Merchant purchase recommendation and fund pre-allocation recommendation method

Abstract

The invention discloses a merchant purchase recommendation and fund pre-allocation recommendation method, and particularly relates to the technical field of market analysis, and the method comprises the following steps: obtaining merchant inventory data, fund data, and market data of various commodities in a preset commodity list; analyzing market demands based on the time sequence prediction model, generating a demand prediction value of the market, screening out all to-be-purchased commodities from a preset commodity list according to the demand prediction value, and summarizing the to-be-purchased commodities into a preliminary purchasing list; performing fund pre-allocation optimization based on the preliminary purchase list, the merchant inventory data and the fund data, and generating a final purchase list and a fund allocation scheme; according to the method, the market demand is accurately predicted through the time sequence prediction model, external factor data is introduced, and the dynamic demand change of the market is effectively captured, so that the demand prediction result is more comprehensive and accurate, unit fund income maximization is realized, and merchants are helped to obtain higher economic returns under the budget constraint.

Description

Technical Field

[0001] The present invention relates to the technical field of market analysis, and more specifically, to a merchant purchase recommendation and fund pre-allocation recommendation method. Background Art

[0002] In the daily operation of merchants, purchasing decisions and capital allocation are key links to ensure stable commodity supply and efficient capital utilization. With the fluctuation of market demand and the shortening of commodity life cycle, merchants are faced with problems such as inventory management, capital optimization and unreasonable purchasing decisions, which lead to rising operating costs, mismatch between commodity supply and demand and waste of resources.

[0003] Traditional procurement methods usually make procurement decisions based on historical sales data or manual experience, and fail to effectively combine market dynamic data, resulting in delayed procurement results and difficulty in responding to changes in market demand. Traditional capital allocation methods usually use fixed ratios or manual experience allocations, and fail to combine the expected returns and procurement costs of goods, resulting in poor decision-making quality. The capital budget is limited, but there is a gap between demand forecasts and actual supply. Existing methods fail to find a balance between maximizing returns and budget constraints. The core goal of capital allocation is to maximize the rate of return per unit of capital, and traditional methods fail to systematically quantify and optimize this goal.

[0004] Therefore, the present invention proposes a merchant procurement recommendation and fund pre-allocation recommendation method, in order to achieve more scientific and efficient procurement recommendation and fund allocation, meet market demand, and improve the overall operational efficiency of merchants. Summary of the invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A merchant purchase recommendation and fund pre-allocation recommendation method comprises the following steps:

[0007] Obtain merchant inventory data, capital data, and market data of various commodities in the preset commodity list;

[0008] Analyze market demand based on the time series forecasting model, generate market demand forecast values, and select all the goods to be purchased from the preset commodity list based on the demand forecast values, and summarize them into a preliminary purchase list;

[0009] Optimize fund allocation based on preliminary purchase list, merchant inventory data and fund data to generate the final purchase list and fund allocation plan.

[0010] In a preferred embodiment, the time series forecasting model generates future commodity demand forecast values ​​through seasonal decomposition and trend analysis. t represents the current time, and h represents the prediction time span.

[0011] In a preferred embodiment, the use principle of the time series prediction model is:

[0012] Get historical demand dataX t , historical error data∈ t , and external factor data F t ;

[0013] Preprocessing of historical data, including seasonal decomposition, smoothing, and normalization;

[0014] Based on the autoregressive term, moving average term and external factor term, a prediction model is established, and the prediction formula is as follows: c is a constant term, i, j, and k represent the serial number of the historical time data in the autoregressive term, the serial number of the historical error in the moving average term, and the serial number of the historical time data in the external factor term, respectively. p is the order of the autoregressive term, q is the order of the moving average term, and m is the order of the external factor term.

[0015] Train the model through historical data and estimate the parameters θ j ,δ k ;

[0016] Using the trained model, input the current data and output the demand forecast value at the next h time points

[0017] The significance of time series forecasting model is reflected in:

[0018] Capturing long-term trends and cyclical changes: The AR part helps understand the long-term trends and cyclical fluctuations in the data, allowing the model to continuously track changes in demand.

[0019] Reduced Error: The MA part can correct the model prediction error and reduce the impact of the model’s failure to fully fit the data.

[0020] Supplement external influences: By introducing external factors, the model can better adapt to non-cyclical fluctuations in market demand, especially when demand is affected by non-time series factors such as external markets and events.

