A Multi-User Group Buying and Bidding Method and System

Abstract

The application discloses a multi-user group purchase collective procurement bidding method and system, wherein the multi-user group purchase collective procurement bidding method comprises the following steps: generating a market price prediction curve of each category and a category pair joint supply reverse index; processing real-time arrived user group purchase demand and generating a demand cumulative quantity prediction sequence and a demand time elasticity distribution diagram; coupling and analyzing the price prediction curve and the demand time elasticity distribution diagram to solve optimal elasticity window parameters with the lowest expected total procurement cost as the target; encapsulating a bidding demand package and executing bidding when the window is cut off; performing income allocation based on an incremental Shapley value approximation method and generating a collective procurement order; and verifying the effectiveness of a joint scheme through cross-category price coordination degree scoring and dynamically adjusting a joint group purchase combination scheme by using an incremental greedy algorithm. The application solves the problems that the existing method ignores price timing, lacks a cross-category unified decision-making framework and has a high calculation complexity in a dynamic environment.

Description

Technical Field

[0001] This invention relates to the field of e-commerce and supply chain management technology for bulk commodities, and more specifically, to a multi-user group buying and bidding method and system. Background Technology

[0002] In bulk commodity procurement scenarios such as steel, chemical raw materials, and non-ferrous metals, multi-user group-buying platforms aggregate the scattered demands of multiple users into bulk purchases, initiate bidding against suppliers to obtain quantity-price discounts, thereby reducing users' procurement costs. Existing multi-user group-buying bidding methods typically employ a flexible aggregation window mechanism, which determines when to trigger bidding by optimizing window parameters and distributes the bulk purchase revenue among users based on the Shapley value.

[0003] However, existing methods have the following technical problems: First, the parameter optimization of the elastic aggregation window only considers the demand aggregation dimension and does not incorporate the timing factor of market price fluctuations into the window decision. When the bidding time coincides with the market price being at a cycle high, the volume and price discount cannot offset the market premium, resulting in a high total procurement cost. Second, when the platform has demand for multiple categories and some suppliers have the ability to jointly supply across categories, the price fluctuation cycles of different categories have phase differences. The trade-off between the benefits of single-category price window optimization and cross-category joint bidding lacks a unified decision-making framework, and the platform cannot make a global trade-off between the benefits of joint discounts and the benefits of price timing. Third, in a dynamic environment where demand arrives in real time, the computational complexity of global traversal and global optimization solutions increases sharply with the scale of active demand, which cannot meet the requirements of real-time response. Furthermore, the revenue sharing based on the accurate Shapley value requires enumerating an exponential number of participant subsets, which is computationally unacceptable in a dynamic environment. Summary of the Invention

[0004] This invention provides a multi-user group-buying and collective procurement bidding method and system, which solves the technical problems in related technologies such as the lack of effective bidding timing optimization, cross-category joint procurement collaboration, and fair profit sharing mechanisms in multi-category bulk commodity group-buying and collective procurement scenarios.

[0005] This invention discloses a multi-user group-buying and collective procurement bidding method, which includes the following steps: obtaining historical market benchmark price time series data and historical group-buying and collective procurement data for each category, generating market price prediction curves for each category within a future preset time range, and constructing a category-to-joint supply inverted index based on the multi-category supply capacity data of candidate suppliers.

[0006] Process real-time user group-buying requests, update the cumulative demand for corresponding product categories, and generate a cumulative demand forecast sequence and a demand time elasticity distribution map for each product category;

[0007] The market price forecast curves of each product category are coupled with the demand time elasticity distribution map for analysis. For each candidate window cutoff time, the supplier's expected quotation benchmark is calculated based on the predicted market benchmark price. The expected discount rate is obtained by querying the quantity-price ladder function in combination with the predicted cumulative demand. The expected total procurement cost is calculated. The optimal elasticity window parameters for each product category are solved with the goal of minimizing the expected total procurement cost.

[0008] When a product category reaches the window deadline corresponding to the optimal elastic window parameter, the current determined demand is packaged into a centralized procurement bidding demand package, bidding is initiated to candidate suppliers, and the optimal supplier is selected based on the discount ratio to generate a centralized procurement transaction plan.

[0009] Based on the group purchase transaction plan, the revenue is distributed according to the proportion of each user's purchase amount, and multi-user group purchase orders are generated.

[0010] Furthermore, the generation of market price prediction curves for each product category within a preset future time range specifically involves:

[0011] For each product category, an autoregressive integral moving average model with exogenous variables is used. The historical market benchmark price time series of each product category is used as the endogenous input, and the exchange rate, upstream raw material price index and industry prosperity index are used as the exogenous input. The output is the predicted market benchmark price value at each time point within the future preset time range and the corresponding upper and lower bounds of the confidence interval.

[0012] The preset time range covers at least one complete price fluctuation cycle;

[0013] The category-pair joint supply inverted index uses category pairs as index keys and a list of suppliers that can be jointly supplied and a joint discount rate as index values. The construction process is as follows:

[0014] Iterate through the multi-category supply capacity data of all candidate suppliers. For each supplier that can supply two or more categories at the same time, extract the category pairs that can be jointly supplied, and add the supplier's identifier and the joint discount rate provided for the category pair to the value list of the corresponding category pair key.

[0015] Furthermore, the generation of the cumulative demand prediction sequence for each category specifically involves: using a Poisson process model to generate the predicted cumulative demand for each time point within a future preset time range based on the historical demand arrival time interval and the current demand arrival rate for that category;

[0016] The demand time elasticity distribution chart uses the time axis as the horizontal axis. At each time point, it marks the number of active users who have not exceeded the procurement deadline and the corresponding cumulative demand limit. As time goes by, the procurement deadline for some users expires, and the number of participating users and the cumulative demand limit gradually decrease.

[0017] Furthermore, the calculation of the supplier's expected quotation benchmark based on the predicted market benchmark price specifically involves: using the supplier's transaction quotation in the historical group purchase data and the corresponding market benchmark price to perform linear regression fitting to obtain the regression coefficient; substituting the predicted market benchmark price at the candidate window cutoff time into the linear regression relationship to obtain the supplier's expected quotation benchmark; the supplier's expected quotation benchmark is equal to the product of the regression slope coefficient and the predicted market benchmark price plus the regression intercept coefficient.

[0018] The expected total procurement cost is equal to the product of the supplier's expected quotation benchmark and the predicted cumulative demand, multiplied by the discount factor of the independent discount rate.

[0019] The optimal elastic window parameters are determined by using the upper bound of the time of the intersection of user purchase deadlines and the minimum group size as constraints. The upper bound of time is the earliest time of the purchase deadline of all users in the current active demand queue, and the minimum group size is the minimum batch size required by the supplier to accept the group purchase bid in the historical group purchase data.

[0020] Furthermore, the quantity-price ladder function is a piecewise function fitted based on the relationship between different bulk purchase quantities and corresponding discount rates in the historical group purchase data of this product category. When the cumulative demand falls into different quantity ranges, it corresponds to different discount rate levels. The larger the cumulative demand, the higher the discount rate.

[0021] The lower and upper bounds of demand and the discount rate for each volume range are obtained by statistical fitting of the actual transaction records of each volume range in the historical group buying and procurement data.

