A Wholesale Demand Forecasting Method Based on Machine Learning Algorithms

Abstract

The present invention discloses a method for predicting wholesale demand based on a machine learning algorithm, including: identifying seasonal commodities according to historical sales data of commodities; identifying complementary commodities of seasonal commodities according to historical sales data of commodities; judging the time synchronization and time lag of demand changes of complementary commodities and seasonal commodities; calculating the seasonal index of seasonal commodities and identifying the market stage of complementary commodities; identifying the complementary pattern of seasonal commodities and their complementary commodities in different market stages according to the sales data of commodities; identifying the market stage of complementary commodities in a preset time period in the future, generating a wholesale purchase plan for seasonal commodities and their complementary commodities; judging the execution effect of the current wholesale purchase plan, and adjusting the parameters of the wholesale purchase plan. The present invention can effectively improve the supply chain response speed and market adaptability in wholesale trade, and help enterprises maintain their competitive advantages in a complex and changing market environment.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce, and in particular to a wholesale demand forecasting method based on a machine learning algorithm. Background Art

[0002] In the field of wholesale trade, the complexity of supply chain management is increasing day by day, and enterprises are facing increasingly severe challenges in market demand fluctuations and commodity sales forecasting. Existing supply chain management and wholesale demand forecasting methods usually rely on simple statistical models and empirical judgments, which are difficult to cope with complex and changing market environments. The limitations of these traditional methods make it difficult for enterprises to make accurate decisions when facing seasonal commodities and their complementary commodities, resulting in problems such as supply and demand imbalance, inventory backlog or out-of-stock, which seriously affect the profitability and market competitiveness of enterprises. First, seasonal commodities have significant demand volatility, and their sales cycle is often closely related to seasonal changes. However, when identifying and responding to these cyclical demand changes, enterprises usually rely on simple average or trend analysis of historical sales data, and are unable to deeply explore the potential sales rules of commodities. This rough forecasting method is difficult to respond quickly when encountering sudden market changes or abnormal consumer behavior, resulting in missed sales opportunities or excessive inventory backlogs. Secondly, seasonal commodities often have a close sales relationship with other complementary commodities, such as beverages and barbecue supplies in summer, heating equipment and warm clothing in winter, etc., and the sales performance of these commodities has a certain correlation. However, traditional demand forecasting and inventory management methods often ignore this complementary relationship between commodities. Even if the correlation between commodities is taken into account, the existing analysis methods are mostly based on static correlation analysis, which cannot accurately capture the time synchronization and lag of commodity demand changes. This practice of ignoring the dynamic relationship between commodities can easily lead to enterprises being caught off guard when dealing with demand fluctuations, which in turn affects the overall supply chain efficiency. In addition, the impact of market stages on commodity demand is also very important. The demand changes in different market stages, such as the introduction period, growth period, maturity period and decline period, are different, but enterprises often find it difficult to accurately identify these market stages in actual operations. Existing market stage identification usually relies on empirical judgment or simple sales data segmentation, lacking dynamic analysis of the commodity life cycle. This results in enterprises being unable to effectively adjust sales strategies and replenishment plans when facing different market stages, thus missing opportunities to optimize inventory and increase sales. Even more complicated is the problem of time synchronization and lag of commodity demand, which refers to whether changes in demand for a certain commodity will affect the demand for its complementary commodity in the same time period, or whether this impact has a time lag effect. Traditional forecasting and analysis tools are usually unable to accurately capture and quantify this complex temporal relationship, which makes it difficult for companies to adjust replenishment and inventory strategies in a timely manner when facing a rapidly changing market, which may eventually lead to a series of problems such as supply and demand mismatch and chaotic inventory management. Therefore, existing supply chain management and wholesale demand forecasting methods have significant limitations when dealing with complex changes in market demand for seasonal goods and their complementary goods.These limitations are mainly reflected in the neglect of the complex relationship between commodities, the insufficient identification of market stages, and the lack of effective analytical tools for the time synchronization and lag of demand changes. These problems seriously restrict the ability of enterprises to improve supply chain efficiency and profitability in the modern competitive market environment. Therefore, a more accurate and efficient solution is urgently needed to meet these challenges. Summary of the invention

[0003] The present invention solves the problems existing in the above-mentioned prior art and provides a wholesale demand forecasting method based on a machine learning algorithm, which mainly includes:

[0004] Based on the historical sales data of the product, determine whether the sales changes of the product have cyclical characteristics and identify seasonal products;

[0005] Based on the historical sales data of the products, identify the complementary products of seasonal products and predict the sales data of seasonal products and complementary products in the future preset time period;

[0006] Based on the sales data of seasonal goods and complementary goods in a preset time period in the future, the cross-correlation coefficients between seasonal goods and complementary goods at different lag times are calculated, and the time synchronization and time lag of the demand changes of complementary goods and seasonal goods are determined;

[0007] Calculate the seasonal index of seasonal commodities based on the historical sales data of seasonal commodities, and build a market stage identification model for complementary commodities to identify the market stage of complementary commodities;

[0008] According to the sales data of commodities, the complementarity coefficients of seasonal commodities and complementary commodities at different time points are calculated, and based on the time series of dynamic complementarity coefficients, the complementarity patterns of seasonal commodities and their complementary commodities in different market stages are identified;

[0009] Based on the sales data of seasonal commodities and their complementary commodities in the future preset time period, the market stage of complementary commodities in the future preset time period is identified, and the wholesale purchase plan of seasonal commodities and their complementary commodities is generated by combining the evaluation results of complementarity pattern, time synchronization and time lag;

[0010] Based on the actual sales data and inventory change data of seasonal and complementary goods, determine the execution effect of the current wholesale purchasing plan, identify the goods and time periods that need to be adjusted, and adjust the parameters of the wholesale purchasing plan.

[0011] Furthermore, judging whether the sales change of the commodity has a periodic characteristic based on the historical sales data of the commodity and identifying seasonal commodities includes:

[0012] Based on the historical sales data of the goods, obtain sales records covering multiple quarters and years, including product name, sales date, daily sales quantity, sales amount and customer information, and arrange them in chronological order; process the obtained historical sales data through data cleaning technology, filter out abnormal values ​​and missing data, unify the data format, and use missing value interpolation methods to process missing items in the data. Missing value interpolation methods include mean interpolation, interpolation method or filling method based on adjacent data points; obtain the cleaned historical sales data, use time series analysis algorithm to perform moving average and exponential smoothing on the historical sales data to obtain the long-term trend of the time series, and use the difference method to remove the trend component to obtain the seasonal component and random error; based on the seasonal component, use the fast Fourier transform method to calculate the amplitude and frequency of each cycle to determine whether the sales changes of the goods have periodic characteristics; if the sales changes of the goods have periodic characteristics, then determine that the goods are seasonal goods.