[0021] Improve forecasting accuracy: By comprehensively considering historical data, errors and external factors, the forecasting accuracy and robustness of the model can be significantly improved, helping decision makers make more accurate business plans and resource allocation.

[0022] By combining autoregression, moving average and external factors, the time series forecasting model can not only better fit historical data, but also adapt to future changes and improve the accuracy and stability of the forecast.

[0023] In a preferred embodiment, the historical error data ∈t The historical error data is usually calculated and fitted by the least squares method. It refers to the deviation between the previous model prediction value and the actual value. These errors can reflect the model's prediction ability and the deviation or deficiency in the prediction process. By using historical error data, the model can better understand past prediction problems and make more accurate adjustments in future predictions.

[0024] In a preferred embodiment, the external factor data F t It is calculated by combining the social media public opinion coefficient and the search heat coefficient. The calculation formula is:

[0025] P t is the number of comments on the product in the current time period, which measures the discussion popularity of the product. t is the positive evaluation rate, reflecting the user's recognition of the product, P avg The historical average number of comments is used as a benchmark for comparison. The comment popularity is combined with the user's positive comments to indicate the market feedback of the product. The higher the positive comment rate, the greater the positive impact on the market demand of the product. t is the search volume of the product in the current time period, indicating the intensity of user attention, S max It is the highest search volume in history. As a standardized benchmark for search popularity, it directly reflects the market's real-time attention to the product. The higher the search volume, the greater the market demand potential. w1 is the public opinion weight, and w2 is the popularity weight, which are used to adjust the contribution ratio of the two parts in the external factor data.

[0026] In a preferred embodiment, the determination logic of the public opinion weight w1 and the heat weight w2 is:

[0027] XGi is the historical correlation between social media public opinion or search popularity and actual commodity demand, which is calculated by the Pearson correlation coefficient. When i is 1, the corresponding w1 is the public opinion weight; when i is 2, the corresponding w2 is the popularity weight. XGH is the sum of the historical correlations between social media public opinion and search popularity and actual commodity demand.

[0028] In a preferred embodiment, screening out all commodities to be purchased from a preset commodity list refers to:

[0029] The future commodity demand forecast value output by the time series forecasting model Compared with the preset safety stock and the current stock, if the demand forecast value If the current inventory exceeds the current inventory and is lower than the safety inventory, procurement is triggered and the product is marked as a product to be purchased.

[0030] In a preferred embodiment, optimizing fund pre-allocation based on the preliminary purchase list, merchant inventory data, and fund data refers to:

[0031] Get the total budget from the preliminary purchase list, merchant inventory data, and capital data. The preliminary purchase list consists of the purchase demand and the unit price of the goods. n ] as chromosome, where qi is the purchase quantity of commodity i;

[0032] Constraint: 0≤q i ≤q max，i ; where qmax,i is the maximum demand gap of commodity i, and the demand forecast value is obtained by The maximum value between the difference between the current inventory and the set value zero is obtained;

[0033] Randomly generate N initial solutions, i.e. chromosomes, to form an initial population;

[0034] The fitness is calculated for each chromosome to measure the pros and cons of the current funding allocation plan. The fitness formula is:

[0035] B is the total budget in the funding data, i represents the product number, ρ i represents the expected unit revenue of commodity i, q i represents the purchase quantity of commodity i, y i Indicates the purchase price of commodity i, f1 and f2 are both preset non-zero proportional coefficients;

[0036] Roulette wheel selection method is used to select the better chromosomes according to fitness to enter the next generation population, and crossover and mutation are performed on the selected chromosomes. At the same time, for the newly generated chromosomes, the total fund allocation is ensured to meet the constraint conditions: the total amount of funds used in the fund allocation plan is less than or equal to the total budget in the fund data;

[0037] When the number of iterations T is met or the fitness converges and no longer improves, the iteration is stopped and the chromosome with the highest fitness is selected as the final fund allocation plan.

[0038] Technical effects and advantages of the present invention:

[0039] The present invention uses a time series forecasting model to accurately forecast market demand, overcoming the defects of traditional methods that rely on historical data and cannot reflect dynamic changes in the market. It introduces external factor data (such as social media public opinion and search popularity) to effectively capture the dynamic changes in market demand, making the demand forecast results more comprehensive and accurate. The demand forecast value is compared with the current inventory and safety inventory, and the goods to be purchased are automatically screened out to form a preliminary purchase list to ensure that the purchase decision meets actual demand and avoids excess or shortage of inventory. Through automated screening, the error and decision-making cost of manual judgment are reduced, and work efficiency is improved.