[0022] Furthermore, after processing real-time user group-buying requests, it also includes:

[0023] Using the category identifier of the newly arriving demand as the query key, perform a search in the category pair joint supply inverted index to obtain a list of all candidate related categories that have joint supply opportunities with the category of the newly arriving demand;

[0024] For each candidate related product category, obtain the cumulative demand of each of the two product categories and query the quantity-price ladder function to obtain their respective independent discount rates. Obtain the joint discount rate corresponding to the product category and calculate the incremental joint net income estimate. The incremental joint net income estimate is equal to the total joint purchase discount of the two product categories calculated according to the joint discount rate minus the sum of the independent purchase discounts of the two product categories calculated according to their respective independent discount rates.

[0025] The total discount for joint procurement is equal to the product of the sum of the cumulative procurement amount benchmarks for the two categories and the joint discount rate. The discount for independent procurement of each category is equal to the product of the cumulative procurement amount benchmark for that category and the independent discount rate for that category.

[0026] When the estimated incremental joint net profit exceeds the preset trigger threshold, the product category pair is marked as a candidate joint pairing. The preset trigger threshold is the mean of the distribution of the net profit difference between joint procurement and independent procurement in historical group purchase data.

[0027] Furthermore, for each candidate joint pairing, the market price prediction curves of the two categories are obtained. On the future time axis, the sum of the standardized distances between the prices of the two categories and their respective predicted troughs is calculated point by point. The sum of the standardized distances is equal to the sum of the two terms obtained by dividing the absolute value of the difference between the predicted market benchmark price and the predicted price trough of the two categories by the predicted price trough of the two categories. The predicted price trough is the minimum value among all predicted values ​​of the market price prediction curve of the corresponding category at all points in the future preset time range.

[0028] The time point with the minimum sum of standardized distances is identified as the optimal time point for price coordination, and the reciprocal of the minimum sum of standardized distances is used as the score for cross-category price coordination.

[0029] Calculate the sum of the expected total procurement cost of the joint scheme at the optimal time point for price coordination and the expected total procurement cost of the independent scheme at the time point corresponding to the optimal elasticity window parameters of each category. The expected total procurement cost of the joint scheme is equal to the sum of the products of the supplier's expected quotation benchmark and the predicted cumulative demand at the optimal time point for price coordination for the two categories, minus the joint discount amount calculated based on the joint discount rate.

[0030] When the sum of the expected total procurement costs of independent solutions minus the expected total procurement costs of joint solutions is greater than zero, the candidate joint pairing is confirmed as valid, and a set of valid joint pairings and the suggested joint bidding time points for each candidate joint pairing are generated.

[0031] Furthermore, it also includes performing local optimization using an incremental greedy algorithm, the steps of which are as follows:

[0032] Obtain all valid joint pairings involving the category of the newly arriving demand, and calculate the total expected cost change after incorporating each valid joint pairing into the existing combination scheme. The total expected cost change is obtained by uniformly adjusting the bidding time of the two categories involved in the candidate pairing to the optimal time point for price coordination corresponding to the candidate pairing, and recalculating the joint expected total procurement cost by replacing the individual discount rates with the joint discount rate, and subtracting the sum of the individual expected total procurement costs of the two categories before inclusion.

[0033] If there is an effective joint pair that reduces the total expected cost, then the effective joint pair that reduces the total expected cost the most is included in the combination scheme and the current combination scheme and the corresponding total expected cost are updated. This process is repeated until there are no candidate pairs that can further reduce the total expected cost, and the current combination scheme is output.

[0034] During the flexible window period, the market price forecast curve and demand accumulation forecast sequence of each product category are simultaneously corrected in a dual-channel sliding manner. When the current actual market benchmark price of a product category is lower than the lower bound of the confidence interval of the market price forecast curve and the current accumulated demand has met the minimum group purchase quantity, the bidding start signal of that product category is triggered.

[0035] When a category with valid joint pairings experiences a sudden price change, the net profit difference between the joint and independent strategies is recalculated. If the independent bid is better, the category is removed from the valid joint pairings and an independent bid is triggered.

[0036] Furthermore, the incremental Shapley value approximation method is used in the revenue sharing, specifically including:

[0037] Maintain the historical cumulative marginal contribution value of each category group. When a new joint match occurs, only recalculate the marginal contribution increment of the category affected by this joint match. The marginal contribution increment is equal to the total revenue of the joint scheme after the category is added to the current joint scheme category set minus the total revenue of the joint scheme before the category is added. The total revenue of the joint scheme is calculated by multiplying the sum of the purchase amount benchmarks of each category participating in the joint scheme by the corresponding joint discount rate. The marginal contribution increment is added to the historical cumulative marginal contribution value of the category.

[0038] The net revenue from joint procurement will be distributed among the participating product category groups based on the proportion of the historical marginal contribution of each product category group to the sum of the historical marginal contribution of all product category groups.

[0039] Within each product category group, the net revenue allocated to that product category group is further distributed among the users based on the proportion of each user's purchase amount to the total purchase amount of that product category group, and the final discount amount and payable amount for each user are calculated.

[0040] This invention provides a multi-user group-buying and bidding system, comprising:

[0041] The price forecasting and index building module is used to obtain historical market benchmark price time series data and historical group purchase data for each category, generate market price forecast curves for each category within a preset time range, and build a category-to-joint supply inverted index based on the multi-category supply capacity data of candidate suppliers.

[0042] The demand processing and forecasting module is used to process real-time user group buying demands, update the cumulative demand for corresponding product categories, and generate a forecast sequence of cumulative demand for each product category and a demand time elasticity distribution map.

[0043] The coupling analysis and window optimization module is used to perform coupling analysis on the market price forecast curves and demand time elasticity distribution maps of each product category. It calculates the expected total procurement cost for each candidate window cutoff time and solves for the optimal elasticity window parameters for each product category with the goal of minimizing the expected total procurement cost.

[0044] The bidding execution module is used to encapsulate the current determined demand into a centralized procurement bidding demand package when a certain product category reaches the window deadline corresponding to the optimal elastic window parameter, initiate bidding to candidate suppliers, and select the optimal supplier based on the discount ratio to generate a centralized procurement transaction plan.

[0045] The revenue sharing and order generation module is used to distribute revenue according to the proportion of each user's purchase amount based on the collective procurement transaction plan, and generate multi-user group purchase orders.

[0046] This invention couples market price forecast curves with demand time elasticity distribution maps for analysis, aiming to minimize expected total procurement costs to solve for optimal elasticity window parameters. This allows the determination of bidding timing to consider both sufficient demand aggregation and favorable market prices, solving the technical problem of existing methods that only optimize the demand aggregation dimension while neglecting price timing, leading to the offsetting of price discounts by market premiums during high-market bidding. It achieves the technical effect of guiding bidding timing to relatively favorable price points and reducing expected total procurement costs, provided that group formation conditions are met. By introducing cross-category price synergy scoring and uniformly comparing the expected total procurement costs of joint and independent schemes, it solves the technical problem of lacking a unified decision-making framework when optimal bidding timings for different categories conflict, achieving the technical effect of quantitatively judging the overall superiority of joint bidding schemes over independent bidding schemes. Furthermore, by utilizing category-based inverted indexes for joint supply, incremental greedy algorithms, and incremental Shapley value approximation methods, it solves the technical problems of excessively high global optimization computational complexity and the unacceptability of enumerating exponential subsets for precise Shapley values ​​in dynamic environments, achieving the technical effect of continuous response and maintaining reasonable revenue sharing in environments with real-time demand arrival. Attached Figure Description

[0047] Figure 1 This is a flowchart of the multi-category bulk commodity group buying and bidding method provided in the embodiments of the present invention;

[0048] Figure 2 This is a schematic diagram of the price forecast curves for hot-rolled coil and styrene monomer for the next 8 weeks provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram illustrating the expected total procurement cost assessment for each candidate window at the deadline provided in an embodiment of the present invention for Category A;

[0050] Figure 4This is a schematic diagram of the cumulative demand forecast (for the next 8 weeks) for product categories A and B provided in an embodiment of the present invention;

[0051] Figure 5 This is a schematic diagram illustrating the relationship between the quantity-price tiered discount rate and demand for product category A, provided in an embodiment of the present invention.

