Business site revenue calculation method based on multi-source data fusion and dynamic calibration

Abstract

This invention discloses a method for calculating the revenue of commercial outlets based on multi-source data fusion and dynamic calibration, belonging to the field of data processing technology. The method includes acquiring multi-source data related to commercial outlet operations and industry-specific operational volume and price data; constructing an online revenue calculation model, a dynamic grid customer flow model, and an average transaction value correction model to calculate online and offline revenue across different dimensions; constructing an industry operational volume and price envelope model to fit industry volume and price patterns and define dynamic relationship intervals to perform model-based calibration on the online and offline revenue calculations; and using a similarity matching algorithm to identify and remove overlapping data in online and offline revenue to generate the total revenue of the commercial outlet. This invention achieves deep fusion of multi-source operational data and dynamically calibrates revenue based on industry patterns, solving the problems of single data source, large calculation deviations, and duplicate statistics in traditional methods, effectively improving the accuracy and rationality of commercial outlet revenue calculation.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for calculating the revenue of commercial outlets based on multi-source data fusion and dynamic calibration. Background Technology

[0002] In existing technologies for calculating the revenue of commercial outlets, the revenue of commercial outlets is usually calculated separately for online and offline revenue. Online revenue calculation relies solely on the operating data of a single platform, while offline revenue calculation relies mainly on manual statistics and partial data collected by a single customer flow counting device. The total revenue of the commercial outlet is obtained through simple numerical calculations.

[0003] The aforementioned existing technologies have not achieved the integration and application of multi-source data on commercial outlet operations, and lack a dynamic calibration process based on industry operating volume and price patterns. As a result, the calculated revenue results of commercial outlets cannot match the actual operating conditions of the outlets and have a large deviation from the real revenue data. The accuracy of revenue calculation is difficult to meet the needs of practical applications. Summary of the Invention

[0004] To address the technical problem that existing technologies fail to accurately reflect the actual operating conditions of commercial outlets and result in significant discrepancies with real revenue data, this invention provides a method for calculating commercial outlet revenue based on multi-source data fusion and dynamic calibration.

[0005] The technical solution adopted in this invention is: a method for calculating the revenue of commercial outlets based on multi-source data fusion and dynamic calibration, comprising the following steps:

[0006] Step 1: Obtain multi-source data related to the operation of commercial outlets and industry-related data on operating volume and price. The multi-source data related to the operation of commercial outlets includes online operating data, customer flow characteristic data, consumer profile data, and operating verification data.

[0007] Step 2: Construct an online revenue calculation model based on online business data, and use the online revenue calculation model to calculate the online revenue of the business outlets.

[0008] Step 3: Construct a dynamic grid passenger flow model based on passenger flow characteristic data, and construct an average transaction value correction model based on consumer profile data; use the dynamic grid passenger flow model to integrate and calculate passenger flow characteristic data and consumer profile data to generate effective passenger flow, and use the average transaction value correction model to model and correct the basic average transaction value to generate accurate average transaction value. Multiply the effective passenger flow by the accurate average transaction value to generate the offline calculated revenue of the commercial outlet.

[0009] Step 4: Construct an industry operating volume-price envelope model based on industry operating volume-price related data. Fit the industry operating volume-price related data using the industry operating volume-price envelope model, define the dynamic relationship range between price and sales volume, and perform model calibration on the volume-price data corresponding to online and offline revenue calculations based on the dynamic relationship range between price and sales volume, generating calibrated online and offline revenue.

[0010] Step 5: Integrate and calculate the online revenue and offline revenue after calibration to generate and output the total revenue of the business outlets.

[0011] The beneficial effects of this invention are: to realize the integrated application of multi-source data on commercial outlet operations, to make the online and offline revenue calculation process conform to the actual operating scenarios of commercial outlets and the operating volume and price rules of the industry, and to avoid duplicate revenue statistics in the integration process, effectively reduce the deviation between revenue calculation results and actual operating data, improve the rationality and reliability of commercial outlet revenue calculation, and provide valuable data support for commercial outlet revenue-related analysis. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0013] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0014] Example

[0015] Existing commercial outlet revenue calculation technologies rely solely on a single type of data source to independently calculate online and offline revenue. They fail to effectively integrate multi-source heterogeneous data related to commercial outlet operations and lack a dynamic calibration mechanism based on objective laws governing industry operating volume and price. They obtain total revenue solely through simple numerical calculations, resulting in discrepancies between the calculation results and the actual operating conditions of commercial outlets. This makes it difficult to meet the accuracy requirements for revenue calculation in practical applications such as regional economic analysis and business district vitality assessment.

[0016] To address the aforementioned technical issues, this embodiment provides a method for calculating the revenue of commercial outlets based on multi-source data fusion and dynamic calibration. All personal information involved in this method has been collected in compliance with regulations and anonymized; it only uses group statistical characteristics and does not identify individuals. Figure 1 As shown, it includes the following steps:

[0017] Step 1: Obtain multi-source data related to the operation of commercial outlets and industry operating volume and price data. The multi-source data related to the operation of commercial outlets includes online operating data, customer flow characteristic data, consumer profile data, and operation verification data.

[0018] It should be noted that multi-source data related to the operation of commercial outlets refers to various heterogeneous data that are directly or indirectly related to the daily operation of commercial outlets. It is the basic data for revenue calculation, including online operation-related data, customer flow characteristic-related data, consumer profile-related data, and operation verification-related data.

[0019] Online business-related data refers to transaction, traffic, and platform rule-related data generated by commercial outlets conducting business activities on various online platforms; customer flow characteristic-related data refers to spatial distribution, temporal fluctuations, and flow trajectories that reflect the characteristics of personnel flow around and within commercial outlets.

[0020] It should be noted that the location trajectory data involving consumers was collected in compliance with regulations and anonymized, retaining only regional and statistical characteristics, and without any individual precise trajectory information.

[0021] Consumer profile data refers to structured aggregated data that reflects the characteristics of consumer groups, based on de-identified consumer behavior, consumption preferences, and consumption capacity, and does not contain consumers' original personal information;

[0022] Operational verification data refers to various types of data used to verify the actual operating status of commercial outlets, including their business qualifications and operational status.

[0023] Industry operating volume and price related data refers to structured data on the operating prices, product sales, and the relationship between the two of various market entities within the industry to which the commercial outlets belong. This includes industry benchmark prices, industry sales ranges, data on volume-price linkage changes, and industry cycle change data.

[0024] This step involves acquiring multi-type, multi-dimensional data related to the operation of commercial outlets and industry-specific data on volume and price through compliant channels. Information involving consumers is immediately anonymized after collection, providing comprehensive and compliant data source support for the construction of subsequent models. Simultaneously, standardized preprocessing transforms heterogeneous raw data into structured data recognizable by the models, resolving issues of inconsistent data formats and noise interference. This ensures that subsequent model construction and revenue calculations are based on standardized and compliant data that aligns with the actual operating conditions of commercial outlets, avoiding the limitations of calculations caused by single data sources and non-standardized data, while also ensuring that data processing complies with regulatory requirements.