[0040] Through the genetic algorithm, the procurement funds are intelligently optimized and allocated to maximize the expected return per unit of funds and achieve efficient use of funds. The fitness formula balances the maximization of procurement returns and budget constraints, ensuring that the fund allocation plan meets the actual operating conditions and reduces the risk of fund waste. According to the unit expected return and procurement cost of the goods, the procurement needs of high-yield goods are prioritized, which improves the economic benefits of the overall procurement plan.

[0041] The public opinion weight and heat weight in the external factor data are dynamically determined based on historical correlation, which can adapt to changes in the market environment and improve the adaptability and accuracy of the prediction model. By dynamically adjusting the weights, market feedback can be captured in real time to optimize the accuracy and flexibility of procurement decisions. The present invention automates the entire process from demand forecasting, commodity screening to fund allocation through algorithms, reducing manual participation and reducing subjective judgment errors. The optimized procurement list and fund allocation plan are scientific, reasonable and feasible, which significantly improves the efficiency and quality of decision-making. By accurately predicting market demand, scientifically screening commodities to be purchased and optimizing fund allocation, the present invention maximizes the unit capital income, helps merchants obtain higher economic returns under budget constraints, and reduces inventory management costs and out-of-stock losses by avoiding inventory backlogs and out-of-stocks. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0043] Figure 1 A schematic diagram of a merchant purchase recommendation and fund pre-allocation recommendation method in the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Reference Figure 1 The following embodiments are obtained:

[0046] Embodiment 1:

[0047] A merchant purchase recommendation and fund pre-allocation recommendation method comprises the following steps:

[0048] Obtain merchant inventory data, capital data, and market data for various commodities in the preset commodity list;Merchant inventory data: Understand the merchant's current inventory level to determine which commodities are in short supply and which are in sufficient supply. Inventory data helps determine whether purchases are needed or whether demand has been met. Capital data: The merchant's capital situation directly determines the amount of available funds, so the capital budget must be clearly defined to ensure that the procurement plan does not exceed the merchant's financial capabilities. Commodity market data: Commodity market data such as demand trends provide a basis for predicting future demand. These data help understand market conditions and consumer preferences, so as to make reasonable purchasing decisions.

[0049] Analyze market demand based on the time series forecasting model, generate market demand forecasts, and based on the demand forecasts, select all the goods to be purchased from the preset list of goods and summarize them into a preliminary purchase list; predict future market demand by analyzing historical demand data and external factors (such as social media sentiment, search popularity, etc.). This is the core part of the procurement decision because it provides data-based market demand predictions to help merchants prepare for upcoming demand. Through the time series analysis model, factors such as seasonality, trend and cyclicality can be identified, so as to make more accurate predictions.

[0050] Filtering of goods to be purchased: Based on demand forecasts, combined with the merchant's existing inventory and safety stock (i.e. the minimum inventory that the merchant believes is required to meet operations), goods with demand greater than the current inventory and less than the safety stock can be automatically filtered out and marked as "goods to be purchased." This step helps merchants identify which goods need to be replenished to avoid sales losses due to out-of-stock situations.

[0051] Based on the preliminary purchase list, merchant inventory data, and fund data, fund pre-allocation optimization is performed to generate the final purchase list and fund allocation plan. After determining which goods need to be purchased, the next step is to consider how to make the best allocation with the existing funds. This process requires comprehensive consideration of the merchant's financial situation (such as the total budget) and the purchase unit price of each product, expected revenue and other factors. The goal of fund pre-allocation optimization is to ensure that the procurement plan can maximize the merchant's profits while avoiding exceeding budget constraints. Commonly used optimization methods (such as genetic algorithms) simulate different combinations of procurement decisions to find the best solution.

[0052] Generate the final purchase list and fund allocation plan: After optimizing the fund allocation, a plan containing the specific commodity purchase quantity and the corresponding fund allocation will be generated. The final purchase list not only takes into account market demand and inventory conditions, but also takes into account the fund budget and the profit potential of the commodity to ensure the scientific nature and financial feasibility of the purchase decision.