[0052] Figure 6 This is a schematic diagram comparing the expected total procurement costs of the independent solution and the combined solution provided in the embodiments of the present invention;

[0053] Figure 7 This is a schematic diagram illustrating the weekly variation of the standardized distance D(t) of cross-category price synergy provided in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram illustrating the ratio of purchase amount to net revenue sharing for Category B group users provided in this embodiment of the invention. Detailed Implementation

[0055] In multi-category bulk commodity group-buying platforms, user procurement demands for various categories such as steel, chemical raw materials, and non-ferrous metals continuously arrive at the platform in real-time. The platform needs to aggregate the scattered demands of multiple users into bulk purchases to obtain volume-based price discounts and initiate bidding with suppliers at appropriate times. However, existing multi-user group-buying bidding methods have the following technical problems:

[0056] First, the parameter optimization of existing flexible aggregation windows only considers the demand aggregation dimension, i.e., when enough users can be aggregated to obtain volume and price discounts, without incorporating the timing factor of commodity market price fluctuations into the window decision. Since the benchmark prices of various product categories fluctuate in different phases and cycles due to factors such as international futures prices, exchange rate changes, and industry cycles, when the window closes at a time when market prices are at a cyclical high, the supplier's benchmark price is too high, and the volume and price discounts cannot offset the market premium, resulting in a higher total cost for centralized procurement.

[0057] Second, when the platform has demand for multiple product categories and some suppliers have the ability to jointly supply across product categories, the price fluctuation cycles of different product categories have different phases, the optimal bidding time for each product category conflicts with each other, and there is no unified decision-making framework for the trade-off between optimizing the price window of a single product category and the benefits of joint bidding across product categories. This makes it impossible for the platform to make a global trade-off between the benefits of joint discounts and the benefits of price timing.

[0058] Third, in dynamic environments where demands arrive in real time, the computational complexity of existing methods using global traversal and global optimization increases dramatically with the scale of active demands, failing to meet real-time response requirements. Furthermore, the continuous changes in demand states during the calculation process cause matching results to become invalid. In addition, revenue sharing based on accurate Shapley values ​​requires enumerating an exponentially large subset of participants, which is even more unacceptable in dynamic environments.

[0059] It should be understood that the operating environment of this implementation method is a centralized procurement platform server with real-time data access capability. The centralized procurement platform server is configured with a time series database to store historical market benchmark price data and historical group purchase data for each category, a message queue to receive real-time user group purchase demand data streams, and a supplier database to store multi-category supply capacity data of candidate suppliers.

[0060] At least one embodiment of the present invention discloses a multi-user group-buying and bidding method, such as... Figure 1 As shown, it includes the following steps:

[0061] Step 1: Obtain historical data and generate market price forecast curves for each product category and inverted indexes of joint supply for each product category;

[0062] We acquire historical market benchmark price time series data for all active commodity categories on the platform, as well as historical group-buying and bulk purchasing data for each category. For each category, we use an autoregressive integral moving average model with exogenous variables (ARIMAX) to generate a market price prediction curve and confidence interval for a preset future time range. Simultaneously, we acquire multi-category supply capacity data from all candidate suppliers and convert this multi-category supply capacity data into a category-to-category inverted supply index.

[0063] It should be noted that the autoregressive integral moving average model with exogenous variables uses historical market benchmark price time series for each commodity category as endogenous input and exchange rates, upstream raw material price indices, and industry prosperity indices as exogenous inputs. It outputs predicted market benchmark price values ​​and corresponding upper and lower bounds of confidence intervals for each point in time within a preset future time range. During training, historical market benchmark price time series data and corresponding historical exogenous variable data are used as training samples. The mean squared error between the predicted value and the actual historical price is used as the loss function, and the maximum likelihood estimation method is employed to fit the parameters of the autoregressive integral moving average model with exogenous variables. It should be understood that the length of the preset time range is related to the typical cycle of price fluctuations for each commodity category and must cover at least one complete fluctuation cycle to capture the time points when price troughs occur.

[0064] It should be noted that the aforementioned category-pair joint supply inverted index refers to a data structure that uses category pairs as index keys and lists of suppliers capable of joint supply and joint discount rates as index values. The specific construction process is as follows: Iterate through the multi-category supply capacity data of all candidate suppliers. For each supplier capable of supplying two or more categories simultaneously, extract the category pairs that can be jointly supplied, and add the supplier's identifier and the joint discount rate offered for that category pair to the value list of the corresponding category pair key. The category-pair joint supply inverted index ensures that subsequent queries to determine whether any category pair has a joint supply opportunity only require a constant-time hash lookup operation.

[0065] Step 2: Process real-time user group-buying requests and generate a cumulative demand forecast sequence and a demand time elasticity distribution map;

[0066] When a new user's group-buying request arrives on the platform in real time, the platform obtains the category identifier, required quantity, and user-declared purchase deadline of the request. The request is then inserted into the active demand queue for the corresponding category, and the cumulative demand for that category is updated. A Poisson process model is used to generate a predicted cumulative demand sequence for the category based on historical demand arrival time intervals and the current demand arrival rate. A demand time elasticity distribution chart is generated by combining each user's purchase deadline, annotating the number of participating users and the corresponding upper limit of cumulative demand at each time point.

[0067] It should be noted that the Poisson process model takes the average arrival rate per unit time obtained from historical demand arrival records and the current active demand queue status as input, and the predicted cumulative demand at each time point within a preset future time range as output. During training, the actual arrival demand at each historical time period is used as the supervision signal, and the mean squared error between the predicted cumulative demand and the actual cumulative demand is used as the loss function. The maximum likelihood estimation method is used to fit the arrival rate parameter of the Poisson process model.

[0068] It should be noted that the aforementioned demand time elasticity distribution chart refers to the distribution data, with the time axis as the horizontal axis, marking the number of active users who have not yet exceeded the procurement deadline at each time point, as well as the corresponding cumulative demand limit for these users. As time progresses, the procurement deadlines for some users expire, and the number of participating users and the cumulative demand limit gradually decrease. The demand time elasticity distribution chart reflects the elastic constraints of demand in the time dimension.

[0069] Step 3: Couple the price forecast curve with the demand time elasticity distribution map and solve for the optimal elasticity window parameters;

[0070] The market price forecast curve for this product category is coupled with the demand time elasticity distribution diagram for analysis. For each candidate window's cutoff time point... Based on the predicted market benchmark price at that point in time The supplier's expected price benchmark is calculated using a linear regression relationship between historical prices and market benchmark prices. :

[0071]

[0072] in, The regression slope coefficient, The regression intercept coefficient, and This was obtained by performing linear regression fitting between supplier transaction prices and corresponding market benchmark prices in historical group-buying data. Because... and All are expressed in monetary units. The coefficient is dimensionless. The units are consistent with the price units, and the dimensions of the formulas are consistent.