[0013] Furthermore, the method of identifying complementary commodities of seasonal commodities based on historical sales data of commodities and predicting sales data of seasonal commodities and complementary commodities in a preset time period in the future includes:

[0014] The historical sales data of commodities are grouped by transactions to form a transaction database, in which each transaction record contains information about all commodities in a purchase; according to the transaction database, a support threshold is set, and the Apriori algorithm is used to identify commodity combinations that appear simultaneously in transactions at a frequency higher than a preset frequency, obtain a list of frequent itemsets, and record the support of each item set; according to the frequent item set list, the confidence of each commodity combination is calculated, the strength of association between commodities is judged, commodity pairs with confidence higher than the set threshold are obtained, complementary commodities of seasonal commodities are screened out, and their support and confidence are recorded; the historical sales data of seasonal commodities and complementary commodities are obtained through the sales record database, and are sorted into a multivariate time series format in chronological order to form time series data; according to the time series data, the long short-term memory network algorithm is used for model training to predict the sales data of seasonal commodities and complementary commodities in a preset time period in the future, and generate a demand forecast curve for seasonal commodities and complementary commodities in a preset time period in the future.

[0015] Furthermore, the cross-correlation coefficients between seasonal commodities and complementary commodities at different lag times are calculated based on the sales data of seasonal commodities and complementary commodities in a preset time period in the future, and the time synchronization and time lag of the demand changes of complementary commodities and seasonal commodities are determined, including:

[0016] According to the sales data of seasonal commodities and their complementary commodities in the future preset time period, the sales volume time series data of each commodity is obtained; the sliding average method is used to set the sliding window size, and the moving average is calculated according to the sales data before and after each time point to obtain the smoothed sales volume time series; according to the smoothed sales volume time series, the formula is used Calculate the sales growth rate of each product at each time point and construct a sales growth rate time series for each product, where G(t) is the sales growth rate at time t, S(t) and S(t-1) are the sales volumes at time t and t-1 respectively; Based on the sales growth rate time series and the preset time lag range, use the cross-correlation coefficient formula Calculate the cross-correlation coefficient R between seasonal products and complementary products at different lag times τ XY (τ), and obtain the cross-correlation coefficient matrix of each commodity pair, where G X (i) and G Y (i) represents the sales growth rate time series of seasonal products and complementary products, and Its average value, σG X and σG Y is its standard deviation; the maximum correlation coefficient R for each pair of commodities in the cross-correlation coefficient matrix ij (τ max ) to determine the lag time τ corresponding to the maximum correlation coefficient max ; If the lag time τ corresponding to the maximum correlation coefficient max =0, and R ij (τ max ) exceeds the preset lag time threshold, it is judged that the demand changes of seasonal products and the complementary products are time synchronized; if τ max >0, and R ij (τ max ) exceeds the set lag time threshold, it is judged that there is a positive time lag between the seasonal product and the complementary product; if τ max <0 and R ij (τ max ) exceeds the set threshold, it is judged that there is a negative time lag between the seasonal product and the complementary product.

[0017] Further, the seasonal index of the seasonal commodity is calculated based on the historical sales data of the seasonal commodity, and a market stage identification model of the complementary commodity is constructed to identify the market stage of the complementary commodity, including:

[0018] Get the historical sales data of seasonal commodities for multiple time periods and arrange them in chronological order to form time series data; use the formula Calculate the overall average sales volume of the entire time series data, where N is the number of time points and S(t) is the sales volume; according to the moving average formula Calculate the moving average sales volume MA(t) within a preset time period, where n is the number of time points in a period; based on the original sales volume S(t) and the moving average sales volume MA(t), use the formula Get the detrended seasonal sales volume D(t) and calculate it through the seasonal index formula Calculate the seasonal index of seasonal commodities, where D(t+u×v) represents the detrended data at the same time point in different years or cycles, v is the length of a cycle, m is the number of complete seasons, and u is the index of each season, representing from the first complete season to the mth complete season; obtain the historical sales data of complementary commodities for multiple time periods, obtain the moving standard deviation, moving average, year-on-year growth rate, month-on-month growth rate, cumulative sales volume of complementary commodities, and the seasonal index of the corresponding seasonal commodities, and mark the market stage of the seasonal commodity, use the random forest algorithm for model training, build a market stage identification model for complementary commodities, and identify the market stage of complementary commodities. The market stage includes the introduction period, growth period, maturity period and decline period.

[0019] Further, the method of calculating the complementarity coefficients of seasonal commodities and complementary commodities at different time points based on the sales data of the commodities, and identifying the complementarity patterns of seasonal commodities and their complementary commodities at different market stages based on the time series of the dynamic complementarity coefficients, includes:

[0020] Based on the sales growth rate time series, determine the second-order derivative of the sales growth rate of each commodity and use the dynamic complementary coefficient formula Calculate the dynamic complementarity coefficient of seasonal products and their complementary products at different time points, where S i (t) and S j (t) represent the sales volume of seasonal product i and complementary product j at time t, G i (t) and G j (t) represents the sales growth rate of commodity i and commodity j respectively, and represents the rate of change of sales growth rate of commodity i and commodity j at time t, λ is the adjustment factor, which is used to control the influence weight of the second-order derivative on the complementarity coefficient and is obtained by fitting historical sales data. ij (t) is the market stage factor, which depends on the current market stage of commodities i and j and is used to adjust the complementarity strength under different market stages. ij (t) to identify the complementarity patterns between seasonal products and their complementary products in multiple market stages.

[0021] Furthermore, the method of identifying the market stage of complementary commodities in the future preset time period based on the sales data of seasonal commodities and their complementary commodities in the future preset time period, and generating a wholesale purchase plan for seasonal commodities and their complementary commodities in combination with the complementarity pattern, time synchronization and time lag evaluation results, includes:

[0022] According to the sales data of seasonal goods and their complementary goods in the future preset time period, the sales volume time series data of complementary goods are obtained, and the market stage identification model of complementary goods is used to identify the market stage of complementary goods in the future preset time period, and the market stage data of seasonal goods in the same time period are obtained; by calculating the complementarity coefficient of seasonal goods and each complementary goods in the time period, the complementary pattern of seasonal goods and each complementary goods in the time period is identified; by the cross-correlation coefficient R between seasonal goods and complementary goods at different lag times τ XY (τ), obtain the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods; formulate wholesale purchase plans for seasonal goods and complementary goods according to the market stage identification results of complementary goods in the future preset time period, the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods, and the complementary patterns of seasonal goods and complementary goods in different market stages, including adjusting the wholesale purchase quantity, purchase time and replenishment frequency.

