causal contributions as inputs, a multi-objective optimization model is established and solved to output the comprehensive score and ranking results of candidate points. Based on the comprehensive scoring and ranking results, combined with the causal contribution results, an explanatory index matrix is ​​constructed, and the site selection scheme is output through a visualization platform. Based on comprehensive feature representation, and through causal inference methods, the system outputs the passenger flow trends and causal contributions of candidate points in different scenarios, as detailed below: Acquire information on external events that affect commercial customer flow, and transform this information into structured data to obtain an event factor dataset; The event factor dataset is hierarchically encoded, and codes for sudden events and periodic events are established separately. Based on the occurrence time and duration of the events, multi-scale time series features are constructed to form a hierarchical event time series factor set. The comprehensive feature representation is temporally aligned and spatially matched with the hierarchical event time series factor set to generate a unified input tensor. Based on the unified input tensor, a multi-task spatiotemporal prediction model containing a main prediction branch and an event impact branch is constructed. Based on the multi-task spatiotemporal prediction model, the baseline scenario data and event scenario data are input respectively, and the passenger flow prediction results of the candidate points under different event scenarios are output. Based on the passenger flow forecast results, the causal effect of the event factors is estimated using the causal inference method to obtain the causal contribution results of the passenger flow changes at the candidate points; Based on the passenger flow forecast results, the confidence interval of the passenger flow forecast results is calculated using the Bayesian approximate inference method to obtain the prediction uncertainty results of candidate points under different event scenarios.

2. The commercial traffic location optimization method based on multi-source data fusion according to claim 1, characterized in that, The process involves acquiring multi-source data of the region to be tested, constructing multi-source features, and fusing these features across modalities using an adaptive weight allocation mechanism to generate a comprehensive feature representation, as detailed below: Acquire multi-source data of the area to be tested and preprocess the multi-source data, which includes: spatial geographic data, spatiotemporal traffic data and consumption behavior data; Based on the spatiotemporal flow data, the passenger flow intensity of the candidate points at different time periods is calculated, and the time series is smoothed by the moving average method to form dynamic flow characteristics. Based on road network and public transportation station data in spatial geographic data, the actual travel time matrix between candidate points and major population gathering areas is calculated, and the travel time is converted into an accessibility index to form a traffic accessibility feature. Based on consumer behavior data, the consumption potential index around the candidate points is calculated, and the data of different dimensions are standardized to obtain consumer behavior characteristics. The dynamic traffic flow characteristics, traffic accessibility characteristics, and consumer behavior characteristics are fused using a weighted method to obtain weighted features, and the weights of each feature are dynamically adjusted based on historical prediction errors. Weighted features are input into a cross-modal embedding model, and deep representation learning methods are used to map features from different modalities to a unified vector space to generate a comprehensive feature representation.

3. The commercial traffic location optimization method based on multi-source data fusion according to claim 2, characterized in that, Based on road network and public transportation station data from spatial geographic data, the actual travel time matrix between candidate points and major population gathering areas is calculated, and the travel time is converted into an accessibility index to form a transportation accessibility feature, as detailed below: Acquire urban road network data and construct a road network model; Collect subway and bus route data, extract station geographical locations and operating timetables, map the stations to the node locations of the road network model, and add waiting time and transfer time parameters to the paths between stations to form a public transportation network model; Based on the road network model, the shortest path algorithm is used to calculate the self-driving travel time from the candidate point to each major population gathering area, and based on the public transportation network model, the public transportation travel time from the candidate point to each major population gathering area is calculated. The proportion of people's travel modes is obtained, and the travel time of self-driving and public transportation is weighted and integrated to obtain the comprehensive travel time from candidate points to various population gathering areas; Starting from candidate points and ending at major population clusters, a travel time matrix between candidate points and population clusters is constructed based on comprehensive travel time. Based on the travel time matrix, the accessibility index of the candidate points is output, i.e., the transportation accessibility feature.