[0073] Furthermore, the specific process of the above linear regression fitting is as follows: using the market benchmark price corresponding to each transaction record in the historical group-buying data as the independent variable and the supplier's transaction price as the dependent variable, a training sample set is constructed, and the least squares method is used to solve for the regression coefficient that minimizes the mean square error between the predicted price and the actual transaction price. and This establishes a mapping relationship between market benchmark prices and suppliers' expected pricing benchmarks.

[0074] Combined with the predicted cumulative demand at this point in time Query the volume-price step function to get the expected discount rate Calculate the expected total procurement cost at that point in time. :

[0075]

[0076] in, This serves as the supplier's expected price quote benchmark (currency / quantity unit). To predict cumulative demand (in units of quantity). The quantity-price step function for predicting cumulative demand The corresponding dimensionless discount rate, The unit is a monetary unit.

[0077] Calculate the aforementioned expected total procurement cost for all candidate window deadlines within a preset future timeframe, generating an evaluation curve for the window deadline versus the expected total procurement cost for this product category. (The expected total procurement cost is then used as the basis for this calculation.) With the objective of minimizing costs and constrained by the intersection of user purchase deadlines and minimum group purchase quantity, we solve for the optimal elastic window parameters for this product category. :

[0078]

[0079]

[0080] in, This is the upper bound of the intersection time of the purchase deadlines of all users in the current active demand queue. This is the minimum batch size for this product category.

[0081] Furthermore, the above Take the earliest time among all the procurement deadlines declared by users in the current active demand queue, which is the upper bound of the time when all users can participate in the bidding. Take the lowest batch size required by the supplier to accept the group purchase bid from the historical group purchase data of this product category. Below If the corresponding time point does not meet the clustering conditions, it will not be included in the candidate range of the optimal elastic window parameters.

[0082] It should be noted that the aforementioned volume-price step function refers to a piecewise function fitted based on the relationship between different bulk purchase quantities and corresponding discount rates in the historical group-buying data of this product category. Different discount rate levels correspond to different volume ranges; the larger the cumulative demand, the higher the discount rate. Specifically, let the volume ranges divided in the historical data total... One, of which The total number of the order of magnitude range, the first The lower bound of the demand for each interval is: The upper bound is The corresponding discount rate is ,in Let be the index of the magnitude interval, then the volume-price step function... Defined as: when hour, ,in , For the first The lower bound of demand in a range of orders of magnitude For the first The upper bound of demand in a range of magnitudes For the first The discount rate corresponding to each order of magnitude range, and Boundary values ​​of each interval , and discount rate All data were obtained by statistical fitting of actual transaction records from various volume ranges in historical group buying and procurement data.

[0083] Step 4: Package the centralized procurement bidding demand package and execute the bidding to generate a centralized procurement transaction plan;

[0084] When a product category reaches its optimal elastic window parameter At the corresponding window deadline, the system retrieves the currently confirmed demand quantity from the active demand queue for that product category and encapsulates this quantity into a centralized procurement bidding demand package. Simultaneously, it obtains the current actual market benchmark price, calculates a reasonable price reference range based on historical discount rates, and includes this range in the centralized procurement bidding demand package. The system then initiates bidding against candidate suppliers for that product category and obtains the price quotes from each supplier.

[0085] The unit price in each supplier's quotation Compared with the current market benchmark price Calculate the discount rate :

[0086]

[0087] in, Provide the supplier's quoted unit price (currency / quantity unit). The current market benchmark price (currency unit / quantity unit) has the same dimensions and a discount rate. This is a dimensionless ratio. It is expressed as a discount rate. As a core ranking indicator, a lower discount rate indicates that the supplier's price is more advantageous relative to the market price. The supplier with the lowest discount rate is selected as the optimal supplier, and a centralized procurement transaction plan is generated.

[0088] Step 5: Distribute revenue according to the proportion of each user's purchase amount and generate multi-user group purchase orders;

[0089] The net profit from the group purchase is determined based on the difference between the total transaction amount in the group purchase agreement and the total amount calculated based on the market benchmark price. Profit is allocated according to the proportion of each user's purchase amount within the group, and the final discount amount and payable amount for each user are calculated. The transaction unit price, market benchmark price reference value, discount rate, and payable amount for each user are integrated to generate a multi-user group purchase order, which is then pushed to the corresponding users and the winning supplier.

[0090] In this embodiment of the application, in order to address the problem of being unable to make a global trade-off between joint discount revenue and price timing revenue when there is a phase conflict in the optimal bidding timing for different product categories, the following step is added to step 2: performing a cross-category joint supply feasibility search:

[0091] Step 6: Retrieve candidate related product categories based on the category-based inverted index of joint supply and calculate incremental joint net revenue;

[0092] Using the category identifier of the newly arriving demand as the query key, a search is performed in the category pair joint supply inverted index to obtain a list of candidate related categories that have joint supply opportunities with the category of the newly arriving demand. For each candidate related category, the cumulative demand of the two categories is obtained, and their independent discount rates are obtained by querying the quantity-price step function. Then, the joint discount rate corresponding to that category pair in the category pair joint supply inverted index is obtained. The estimated incremental joint net revenue is calculated. :

[0093]

[0094] in, The total discount for joint procurement of two product categories, calculated based on the joint discount rate. and These are the independent purchase discounts calculated for each of the two product categories using their respective independent discount rates. , , The calculation method is as follows: Let the category be... The current cumulative demand is The current cumulative procurement amount is based on Category The current cumulative demand is The current cumulative procurement amount is based on Category The independent discount rate is (Based on the price-volume step function) (Value obtained from the location), category The independent discount rate is (Based on the price-volume step function) (The value is obtained from the purchase), the combined discount rate for this category is... (The value is obtained by retrieving the corresponding record from the inverted index of the joint supply for each category), then , , .in , The unit is the monetary unit. , , All are dimensionless discount rates. , , and All units are monetary units, and all have the same dimensions.

[0095] Furthermore, when the incremental joint net income When the preset trigger threshold is exceeded, the product category pair is marked as a candidate for joint pairing. The preset trigger threshold is obtained by statistically analyzing the distribution of the net profit difference between joint procurement and independent procurement in historical group-buying data. Specifically, the mean of the historical net profit difference distribution is taken as the preset trigger threshold to filter product category pairs with insignificant incremental joint net profit, ensuring that only product category pairs with significantly better joint profit than independent procurement are included in the subsequent joint pairing process.

[0096] In this embodiment of the application, in order to further conduct a unified comparison based on price timing between the joint solution and the independent solution, in addition to step 6, the following step is also included to perform a cross-category price synergy assessment:

[0097] Step 7: Calculate the cross-category price synergy score of the candidate joint pairings and verify the effectiveness of the joint scheme;

[0098] For each candidate joint pairing, obtain the two categories in the candidate joint pairing. and Each product's market price forecast curve. The sum of standardized distances from the predicted troughs of the two product categories' prices is calculated point-in-time on a future time axis. :

[0099]

[0100] in, and Category and categories At the point of time Predicted market benchmark price, and Category and categories The predicted price trough within a predetermined future timeframe. Both of the above are ratios of price deviation to the trough price; after canceling out their dimensions, they become dimensionless ratios. When the two are added together, their dimensions are identical. It is the sum of dimensionless standardized distances.

[0101] Furthermore, the aforementioned predicted price trough and Take the minimum value among all predicted values ​​of the market price forecast curve generated in step 1 for each product category within a preset future time range.