[0023] Furthermore, judging the execution effect of the current wholesale purchase plan based on the actual sales data and inventory change data of seasonal goods and complementary goods, identifying the goods and time periods that need to be adjusted, and adjusting the parameters of the wholesale purchase plan include: using the sales management system and the inventory management system to obtain the actual sales data and inventory change data of the seasonal goods and complementary goods after the execution of the wholesale purchase plan; calculating the actual sales growth rate and inventory turnover rate through the actual sales data and inventory change data, and comparing them with the predicted sales growth rate and inventory level to judge the execution effect of the current wholesale purchase plan; if the deviation between the actual sales growth rate and the predicted value is greater than the preset deviation value, or the inventory turnover does not meet expectations, identifying the sales growth rate and inventory turnover rate through time series analysis; and judging the execution effect of the current wholesale purchase plan based on the actual sales growth rate and inventory change data. The system can identify abnormal phenomena in the data or inventory data, including sudden increase in demand, inventory backlog or sales decline, to determine whether there is a mismatch between demand changes and market stages, and identify the products and time periods that need to be adjusted; recalculate the demand forecast curve of the products according to the latest sales data and market trends, re-evaluate the market stage of the products, and adjust the parameters of the wholesale purchase plan, including purchase quantity, purchase time and replenishment frequency; evaluate whether the adjusted wholesale purchase plan is effective through the set key performance indicators, use data visualization tools to generate regular reports, monitor the sales and inventory status of each product, and adjust the wholesale purchase plan based on the effectiveness evaluation of the adjusted wholesale purchase plan. Key performance indicators include inventory turnover rate, out-of-stock rate, and sales achievement rate.

[0024] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0025] The present invention provides a method for predicting wholesale demand based on a machine learning algorithm. The present invention significantly improves the accuracy of demand forecasting and the efficiency of supply chain management by conducting an in-depth analysis of the sales data of seasonal commodities and their complementary commodities in the wholesale trade field. By identifying the periodic sales characteristics of commodities and the correlation between complementary commodities, future market demand can be better predicted to ensure that the optimal inventory and replenishment decisions are made in different market stages. The present invention also accurately captures the dynamic relationship between commodities by calculating the time synchronization and lag of commodity demand, thereby optimizing wholesale purchase plans and reducing inventory backlogs and out-of-stock risks. In addition, based on the time series analysis of the dynamic complementarity coefficient, the present invention can identify the complementary change patterns of commodities in different market stages, and provide support for formulating more targeted wholesale strategies. The present invention can effectively improve the supply chain response speed and market adaptability in wholesale trade, and help enterprises maintain their competitive advantages in a complex and changing market environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of a wholesale demand forecasting method based on a machine learning algorithm of the present invention;

[0027] Figure 2 A schematic diagram of a wholesale demand forecasting method based on a machine learning algorithm of the present invention;

[0028] Figure 3 It is another schematic diagram of a wholesale demand forecasting method based on a machine learning algorithm of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] like Figure 1-3 In this embodiment, a method for predicting wholesale demand based on a machine learning algorithm may specifically include:

[0031] Step S101, judging whether the sales change of the commodity has periodic characteristics based on the historical sales data of the commodity, and identifying seasonal commodities.

[0032] According to the historical sales data of the goods, obtain sales records covering multiple quarters and years, including product name, sales date, daily sales quantity, sales amount and customer information, and arrange them in chronological order. Process the acquired historical sales data through data cleaning technology, filter out abnormal values ​​and missing data, unify the data format, and use missing value interpolation methods to process missing items in the data. Missing value interpolation methods include mean interpolation, interpolation method or filling method based on adjacent data points. Obtain the cleaned historical sales data, use the time series analysis algorithm to perform moving average and exponential smoothing on the historical sales data to obtain the long-term trend of the time series, and use the difference method to remove the trend component to obtain the seasonal component and random error. According to the seasonal component, use the fast Fourier transform method to calculate the amplitude and frequency of each cycle to determine whether the sales change of the product has periodic characteristics. If the sales change of the product has periodic characteristics, the product is judged to be a seasonal product.

[0033] For example, there is historical sales data of commodity A, covering the records of the past three years. The data includes the name of commodity A, sales date, daily sales quantity, sales amount and customer information, and is arranged in chronological order. If on January 1, 2021, the sales quantity of commodity A is 50 pieces, the sales amount is 500 yuan, and the purchasing customer is customer X, these historical sales data are cleaned, outliers and missing data are filtered out, and the data format is unified. For missing data items, such as the missing sales quantity on February 15, 2021, the mean interpolation method is used to complete it to the average sales quantity of 30 pieces on that day. After the cleaning is completed, the cleaned historical sales data is obtained, and a time series analysis is performed on it. The average sales quantity of commodity A each month is calculated by the moving average method. It is found that its long-term sales trend is an upward trend. The exponential smoothing method is used to further smooth the time series data, eliminate short-term fluctuations, and obtain the smoothed long-term trend curve. The long-term trend component in the sales data of commodity A is removed by the difference method, and its seasonal component and random error are extracted. After obtaining the seasonal component, the fast Fourier transform is used to perform spectrum analysis on the sales data of commodity A. Through FFT analysis, it is identified that the sales data of commodity A has significant sales peaks in March and September each year, with an amplitude of 100 pieces and a frequency of twice a year, which shows that the sales changes of commodity A have cyclical characteristics. Since the sales data of commodity A shows repeated peak characteristics in different quarters, it can be judged that commodity A is a seasonal commodity.

[0034] Step S102, identifying complementary commodities of seasonal commodities based on historical sales data of commodities, and predicting sales data of seasonal commodities and complementary commodities in a preset time period in the future.

[0035] The historical sales data of commodities are grouped by transactions to form a transaction database, in which each transaction record contains information about all commodities in a purchase. According to the transaction database, a support threshold is set, and the Apriori algorithm is used to identify the commodity combinations that appear in transactions at a frequency higher than the preset frequency, obtain a list of frequent itemsets, and record the support of each item set. According to the frequent item set list, the confidence of each commodity combination is calculated, the association strength between commodities is judged, commodity pairs with confidence higher than the set threshold are obtained, complementary commodities of seasonal commodities are screened out, and their support and confidence are recorded. The historical sales data of seasonal commodities and complementary commodities are obtained through the sales record database, and are sorted into a multivariate time series format in chronological order to form time series data. According to the time series data, the long short-term memory network algorithm is used for model training to predict the sales data of seasonal commodities and complementary commodities in the future preset time period, and generate the demand forecast curve of seasonal commodities and complementary commodities in the future preset time period.