[0025] In the specific implementation process, online business-related data is collected from multiple platforms such as e-commerce platforms, social marketing platforms, and local life platforms that have commercial outlets. Data such as transaction flow, order number, order amount, payment time, platform commission rate, traffic recommendation rules, promotional activity rules, and changes in online traffic during time periods are collected through the compliant data interfaces opened by the platforms. The original personal payment information of consumers is not collected. Only the anonymized transaction association identifier is obtained. The collection frequency is consistent with the platform data update frequency, which is hourly / daily.

[0026] Passenger flow characteristic data is collected from compliant channels such as mobile signaling systems, store passenger flow monitoring equipment, and geographic information platforms. This includes data on the spatial distribution density of people within 1-3 kilometers around commercial outlets, the number of people entering and exiting store entrances, group statistics of people's walking directions, aggregation characteristics of dwell time, passenger flow fluctuations during time periods / days, weeks, and months, and changes in passenger flow during holidays. The collected individual location trajectory data is immediately aggregated and desensitized, retaining only group and regional passenger flow characteristics, without precise individual information. The passenger flow monitoring equipment collects data at the minute level, while the mobile signaling data is collected at the hour level.

[0027] Consumer profile data is collected from compliant channels such as consumer profile platforms, payment platforms, and industry consumer survey databases with legal data processing qualifications. The data is anonymized and structured aggregated profile data, including age distribution, gender ratio, consumption frequency aggregation, consumption amount range statistics, consumption intention characteristics, consumption capacity aggregation, and consumption scenario preferences of consumer groups. No original personal information of any individual consumer is collected, and the collection frequency is daily.

[0028] The data related to business verification is collected from the enterprise credit information disclosure system and the internal operation system of commercial outlets, including business registration information, business qualifications, number of tables / shelves in stores, and basic data on food / commodity procurement. First-hand data such as actual operating hours of stores, aggregated results of offline customer feedback, and actual business scale are collected through on-site visits and questionnaire surveys. The collection frequency is weekly / monthly.

[0029] Industry operating volume and price-related data are collected from industry association databases, market research institutions, and industry big data platforms. The data includes industry benchmark average transaction value, correlation data between price adjustments and sales volume changes, industry seasonal volume and price fluctuations, and reasonable sales volume corresponding to different price ranges. The data collection frequency is monthly.

[0030] The personal identification information of consumers involved in the above-collected data was subject to mandatory anonymization: mobile phone numbers were masked with the middle 4 digits, member IDs / payment accounts were processed using MD5 hashing, precise locations were aggregated to a 1 km × 1 km geographic grid, individual consumption behaviors were integrated into group statistical characteristics, and the anonymized data could not be linked to specific individual consumers and did not store any original personal information; data from different channels and in different formats were converted into CSV structured format, with unified timestamps in Beijing, monetary units in yuan, and customer flow units in person-times;

[0031] For missing numerical data, mean imputation is used to complete the data, with the completion window being the corresponding data collection period. For missing categorical data, mode imputation is used to complete the data. Outliers in numerical data are removed using the 3σ principle, with σ=3, meaning outliers exceeding the mean ± 3 times the standard deviation are removed.

[0032] All numerical feature data are standardized using the Min-Max normalization algorithm, mapping the data to the [0,1] interval. The algorithm formula is as follows:

[0033]

[0034] in, These are the normalized eigenvalues. These are the original eigenvalues. This is the minimum value of the feature data. The maximum value of this feature data is given; the output is a standardized, compliant, multi-source dataset related to the operation of commercial outlets and a dataset related to the volume and price of industry operations, which are free of original consumer personal information, and serves as the input data for all subsequent model construction and calculation.

[0035] For example, when collecting online business data for a chain of offline Chinese restaurants, the data collected includes the store's online order number, average order amount of 35 yuan, average daily order volume of 200 orders, Meituan platform commission rate of 18%, Douyin's traffic recommendation rules for catering hours, and discount activities of 10 yuan for purchases over a certain amount, from the compliant data interfaces of platforms such as Meituan, Dianping, and Douyin. The original payment account of consumers is not collected, and only the anonymized order association identifier is obtained.

[0036] When collecting customer flow characteristic data, the spatial distribution density of people within 1 kilometer around the store is collected from the mobile phone signaling system, and the number of people entering / leaving the store is collected minute by minute from the customer flow camera at the store entrance. The time fluctuation data of the peak customer flow from 11:30 to 13:30 during lunch and from 17:30 to 19:30 during dinner are obtained. The collected individual trajectory data is immediately aggregated and desensitized, and there is no precise location information of specific consumers.

[0037] When collecting consumer profile data, we collect anonymized aggregated profile data from qualified payment platforms. The data shows that the age of the in-store consumers is concentrated between 20 and 40 years old, the average consumption is 30-50 yuan, and the consumption scenario is mainly single dining / double dining. There is no original personal information such as consumer name, mobile phone number, or payment account.

[0038] When collecting data related to operational verification, we collected the store's business registration information from the enterprise credit information disclosure system and recorded first-hand data through on-site visits, showing that the store's actual operating hours were 10:00-21:00 and the number of tables was 20. When collecting data related to industry operating volume and price, we collected volume and price correlation data from the local catering industry association, which showed that the average customer spending per person was 45 yuan and the average customer spending was 30-60 yuan, corresponding to an average daily sales volume of 200-500 orders.

[0039] The original data was anonymized. A second MD5 hash operation was performed on a small number of incompletely anonymized related identifiers in the data. Precise location information was aggregated into a 1-kilometer grid. All timestamps were then unified to Beijing timestamps, and the monetary unit was unified to yuan. Missing two days of passenger flow data were imputed using the 7-day average. The 3σ principle was used to remove 100 abnormal data points from a single accidental touch collected by the passenger flow camera. Numerical features such as order volume, passenger flow density, and consumption amount were normalized to the [0,1] interval using Min-Max normalization. Finally, a standardized, compliant, multi-source operational dataset and industry volume and price dataset for the restaurant without original consumer personal information were obtained.

[0040] Step 2: Construct an online revenue calculation model based on online business data, and use the online revenue calculation model to calculate the online revenue of the business outlets.

[0041] Existing technologies rely solely on simple summation of transaction data from a single platform to calculate online revenue. They fail to explore the multi-dimensional characteristics of online business data, filter out invalid transaction data, and cannot adapt to the temporal fluctuations of online transactions, such as peak hours in the afternoon and evening, holiday fluctuations, and the influence of platform rules, resulting in significant deviations in online revenue calculation results.