[0102] Sum of standardized distances Minimum time point Identified as the optimal time point for price coordination:

[0103]

[0104] The reciprocal of the sum of the minimum distances is used as the score for cross-category price synergy. :

[0105]

[0106] in, To prevent extremely small positive numbers from being divided by zero, It is a dimensionless quantity. It is also a dimensionless score. The larger the value, the better the timing of price coordination between the two product categories. The closer the prices at any given moment are to their respective troughs, the higher the degree of price coordination. The smaller the value, the greater the difference in the phase of price fluctuations between the two categories, making it difficult for them to approach their respective troughs at the same time.

[0107] Calculate the expected total procurement cost for the combined solution and the independent solutions separately. Expected total procurement cost of the combined solution. To achieve the optimal time for price coordination Total cost of joint bidding:

[0108]

[0109] in, and The optimal time points for price coordination for the two product categories. The supplier's expected quotation benchmark at any given time (currency / quantity unit). and The optimal time points for price coordination for the two product categories. Forecasted cumulative demand (in units) at any given time. The combined discount amount (in currency) is calculated based on the combined discount rate. The calculation method for the combined discount amount is as follows: ,in This refers to the dimensionless joint discount rate corresponding to this product category in the inverted index of joint supply for product categories. The unit is a monetary unit, and all terms in the formula have the same dimensions.

[0110] Expected total procurement cost of standalone solution The sum of the total costs of independent bidding for each product category at the time point corresponding to its optimal elastic window parameters:

[0111]

[0112] in, and Category and categories The expected total procurement cost (in currency) calculated based on their respective optimal flexibility window parameters. and Category and categories Their respective optimal elastic window parameters, and The calculation method is the same as the expected total procurement cost in step 3. The calculation method is the same.

[0113] When the net profit difference between the joint plan and the independent plan If the result is positive, the candidate joint pairing is confirmed as valid. Perform the above verification on all candidate joint pairs to generate a set of valid joint pairs verified by price coordination and a suggested joint bidding time point for each candidate joint pairing. .

[0114] Furthermore, the aforementioned net income difference The criteria for determining positive or negative are as follows: This indicates that the expected total procurement cost of the joint proposal is lower than that of the independent proposal, and that the joint bidding is superior to the independent bidding in terms of the combined effect of price timing and discount benefits. Therefore, the candidate joint pairing is confirmed as valid and included in the set of valid joint pairings. This indicates that the independent solution is better, and the candidate joint pairing is not included in the set of valid joint pairs.

[0115] In this embodiment of the application, in order to reduce computational complexity and support real-time response in a dynamic environment where demands arrive in real time, the following steps are included in addition to step 7: incremental local optimization and dynamic adjustment are performed:

[0116] Step 8: Use an incremental greedy algorithm to perform local optimization and dynamically adjust the joint purchase combination scheme based on real-time data;

[0117] The incremental greedy algorithm takes the effective set of joint pairings and the total expected cost of the current combination scheme as input, and the updated combination scheme as output. The steps of the incremental greedy algorithm are as follows:

[0118] Step 8-1: Obtain all valid joint pairings related to the category of the newly arriving demand, and calculate the total expected cost change after incorporating each valid joint pairing into the existing combination scheme.

[0119] Step 8-2: If there is an effective combination that reduces the total expected cost, then include the effective combination that reduces the total expected cost into the combination scheme and update the current combination scheme and the corresponding total expected cost; otherwise, keep the existing scheme unchanged.

[0120] Step 8-3: Repeat step 8-2 until there are no candidate pairs that can further reduce the total expected cost, and output the current combination scheme.

[0121] Furthermore, the calculation method for the total expected cost change in step 8-1 above is as follows: after incorporating the candidate pairing into the portfolio scheme, the bidding time points of the two categories involved in the candidate pairing are uniformly adjusted to the optimal time point for price coordination corresponding to the candidate pairing. and with a combined discount rate Replace the individual discount rates with the expected total procurement cost in step 3. The calculation method is to recalculate the combined expected total procurement cost of the two product categories and subtract it from the sum of the individual expected total procurement costs of the two product categories to obtain the change in total expected cost.

[0122] During the flexible window period, the latest market benchmark price data is acquired in real time, and the market price forecast curves and cumulative demand forecast sequences for each product category are simultaneously corrected using a dual-channel sliding method. This dual-channel sliding method means that the first channel uses the latest acquired actual market benchmark price data to continuously correct the market price forecast curve, while the second channel uses the latest arriving user demand data to continuously correct the cumulative demand forecast sequence; both channels execute simultaneously.

[0123] When a product category experiences an unexpected sharp price drop, meaning the current actual market benchmark price is below the lower bound of the confidence interval of the market price forecast curve, and the current accumulated demand has already met the minimum group-buying quantity... When a price change occurs in a category with valid joint purchase pairings, the net profit difference between the joint and independent purchase plans is recalculated. If the independent bid is more favorable, the category with the price change is removed from the valid joint purchase pairings and an independent bid is triggered. Dynamically adjusted joint purchase combinations and bid trigger instructions are generated.

[0124] Furthermore, the criteria for judging a sharp price drop are as follows: the current actual market benchmark price is lower than the lower bound of the confidence interval at the corresponding time point output by the autoregressive integral moving average model with exogenous variables in step 1, and the cumulative demand in the current active demand queue is not less than... When both conditions are met simultaneously, it is considered an unexpected sharp price drop, immediately triggering a bidding start signal without waiting for the optimal elastic window parameter. The corresponding moment has arrived.

[0125] In this embodiment of the application, in order to reduce the exponential complexity of accurate Shapley value calculation and support dynamic updates in a joint procurement scenario, step 5 uses an incremental Shapley value approximation method to replace accurate Shapley value calculation for revenue sharing. The specific processing includes the following sub-steps:

[0126] Step 501: Maintain the historical cumulative marginal contribution value for each product category group. When a new joint match occurs, only recalculate the marginal contribution increment for the product categories affected by this joint match. The aforementioned marginal contribution increment refers to the increase in the total revenue of the joint scheme after the product category is added to the joint match, relative to the total revenue of the joint scheme without the product category. Update the marginal contribution increment to the historical cumulative marginal contribution value of the product category.

[0127] Furthermore, the specific calculation method for the aforementioned marginal contribution increment is as follows: assuming the current joint scheme includes a set of categories. Category The total revenue of the joint program before joining was Category The total benefit of the joint scheme after joining is Then the category The marginal contribution increment of this joint matching is ,in The set of categories included in the current joint scheme. To be from the category set Removed categories Subset after, For category collection The corresponding total benefit of the joint scheme, For category collection The corresponding total revenue from the joint scheme is calculated based on the joint discount amount, which is the sum of the benchmark purchase amounts for each category participating in the joint scheme multiplied by the corresponding joint discount rate. This marginal contribution increment is then added to the category. The historical cumulative value of marginal contribution.

[0128] Step 502: Based on the updated historical marginal contribution cumulative value of each category group, the net revenue under the joint procurement scenario is distributed among the participating category groups according to the proportion of the historical marginal contribution cumulative value of each category group to the sum of the historical marginal contribution cumulative values ​​of all category groups.

[0129] Step 503: Within each product category group, the net revenue allocated to that product category group is redistributed among the users based on the proportion of each user's purchase amount to the total purchase amount of that product category group. The final discount amount and payable amount for each user are then calculated. The transaction unit price, market benchmark price reference value, discount rate, and payable amount for each user are integrated to generate a multi-user group-buying order, which is then pushed to the corresponding users and the winning supplier.