[0036] Exemplarily, the historical sales data of three commodities B, C, and D are obtained, and the historical sales data are grouped according to the transactions of each customer to form a transaction database. In this database, each transaction record contains the information of all commodities purchased by the customer in one purchase, such as transaction 1 contains the purchase records of commodities B, C, and D, and transaction 2 contains the purchase records of commodities B and D. The support threshold is set to 20%, and the Apriori algorithm is used to analyze these transaction data. It is found that the frequency of simultaneous occurrence of the combination of commodities B and C is 30%, which is higher than the preset support threshold. Therefore, commodities B and C are identified as a frequent item set with a support of 30%. According to the frequent item set list, the confidence of the combination of commodities B and C is calculated, and it is obtained that 50% of all purchases of commodity B also purchase commodity C, with a confidence of 50%. The strength of the association between commodities B and C is judged, and the confidence is higher than the set threshold of 40%. Therefore, commodities B and C are screened as complementary commodities of seasonal commodities, and their support is recorded as 30% and the confidence is 50%. Through the sales record database, the historical sales data of commodity B and commodity C are obtained and sorted into a multivariate time series format in chronological order to form time series data. Based on this time series data, the long short-term memory network algorithm is used for model training. Through the training model, it is predicted that the sales volume of commodity B will increase month by month in the next three months, reaching 120 pieces, 150 pieces and 180 pieces respectively, and the sales volume of commodity C will also increase accordingly, reaching 60 pieces, 75 pieces and 90 pieces respectively. The demand forecast curve of commodity B and commodity C in the next three months is generated, showing that the sales volume of both shows a seasonal upward trend.

[0037] Step S103, based on the sales data of seasonal goods and complementary goods in a preset time period in the future, calculate the cross-correlation coefficients between seasonal goods and complementary goods at different lag times, and determine the time synchronization and time lag of demand changes of complementary goods and seasonal goods.

[0038] Based on the sales data of seasonal products and their complementary products in the future preset time period, obtain the sales volume time series data of each product. Use the sliding average method, set the sliding window size, calculate the moving average based on the sales data before and after each time point, and obtain the smoothed sales volume time series. Based on the smoothed sales volume time series, use the formula Calculate the sales growth rate of each product at each time point and construct a sales growth rate time series for each product, where G(t) is the sales growth rate at time t, S(t) and S(t-1) are the sales volume at time t and t-1 respectively. Based on the sales growth rate time series and the preset time lag range, use the cross-correlation coefficient formula Calculate the cross-correlation coefficient R between seasonal products and complementary products at different lag times τ XY (τ), and obtain the cross-correlation coefficient matrix of each commodity pair, where G X (i) and G Y (i) represents the sales growth rate time series of seasonal products and complementary products, and Its average value, σG X and σG Y is its standard deviation. The maximum correlation coefficient R for each pair of commodities in the cross-correlation coefficient matrix ij (τ max ) to determine the lag time τ corresponding to the maximum correlation coefficient max If the lag time τ corresponding to the maximum correlation coefficient max =0, and R ij (τ max ) exceeds the preset lag time threshold, it is judged that the demand changes of seasonal products and the complementary products are time synchronized. max >0, and R ij (τ max ) exceeds the set lag time threshold, it is judged that there is a positive time lag between the seasonal product and the complementary product. max <0 and R ij (τ max ) exceeds the set threshold, it is judged that there is a negative time lag between the seasonal product and the complementary product.

[0039] For example, the daily sales data of seasonal product E and complementary product F for the next three months are available, and the sales time series data of product E and product F are obtained, where product E sold 100 pieces on the first day and 110 pieces on the second day, and product F sold 50 pieces on the first day and 55 pieces on the second day. Using the sliding average method, set the sliding window size to 3 days, and calculate the moving average of product E and product F based on the sales data before and after each time point. If on the third day, the sales volume of product E is 120 pieces and the sales volume of product F is 60 pieces, then the moving average sales volume of product E is (100+110+120) / 3=110 pieces, and the moving average sales volume of product F is (50+55+60) / 3=55 pieces, and the smoothed sales time series is obtained. According to the smoothed sales time series, use the sales growth rate formula Calculate the sales growth rate of each product at each time point, and get the sales growth rate of product E on the second day is (110-100) / 100=0.1, that is, 10%, and the sales growth rate of product F on the second day is (55-50) / 50=0.1, that is, 10%, where G(t) is the sales growth rate at time t, S(t) and S(t-1) are the sales volume at time t and t-1 respectively. Through this method, construct the sales growth rate time series of product E and product F. According to the sales growth rate time series and the preset time lag range of -2 to 2 days, use the cross-correlation coefficient formula Calculate the cross-correlation coefficient between seasonal product E and complementary product F at different lag times, where G X (i) and G Y (i) represents the sales growth rate time series of seasonal products and complementary products, and Its average value, σG X and σG Y is its standard deviation. If the calculated cross-correlation coefficient is 0.85 when the lag time τ=0, the cross-correlation coefficient is 0.75 when the lag time τ=1, and the cross-correlation coefficient is 0.65 when the lag time τ=-1. Comparing the maximum correlation coefficients of each pair of commodities in the cross-correlation coefficient matrix, it is found that the maximum correlation coefficient of commodity E and commodity F is 0.85, and the corresponding lag time τ=0. Since the lag time τ=0 corresponding to the maximum correlation coefficient, and 0.85 exceeds the preset lag time threshold of 0.8, it is judged that the demand changes of seasonal commodity E and complementary commodity F are time synchronized. If the lag time τ corresponding to the maximum correlation coefficient is 1 and exceeds the preset lag time threshold of 0.8, it is judged that there is a positive time lag between seasonal commodity E and complementary commodity F. If τ is -1 and exceeds the preset lag time threshold of 0.8, it is judged that there is a negative time lag between seasonal commodity E and complementary commodity F.

[0040] Step S104, calculating the seasonal index of the seasonal commodity based on the historical sales data of the seasonal commodity, and constructing a market stage identification model for the complementary commodity to identify the market stage of the complementary commodity.