[0042] To address the aforementioned technical issues, in one possible implementation, the online revenue calculation model constructed based on online business-related data includes the following:

[0043] Step 2.1: Extract transaction features, platform rule features, and time-series change features from online business-related data to construct a three-dimensional feature dataset.

[0044] It should be noted that the three-dimensional feature dataset refers to a three-dimensional tensor dataset formed by structured encoding of three types of features extracted from anonymized standardized online business-related data: transaction features, platform rule features, and time-series change features.

[0045] In the specific implementation process, transaction features are extracted from the standardized and anonymized online business-related dataset output in step 1. Platform rules and features Temporal variation characteristics ;

[0046] Transaction characteristics This includes order amount, order quantity, and transaction completion rate, with three dimensions; platform rule characteristics. This includes platform commission rates, promotional discount rates, and traffic recommendation weights, with three dimensions; time-series variation characteristics. This includes hourly trading volume, daily, weekly, and monthly trading volatility coefficients, and holiday trading increases, with three dimensions.

[0047] Tensor encoding is performed on the three types of features to construct a three-dimensional feature dataset. The dataset has the following dimensions: , This represents the sample size.

[0048] Step 2.2: Using the time-series features of the entire transaction process as anchor features, perform abnormal feature filtering on the three-dimensional feature dataset to remove invalid feature data.

[0049] The time sequence characteristics of the entire transaction process refer to the time sequence characteristics of each stage of the entire online transaction process from order placement to completion, including order placement time sequence, payment time sequence, confirmation of receipt time sequence, and no-refund time sequence.

[0050] In the specific implementation process, the time sequence characteristics of the entire transaction process are used as anchor features to set a complete threshold for the entire transaction process. This means that a transaction is considered valid only if at least 90% of the transaction steps are completed. Abnormal data indicating incomplete transactions, such as fraudulent orders, testing, or invalid refunds, are removed, and the filtered 3D feature dataset is output. .

[0051] Step 2.3: Dynamically assign feature weights to the filtered 3D feature dataset, complete the weight assignment based on transaction time priority, and construct a feature mapping matrix.

[0052] The feature mapping matrix refers to a matrix constructed by dynamically assigning feature weights based on transaction time-series priority to the filtered three-dimensional feature dataset. This matrix represents the linear mapping relationship between multi-dimensional features and online revenue. The dimensions are... ,in This represents the total number of dimensions for all features.

[0053] In the specific implementation process, the filtered 3D feature dataset is input. Feature weights are dynamically allocated based on transaction time priority. The higher the completion rate of each stage in the entire transaction process, the higher the weight of the corresponding feature. Basic weights are set for three types of features, and these weights are dynamically adjusted according to the transaction completion rate. The basic weights satisfy… Among them, transaction feature weights Platform rule feature weights Weights of time-series change features If the transaction completion rate is below 95%, the weight of the transaction features will be reduced proportionally, with an adjustment factor of [value missing]. ; Construct a feature mapping matrix by linearly combining all features according to their weights. The matrix dimension is 9 represents the total number of dimensions for the three types of features, and the matrix elements are the weight values ​​of each feature, used to achieve a linear mapping between multi-dimensional features and online revenue; finally, the feature mapping matrix is ​​output. .

[0054] Step 2.4: Build the computational architecture of the online revenue calculation model based on the feature mapping matrix, embed time-series prediction operators into the computational architecture, and complete the construction of the online revenue calculation model. The time-series prediction operators are used to adapt to the time-series fluctuation characteristics of online transactions.

[0055] It should be noted that the time series prediction operator refers to the time series prediction algorithm operator embedded in the computing architecture of the online revenue calculation model. It is implemented using the ARIMA algorithm and is used to capture the time series fluctuation characteristics of online transactions and adjust the model operation parameters in real time. The input of the operator is the time series change characteristics, and the output is the time series fluctuation correction coefficient.

[0056] In the specific implementation process, a three-layer operation architecture consisting of a feature input layer, a feature operation layer, and a result output layer is built; the ARIMA time series prediction operator is embedded in the feature operation layer.

[0057] The feature input layer receives standardized and anonymized online business-related data and feature datasets, and performs feature dimension matching; the input to the feature input layer is the standardized and anonymized online business-related dataset. The output of the feature input layer is a matched one-dimensional feature vector. (dimension is) The data stream from the feature input layer is output to the feature computation layer.

[0058] The feature operation layer is used to perform feature mapping operations and temporal fluctuation correction. The input of the feature operation layer is the one-dimensional feature vector output by the feature input layer. eigenmap matrix Then, the ARIMA time series prediction operator is embedded, and the parameters of the time series prediction operator are set to the autoregressive order. Difference order moving average order The input to the time series prediction operator is the time series variation characteristics, and the output of the operator is the time series fluctuation correction coefficient. value range ;

[0059] The operation rule is to first perform feature mapping operation. Then the calculation result is compared with the timing fluctuation correction coefficient. Multiply; the output is a preliminary estimate of online revenue, which is then sent to the result output layer.

[0060] The result output layer is used to restore the unit of the preliminary calculation value and output the result. The input of the result output layer is the preliminary calculation value of online revenue output by the feature operation layer; then the normalized calculation value is restored to the actual amount unit (yuan); the output is the online revenue.

[0061] The time-series prediction operator captures the time-series fluctuation characteristics of online transactions in real time and outputs correction coefficients. The computation results of the feature operation layer are dynamically adjusted to adapt the model to the fluctuations in the time series of online transactions.

[0062] Completed online revenue calculation model:

[0063] (2)

[0064] Standardized and anonymized data is input into the online revenue calculation model. After feature input layer matching, feature operation layer mapping and time sequence correction, and result output layer unit restoration, the online revenue is calculated.

[0065] The formula for calculating online revenue:

[0066] (3)

[0067] in For standardized, anonymized online business-related datasets, This is the time series fluctuation correction factor. For feature vectors, This is the feature mapping matrix.

[0068] For example, based on the standardized and anonymized online operational dataset of the aforementioned Chinese-style restaurants, transaction features were extracted, such as an average order value of 35 yuan, an average of 200 orders per day, and a transaction completion rate of 98%; platform rule features were extracted, such as a commission rate of 18%, a discount rate of 0.9, and a traffic recommendation weight of 0.6; and time-series variation features were extracted, such as lunchtime transactions accounting for 60%, weekend transaction growth of 50%, and holiday transaction growth of 100%.

[0069] A three-dimensional feature dataset with dimensions [30, 3, 3] was constructed, where 30 represents the number of days collected. The dataset contains no original personal information of consumers. Using the anonymized transaction time-series features as anchors, a completeness threshold of 0.9 was set, and 50 invalid order feature data generated by fraudulent transactions were removed. Weights were assigned according to transaction time-series priority, with a transaction completion rate of 98% > 95%, using basic weights. , , Construct a 9x1 feature mapping matrix; build a three-layer computational architecture and embed the ARIMA time series prediction operator. The time-series prediction operator outputs time-series correction coefficients based on the midday and evening market peak hours. Standardized and anonymized online business data is input into the model and processed to obtain... Multiplying this by a correction factor of 1.1 and restoring the units, we get the store's estimated online revenue of approximately 6,800 yuan per day.