4. The commercial traffic location optimization method based on multi-source data fusion according to claim 3, characterized in that, The process involves calculating the consumption potential index around candidate locations based on consumer behavior data, and standardizing data from different dimensions to obtain consumer behavior characteristics, as detailed below: Obtain census data and publicly available data from the statistics bureau, extract the number of permanent residents, age structure, family size and income level within the neighborhood of candidate points, and form a population statistics set for the neighborhood of candidate points; Based on user surveys, we obtained consumption preference data, and then grouped the consumption preference data according to demographic attribute tags to obtain consumption preference parameter sets for different groups. The population statistics set of the candidate point neighborhood is matched with the consumption preference parameter set. For the number of various groups in the population data, the corresponding consumption tendency coefficient is matched to obtain the weighted consumption vector of the candidate point neighborhood. Based on the weighted consumption vector, the consumption potential index of the candidate points is output, and the consumption potential index of different candidate points is normalized by the interval scaling method to obtain the consumption behavior characteristics.

5. The commercial traffic location optimization method based on multi-source data fusion according to claim 4, characterized in that, The process involves using passenger flow trends and causal contributions as inputs to establish a multi-objective optimization model, solving it, and outputting the comprehensive score and ranking results of the candidate points, as detailed below: A multi-objective optimization model is established based on passenger flow forecast results, causal contribution results, and forecast uncertainty results. The site selection scheme of the candidate points is represented as a chromosome coding structure, and an initial population is randomly generated based on the chromosome coding structure of the individuals. Based on the multi-objective optimization model, the fitness value of each individual in the initial population is output, and individuals with fitness values ​​higher than a preset threshold are selected. The selected individuals exchange candidate points to select genes, generate new offspring individuals, and re-output the fitness values ​​of the offspring individuals; Based on the re-output fitness values, perform non-dominated sorting on the evolved population and calculate the crowding distance of individuals. Individuals are selected based on non-dominated ranking and crowding distance to form a new generation of population, which serves as input for the next round of evolutionary iteration; If the number of iterations reaches a preset threshold or the convergence of the Pareto optimal solution set meets a preset standard, then the final Pareto optimal solution set is output.

6. The commercial traffic location optimization method based on multi-source data fusion according to claim 5, characterized in that, Based on the comprehensive scoring and ranking results, combined with the causal contribution results, an explanatory index matrix is ​​constructed, and the site selection scheme is output through a visualization platform, as detailed below: An explanatory index matrix is ​​constructed by combining the comprehensive scores, ranking results, and causal contribution results; For each candidate point in the explanatory index matrix, construct a set of factor combinations; Based on the factor combination set, a factor permutation sequence is generated by Monte Carlo sampling, and a single factor is added one by one. The marginal improvement value of the factor on the comprehensive score of the candidate point before and after the addition is calculated to obtain the marginal contribution result of each factor in each permutation. The marginal contribution results are weighted and averaged to obtain the Shapley values ​​of the flow-driving factor, traffic accessibility factor, and competition pressure factor for each candidate point. The Shapley values ​​are correlated with the overall scores and rankings of the candidate points to generate an interpretive quantitative result set.

7. The commercial traffic location optimization method based on multi-source data fusion according to claim 6, characterized in that, The Shapley value results are correlated with the comprehensive scores and ranking results of the candidate points to generate an explanatory quantitative result set, as detailed below: The marginal contribution results of factors are mapped and correlated with the comprehensive scores and ranking results corresponding to the candidate points to form a set of explanatory factors for candidate points; Based on the set of explanatory factors, a set of explanatory quantification results for candidate points is generated, and causal link tuples are constructed. The explanatory quantification result set is stored in a distributed ledger system, and the hash value of the candidate point explanatory factor set is recorded through a blockchain smart contract; The causal link tuples are mapped to a pre-defined business location knowledge graph, and the query and visualization of causal links are realized based on the graph database; The interpretive quantification result set and causal link inference results are output to the visualization module through the preset data interface.

Images