[0041] Get the historical sales data of seasonal products for multiple time periods and arrange them in chronological order to form time series data. Use the formula Calculate the overall average sales volume of the entire time series data, where N is the number of time points and S(t) is the sales volume. According to the moving average formula Calculate the moving average sales volume MA(t) within a preset time period, where n is the number of time points in a period. Based on the original sales volume S(t) and the moving average sales volume MA(t), use the formula Get the detrended seasonal sales volume D(t) and calculate it through the seasonal index formula Calculate the seasonal index of seasonal commodities, where D(t+u×v) represents the detrended data at the same time point in different years or cycles, v is the length of a cycle, m is the number of complete seasons, and u is the index of each season, representing the period from the first complete season to the mth complete season. Obtain the historical sales data of complementary commodities for multiple time periods, obtain the moving standard deviation, moving average, year-on-year growth rate, month-on-month growth rate, cumulative sales volume, and seasonal index of the corresponding seasonal commodities, and mark the market stage of the seasonal commodity. Use the random forest algorithm for model training, build a market stage identification model for complementary commodities, and identify the market stage of complementary commodities. The market stage includes the introduction period, growth period, maturity period, and decline period.

[0042] For example, there is a historical sales data of seasonal product G, covering the past 4 years. Statistics are collected monthly each year, and these sales data are obtained and arranged in chronological order to form time series data. The sales data of product G includes 120 pieces sold in January of the first year, 150 pieces sold in January of the second year, 180 pieces sold in January of the third year, and 160 pieces sold in January of the fourth year. Use the formula Calculate the overall average sales volume of the entire time series data. If there are 48 time points in the sales data of each month in these 4 years, and the total sales volume of all months is 6,000 units, then the overall average sales volume MA = 6,000 / 48 = 125 units. According to the moving average formula Calculate the moving average sales volume for each quarter. If you need to calculate the moving average sales volume for January of the first year and use the data of the previous and next month, the moving average sales volume for January will be the average of December, January and February. If the sales volume in December of the previous year is 100 pieces, January is 120 pieces, and February is 130 pieces, then the moving average for January is (100+120+130) / 3=116.67 pieces. Based on the original sales data and the moving average, calculate the detrended seasonal sales volume and obtain the detrended sales volume D(1)=1.029 for January of the first year. Similarly, calculate the detrended sales volume in January of other years and obtain the detrended sales volume D(1+12)=1.071 for January of the second year, the detrended sales volume D(1+24)=1.125 for January of the third year, and the detrended sales volume D(1+36)=1.032 for January of the fourth year. Using these detrended data, the seasonal index is calculated according to the formula The seasonal index SI(1) for January is calculated to be 1.06425, which reflects the average sales performance of January relative to its long-term trend in all four years, where D(t+u×v) represents the detrended data at the same time point in different years or cycles, m is the number of complete seasons, u is the index of each season, representing the period from the first complete season to the mth complete season, and v is the length of a cycle, such as 12 months. The historical sales data of complementary product H of product G is obtained, including sales data of 60, 65, 75 and 70 pieces in January of each year. The moving average of product H is calculated to be 67.5 pieces, and the moving standard deviation is 5 pieces. Based on these data, the year-on-year growth rate, month-on-month growth rate and cumulative sales volume of product H are calculated, and the year-on-year growth rate is 8%, the month-on-month growth rate is 5%, and the cumulative sales volume is 270 pieces. Using the seasonal index of product G and these sales indicators of product H, the market stage of product G is marked, such as January of the first year is the introduction period, January of the second year is the growth period, etc. These data are input into the random forest algorithm for model training, and a market stage recognition model for complementary product H is constructed to identify the market stage of product H in the future time period, and obtain that the next month of product H is the growth stage.

[0043] Step S105, calculating the complementarity coefficients of seasonal commodities and complementary commodities at different time points according to the sales data of the commodities, and identifying the complementarity patterns of seasonal commodities and their complementary commodities in different market stages based on the time series of the dynamic complementarity coefficients.

[0044] Based on the sales growth rate time series, determine the second-order derivative of the sales growth rate of each commodity and use the dynamic complementary coefficient formula Calculate the dynamic complementarity coefficient of seasonal products and their complementary products at different time points, where S i (t) and S j (t) represent the sales volume of seasonal product i and complementary product j at time t, G i (t) and G j (t) represents the sales growth rate of commodity i and commodity j respectively, and represents the rate of change of sales growth rate of commodity i and commodity j at time t, λ is the adjustment factor, which is used to control the influence weight of the second-order derivative on the complementarity coefficient and is obtained by fitting historical sales data. ij (t) is the market stage factor, which depends on the current market stage of commodities i and j and is used to adjust the complementarity strength under different market stages. ij (t) to identify the complementarity patterns between seasonal products and their complementary products in multiple market stages.

[0045] For example, there is historical sales data of a seasonal product X and its complementary product Y. At a certain time point t1 in March 2020, the sales volume of product X was 200 pieces and the sales volume of product Y was 150 pieces. At the same time point t1, the sales growth rate of product X was 5%, and the sales growth rate of product Y was 3%. By analyzing the sales growth rate time series data, it is calculated that the second-order derivative of the sales growth rate of product X at time point t1 is 0.002, and the second-order derivative of product Y is 0.001. The adjustment factor λ is obtained by fitting historical data to 2.0, indicating that the influence of the second-order derivative in the dynamic complementarity coefficient is more important. Market stage factor M XY (t) depends on the performance of commodity X and commodity Y in the current market stage. If at this time point t1, commodity X and commodity Y are in the mature stage, the market stage factor M XY The value of (t) is 1.5, which is used to enhance the complementarity strength at this stage. The dynamic complementarity coefficient formula is used. Calculate the dynamic complementarity coefficient between seasonal products and their complementary products at time point t1. The calculated dynamic complementarity coefficient C XY (t1) is 128.572, which means that at time point t1, the complementarity between seasonal product X and complementary product Y is relatively strong, especially in the current mature stage of the market. According to historical sales data, at time point t2 in June 2020, the sales volume of product X increased to 250 pieces, and the sales volume of product Y also increased to 180 pieces, and their sales growth rates were 6% and 4%, respectively. The calculated second-order derivatives are 0.003 and 0.0015, respectively, and the market stage factor is still 1.5 because they are still in the mature stage. Substituting the dynamic complementarity coefficient formula into the dynamic complementarity coefficient C XY (t2) = 140.784. This coefficient further indicates that in the mature stage, the complementarity between commodities X and Y is still strong. At time point t3 in December 2020, commodities X and Y entered the recession stage, with their sales volume falling to 180 and 140 pieces respectively, and sales growth rates falling to 2% and 1.5%. The second-order derivatives are 0.001 and 0.0008, and the market stage factor is adjusted to 1.2, reflecting the reduction of complementarity in the recession stage. The calculated dynamic complementarity coefficient C XY (t3) = 85.2, which indicates that in the recession period, the complementarity between commodities X and Y is significantly weakened. XY (t1), C XY (t2), C XY(t3) Identify the complementary patterns between seasonal product X and its complementary product Y in different market stages. In the growth and mature stages, the complementarity between products X and Y is strong, promoting each other's sales. However, in the recession stage, their complementarity weakens and may even have a negative impact. Based on this analysis, the joint promotion and inventory management strategies of products X and Y can be strengthened in the growth and mature stages, while in the recession stage, inventory can be reduced to avoid excessive backlogs, thereby optimizing the supply chain and marketing strategies and improving overall efficiency.

