WO2019001120A1 - 商品动态定价数据处理方法和系统 - Google Patents

商品动态定价数据处理方法和系统 Download PDF

Info

Publication number
WO2019001120A1
WO2019001120A1 PCT/CN2018/084382 CN2018084382W WO2019001120A1 WO 2019001120 A1 WO2019001120 A1 WO 2019001120A1 CN 2018084382 W CN2018084382 W CN 2018084382W WO 2019001120 A1 WO2019001120 A1 WO 2019001120A1
Authority
WO
WIPO (PCT)
Prior art keywords
price
expected
commodity
sales
inventory
Prior art date
Application number
PCT/CN2018/084382
Other languages
English (en)
French (fr)
Inventor
宋磊
Original Assignee
北京京东尚科信息技术有限公司
北京京东世纪贸易有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京京东尚科信息技术有限公司, 北京京东世纪贸易有限公司 filed Critical 北京京东尚科信息技术有限公司
Publication of WO2019001120A1 publication Critical patent/WO2019001120A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0283—Price estimation or determination
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • the present disclosure relates to the field of data processing, and in particular, to a method and system for processing commodity dynamic pricing data.
  • an algorithm can be used to fit the price elasticity parameter of each commodity, and accordingly, the sales volume of a certain commodity under a given price condition can be predicted, and then, the existing inventory level and the expectation are combined.
  • the length of the clearance automatically recommends a most reasonable pricing, but the relevant algorithm cannot dynamically adjust the suggested price.
  • the price elasticity parameter After the price elasticity parameter is determined, it will not change during the entire slow-moving clearance period, and thus lacks the ability to self-update according to the latest data. Since the volume-price relationship of goods during the normal sales period and the slow-moving clearance period is likely to be different, and the relevant algorithm uses the price elasticity parameter during the normal sales period, this will have a certain impact on the pricing accuracy.
  • the existing static model can not get feedback and make corresponding adjustments, so even if the volume-price relationship of the commodity does not change much during the slow-moving clearance period.
  • the use of static volume-price relationships also affects the accuracy of pricing.
  • the algorithm does not support human intervention in the proposed price. After the algorithm gives the suggested price, for various reasons, the user may adjust the price, which causes the relevant algorithm to deviate from the evaluation of the next clearance effect. At the same time, the related algorithm cannot adjust the future pricing strategy according to the current price.
  • One technical problem to be solved by the present disclosure is to provide a method and system for processing dynamic pricing data of commodities, and to maximize the profit of commodities under the premise of clearing inventory.
  • a commodity dynamic pricing data processing method including: acquiring historical sales data of an item; and determining, according to historical sales data, an item from a current inventory status to an expected inventory in a case of executing a predetermined price policy The state transition probability of the quantity state; using the expected target as the reward function, constructing the Markov decision model based on the state transition probability, the expected cumulative return and the reward function, and determining the expected cumulative return expression based on the Markov decision model, with the predetermined price as the variable
  • the iterative calculation is performed by using an enhanced learning algorithm to determine an optimal price of the commodity; wherein the expected target is an expected return of the commodity from the current inventory state to the expected inventory state in the case of executing the predetermined price strategy.
  • determining an average sales value of the commodity determining an average sales value of the commodity; determining a state transition probability of the commodity from the current inventory state to the expected inventory state in the case of executing the predetermined price policy based on the current inventory amount, the expected inventory amount, and the sales average value.
  • determining a first price interval according to a predetermined price determining whether the quantity of historical sales data in the first price interval is greater than or equal to a threshold; and if the quantity of historical sales data in the first price interval is greater than or equal to a threshold, according to the first
  • the historical sales data of the price range determines the average sales value of the commodity; if the quantity of the historical sales data in the first price interval is less than the threshold, the first price interval is gradually expanded to the second price interval, so that the second price interval
  • the quantity of historical sales data is greater than or equal to the threshold value, and the price elasticity model is used to determine the price elasticity parameter of the commodity in the second price range, and the average value of the sales volume of the commodity is determined according to the price elasticity parameter.