[0053] The time series forecasting model generates future commodity demand forecasts through seasonal decomposition and trend analysis. t represents the current moment, and h represents the forecast time span. Seasonal decomposition in the time series model breaks down the data into three parts: the trend part describes the direction of long-term changes, such as the growth or decline of demand, the seasonal part reflects cyclical changes, such as the impact of festivals and seasons on demand, and the residual part represents random fluctuations or outliers.

[0054] The principle of using the time series forecasting model is:

[0055] Get historical demand dataX t , historical error data∈ t , and external factor data F t ;

[0056] Preprocess historical data, including seasonal decomposition, smoothing and standardization; seasonal decomposition is to decompose the original data into trend, seasonality and residual, data smoothing is to remove noise, data standardization or normalization to ensure model training convergence.

[0057] Based on the autoregressive term, moving average term and external factor term, a prediction model is established, and the prediction formula is as follows: c is a constant term, i, j, and k represent the serial number of the historical time data in the autoregressive term, the serial number of the historical error in the moving average term, and the serial number of the historical time data in the external factor term, respectively. p is the order of the autoregressive term, q is the order of the moving average term, and m is the order of the external factor term.

[0058] is the autoregressive term, AR part, through the historical data X of the previous p time points t+h-i To predict future demand, here X t Includes trend and seasonal information, because the original data X t The decomposed characteristics are already implied in the model, and the autoregressive term can capture the long-term trend and continuous changes of the data.

[0059] is the moving average term, the MA part, taking into account the historical forecast error ∈ t+h-jImpact on the future. The residual can be regarded as a random fluctuation part, which reflects the part that the model cannot fit, and the moving average term is used to correct the uncertainty of this part.

[0060] It is an external factor item, which takes into account external factors related to commodity demand, such as social media public opinion and search popularity. External factors supplement the information that cannot be captured in time series data, especially when market demand is affected by non-cyclical factors. External factors reflect nonlinear dynamic effects other than seasonality. It plays a complementary role, further improving the accuracy of the forecast results and enabling the model to better adapt to market fluctuations.

[0061] Train the model through historical data and estimate the parameters θ j , δ k ;

[0062] Using the trained model, input the current data and output the demand forecast value at the next h time points

[0063] Historical error data∈ t The calculation and fitting were performed using the least squares method.

[0064] External Factor Data F t It is calculated by combining the social media public opinion coefficient and the search heat coefficient. The calculation formula is:

[0065] P t is the number of product reviews in the current time period, R t is the positive evaluation rate, P avg is the historical average number of comments, S t is the search volume of the product in the current time period, S max is the highest search volume in history, w1 is the public opinion weight, and w2 is the heat weight.

[0066] The logic for determining the public opinion weight w1 and the heat weight w2 is:

[0067] XGi is the historical correlation between social media public opinion or search heat and actual commodity demand, which is calculated by Pearson correlation coefficient. When i is 1, the corresponding w1 is the public opinion weight, and when i is 2, the corresponding w2 is the heat weight. XGH is the sum of the historical correlations between social media public opinion and search heat and actual commodity demand. Through historical correlation analysis, the actual impact of public opinion and heat on commodity demand is determined, so as to assign reasonable weights and ensure that the formula is more scientific and dynamic. By comprehensively considering social media public opinion (comment volume and evaluation rate) and search heat (user attention), the dynamic impact of external factors on commodity demand is effectively captured. Through historical correlation analysis, the public opinion weight w1 and heat weight w2 are adaptively adjusted to ensure that the weight distribution matches the actual market demand law. The data of comment volume, evaluation rate and search volume are standardized to eliminate the dimensionality effect, make different data sources comparable, and improve the reliability of the external factor index. Inputting external factor data into the demand forecasting model and combining it with historical demand data can achieve more accurate future demand forecasts, which helps to optimize procurement decisions and inventory management.

[0068] Filtering all the items to be purchased from the preset item list means:

[0069] The future commodity demand forecast value output by the time series forecasting model Compared with the preset safety stock and the current stock, if the demand forecast value If the current inventory exceeds the current inventory and the current inventory is lower than the safety inventory, procurement is triggered and the product is marked as a product to be purchased. Through the future commodity demand forecast output by the time series forecasting model, the enterprise can accurately predict the future commodity demand. By comparing with the current inventory and safety inventory, it is ensured that the commodity inventory can meet future demand.