[0046] Step S106, based on the sales data of seasonal commodities and their complementary commodities in a future preset time period, the market stage of complementary commodities in a future preset time period is identified, and a wholesale purchase plan for seasonal commodities and their complementary commodities is generated in combination with the complementarity pattern, time synchronization and time lag evaluation results.

[0047] According to the sales data of seasonal goods and their complementary goods in the future preset time period, the sales volume time series data of complementary goods are obtained. The market stage identification model of complementary goods is used to identify the market stage of complementary goods in the future preset time period, and the market stage data of seasonal goods in the same time period are obtained. By calculating the complementarity coefficient of seasonal goods and each complementary goods in the time period, the complementary pattern of seasonal goods and each complementary goods in the time period is identified. The cross-correlation coefficient R between seasonal goods and complementary goods at different lag times τ is used XY (τ), obtain the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods. According to the market stage identification results of complementary goods in the future preset time period, the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods, and the complementary patterns of seasonal goods and complementary goods in different market stages, formulate the wholesale purchase plan of seasonal goods and complementary goods, including adjusting the wholesale purchase quantity, purchase time and replenishment frequency.

[0048] Exemplarily, there is sales data for a seasonal product M and its complementary product N in a preset future time period, covering sales from January to June 2020. Obtain the sales volume time series data of product N, and obtain that the sales volume of product N in January is 150 pieces, 200 pieces in February, 250 pieces in March, 180 pieces in April, 160 pieces in May, and 140 pieces in June. Using the market stage identification model for complementary products, it is identified that product N is in the growth stage from January to March, and enters the mature stage from April to June. Obtain the market stage data of seasonal product M in the same time period, and obtain that product M is in the growth stage in January and February, enters the mature stage in March, and remains in the mature stage from April to June. By comparing the market stages of commodity M and commodity N in these time periods, the complementarity coefficients of commodity M and commodity N in different time periods are calculated, and the complementarity coefficients in January are 1.8, February is 2.0, March is 1.6, April is 1.2, May is 1.1, and June is 1.0. This shows that in the growth period from January to February, the complementarity of commodities M and N is strong, while in the mature period from March to June, the complementarity gradually weakens. In order to further analyze the demand change relationship between commodity M and commodity N, the cross-correlation coefficients of seasonal commodity M and complementary commodity N at different lag times τ are calculated to evaluate their time synchronization and time lag. If the lag time range τ is set to -2 to 2 months, after calculation, it is obtained that when the lag time τ = 0, the cross-correlation coefficient is 0.92, when the lag time τ = 1, the cross-correlation coefficient is 0.75, and when the lag time τ = -1, the cross-correlation coefficient is 0.8. Therefore, the lag time τ=0 corresponding to the maximum correlation coefficient of 0.92 is obtained, which exceeds the preset lag time threshold of 0.8. It is judged that in these time periods, the demand changes of commodity M and commodity N are time-synchronized, especially in the growth period and the first half of the mature period, the synchronization of demand changes is particularly obvious. Based on the market stage identification results of complementary commodity N in the future preset time period, the evaluation results of the time synchronization and time lag of the demand changes of seasonal commodity M and complementary commodity N, and the complementary pattern of commodity M and commodity N in different market stages, the wholesale purchase plan of commodities M and N is formulated, including in the growth period of January and February. Due to the strong complementarity of commodities M and N and the high synchronization of demand changes, the wholesale purchase quantity of commodity M is increased to 600 pieces per month, and the purchase quantity of commodity N is increased to 400 pieces per month. At the same time, the purchase frequency is set to once every two weeks to ensure that the inventory can be replenished in time and meet market demand. After entering the mature period from March, the complementarity coefficient of commodities M and N began to decline, and the synchronization of demand still exists, but the intensity has weakened. Therefore, from March to June, the wholesale purchase quantity of commodity M was adjusted to 500 pieces per month, the purchase quantity of commodity N was adjusted to 300 pieces per month, and the purchase frequency was adjusted to once a month. The purpose is to prevent excess inventory and reduce cost risks while meeting market demand.Through this adjustment, the supply strategy of commodities M and N in different market stages can be optimized, which not only meets the high demand in the growth stage, but also avoids the inventory backlog that may occur in the mature stage, thereby improving the overall supply chain efficiency and market responsiveness.

[0049] Step S107, judging the execution effect of the current wholesale purchase plan based on the actual sales data and inventory change data of seasonal commodities and complementary commodities, identifying the commodities and time periods that need to be adjusted, and adjusting the parameters of the wholesale purchase plan.

[0050] Use the sales management system and inventory management system to obtain the actual sales data and inventory change data of seasonal and complementary goods after the implementation of the wholesale purchase plan. Calculate the actual sales growth rate and inventory turnover rate through the actual sales data and inventory change data, and compare them with the predicted sales growth rate and inventory level to determine the implementation effect of the current wholesale purchase plan. If the deviation between the actual sales growth rate and the predicted value is greater than the preset deviation value, or the inventory turnover does not meet expectations, identify abnormal phenomena in the sales data or inventory data through time series analysis, including sudden increase in demand, inventory backlog or sales decline, determine whether there is a phenomenon of mismatch between demand changes and market stages, and identify the goods and time periods that need to be adjusted. Recalculate the demand forecast curve of the goods based on the latest sales data and market trends, re-evaluate the market stage of the goods, and adjust the parameters of the wholesale purchase plan, including purchase quantity, purchase time and replenishment frequency. Through the set key performance indicators, evaluate whether the adjusted wholesale purchasing plan is effective, use data visualization tools to generate regular reports, monitor the sales and inventory status of each commodity, and adjust the wholesale purchasing plan based on the effectiveness evaluation of the adjusted wholesale purchasing plan. Key performance indicators include inventory turnover rate, out-of-stock rate, and sales achievement rate.