[0130] This implementation method couples the market price forecast curve with the demand time elasticity distribution map in step 3. For each candidate window's cutoff time, it simultaneously evaluates the discount rate corresponding to the predicted market price level and the predicted cumulative demand, calculates the comprehensive expected total procurement cost, and solves for the optimal elasticity window parameters using the lowest expected total procurement cost. This ensures that the determination of bidding timing considers both the sufficiency of demand aggregation and the favorable market price. Therefore, when the market price is at a cyclical high, even if the demand meets the group formation conditions, the expected total procurement cost is still not optimal due to the high price benchmark. The window cutoff time will be postponed to a more favorable price point, thus overcoming the problem of existing methods that only optimize the demand aggregation dimension while neglecting price timing, leading to the market premium offsetting the price discount during high-market bidding.

[0131] Furthermore, this implementation method quantifies the alignment of price trends across different product categories by introducing a cross-category price synergy score in step 7. It also compares the expected total procurement cost of the joint solution at the optimal time point for price synergy with the expected total procurement cost of the independent solutions at their respective optimal time points. This allows the platform to quantitatively weigh the benefits of joint discounts against the benefits of price timing. When the price fluctuation cycles of the two product categories differ significantly in phase, causing the optimal time point for price synergy to deviate far from their respective troughs, the price disadvantage of the joint solution will outweigh the benefits of joint discounts. In this case, the system automatically selects the independent solution, thus resolving the problem of lacking a unified decision-making framework when there is a phase conflict in the optimal bidding timing for different product categories.

[0132] Furthermore, this implementation reduces the computational complexity of the joint feasibility query to be related to the number of category pairs involved in the newly arriving demand rather than the total scale of active demand by using the category pair joint supply inverted index in step 6. In step 8, it performs local optimization instead of global re-optimization based on the existing scheme by using an incremental greedy algorithm. In steps 501 to 503, it uses the incremental Shapley value approximation method to update only the marginal contribution increment of the affected category instead of exponential subset enumeration. This enables the method to continuously respond in the dynamic environment of real-time demand arrival and maintain the fairness of the revenue sharing results.

[0133] The following is an example of an application of the present invention, such as... Figure 2-8 As shown, the implementation process is as follows:

[0134] A bulk commodity procurement platform (hereinafter referred to as "Platform A") serves multiple enterprise users in the construction engineering and chemical manufacturing industries. On a certain working day in March 20XX, Platform A's active product categories included hot-rolled coil (Category A) and styrene monomer (Category B). Platform A's time-series database has accumulated benchmark market price data for both categories over the past 24 months. Its message queue receives real-time procurement requests from 7 users, and its supplier database has 12 candidate suppliers registered, 3 of which have the cross-category joint supply capability to simultaneously supply hot-rolled coil and styrene monomer.

[0135] Platform A extracts weekly benchmark market price data for hot-rolled coil and styrene monomer from a time-series database for the past 24 months. It also uses the exchange rate, upstream iron ore / styrene raw material price index, and steel / chemical industry prosperity index as exogenous variables to train two ARIMAX models, generating price forecast curves and 95% confidence intervals for the next 8 weeks. Simultaneously, it traverses the multi-category supply capacity data of 12 candidate suppliers, identifying suppliers SUP-07, SUP-09, and SUP-11 as having joint supply capacity for both hot-rolled coil and styrene monomer. Using the category pair (hot-rolled coil, styrene monomer) as the index key, it writes the three supplier identifiers and their joint discount rates into the category pair joint supply inverted index.

[0136] Table 1 shows the results of constructing the inverted index for joint supply by product category.

[0137]

[0138] In the second week of March 20XX, Platform A received seven group-buying requests from users in its message queue. Four of these requests were for hot-rolled coil (Category A), and three were for styrene monomer (Category B). Upon receiving each request, the system immediately updated the active demand queue and cumulative demand for the corresponding category. Based on a Poisson process model, it generated a forecast sequence of cumulative demand for the next eight weeks and simultaneously generated a demand time elasticity distribution chart based on the procurement deadlines declared by each user.

[0139] Table 2 Active Demand Queue Data

[0140]

[0141] The current cumulative demand for Category A is 1,200 tons, with a cumulative purchase amount of RMB 6.48 million (calculated at the current market benchmark price of RMB 5,400 / ton); the current cumulative demand for Category B is 260 tons, with a cumulative purchase amount of RMB 2.86 million (calculated at the current market benchmark price of RMB 11,000 / ton).

[0142] Taking hot-rolled coil (category A) as an example, this paper illustrates the process of solving for the optimal elastic window parameters. Historical data is used to obtain the linear regression coefficients of supplier quotations for category A. , (RMB / ton). Minimum batch size for Category A. Tons, the upper bound of the intersection of the purchase deadlines of all users in the current active demand queue is week 4 ( ).

[0143] In the volume-price step function, category A is divided into 3 volume ranges: Discount rate per ton , Discount rate per ton , Discount rate per ton .

[0144] Using the end of the third week as the candidate window cutoff time, the ARIMAX model predicts the market benchmark price at that point in time. Yuan / ton, Poisson process model predicts cumulative demand tons, falling into the second order of magnitude, discount rate .

[0145] Supplier's expected quotation benchmark:

[0146]

[0147] Expected total procurement cost:

[0148]

[0149] The evaluation curves are generated by calculating the values ​​for each candidate time point in the future:

[0150] Table 3. Expected Total Procurement Cost Assessment for Each Candidate Window in Category A

[0151]

[0152] Because the expected total procurement cost of RMB 5.087 million at the end of the second week is the lowest, and meets the requirements... and The optimal elasticity window parameter for category A is found to be at the end of the second week, corresponding to the expected total procurement cost. Ten thousand yuan.

[0153] Similarly, the optimal elasticity window parameter for category B (styrene monomer) is found to be at the end of the 4th week, corresponding to the expected total procurement cost. Ten thousand yuan.

[0154] When a new demand for Category B (styrene monomer) arrives (user U07, 65 tons), the system performs a hash lookup in the category pair's combined supply inverted index using "styrene monomer" as the query key. If the target category pair (hot-rolled coil, styrene monomer) is found, the system returns three suppliers: SUP-07, SUP-09, and SUP-11, and selects the best combined discount rate. (Provided by SUP-11).

[0155] At this point, the cumulative purchase amount benchmark for Category A 10,000 yuan, independent discount rate Cumulative purchase amount benchmark for Category B 10,000 yuan, independent discount rate (The discount rate corresponding to the quantity-price step function of product category B at 260 tons).

[0156] The discount amounts are calculated as follows:

[0157]

[0158]

[0159]

[0160]

[0161] The historical net profit difference distribution mean (preset trigger threshold) is 120,000 yuan. Because... The product category pair (hot-rolled coil, styrene monomer) is marked as a candidate joint pairing and proceeds to step 7.

[0162] For candidate joint pairings (Category A, Category B), the system extracts the ARIMAX price prediction curves for each of the two categories and calculates the sum of standardized distances week by week over the next 8 weeks. The predicted price troughs for the two categories are 5210 yuan / ton for category A (occurring in week 5) and 10450 yuan / ton for category B (occurring in week 4).

[0163] Taking the end of the fourth week as an example, the predicted price for product category A is 5290 yuan / ton, and the predicted price for product category B is 10450 yuan / ton (exactly the lowest price).

[0164]

[0165] After iterating through all candidate time points, the optimal time point for price coordination was identified as the end of the fourth week. .