[0070] The preset safety stock is to prevent out-of-stock situations caused by supply chain fluctuations or other unpredictable factors. If the demand forecast exceeds the current inventory and the current inventory is lower than the safety stock, it means that there is an imminent risk of insufficient inventory, so procurement is required. The goal of procurement management is to avoid low inventory, which leads to failure to meet customer needs and loss of sales opportunities; at the same time, it is also necessary to avoid excessive inventory, which leads to capital occupation, inventory expiration, storage costs and other problems. If the current inventory is lower than the safety stock and the demand forecast exceeds the existing inventory, procurement action is triggered, so that inventory can be replenished in time to maintain a balance between supply and demand. Through accurate demand forecasting and timely procurement decisions, enterprises can effectively reduce unnecessary procurement and optimize inventory levels. Avoiding blind over-purchasing can reduce warehousing and management costs. The mechanism of automated screening of goods to be purchased can improve the response speed of the supply chain, shorten the supply chain cycle, ensure the timely supply of goods, and improve overall operational efficiency.

[0071] Timely procurement can reduce the risk of supply disruptions caused by inventory shortages, ensure that products can always meet market demand, and avoid customer loss due to out-of-stock. At the same time, ensuring that goods are delivered on time helps improve customer satisfaction and enhance customer loyalty to the brand. This process relies on a time series forecasting model, which analyzes future demand changes through historical data and can accurately predict demand fluctuations. This avoids the bias caused by manual judgment and improves the scientificity and accuracy of decision-making. Systematic forecasting and procurement strategies help improve the responsiveness and flexibility of enterprises and enable them to respond quickly to market changes. Through refined management of inventory and procurement, capital occupation can be reduced and capital liquidity can be improved. Through reasonable procurement strategies and inventory management, enterprises can reasonably control costs and optimize financial conditions while avoiding over-purchasing.

[0072] Funds pre-allocation optimization based on preliminary purchase list, merchant inventory data, and fund data refers to:

[0073] Get the total budget from the preliminary purchase list, merchant inventory data, and capital data. The preliminary purchase list consists of the purchase demand and the unit price of the goods. n ] as chromosome, where qi is the purchase quantity of commodity i;

[0074] Constraint: 0≤q i ≤q max，i ; where qmax,i is the maximum demand gap of commodity i, and the demand forecast value is obtained by The maximum value between the difference between the current inventory and the set value zero is obtained;

[0075] Randomly generate N initial solutions, i.e. chromosomes, to form an initial population;

[0076] The fitness function measures the pros and cons of the current funding allocation plan. The goal is to maximize the unit procurement revenue and ensure that the use of funds meets the constraints. The fitness is calculated for each chromosome to measure the pros and cons of the current funding allocation plan. The fitness formula is:

[0077] B is the total budget in the funding data, i represents the product number, ρ i represents the expected unit revenue of commodity i, q i represents the purchase quantity of commodity i, y i Indicates the purchase price of commodity i, f1 and f2 are both preset non-zero proportional coefficients;

[0078] Roulette wheel selection method is used to select the better chromosomes according to fitness to enter the next generation population, and the selected chromosomes are crossovered and mutated to generate new chromosomes to maintain population diversity. At the same time, for the newly generated chromosomes, the total fund allocation is ensured to meet the constraint conditions: the total amount of funds used in the fund allocation plan is less than or equal to the total budget in the fund data;

[0079] When the number of iterations T is met or the fitness converges and no longer improves, the iteration is stopped and the chromosome with the highest fitness is selected as the final fund allocation plan.

[0080] Through genetic algorithms, we ensure that high-yield commodities are purchased first under a limited budget to achieve optimal allocation of funds. Through the fitness formula, we not only pursue the maximization of unit revenue, but also use the penalty item constraint scheme to meet the budget conditions to ensure the feasibility of the scheme. The crossover and mutation operations of genetic algorithms enable the scheme to evolve continuously, overcome the local optimal problem, and find the global optimal solution. Combined with the preliminary purchase list, demand forecast and inventory data, the optimal purchase list is generated through intelligent optimization to avoid waste of funds and inventory backlog.

[0081] The final purchase list and fund allocation plan refers to the intelligent optimization of the commodity demand and fund allocation in the preliminary purchase list, under the premise of meeting the capital budget constraints, merchant inventory requirements and commodity procurement optimization goals, to determine the final commodities to be purchased and their purchase quantities, and to reasonably allocate funds to ensure the maximum utilization of resources and the best benefits. Purchase list: refers to a list of commodities to be purchased and their specific purchase quantities, ensuring that the conditions for meeting market demand and safe inventory replenishment are met.