  • the sales probability density function is determined based on the sales average and the sales standard deviation, and the difference between the current inventory quantity and the expected inventory quantity minus the predetermined value is a lower limit, The difference between the current inventory quantity and the expected inventory quantity is an upper limit, and the sales probability density function is integrated to calculate a state transition probability of the commodity from the current inventory quantity state to the expected inventory quantity state in the case of executing the predetermined price strategy;
  • the sales standard deviation shown is determined by the standard deviation formula based on the sales average.
  • S is the current inventory quantity
  • S' is the expected inventory quantity
  • L is the remaining clearance time
  • ⁇ is the possible price of the commodity
  • P i is the optimal price of the commodity i day
  • P min is the minimum price of the commodity
  • P max For the maximum price of the commodity
  • T(S, ⁇ , S') is the state transition probability from the current inventory state S to the expected inventory state S' in the case where the price ⁇ is executed
  • T(S, P i , S') The state transition probability from the current stock quantity state S to the expected stock quantity state S' in the case of performing the optimal price P i
  • F(S, i) is the cumulative profit when the remaining clearance time is i
  • the current stock quantity is S
  • F(S', i-1) is the expected cumulative return when the remaining clearance period is i-1
  • the expected maximum benefit of the quantity state S to the expected inventory quantity state S', R(S, P i , S') is the expected from the current inventory quantity state S to the expected inventory quantity state S' in the case of performing the optimal price P i
  • the maximum return, F(S, 0) and F(0, i) are initial values.
  • a commodity dynamic pricing data processing system comprising: a data obtaining unit, configured to acquire historical sales data of the commodity; and a probability determining unit, configured to determine, according to the historical sales data, that the commodity is executed The state transition probability from the current inventory status to the expected inventory status in the case of a predetermined price strategy; an optimal price determination unit for constructing a Marco based on the state transition probability, the expected cumulative return, and the reward function with the expected target as a reward function
  • the decision-making model is based on the Markov decision model to determine the expected cumulative return expression, with the predetermined price as the variable, and the iterative calculation using the enhanced learning algorithm to determine the optimal price of the commodity; wherein the expected target is the commodity in the execution of the predetermined price strategy.
  • the probability determining unit is configured to determine a sales average value of the commodity; and determine a state transition probability from the current inventory state to the expected inventory state in the case of executing the predetermined price policy based on the current inventory amount, the expected inventory amount, and the sales average value .
  • the probability determining unit is further configured to determine a first price interval according to the predetermined price; determine whether the quantity of historical sales data in the first price interval is greater than or equal to the threshold; if the quantity of historical sales data in the first price interval is greater than or equal to The threshold value is used to determine the average sales value of the commodity according to the historical sales data of the first price range; if the quantity of the historical sales data in the first price interval is less than the threshold, the first price interval is gradually expanded to the second price interval to The quantity of historical sales data of the second price interval is greater than or equal to the threshold value, and the price elasticity model is used to determine the price elasticity parameter of the commodity in the second price interval, and the average value of the sales volume of the commodity is determined according to the price elasticity parameter.
  • the probability determining unit is further configured to determine, by using a maximum likelihood estimation algorithm, if the sales volume of the commodity obeys the positive distribution, determine a sales probability density function based on the sales average and the sales standard deviation, and subtract the difference between the current inventory quantity and the expected inventory quantity.
  • the predetermined value is the lower limit
  • the difference between the current inventory quantity and the expected inventory quantity is the upper limit
  • the sales probability density function is integrated to determine the state transition of the commodity from the current inventory state to the expected inventory state in the case of executing the predetermined price strategy. Probability; wherein the indicated sales standard deviation is determined based on the sales mean using the standard deviation formula.