[0051] Exemplarily, a wholesale purchase plan for seasonal product P and its complementary product Q was implemented from January to March 2020. Using the sales management system and inventory management system, the actual sales data and inventory change data during this period are obtained, including that the actual sales volume of product P in January was 500 pieces, 600 pieces in February, and 550 pieces in March, while the actual sales volume of product Q was 300 pieces in January, 350 pieces in February, and 320 pieces in March. Based on these data, the actual sales growth rates of products P and Q are calculated, and the actual sales growth rate of product P from January to February is 20%, while the growth rate from February to March is -8.3%. For product Q, the sales growth rate from January to February is 16.7%, and the growth rate from February to March is -8.6%. According to the records of the inventory management system, the inventory turnover rate was calculated, and the average inventory of commodity P in January was 700 pieces, 750 pieces in February, and 720 pieces in March. The inventory turnover rate from January to February was 0.8, and the inventory turnover rate from February to March was 0.764. Similarly, the average inventory of commodity Q in January was 400 pieces, 420 pieces in February, and 410 pieces in March. The inventory turnover rate from January to February was 0.833, and the inventory turnover rate from February to March was 0.78. Comparing these actual data with the forecast data, it was found that the deviation of the actual sales growth rate of commodity P and commodity Q from the forecast value exceeded the preset deviation value of 5%, and the inventory turnover rate did not reach the expected 1.0. The actual sales growth rate of commodity P fell by 8.3% from February to March, while the forecasted growth rate was only 0%. Therefore, through time series analysis, it was identified that during the period from February to March, the sales data of goods P and goods Q showed abnormalities, indicating that demand had declined and there was a certain backlog of inventory. In order to solve these problems, the latest sales data and market trends were re-analyzed, and it was found that the demand changes of goods P and Q in March did not match the previous market stage assessment. Therefore, the demand forecast curves of goods P and Q were recalculated, and their market stages were re-evaluated. In view of these changes, the parameters of the wholesale purchase plan need to be adjusted. The specific adjustments include reducing the purchase quantity of goods P in April from 700 to 500 pieces, reducing the purchase quantity of goods Q from 400 to 300 pieces, adjusting the purchase time to once a month, and changing the replenishment frequency from once every two weeks to once a month to prevent further inventory backlogs. In order to evaluate the effectiveness of the adjusted wholesale purchase plan, key performance indicators were set, including inventory turnover rate, out-of-stock rate, and sales achievement rate. Through continuous monitoring, it was found that in April, the inventory turnover rate of commodity P increased to 0.95, close to the target value of 1.0, the out-of-stock rate of commodity Q decreased to 2%, and the sales achievement rates reached 90% and 85% respectively, indicating that the adjusted purchasing plan effectively adapted to changes in market demand and improved inventory management and sales performance.Through these adjustments and evaluations, a regular report is generated every month, using data visualization tools to monitor the sales and inventory status of products P and Q. Based on the effectiveness evaluation results of the adjusted wholesale purchase plan, the purchase strategy is continuously optimized to ensure that the inventory level is consistent with market demand, thereby maximizing sales profits and inventory turnover.

[0052] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A wholesale demand forecasting method based on machine learning algorithm, characterized in that: The method comprises: Based on the historical sales data of the product, determine whether the sales changes of the product have cyclical characteristics and identify seasonal products; Identify complementary products of seasonal products based on historical sales data of products, and predict sales data of seasonal products and complementary products in a preset time period in the future; Based on the sales data of seasonal goods and complementary goods in a preset time period in the future, the cross-correlation coefficients between seasonal goods and complementary goods at different lag times are calculated, and the time synchronization and time lag of the demand changes of complementary goods and seasonal goods are determined; Calculate the seasonal index of seasonal commodities based on the historical sales data of seasonal commodities, and build a market stage identification model for complementary commodities to identify the market stage of complementary commodities; According to the sales data of commodities, the complementarity coefficients of seasonal commodities and complementary commodities at different time points are calculated, and based on the time series of dynamic complementarity coefficients, the complementarity patterns of seasonal commodities and their complementary commodities in different market stages are identified; Based on the sales data of seasonal commodities and their complementary commodities in the future preset time period, the market stage of complementary commodities in the future preset time period is identified, and the wholesale purchase plan of seasonal commodities and their complementary commodities is generated by combining the evaluation results of complementarity pattern, time synchronization and time lag; Based on the actual sales data and inventory change data of seasonal and complementary goods, determine the execution effect of the current wholesale purchasing plan, identify the goods and time periods that need to be adjusted, and adjust the parameters of the wholesale purchasing plan.

2. The method according to claim 1, wherein: The method of judging whether the sales change of a commodity has a periodic characteristic based on the historical sales data of the commodity and identifying seasonal commodities includes: Based on the historical sales data of the goods, obtain sales records covering multiple quarters and years, including product name, sales date, daily sales quantity, sales amount and customer information, and arrange them in chronological order; process the obtained historical sales data through data cleaning technology, filter out abnormal values ​​and missing data, unify the data format, and use missing value interpolation methods to process missing items in the data. Missing value interpolation methods include mean interpolation, interpolation method or filling method based on adjacent data points; obtain the cleaned historical sales data, use time series analysis algorithm to perform moving average and exponential smoothing on the historical sales data to obtain the long-term trend of the time series, and use the difference method to remove the trend component to obtain the seasonal component and random error; based on the seasonal component, use the fast Fourier transform method to calculate the amplitude and frequency of each cycle to determine whether the sales changes of the goods have periodic characteristics; if the sales changes of the goods have periodic characteristics, then determine that the goods are seasonal goods.

3. The method according to claim 1, wherein: The method of identifying complementary commodities of seasonal commodities based on historical sales data of commodities and predicting sales data of seasonal commodities and complementary commodities in a future preset time period includes: The historical sales data of commodities are grouped by transactions to form a transaction database, in which each transaction record contains information about all commodities in a purchase; according to the transaction database, a support threshold is set, and the Apriori algorithm is used to identify commodity combinations that appear simultaneously in transactions at a frequency higher than a preset frequency, obtain a list of frequent itemsets, and record the support of each item set; according to the frequent item set list, the confidence of each commodity combination is calculated, the strength of association between commodities is judged, commodity pairs with confidence higher than the set threshold are obtained, complementary commodities of seasonal commodities are screened out, and their support and confidence are recorded; the historical sales data of seasonal commodities and complementary commodities are obtained through the sales record database, and are sorted into a multivariate time series format in chronological order to form time series data; according to the time series data, the long short-term memory network algorithm is used for model training to predict the sales data of seasonal commodities and complementary commodities in a preset time period in the future, and generate a demand forecast curve for seasonal commodities and complementary commodities in a preset time period in the future.