[0166] Cross-category price synergy score (take) ):

[0167]

[0168] Calculate the expected total procurement cost of the joint solution at the end of the fourth week, the optimal time for price coordination. The supplier's expected quote benchmark for Category A at the end of the fourth week:

[0169]

[0170] Supplier expected price benchmark for Category B at the end of the 4th week (Category B regression coefficient) , ):

[0171]

[0172] Expected purchase amount for Category A (cumulative demand forecast at the end of week 4: 1380 tons):

[0173]

[0174] Expected purchase amount for Category B (cumulative demand of 260 tons):

[0175]

[0176] Combined discount amount:

[0177]

[0178] The expected total procurement cost of the joint solution:

[0179]

[0180] Expected total procurement cost for standalone solutions:

[0181]

[0182] Because the joint scheme postpones the bidding period for Category A from the end of the second week to the end of the fourth week, the expected procurement cost for Category A at the end of the fourth week is RMB 6.994 million (see Table 3). The expected procurement cost for Category B under the joint discount is... The amount is 10,000 yuan, and it is recalculated by combining the two amounts:

[0183]

[0184]

[0185] because The joint proposal is expected to have a higher total procurement cost than the independent proposal (the price disadvantage caused by the delayed bidding of category A outweighs the benefits of the joint discount). The candidate joint pairing verification fails and is not included in the effective joint pairing set. The two categories will each execute bidding according to their independent optimal elastic window parameters.

[0186] Since the set of valid joint pairings is empty after verification in step 7, the incremental greedy algorithm determines that there is no valid joint pairing that reduces the total expected cost. The existing independent scheme remains unchanged, and the current combination scheme is directly output: Category A will be auctioned independently at the end of the second week, and Category B will be auctioned independently at the end of the fourth week. The total expected cost remains at 7.789 million yuan.

[0187] During the flexible window period, the system's dual-channel sliding correction continued to operate. In the middle of week 3, the system obtained real-time market price data for Category B at 10,310 yuan / ton, which was lower than the lower bound of the confidence interval of 10,390 yuan / ton output by the ARIMAX model for week 3. At the same time, the cumulative demand for Category B of 260 tons had met the minimum batching quantity of 200 tons. With both conditions met, the bidding start signal for Category B was triggered, and Category B initiated independent bidding ahead of schedule in the middle of week 3, without having to wait for the originally scheduled end of week 4.

[0188] If a valid joint pairing scenario occurs after the independent bidding for Category B, the incremental Shapley value approximation will be performed as follows: Assume the joint scheme includes a set of categories. The total revenue of the joint program before category B was added was equal to the independent discount amount of category A. Ten thousand yuan, total revenue from joint discounts after category B is added. The marginal contribution increment of category B is 10,000 yuan. The amount is 10,000 yuan, which is added to the historical marginal contribution cumulative value of Category B.

[0189] Step 502 completes the initial allocation between product categories based on the historical cumulative marginal contribution value of each product category group. Step 503 then performs a secondary allocation within Product Category B group based on each user's purchase amount percentage. Taking Product Category B group as an example, the proportions of net revenue allocated to users U05, U06, and U07 are as follows:

[0190]

[0191]

[0192]

[0193] Based on this, the final discount amount and amount payable for each user are calculated, and the transaction unit price, market benchmark price reference value and discount rate are integrated to generate a multi-user group purchase order and push it to the corresponding user and the winning supplier.

[0194] Table 4. Results of Secondary Allocation of Net Revenue for Category B Group Users (Example)

[0195]

[0196] Throughout the process, data starts from historical price time series and supplier capability data in step 1. A price forecast curve is generated using the ARIMAX model, and a joint supply mapping is constructed using an inverted index. This data flows into the real-time demand queue in step 2, forming cumulative demand and deadline constraints. In step 3, the two data streams are coupled to calculate the expected total procurement cost at each candidate time point and locate the optimal window (5.087 million yuan at the end of the second week for category A, and 2.702 million yuan at the end of the fourth week for category B). Step 6 utilizes the inverted index to complete the joint feasibility search in constant time and... The quantification of joint gains per 10,000 yuan: Step 7 calculates the difference between the actual comprehensive cost of the joint solution (9,515,000 yuan) and the independent solution (7,789,000 yuan) under a unified framework, which is -1,726,000 yuan. The verification fails. Step 8 maintains the independent solution and responds to the price change trigger signal to make category B bid earlier in the middle of week 3. Finally, steps 501 to 503 use incremental Shapley approximation to complete the two-level revenue sharing across categories and users. Data status is continuously transmitted throughout the process and the input of each step comes directly from the output of the previous step, ensuring the integrity and coherence of the decision-making chain.

[0197] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A multi-user group-buying and bidding method, characterized in that, Includes the following steps: Obtain historical market benchmark price time series data and historical group purchase data for each category, generate market price prediction curves for each category within a preset future time range, and construct a category-to-joint supply inverted index based on the multi-category supply capacity data of candidate suppliers. Process real-time user group-buying requests, update the cumulative demand for corresponding product categories, and generate a cumulative demand forecast sequence and a demand time elasticity distribution map for each product category; The market price forecast curves of each product category are coupled with the demand time elasticity distribution map for analysis. For each candidate window cutoff time, the supplier's expected quotation benchmark is calculated based on the predicted market benchmark price. The expected discount rate is obtained by querying the quantity-price ladder function in combination with the predicted cumulative demand. The expected total procurement cost is calculated. The optimal elasticity window parameters for each product category are solved with the goal of minimizing the expected total procurement cost. When a product category reaches the window deadline corresponding to the optimal elastic window parameter, the current determined demand is packaged into a centralized procurement bidding demand package, bidding is initiated to candidate suppliers, and the optimal supplier is selected based on the discount ratio to generate a centralized procurement transaction plan. Based on the group purchase transaction plan, the revenue is distributed according to the proportion of each user's purchase amount, and multi-user group purchase orders are generated.

2. The multi-user group-buying and bidding method according to claim 1, characterized in that, The generation of market price prediction curves for each product category within a preset future time range is specifically as follows: For each product category, an autoregressive integral moving average model with exogenous variables is used. The historical market benchmark price time series of each product category is used as the endogenous input, and the exchange rate, upstream raw material price index and industry prosperity index are used as the exogenous input. The output is the predicted market benchmark price value at each time point within the future preset time range and the corresponding upper and lower bounds of the confidence interval. The preset time range covers at least one complete price fluctuation cycle; The category-pair joint supply inverted index uses category pairs as index keys and a list of suppliers that can be jointly supplied and a joint discount rate as index values. The construction process is as follows: Iterate through the multi-category supply capacity data of all candidate suppliers. For each supplier that can supply two or more categories at the same time, extract the category pairs that can be jointly supplied, and add the supplier's identifier and the joint discount rate provided for the category pair to the value list of the corresponding category pair key.

3. The multi-user group-buying and bidding method according to claim 1, characterized in that, The process of generating the cumulative demand forecast sequence for each product category is as follows: using a Poisson process model to generate the predicted cumulative demand for each time point within a future preset time range based on the historical demand arrival time interval and the current demand arrival rate of the product category. The demand time elasticity distribution chart uses the time axis as the horizontal axis. At each time point, it marks the number of active users who have not exceeded the procurement deadline and the corresponding cumulative demand limit. As time goes by, the procurement deadline for some users expires, and the number of participating users and the cumulative demand limit gradually decrease.