  • the probability determining unit is further configured to use the maximum likelihood estimation algorithm to determine the sales volume of the commodity to obey the Poisson distribution, and then use the formula A state transition probability from a current inventory status to an expected inventory status in the case of executing a predetermined price policy; wherein S is the current inventory quantity, S' is the expected inventory quantity, and ⁇ is the sales volume average.
  • the expected maximum return of the inventory state S to the expected inventory state S', R(S, P i , S') is the current inventory quantity state S to the expected inventory quantity state S' in the case where the optimal price P i is executed
  • Expected maximum benefit, F(S, 0) and F(0, i) are initial values.
  • a commodity dynamic pricing data processing system comprising: a memory; and a processor coupled to the memory, the processor being configured to perform the method as described above based on the instructions stored in the memory.
  • a computer readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the steps of the method described above.
  • the embodiments of the present disclosure use the historical sales data to perform data processing based on the Markov decision model and the enhanced learning algorithm, and obtain an optimal pricing strategy for the commodity, which can maximize the profit of the commodity.
  • FIG. 1 is a schematic flow chart of some embodiments of a method for processing dynamic pricing data of a commodity according to the present disclosure.
  • FIG. 2 is a schematic flow chart of still another embodiment of a method for processing dynamic pricing data of a commodity according to the present disclosure.
  • FIG. 3 is a schematic structural diagram of some embodiments of a commodity dynamic pricing data processing system according to the present disclosure.
  • FIG. 4 is an application architecture diagram of some embodiments of a commodity dynamic pricing data processing system of the present disclosure.
  • FIG. 5 is a schematic structural diagram of another embodiment of a commodity dynamic pricing data processing system according to the present disclosure.
  • FIG. 6 is a schematic structural diagram of still another embodiment of a commodity dynamic pricing data processing system according to the present disclosure.
  • FIG. 1 is a schematic flow chart of some embodiments of a method for processing dynamic pricing data of a commodity according to the present disclosure. The method includes the following steps 110-140.
  • historical sales data for the item is obtained. For example, a sales record in which the commodity transaction price is between ⁇ -0.5 and ⁇ +0.5 in a predetermined period of time may be extracted, wherein ⁇ is a price at which the merchandise is sold, and the sales data includes sales data, a selling price, and the like.
  • a state transition probability from the current inventory status to the expected inventory status is determined for the item in the event that the predetermined price strategy is executed.
  • the Markov decision model is constructed based on the state transition probability, the expected cumulative return, and the reward function with the expected goal as a reward function.
  • the expected target is the expected return of the commodity from the current inventory status to the expected inventory status in the event that the predetermined price strategy is executed, wherein the expected return may be the expected sales or the expected maximum gross profit.
  • the Markov Decision Process is a mathematical model used to deal with the optimal scheduling problem of partially random, partially decision maker control systems.
  • the Markov decision process can be expressed as a tuple as follows:
  • Sta is a set of states
  • Act is a set containing all decisions
  • T assigns a suitable probability to all state transition rules of the form (S, ⁇ , S'), that is, given a state S, a decision ⁇ And a subsequent state S', T(S, ⁇ , S') represents the probability of moving from S to S' in the case of executing the decision ⁇
  • R is a reward function, that is, when a certain rule is executed, the system can obtain Reward.
  • the status is all possible stocks for a commodity, and the decision is all possible pricing for the commodity.
  • the setting of the reward function R depends on the given goal.
  • the reward corresponding to the rule (S, ⁇ , S') is (S-S') ⁇ ⁇ , where (S-S' ) indicates the change in stock quantity, that is, the quantity of the item sold on the day, and ⁇ is the price of the item on the day. If the goal is set to maximize gross profit, the corresponding reward should be (S-S') x ( ⁇ -P cost ), where P cost is the cost price of the item.
  • the main task of the Markov decision model is to determine the state transition probability T(S, ⁇ , S'), that is, the probability corresponding to each rule (S, ⁇ , S'), that is, in the case where the price is ⁇ .