4. The method according to claim 1, wherein: The method of calculating the cross-correlation coefficients between seasonal products and complementary products at different lag times based on the sales data of seasonal products and complementary products in a preset time period in the future, and judging the time synchronization and time lag of the demand changes of complementary products and seasonal products, includes: According to the sales data of seasonal commodities and their complementary commodities in the future preset time period, the sales volume time series data of each commodity is obtained; the sliding average method is used to set the sliding window size, and the moving average is calculated according to the sales data before and after each time point to obtain the smoothed sales volume time series; according to the smoothed sales volume time series, the formula is used Calculate the sales growth rate of each product at each time point and construct a sales growth rate time series for each product, where G(t) is the sales growth rate at time t, S(t) and S(t-1) are the sales volumes at time t and t-1 respectively; Based on the sales growth rate time series and the preset time lag range, use the cross-correlation coefficient formula Calculate the cross-correlation coefficient R between seasonal products and complementary products at different lag times τ XY (τ), and obtain the cross-correlation coefficient matrix of each commodity pair, where G X (i) and G Y (i) represents the sales growth rate time series of seasonal products and complementary products, and Its average value, σG X and σG Y is its standard deviation; the maximum correlation coefficient R for each pair of commodities in the cross-correlation coefficient matrix ij (τ max ) to determine the lag time τ corresponding to the maximum correlation coefficient max ; If the lag time τ corresponding to the maximum correlation coefficient max =0, and R ij (τ max ) exceeds the preset lag time threshold, it is judged that the demand changes of seasonal products and the complementary products are time synchronized; if τ max >0, and R ij (τ max ) exceeds the set lag time threshold, it is judged that there is a positive time lag between the seasonal product and the complementary product; if τ max <0 and R ij (τ max ) exceeds the set threshold, it is judged that there is a negative time lag between the seasonal product and the complementary product.

5. The method according to claim 1, wherein: The seasonal index of seasonal commodities is calculated based on the historical sales data of seasonal commodities, and a market stage identification model of complementary commodities is constructed to identify the market stage of complementary commodities, including: Get the historical sales data of seasonal commodities for multiple time periods and arrange them in chronological order to form time series data; use the formula Calculate the overall average sales volume of the entire time series data, where N is the number of time points and S(t) is the sales volume; according to the moving average formula Calculate the moving average sales volume MA(t) within a preset time period, where n is the number of time points in a period; based on the original sales volume S(t) and the moving average sales volume MA(t), use the formula Get the detrended seasonal sales volume D(t) and calculate it through the seasonal index formula Calculate the seasonal index of seasonal commodities, where D(t+u×v) represents the detrended data at the same time point in different years or cycles, v is the length of a cycle, m is the number of complete seasons, and u is the index of each season, representing from the first complete season to the mth complete season; obtain the historical sales data of complementary commodities for multiple time periods, obtain the moving standard deviation, moving average, year-on-year growth rate, month-on-month growth rate, cumulative sales volume of complementary commodities, and the seasonal index of the corresponding seasonal commodities, and mark the market stage of the seasonal commodity, use the random forest algorithm for model training, build a market stage identification model for complementary commodities, and identify the market stage of complementary commodities. The market stage includes the introduction period, growth period, maturity period and decline period.

6. The method according to claim 1, wherein: The method of calculating the complementarity coefficients of seasonal commodities and complementary commodities at different time points based on the sales data of commodities, and identifying the complementarity patterns of seasonal commodities and their complementary commodities at different market stages based on the time series of the dynamic complementarity coefficients, includes: Based on the sales growth rate time series, determine the second-order derivative of the sales growth rate of each commodity and use the dynamic complementary coefficient formula Calculate the dynamic complementarity coefficient of seasonal products and their complementary products at different time points, where S i (t) and S j (t) represent the sales volume of seasonal product i and complementary product j at time t, G i (t) and G j (t) represents the sales growth rate of commodity i and commodity j respectively, and represents the rate of change of sales growth rate of commodity i and commodity j at time t, λ is the adjustment factor, which is used to control the influence weight of the second-order derivative on the complementarity coefficient and is obtained by fitting historical sales data. ij (t) is the market stage factor, which depends on the current market stage of commodities i and j and is used to adjust the complementarity strength under different market stages. ij (t) to identify the complementarity patterns between seasonal products and their complementary products in multiple market stages.

7. The method according to claim 1, wherein: The method of identifying the market stage of complementary commodities in the future preset time period based on the sales data of seasonal commodities and their complementary commodities in the future preset time period, and generating a wholesale purchase plan for seasonal commodities and their complementary commodities in combination with the complementarity pattern, time synchronization and time lag evaluation results, includes: According to the sales data of seasonal goods and their complementary goods in the future preset time period, the sales volume time series data of complementary goods are obtained, and the market stage identification model of complementary goods is used to identify the market stage of complementary goods in the future preset time period, and the market stage data of seasonal goods in the same time period are obtained; by calculating the complementarity coefficient of seasonal goods and each complementary goods in the time period, the complementary pattern of seasonal goods and each complementary goods in the time period is identified; by the cross-correlation coefficient R between seasonal goods and complementary goods at different lag times τ XY (τ), obtain the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods; formulate wholesale purchase plans for seasonal goods and complementary goods according to the market stage identification results of complementary goods in the future preset time period, the time synchronization and time lag evaluation results of the demand changes of seasonal goods and complementary goods, and the complementary patterns of seasonal goods and complementary goods in different market stages, including adjusting the wholesale purchase quantity, purchase time and replenishment frequency.

8. The method according to claim 1, wherein: The method of judging the execution effect of the current wholesale purchase plan based on the actual sales data and inventory change data of seasonal and complementary products, identifying the products and time periods that need to be adjusted, and adjusting the parameters of the wholesale purchase plan includes: Utilize the sales management system and inventory management system to obtain the actual sales data and inventory change data of seasonal and complementary goods after the implementation of the wholesale purchase plan; calculate the actual sales growth rate and inventory turnover rate through the actual sales data and inventory change data, and compare them with the predicted sales growth rate and inventory level to determine the implementation effect of the current wholesale purchase plan; if the deviation between the actual sales growth rate and the predicted value is greater than the preset deviation value, or the inventory turnover does not meet expectations, identify abnormal phenomena in the sales data or inventory data through time series analysis, including sudden increase in demand, inventory backlog or sales decline, and determine whether there is a change in demand Recognize the phenomenon of mismatch with the market stage and identify the products and time periods that need to be adjusted; recalculate the demand forecast curve of the products according to the latest sales data and market trends, re-evaluate the market stage of the products, and adjust the parameters of the wholesale purchase plan, including purchase quantity, purchase time and replenishment frequency; evaluate whether the adjusted wholesale purchase plan is effective through the set key performance indicators, use data visualization tools to generate regular reports, monitor the sales and inventory status of each product, and adjust the wholesale purchase plan based on the effectiveness evaluation of the adjusted wholesale purchase plan. Key performance indicators include inventory turnover rate, out-of-stock rate, and sales achievement rate.

Images