[0070] This implementation method makes the data source for online revenue calculation more authentic by filtering out invalid and abnormal transaction data; the embedding of time-series prediction operators and collaborative operation rules enable the online revenue calculation model to accurately adapt to the time-series fluctuation characteristics of online transactions, improving the model's adaptability to the online operation scenarios of commercial outlets; and the fusion operation of multi-dimensional features through feature mapping matrix effectively reduces the deviation of online revenue calculation results, resulting in online revenue calculations that are more in line with the actual online operation status of commercial outlets.

[0071] Step 3: Construct a dynamic grid passenger flow model based on passenger flow characteristic data, and construct an average transaction value correction model based on consumer profile data. The dynamic grid passenger flow model is used to integrate and calculate the passenger flow characteristic data and consumer profile data to generate effective passenger flow. The average transaction value correction model is used to model and correct the basic average transaction value to generate an accurate average transaction value. The effective passenger flow and the accurate average transaction value are multiplied to generate the offline calculated revenue of the commercial outlet.

[0072] Effective customer traffic refers to the number of customers selected from the overall customer flow of a commercial outlet through a dynamic grid customer flow model that integrates desensitized consumer profile data, matches the characteristics of the consumer profile, and has the potential to make a purchase. It is a numerical result, measured in person-times, and is one of the indicators for offline revenue calculation. Based on group statistical data, without individual customer flow identification, the basic average transaction value (ADR) refers to the benchmark ADR for the commercial outlet's industry or the historical average ADR for the commercial outlet. It is a numerical result, measured in yuan per person, and is the basic input value for ADR correction. Precise ADR refers to the ADR output by the ADR correction model after multi-dimensional desensitization of the basic ADR, matching consumer profile characteristics, and calibrating for deviations. It is a numerical result, measured in yuan per person.

[0073] Offline revenue calculation refers to the preliminary estimate of the offline operating revenue of a commercial outlet, obtained by multiplying the effective customer traffic and the accurate average transaction value. It is a numerical result, and the unit is yuan.

[0074] Current technologies for offline revenue calculation heavily rely on manual statistics, sampling surveys, or single customer flow counting devices. They fail to explore the spatiotemporal linkage characteristics of customer flow data and do not effectively integrate customer flow data with consumer profile data. Furthermore, the average transaction value is often based solely on industry benchmarks or historical averages without adjustments for the actual consumer profile characteristics and customer flow scale of the commercial outlet. This results in significant discrepancies between the calculated effective customer flow and average transaction value, thus affecting the accuracy of offline revenue calculation. To address these issues, in one possible implementation, the construction of a dynamic grid customer flow model based on customer flow characteristic data includes the following:

[0075] Step 3.1: Extract spatial distribution features, temporal fluctuation features, and flow trajectory features from passenger flow characteristic data, normalize the three types of features, and generate a standardized passenger flow feature set.

[0076] In the specific implementation process, spatial distribution characteristics include passenger flow density and passenger flow distribution at store entrances; temporal fluctuation characteristics include time period fluctuations and holiday changes; flow trajectory characteristics include group statistics of travel direction and aggregation characteristics of dwell time; a total of 3 categories and 9 dimensions of features are used, which are mapped to [0,1] using Min-Max normalization to generate a standardized passenger flow feature set. , dimension .

[0077] Step 3.2: Perform cross-domain correlation mapping between spatial distribution features and temporal fluctuation features to construct a spatiotemporal linkage feature vector.

[0078] In the specific implementation process, feature splicing and Pearson correlation coefficient are used to calculate the correlation of spatiotemporal characteristics of passenger flow, and to construct a spatiotemporally linked feature vector. , dimension This characterizes the interaction between the spatial and temporal characteristics of passenger flow.

[0079] Step 3.3: Based on the spatiotemporal linkage feature vector, an adaptive grid partitioning algorithm is adopted to dynamically adjust the grid granularity according to the passenger flow density, thereby completing the dynamic partitioning of the geographic grid.

[0080] based on An adaptive mesh generation algorithm is adopted, and a mesh granularity adjustment threshold is set. Areas with a normalized passenger flow density value ≥0.5 are considered high-density areas. The grid size for high-density areas is 10m×10m, while for low-density areas it is 50m×50m. This completes the dynamic division of the geographic grid within a 1-3km radius of commercial outlets and within stores.

[0081] Step 3.4: Embed the flow trajectory features into the dynamically divided geographic grid, build a passenger flow calculation layer and a feature feedback layer to form a dynamic grid passenger flow model. The feature feedback layer is used to receive changes in passenger flow features in real time and adjust the model parameters.

[0082] The passenger flow calculation layer is used to perform passenger flow feature calculation, fusion of desensitized consumer profile features, and determination of passenger flow validity. The input of the passenger flow calculation layer is a standardized passenger flow feature set and a consumer feature vector. The output of the passenger flow calculation layer is the effective passenger flow and the passenger flow feature change value.

[0083] The input to the feature feedback layer is the change value of passenger flow characteristics output by the passenger flow calculation layer; the output of the feature feedback layer is the model parameter adjustment coefficient, such as the grid granularity adjustment coefficient and the operator decision threshold adjustment coefficient.

[0084] The passenger flow calculation layer and the feature feedback layer are bidirectionally connected. The feature feedback layer feeds back the parameter adjustment coefficients to the passenger flow calculation layer, enabling real-time dynamic adjustment of the model parameters and outputting the constructed dynamic grid passenger flow model. .

[0085] In one possible implementation, the step of generating effective passenger flow by fusing passenger flow characteristic data and consumer profile data through a dynamic grid passenger flow model includes the following:

[0086] Step 3.5: Extract consumption intention features and consumption capacity features from the relevant consumer profile data to construct a consumption feature vector.

[0087] Step 3.6: Expand the dynamic grid passenger flow model by fusion module and build a reverse coupling unit for passenger flow consumption.

[0088] Among them, the passenger flow consumption reverse coupling unit refers to the feature fusion module that extends the dynamic grid passenger flow model. It is used to realize the reverse matching and verification of the desensitized passenger flow feature-related data and the consumption profile-related data. The module input is the consumption feature vector and the standardized passenger flow feature set, and the output is the matched passenger flow data. The input data are all group aggregation data.

[0089] Step 3.7: Input the consumption feature vector into the reverse coupling unit and perform reverse matching verification with the standardized passenger flow feature set in the dynamic grid passenger flow model to filter out passenger flow data that matches the consumption features.