4. The multi-user group-buying and bidding method according to claim 1, characterized in that, The calculation of the supplier's expected quotation benchmark based on the predicted market benchmark price is specifically as follows: the supplier's transaction quotation in the historical group purchase data is used to perform linear regression fitting with the market benchmark price at the corresponding time to obtain the regression coefficient. The predicted market benchmark price at the candidate window cutoff time is substituted into the linear regression relationship to obtain the supplier's expected quotation benchmark. The supplier's expected quotation benchmark is equal to the product of the regression slope coefficient and the predicted market benchmark price plus the regression intercept coefficient. The expected total procurement cost is equal to the product of the supplier's expected quotation benchmark and the predicted cumulative demand, multiplied by the discount factor of the independent discount rate. The optimal elastic window parameters are determined by using the upper bound of the time of the intersection of user purchase deadlines and the minimum group size as constraints. The upper bound of time is the earliest time of the purchase deadline of all users in the current active demand queue, and the minimum group size is the minimum batch size required by the supplier to accept the group purchase bid in the historical group purchase data.

5. The multi-user group-buying and bidding method according to claim 4, characterized in that, The quantity-price ladder function is a piecewise function fitted based on the relationship between different bulk purchase batches and corresponding discount rates in the historical group purchase data of this product category. When the cumulative demand falls into different quantity ranges, it corresponds to different discount rate levels. The larger the cumulative demand, the higher the discount rate. The lower and upper bounds of demand and the discount rate for each volume range are obtained by statistical fitting of the actual transaction records of each volume range in the historical group buying and procurement data.

6. The multi-user group-buying and bidding method according to claim 1, characterized in that, After processing real-time user group-buying requests, it also includes: Using the category identifier of the newly arriving demand as the query key, perform a search in the category pair joint supply inverted index to obtain a list of all candidate related categories that have joint supply opportunities with the category of the newly arriving demand; For each candidate related product category, obtain the cumulative demand of each of the two product categories and query the quantity-price ladder function to obtain their respective independent discount rates. Obtain the joint discount rate corresponding to the product category and calculate the incremental joint net income estimate. The incremental joint net income estimate is equal to the total joint purchase discount of the two product categories calculated according to the joint discount rate minus the sum of the independent purchase discounts of the two product categories calculated according to their respective independent discount rates. The total discount for joint procurement is equal to the product of the sum of the cumulative procurement amount benchmarks for the two categories and the joint discount rate. The discount for independent procurement of each category is equal to the product of the cumulative procurement amount benchmark for that category and the independent discount rate for that category. When the estimated incremental joint net profit exceeds the preset trigger threshold, the product category pair is marked as a candidate joint pairing. The preset trigger threshold is the mean of the distribution of the net profit difference between joint procurement and independent procurement in historical group purchase data.

7. The multi-user group-buying and bidding method according to claim 6, characterized in that, For each candidate joint pairing, obtain the market price prediction curves for each of the two categories. Calculate the sum of standardized distances between the prices of the two categories and their respective predicted troughs on the future time axis. The sum of standardized distances is equal to the sum of the absolute values ​​of the difference between the predicted market benchmark price and the predicted price trough for each of the two categories, divided by the predicted price trough. The predicted price trough is the minimum value among all predicted values ​​of the corresponding category's market price prediction curve within a preset future time range. The time point with the minimum sum of standardized distances is identified as the optimal time point for price coordination, and the reciprocal of the minimum sum of standardized distances is used as the score for cross-category price coordination. Calculate the sum of the expected total procurement cost of the joint scheme at the optimal time point for price coordination and the expected total procurement cost of the independent scheme at the time point corresponding to the optimal elasticity window parameters of each category. The expected total procurement cost of the joint scheme is equal to the sum of the products of the supplier's expected quotation benchmark and the predicted cumulative demand at the optimal time point for price coordination for the two categories, minus the joint discount amount calculated based on the joint discount rate. When the sum of the expected total procurement costs of independent solutions minus the expected total procurement costs of joint solutions is greater than zero, the candidate joint pairing is confirmed as valid, and a set of valid joint pairings and the suggested joint bidding time points for each candidate joint pairing are generated.

8. The multi-user group-buying and bidding method according to claim 7, characterized in that, It also includes using an incremental greedy algorithm to perform local optimization, the steps of which are as follows: Obtain all valid joint pairings involving the category of the newly arriving demand, and calculate the total expected cost change after incorporating each valid joint pairing into the existing combination scheme. The total expected cost change is obtained by uniformly adjusting the bidding time of the two categories involved in the candidate pairing to the optimal time point for price coordination corresponding to the candidate pairing, and recalculating the joint expected total procurement cost by replacing the individual discount rates with the joint discount rate, and subtracting the sum of the individual expected total procurement costs of the two categories before inclusion. If there is an effective joint pair that reduces the total expected cost, then the effective joint pair that reduces the total expected cost the most is included in the combination scheme and the current combination scheme and the corresponding total expected cost are updated. This process is repeated until there are no candidate pairs that can further reduce the total expected cost, and the current combination scheme is output. During the flexible window period, the market price forecast curve and demand accumulation forecast sequence of each product category are simultaneously corrected in a dual-channel sliding manner. When the current actual market benchmark price of a product category is lower than the lower bound of the confidence interval of the market price forecast curve and the current accumulated demand has met the minimum group purchase quantity, the bidding start signal of that product category is triggered. When a category with valid joint pairings experiences a sudden price change, the net profit difference between the joint and independent strategies is recalculated. If the independent bid is better, the category is removed from the valid joint pairings and an independent bid is triggered.

9. The multi-user group-buying and bidding method according to claim 7, characterized in that, The incremental Shapley value approximation method is used in the revenue sharing, specifically including: Maintain the historical cumulative marginal contribution value of each category group. When a new joint match occurs, only recalculate the marginal contribution increment of the category affected by this joint match. The marginal contribution increment is equal to the total revenue of the joint scheme after the category is added to the current joint scheme category set minus the total revenue of the joint scheme before the category is added. The total revenue of the joint scheme is calculated by multiplying the sum of the purchase amount benchmarks of each category participating in the joint scheme by the corresponding joint discount rate. The marginal contribution increment is added to the historical cumulative marginal contribution value of the category. The net revenue from joint procurement will be distributed among the participating product category groups based on the proportion of the historical marginal contribution of each product category group to the sum of the historical marginal contribution of all product category groups. Within each product category group, the net revenue allocated to that product category group is further distributed among the users based on the proportion of each user's purchase amount to the total purchase amount of that product category group, and the final discount amount and payable amount for each user are calculated.

10. A multi-user group-buying and bidding system, used to execute the multi-user group-buying and bidding method according to any one of claims 1 to 9, characterized in that, include: The price forecasting and index building module is used to obtain historical market benchmark price time series data and historical group purchase data for each category, generate market price forecast curves for each category within a preset time range, and build a category-to-joint supply inverted index based on the multi-category supply capacity data of candidate suppliers. The demand processing and forecasting module is used to process real-time user group buying demands, update the cumulative demand for corresponding product categories, and generate a forecast sequence of cumulative demand for each product category and a demand time elasticity distribution map. The coupling analysis and window optimization module is used to perform coupling analysis on the market price forecast curves and demand time elasticity distribution maps of each product category. It calculates the expected total procurement cost for each candidate window cutoff time and solves for the optimal elasticity window parameters for each product category with the goal of minimizing the expected total procurement cost. The bidding execution module is used to encapsulate the current determined demand into a centralized procurement bidding demand package when a certain product category reaches the window deadline corresponding to the optimal elastic window parameter, initiate bidding to candidate suppliers, and select the optimal supplier based on the discount ratio to generate a centralized procurement transaction plan. The revenue sharing and order generation module is used to distribute revenue according to the proportion of each user's purchase amount based on the collective procurement transaction plan, and generate multi-user group purchase orders.

Images