[0082] Fund allocation plan: refers to the reasonable allocation of procurement funds within the merchant's fund budget B to ensure that the total funds for the purchased goods do not exceed the budget, while maximizing the unit fund income.

[0083] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0084] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0085] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0087] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A merchant purchase recommendation and fund pre-allocation recommendation method, characterized in that: The following steps are involved: Obtain merchant inventory data, capital data, and market data of various commodities in the preset commodity list; Analyze market demand based on the time series forecasting model, generate market demand forecast values, and select all the goods to be purchased from the preset commodity list based on the demand forecast values, and summarize them into a preliminary purchase list; Optimize fund allocation based on preliminary purchase list, merchant inventory data and fund data to generate the final purchase list and fund allocation plan.

2. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 1, characterized in that: The time series forecasting model generates future commodity demand forecasts through seasonal decomposition and trend analysis. t represents the current time, and h represents the prediction time span.

3. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 2, characterized in that: The working principle of the time series prediction model is as follows: Get historical demand dataX t , historical error data∈ t , and external factor data F t ; Preprocessing of historical data, including seasonal decomposition, smoothing, and normalization; Based on the autoregressive term, moving average term and external factor term, a prediction model is established, and the prediction formula is as follows: c is a constant term, i, j, and k represent the serial number of the historical time data in the autoregressive term, the serial number of the historical error in the moving average term, and the serial number of the historical time data in the external factor term, respectively. p is the order of the autoregressive term, q is the order of the moving average term, and m is the order of the external factor term. Train the model through historical data and estimate the parameters θ j , δ k ; Using the trained model, input the current data and output the demand forecast value at the next h time points 4. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 3, characterized in that: Historical error data∈ t The calculation and fitting were performed using the least squares method.

5. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 4, characterized in that: External Factor Data F t It is calculated by combining the social media public opinion coefficient and the search heat coefficient. The calculation formula is: P t is the number of product reviews in the current time period, R t is the positive evaluation rate, P avg is the historical average number of comments, S t is the search volume of the product in the current time period, S max is the highest search volume in history, w1 is the public opinion weight, and w2 is the heat weight.

6. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 5, characterized in that: The logic for determining the public opinion weight w1 and the heat weight w2 is: XGi is the historical correlation between social media public opinion or search popularity and actual commodity demand, which is calculated by the Pearson correlation coefficient. When j is 1, the corresponding w1 is the public opinion weight. When i is 2, the corresponding w2 is the heat weight. XGH is the sum of the historical correlations between social media public opinion and search popularity and actual commodity demand.

7. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 6, characterized in that: Filtering all the items to be purchased from the preset item list means: The future commodity demand forecast value output by the time series forecasting model Compared with the preset safety stock and the current stock, if the demand forecast value If the current inventory exceeds the current inventory and is lower than the safety inventory, procurement is triggered and the product is marked as a product to be purchased.

8. A merchant purchase recommendation and fund pre-allocation recommendation method according to claim 7, characterized in that: Funds pre-allocation optimization based on preliminary purchase list, merchant inventory data, and fund data refers to: Get the total budget from the preliminary purchase list, merchant inventory data, and capital data. The preliminary purchase list consists of the purchase demand and the unit price of the goods. n ] as chromosome, where qi is the purchase quantity of commodity i; Constraint: 0≤q i ≤q max，i ; where qmax,i is the maximum demand gap of commodity i, and the demand forecast value is obtained by The maximum value between the difference between the current inventory and the set value zero is obtained; Randomly generate N initial solutions, i.e. chromosomes, to form an initial population; The fitness is calculated for each chromosome to measure the pros and cons of the current funding allocation plan. The fitness formula is: B is the total budget in the funding data, i represents the product number, ρ i represents the expected unit revenue of commodity i, q i represents the purchase quantity of commodity i, y i Represents the purchase price of commodity i, where f1 and f2 are both preset non-zero proportional coefficients; Roulette wheel selection method is used to select the better chromosomes according to fitness to enter the next generation population, and crossover and mutation are performed on the selected chromosomes. At the same time, for the newly generated chromosomes, the total fund allocation is ensured to meet the constraint conditions: the total amount of funds used in the fund allocation plan is less than or equal to the total budget in the fund data; When the number of iterations T is met or the fitness converges and no longer improves, the iteration is stopped and the chromosome with the highest fitness is selected as the final fund allocation plan.

Images