[0090] Step 3.8: Embed the effective passenger flow determination operator in the dynamic grid passenger flow model, verify the validity of the matched passenger flow data, and generate the effective passenger flow of commercial outlets.

[0091] Among them, the effective passenger flow determination operator refers to the classification and determination algorithm operator embedded in the dynamic grid passenger flow model. It is implemented using the logistic regression algorithm and is used to verify the effectiveness of passenger flow data after matching consumption profile features. The input of the effective passenger flow determination operator is the matched passenger flow data, and the output is the passenger flow effectiveness determination result.

[0092] In the specific implementation process, consumption intention and consumption capacity characteristics of group aggregation are extracted from the de-identified consumer profile data to construct a consumer feature vector. , dimension There is no individual consumer information. The passenger flow consumption reverse coupling unit is fully embedded and extended into the passenger flow calculation layer as an independent sub-functional module, forming a serial calculation relationship with the feature calculation unit and validity determination unit of the passenger flow calculation layer.

[0093] The reverse coupling unit adopts a three-layer structured operation architecture, consisting of a coupled input layer, a reverse matching operation layer, and a coupled output layer.

[0094] The coupled input layer receives two sets of feature data from external input. The fixed input consists of two types of desensitized group clustering data: one is a consumer feature vector generated from consumer profile data. Secondly, the standardized passenger flow feature set output by the dynamic grid passenger flow model. ;

[0095] The reverse matching layer is used to perform reverse matching verification between passenger flow features and consumption features. The operation logic is as follows: using the consumption feature vector as the matching benchmark, the standardized passenger flow feature set is traversed and compared feature by feature and grid by grid. The cosine similarity algorithm is used to calculate the matching degree between passenger flow features and consumption features. The matching degree calculation formula is:

[0096]

[0097] Set the reverse matching similarity threshold Only retain passenger flow feature data with a matching degree greater than or equal to the threshold to complete the reverse matching verification:

[0098] The coupling output layer is used to output the passenger flow data after reverse matching and verification, and transmit the data directly to the next unit of the passenger flow calculation layer, namely the input port of the effective passenger flow determination operator.

[0099] The passenger flow consumption reverse coupling unit only interacts with the passenger flow calculation layer internally and is not directly connected to the feature feedback layer. The input data comes from the consumption profile features and passenger flow features, and the output data flows to the effective passenger flow judgment operator, forming a complete fusion link of passenger flow features, reverse coupling, reverse verification of consumption features, and effective passenger flow screening.

[0100] In the passenger flow calculation layer, after the output of the passenger flow consumption reverse coupling unit, a logistic regression effective passenger flow determination operator is embedded. The parameter of the effective passenger flow determination operator is set to the learning rate. Number of iterations , Decision threshold Predicted probability For effective passenger flow.

[0101] The passenger flow calculation layer statistically analyzes the effective passenger flow data and dynamically corrects it using parameter adjustment coefficients from the feature feedback layer to obtain the effective passenger flow. The calculation formula is as follows:

[0102] (5)

[0103] in, For determining effective passenger flow, This represents the parameter adjustment coefficient for the feature feedback layer, with a value range of [0.9, 1.1]. Effective customer traffic for commercial outlets.

[0104] In one possible implementation, the step of modeling and correcting the basic average order value using an average order value correction model to generate an accurate average order value includes the following:

[0105] Step 3.9: Expand the functionality of the average order value correction model by building a customer flow-average order value matching and verification module. The customer flow-average order value matching and verification module is used to associate the feature matching relationship between effective customer flow and basic average order value.

[0106] Among them, the passenger flow-average transaction price matching and verification module refers to the feature matching module that extends the average transaction price correction model. It is used to associate the feature matching relationship between effective passenger flow and basic average transaction price. The module input is effective passenger flow and preliminary corrected average transaction price, and the output is the matching degree γ between the two.

[0107] Step 3.10: Input the basic average order value into the feature adaptation layer of the average order value correction model, perform feature matching with the mirror feature matrix, and generate the preliminary corrected average order value.

[0108] Step 3.11: Input the effective passenger flow into the passenger flow-average order price matching and verification module, and perform feature matching verification with the preliminary corrected average order price to determine whether the matching degree between the two meets the feature adaptation threshold of the average order price correction model. If the matching degree meets the feature adaptation threshold, output the preliminary corrected average order price as the accurate average order price. If it does not meet the threshold, perform dynamic deviation adjustment through the deviation calibration layer until the matching degree meets the feature adaptation threshold, and output the accurate average order price.

[0109] In the specific implementation process, the average order value correction model is a unidirectional serial architecture. The correction calculation layer includes a feature adaptation layer and a deviation calibration layer. The customer flow-average order value matching and verification module is extended to the correction calculation layer.

[0110] The process of constructing the average order value correction model is as follows: Input the standardized and de-identified consumer profile dataset and the standardized industry operating volume and price dataset output from step 1; extract the group aggregation customer characteristics, consumption preference characteristics, and consumption scenario characteristics (3 categories and 9 dimensions in total) from the de-identified consumer profile data; and extract the industry benchmark average order value characteristics from the industry volume and price data. (Numerical data, unit: yuan / person);

[0111] Will The mirror image is embedded into the feature space composed of three types of de-identified consumer profile features, and a feature expansion algorithm is used to generate the mirror feature matrix. , dimension The matrix elements represent the correlation values ​​between consumer profile features and the industry benchmark average order value.

[0112] A unidirectional, cascaded correction operation layer is constructed, consisting of a feature adaptation layer and a deviation calibration layer. The function of the feature adaptation layer is to match the differences between the customer group characteristics of the aggregated customer group and the industry benchmark average order value characteristics, generating a preliminary corrected average order value. The input of the feature adaptation layer is the benchmark average order value. The output of the feature adaptation layer is the initial correction of the average order value. Characteristic difference values;

[0113] The function of the deviation calibration layer is to correct the deviation in the feature adaptation process based on the feature difference value; the input of the deviation calibration layer is... Characteristic difference values; the output of the deviation calibration layer is the calibrated unit price;

[0114] Will The input correction layer uses an analysis of variance algorithm to automatically generate feature adaptation thresholds based on the quantized feature difference values ​​in the mirror feature matrix. value range This serves as a basis for judging the match between average order value and customer traffic.

[0115] To correct the error in the calculation layer, an extended customer flow-average transaction price matching and verification module is added. This module calculates the matching degree γ between the effective customer flow and the initially corrected average transaction price. The correction calculation layer, feature adaptation threshold θ, and customer flow-average transaction price matching and verification module are integrated to complete the construction of the average transaction price correction model, outputting the completed model. .

[0116] Will Input feature adaptation layer, and Perform feature matching to generate a preliminary revised average order value. ;

[0117] Will and The input passenger flow-average transaction price matching and verification module uses a cosine similarity algorithm to calculate the matching degree between the two. ;

[0118] Then, a matching degree determination is performed; if... Direct output To accurately determine the average order value, if ,Will Input deviation calibration layer, dynamically generates calibration values ​​based on feature difference values. ,right Make adjustments until ;

[0119] Specifically, precise average order value The calculation formula is:

[0120] (6)

[0121] in The calibration values ​​generated for the deviation calibration layer are obtained by linear calculation of the characteristic difference values.

[0122] Offline revenue generation calculation formula: ,in For effective passenger flow.

[0123] Current technologies for offline revenue calculation heavily rely on manual statistics, sampling surveys, or single customer flow counting devices. They fail to explore the spatiotemporal linkage characteristics of customer flow data and do not effectively integrate customer flow data with consumer profile data. Furthermore, the average transaction value is based solely on industry benchmarks or historical averages without being adjusted for the actual consumer profile characteristics and customer flow scale of the commercial outlet. This results in significant discrepancies between the calculated effective customer flow and average transaction value, thus affecting the accuracy of offline revenue calculation. In addition, some methods calculate revenue by identifying individual customer flow, directly collecting consumers' precise trajectories and personal information, which raises serious compliance and data security issues.

[0124] This step targets the actual scenario of offline customer traffic consumption and pricing in commercial outlets. It achieves deep integration of customer traffic characteristic data and consumer profile data through a dynamic grid customer traffic model, and accurately filters effective customer traffic based on group statistics. It achieves coordinated correction of industry benchmarks, consumer profiles, and customer traffic scale through an average transaction value correction model to obtain an accurate average transaction value. Finally, it generates offline revenue through the product of the two, realizing deep integration of multi-source data in the offline revenue calculation process.

[0125] Step 4: Construct an industry operating volume-price envelope model based on industry operating volume and price related data. Fit the industry operating volume-price related data using the industry operating volume-price envelope model, define the dynamic relationship range between price and sales volume, and perform model calibration on the volume and price data corresponding to online and offline revenue calculations based on the dynamic relationship range between price and sales volume, generating calibrated online and offline revenue.

[0126] Existing technologies lack a dynamic calibration mechanism for revenue calculation results. Preliminary calculations of online and offline revenue may deviate from the non-linear correlation and cyclical characteristics of the industry's volume and price, resulting in calculation results that, while conforming to local data logic, do not conform to the overall operational logic of the industry and lack rationality. To solve this technical problem, in one possible implementation, the construction of an industry operational volume-price envelope model based on industry operational volume-price related data includes the following:

[0127] Step 4.1: Extract the volume-price correlation characteristics and industry cycle characteristics from the industry's operating volume-price related data. At the same time, extract the real-time operating volume-price characteristics of commercial outlets from online and offline revenue calculations.

[0128] Step 4.2: Using the real-time operating volume and price characteristics of commercial outlets as the iterative benchmark, dynamically calibrate the industry volume and price correlation characteristics and industry cycle characteristics to generate an industry volume and price characteristic set adapted to commercial outlets;

[0129] Step 4.3: Build a nonlinear fitting operation architecture and embed an adaptive fitting operator into the nonlinear fitting operation architecture. The adaptive fitting operator automatically selects a fitting algorithm based on the industry volume and price feature set and performs fitting processing on the adapted industry volume and price feature set.

[0130] Step 4.4: Based on the fitting results, a dynamic volume-price interval definition module is built to automatically define the dynamic relationship interval between price and sales volume, and to complete the construction of the industry's operating volume-price envelope model. The dynamic volume-price interval definition module is used to receive changes in industry characteristics in real time and adjust the interval range.

[0131] It should be noted that the dynamic relationship range between price and sales volume refers to a reasonable range of price and sales volume values ​​that conforms to industry operating rules, defined by the non-linear fitting results of relevant industry operating volume and price data. This range is expressed as... ,in The minimum / maximum reasonable price in the industry. The minimum / maximum reasonable sales volume for the corresponding price range in the industry serves as an objective basis for calibrating online and offline revenue calculations.

[0132] An adaptive fitting operator refers to an algorithm selection operator embedded in a nonlinear fitting computational architecture. It can automatically select a suitable nonlinear fitting algorithm based on the characteristic distribution law of the corrected industry volume and price feature set.

[0133] In the specific implementation process, the construction process of the industry's operating volume-price envelope model is as follows:

[0134] Input the standardized industry operating volume and price related dataset from step 1, and the online revenue calculation dataset from step 2. Step 3 output: Offline revenue calculation Extract volume-price correlation features from industry volume-price data, such as industry price adjustment range, sales volume change rate, and volume-price correlation coefficient; extract industry cycle features from industry volume-price data, such as seasonal volume-price fluctuation coefficient and annual volume-price change rate, for a total of 2 categories and 6 dimensions of features;

[0135] from , Extract real-time operating volume and price characteristics of commercial outlets (e.g., online / offline average order value, sales volume); As an iterative benchmark, a feature-weighted correction algorithm is used to dynamically correct the industry volume and price characteristics, eliminating industry data that does not match the actual operation of commercial outlets, and generating an industry volume and price feature set adapted to commercial outlets. , dimension ;

[0136] A nonlinear fitting computation architecture is constructed, comprising a feature input layer, a fitting computation layer, and a result output layer. Each layer is connected by a unidirectional data flow, and an adaptive fitting operator is embedded in the fitting computation layer. If the Pearson correlation coefficient of industry volume and price characteristics Choose a polynomial fitting algorithm (order 2); if the correlation coefficient Choose the exponential fitting algorithm; adaptive fitting operator. It can automatically retrieve the fitting algorithm type and corresponding algorithm parameters without requiring manual specification;

[0137] A module for dynamically defining price and volume ranges is built. This module uses the quartile method to define a reasonable range between price and sales volume, and sets the range definition parameter to the upper quartile. Lower quartiles The interval is IQR stands for interquartile range. The module for dynamically defining volume and price ranges can receive changes in industry characteristics in real time and can also dynamically update the industry volume and price ranges on a monthly / quarterly basis, and dynamically adjust the range boundary values.

[0138] Industry volume and price characteristics This system integrates a nonlinear fitting computational architecture, an adaptive fitting operator Z, and a module for dynamically defining the price-volume interval, to construct an industry operating price-volume envelope model and output the completed model. .

[0139] The process of defining the dynamic range between price and sales volume is as follows: Input industry operating volume-price envelope model The feature input layer, after dimension matching, outputs to the fitting operation layer; the adaptive fitting operator Z is based on... The feature distribution law automatically selects the fitting algorithm, for Nonlinear fitting is performed to obtain the nonlinear correlation curve between industry volume and price;

[0140] The fitted curve is input into the dynamic price-volume range definition module. The module uses the quartile method to define a reasonable range of price and sales volume, generating a dynamic relationship range between price and sales volume. Output the dynamic range of industry price and sales volume.

[0141] The process of calibrating the online and offline revenue model involves: inputting the dynamic relationship range, , , Corresponding anonymized online volume and price data , Corresponding anonymized offline volume and price data ;

[0142] Will Each data point is compared to the dynamic relationship interval to determine if it falls within that interval. If the price and volume data exceed the interval, the model uses a distance decay algorithm to generate a calibration coefficient based on the Euclidean distance *d* between the price and volume data and the interval. , ,in The calibration coefficient range represents the maximum Euclidean distance between industry volume and price data. If the volume and price data are within the range, the calibration coefficient ;

[0143] Multiply the calculated revenue by the corresponding calibration factor to obtain the calibrated revenue; the formula for calculating the calibrated revenue is:

[0144] (7)

[0145] (8)

[0146] in These are the automatically generated online and offline calibration coefficients.

[0147] The existing technology lacks a dynamic calibration mechanism for revenue calculation results. The preliminary calculation of online and offline revenue may deviate from the non-linear correlation between volume and price and the cyclical change characteristics of the industry. As a result, although the calculation results are consistent with the local data logic, they are not consistent with the overall business logic of the industry and lack rationality.

[0148] This step involves constructing a proprietary industry-specific volume-price envelope model based on the operational patterns of the business outlets' respective industries. The model is then used to calibrate the initially calculated online and offline revenue based on these industry patterns. This ensures that the revenue calculation results not only align with the actual operating conditions of the business outlets but also conform to the objective laws of industry operations, further enhancing the rationality of the revenue calculation. Furthermore, the model is highly compatible with the industry scenario.

[0149] Step 5: Integrate and calculate the online revenue and offline revenue after calibration to generate and output the total revenue of the business outlets.

[0150] Current technologies for integrating online and offline revenue simply use arithmetic addition, failing to consider the overlap in online and offline consumer behavior at retail outlets. For example, if a customer places an order through an online mini-program while in-store, the transaction is simultaneously counted in both online and offline revenue, leading to duplicate revenue statistics and distorting the total revenue calculation. To address this issue, in one possible implementation, the process of integrating and calculating the calibrated online and offline revenue to generate and output the total revenue of the retail outlet includes the following:

[0151] Step 5.1: Extract the online transaction identifier information and consumer entity identifier information corresponding to the calibrated online revenue, and extract the offline transaction identifier information and consumer entity identifier information corresponding to the calibrated offline revenue.

[0152] Step 5.2: Match the online transaction identification information and consumer identification information with the offline transaction identification information and consumer identification information to identify overlapping revenue data corresponding to the same consumption behavior.

[0153] Step 5.3: Based on the calibrated online revenue, calibrated offline revenue, and the identified overlapping revenue data, perform a deduplication and summation operation to generate and output the total revenue of the business outlets.

[0154] It should be noted that transaction identification information refers to the anonymized structured data used to uniquely identify a transaction, including the anonymized order number, the hashed payment serial number, the transaction time, and the transaction amount.

[0155] Consumer entity identification information refers to the de-identified group aggregation identifier or hash-processed association identifier used to match consumption behavior, including the member ID after MD5 hash operation and the payment account after masking.

[0156] Identity matching algorithms based on de-identified labels refer to algorithms that use multi-feature similarity matching to compare de-identified online and offline label information to identify transaction data corresponding to the same consumer behavior.

[0157] Overlapping revenue data refers to revenue data that is counted twice due to the same consumption behavior being included in both online and offline revenue. It is a numerical result, and the unit is yuan.

[0158] In the specific implementation process, the online revenue after calibration is input from step 4. Offline revenue after calibration , The corresponding anonymized full volume of online transaction data, The corresponding anonymized full volume of offline transaction data;

[0159] The unique desensitized identification information is extracted from the full volume of desensitized online / offline transaction data. It is divided into transaction identification information and consumer identification information. All identification information is the information that has been desensitized in step 1. There is no need to process the original information again. The desensitized identification feature vector is directly formed.

[0160] Online desensitization identifier feature vector It includes the anonymized order number, the hashed payment serial number, the transaction time, the transaction amount, and the member ID after MD5 hashing, with dimensions [1,5].

[0161] Offline desensitization identifier feature vector It includes the offline de-identified consumption transaction number, the payment transaction number after hash processing, the transaction time, the transaction amount, and the member ID after MD5 hashing, with dimensions [1,5].

[0162] Output online desensitized identifier feature vector set Offline de-identification feature vector set .

[0163] An identity matching algorithm based on multi-feature similarity matching is adopted, and the core matching features and weights are clearly defined: the weight of transaction time is 0.3, the weight of transaction amount is 0.3, the weight of the de-identified payment serial number / member ID is 0.4, and the total weight is 1; the matching process is based only on the de-identified identifier and does not involve any original personal information of consumers.

[0164] A multi-feature comprehensive similarity threshold ε=0.9 is set, that is, if the comprehensive similarity is ≥0.9, it is judged as the same consumption behavior; and Align each pair of vectors one by one, and calculate the single-feature similarity of each pair of vectors. For example, transaction time is calculated using time difference similarity, transaction amount is calculated using numerical similarity, and de-identification label is calculated using hash value matching similarity. Multiply the single-feature similarity of each pair of vectors with the corresponding feature weights and sum them to calculate the multi-feature comprehensive similarity S. If S≥ε, it is determined to be the same consumption behavior, and the corresponding revenue data is overlapping revenue data.

[0165] The revenue amounts of all transactions identified as belonging to the same consumer behavior are summed to obtain the total overlapping revenue data. ;from or Remove from The elimination rule is to eliminate overlapping data only once to avoid repeated elimination;

[0166] Summation operation generates total revenue The calculation formula is:

[0167] (9)

[0168] This step addresses the practical scenario of integrated online and offline operations for commercial outlets. It employs a identity matching algorithm based on de-identified tags to integrate calibrated online and offline revenue. The algorithm identifies and eliminates overlapping revenue data before performing a summation operation to resolve the issue of duplicate statistics and achieve accurate calculation of total revenue.

[0169] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A business site revenue calculation method based on multi-source data fusion and dynamic calibration, characterized in that, Includes the following steps: Step 1: Obtain multi-source data related to the operation of commercial outlets and industry-related data on operating volume and price. The multi-source data related to the operation of commercial outlets includes online operating data, customer flow characteristic data, consumer profile data, and operating verification data. Step 2: Construct an online revenue calculation model based on online business data, and use the online revenue calculation model to calculate the online revenue of the business outlets. Step 3: Construct a dynamic grid passenger flow model based on passenger flow characteristic data, and construct an average transaction value correction model based on consumer profile data; use the dynamic grid passenger flow model to integrate and calculate passenger flow characteristic data and consumer profile data to generate effective passenger flow, and use the average transaction value correction model to model and correct the basic average transaction value to generate accurate average transaction value. Multiply the effective passenger flow by the accurate average transaction value to generate the offline calculated revenue of the commercial outlet. The average order value correction model constructed based on consumer profile data includes the following: Extract customer group characteristics, consumption preference characteristics, and consumption scenario characteristics from consumer profile data, and extract industry benchmark average order value characteristics from industry operating volume and price data; The industry benchmark average order value is mirrored and embedded into the feature space composed of customer group characteristics, consumption preference characteristics, and consumption scenario characteristics to form a mirrored feature matrix. A correction calculation layer is constructed, which includes a feature adaptation layer and a deviation calibration layer. The feature adaptation layer is used to match the differences between customer group characteristics and industry benchmark order value characteristics, and the deviation calibration layer is used to correct the deviation in the feature adaptation process. The mirror feature matrix is ​​input into the correction operation layer. The correction operation layer automatically generates a feature adaptation threshold based on the feature differences in the mirror feature matrix. The correction operation layer and the feature adaptation threshold are integrated to form the average order value correction model. The process of using a model to adjust the base average order value to generate an accurate average order value includes the following: The average order value correction model is extended to include a customer flow-average order value matching and verification module. This module is used to associate the feature matching relationship between effective customer flow and basic average order value. The base average order value is input into the feature adaptation layer of the average order value correction model, and feature matching is performed with the mirror feature matrix to generate the preliminary corrected average order value. The effective passenger flow is input into the passenger flow-average transaction price matching and verification module, and a feature matching and verification is performed with the initially corrected average transaction price to determine whether the matching degree between the two meets the feature adaptation threshold of the average transaction price correction model. If the matching degree meets the feature adaptation threshold, the initial corrected average order value is output as the accurate average order value; if it does not meet the threshold, dynamic deviation adjustment is performed through the deviation calibration layer until the matching degree meets the feature adaptation threshold, and the accurate average order value is output. Step 4: Construct an industry operating volume-price envelope model based on industry operating volume-price related data. Fit the industry operating volume-price related data using the industry operating volume-price envelope model, define the dynamic relationship range between price and sales volume, and perform model calibration on the volume-price data corresponding to online and offline revenue calculations based on the dynamic relationship range between price and sales volume, generating calibrated online and offline revenue. Step 5: Integrate and calculate the online revenue and offline revenue after calibration to generate and output the total revenue of the business outlets.

2. The method for calculating revenue of commercial outlets based on multi-source data fusion and dynamic calibration according to claim 1, characterized in that, The online revenue calculation model built based on online business-related data includes the following: Extract transaction features, platform rule features, and time-series change features from online business-related data to construct a three-dimensional feature dataset; Using the time-series features of the entire transaction process as anchor features, anomaly feature filtering is performed on the three-dimensional feature dataset to remove invalid feature data; The filtered 3D feature dataset is dynamically weighted, and the weight allocation is completed based on the transaction time priority to construct a feature mapping matrix. The computational architecture of the online revenue calculation model is built based on the feature mapping matrix. A time-series prediction operator is embedded in the computational architecture to complete the construction of the online revenue calculation model. The time-series prediction operator is used to adapt to the time-series fluctuation characteristics of online transactions.

3. The method for calculating revenue of commercial outlets based on multi-source data fusion and dynamic calibration according to claim 1, characterized in that, The construction of the dynamic grid passenger flow model based on passenger flow characteristic-related data includes the following: Spatial distribution features, temporal fluctuation features, and flow trajectory features are extracted from passenger flow characteristic data. The three types of features are normalized to generate a standardized passenger flow feature set. By performing cross-domain correlation mapping between spatial distribution characteristics and temporal fluctuation characteristics, a spatiotemporal linkage feature vector is constructed. Based on spatiotemporal linkage feature vectors, an adaptive grid partitioning algorithm is adopted to dynamically adjust the grid granularity according to passenger flow density, thereby completing the dynamic partitioning of the geographic grid. By embedding flow trajectory features into a dynamically divided geographic grid, a passenger flow calculation layer and a feature feedback layer are built to form a dynamic grid passenger flow model. The feature feedback layer is used to receive changes in passenger flow features in real time and adjust the model parameters.

4. The method for calculating revenue of commercial outlets based on multi-source data fusion and dynamic calibration according to claim 3, characterized in that, The process of generating effective passenger flow by fusing passenger flow characteristic data and consumer profile data through a dynamic grid passenger flow model includes the following: Extract consumption intention and consumption capacity features from relevant consumer profile data to construct a consumption feature vector; The dynamic grid passenger flow model is expanded by fusion modules to build a reverse coupling unit for passenger flow consumption; The consumption feature vector is input into the reverse coupling unit and back-matched with the standardized passenger flow feature set in the dynamic grid passenger flow model to filter out passenger flow data that matches the consumption features. An effective passenger flow determination operator is embedded in the dynamic grid passenger flow model to verify the validity of the matched passenger flow data and generate the effective passenger flow of commercial outlets.

5. The method for calculating revenue of commercial outlets based on multi-source data fusion and dynamic calibration according to claim 1, characterized in that, The construction of the industry operating volume-price envelope model based on industry operating volume and price related data includes the following: Extract volume-price correlation characteristics and industry cycle characteristics from industry operating volume-price related data, and extract real-time operating volume-price characteristics of commercial outlets from online and offline revenue calculations. Using the real-time operating volume and price characteristics of commercial outlets as the iterative benchmark, the industry volume and price correlation characteristics and industry cycle characteristics are dynamically calibrated to generate an industry volume and price characteristic set that is suitable for commercial outlets. A nonlinear fitting computation architecture is built, and an adaptive fitting operator is embedded in the nonlinear fitting computation architecture. The adaptive fitting operator automatically selects the fitting algorithm based on the industry volume and price feature set and performs fitting processing on the adapted industry volume and price feature set. Based on the fitting results, a dynamic volume-price range definition module is built to automatically define the dynamic relationship range between price and sales volume, and to complete the construction of the industry's operating volume-price envelope model. The dynamic volume-price range definition module is used to receive changes in industry characteristics in real time and adjust the range range.

6. The method for calculating revenue of commercial outlets based on multi-source data fusion and dynamic calibration according to claim 1, characterized in that, The process involves integrating and calculating the calibrated online and offline revenues to generate and output the total revenue of the business outlets. Includes the following: Extract the online transaction identifier information and consumer entity identifier information corresponding to the calibrated online revenue; extract the offline transaction identifier information and consumer entity identifier information corresponding to the calibrated offline revenue. The system performs identity matching between online transaction identification information and consumer identification information and offline transaction identification information and consumer identification information to identify overlapping revenue data corresponding to the same consumption behavior. Based on the calibrated online revenue, calibrated offline revenue, and the identified overlapping revenue data, a deduplication and summation operation is performed to generate and output the total revenue of the business outlets.

Images