  • an expected cumulative return expression is determined based on the Markov decision model, with the predetermined price as a variable, and an iterative calculation using the enhanced learning algorithm to dynamically determine the optimal price of the commodity.
  • Enhanced learning also known as reinforcement learning, is a machine learning field that emphasizes how to dynamically adjust strategies based on changes in the external environment to maximize expected benefits. For example, taking the price ⁇ as a variable, thinking that the remaining clearance time is i, and the cumulative return F(S, i) when the inventory is S is an iterative calculation for the objective function, so that F(S, i) takes the maximum value, when F( S, i) The value of ⁇ when the maximum value is obtained is the optimal price of the commodity.
  • the optimal price of the commodity may be calculated using the formula, which may be referred to as the Bellman (shortest path algorithm) equation.
  • S is the current inventory quantity
  • S' is the expected inventory quantity
  • L is the remaining clearance time
  • ⁇ is the possible price of the commodity
  • P i is the optimal price of the commodity i day
  • P min is the minimum price of the commodity
  • P max For the maximum price of the commodity
  • T(S, ⁇ , S') is the state transition probability from the current inventory state S to the expected inventory state S' in the case where the price ⁇ is executed
  • T(S, P i , S') The state transition probability from the current stock quantity state S to the expected stock quantity state S' in the case where the optimal price P i is executed
  • F(S, i) is the cumulative profit when the remaining clearance time is i and the stock quantity is S
  • F(S', i-1) is the expected cumulative return when the remaining clearance period is i-1 and the stock quantity is S'
  • R(S, ⁇ , S') is the current stock status when the execution price is ⁇ .
  • R(S, P i , S') is the expected maximum return from the current stock quantity state S to the expected stock quantity state S' in the case of performing the optimal price P i , F(S, 0) and F(0, i) are initial values.
  • historical sales data is used for data processing to obtain an optimal pricing strategy for the commodity, and when the product is slow-moving, it can be maximized under the premise of clearing the inventory. The proceeds of the goods.
  • FIG. 2 is a schematic flow chart of still another embodiment of a method for processing dynamic pricing data of a commodity according to the present disclosure.
  • the present disclosure can achieve good results in the sales of slow-moving products. The following description will be made by taking a slow-moving product as an example, and the method includes the following steps 210-290.
  • step 210 historical sales data for the slow-moving item is obtained.
  • a first price interval is determined based on the predetermined price. For example, if the predetermined price is ⁇ , it may be determined that the first price interval is ⁇ -0.5 to ⁇ +0.5. Those skilled in the art will appreciate that different price ranges can be set depending on the value of different slow-moving items.
  • step 230 it is determined whether the number of historical sales data in the first price interval is greater than or equal to the threshold. If the threshold is greater than the threshold, step 240 is performed; otherwise, step 250 is performed. For example, if the number of sales records exceeds 30 days, the data is considered available, otherwise, the data is considered too small.
  • the average sales volume of the unsalable item is determined based on the historical sales data for each day of the first price range.
  • the maximum likelihood estimation algorithm can be used to determine that the sales volume of the slow-moving products obeys the positive distribution. If the historical sales volume is expressed as (Q N , Q N-1 , ..., Q 1 ), where N is the number of days of sales.
  • the mean ⁇ of the fitted positive distribution is as follows:
  • the first price interval is gradually expanded to a second price interval such that the amount of historical sales data for the second price interval is greater than or equal to the threshold. For example, when the number of sales records in a given price range is less than 30 and a reliable distribution cannot be fitted, the price range is gradually increased until there is a sufficient number of historical records.
  • the price elasticity model is used to determine the price elasticity parameter of the slow-moving item in the second price range.
  • the volume model is an economic model used to measure the sensitivity of slow-moving products to price changes.
  • the volume and price model is as follows:
  • the average sales volume of the unsalable item is determined based on the price elasticity parameter.
  • the sales standard deviation ⁇ is calculated using the standard deviation formula based on the sales average.
  • a state transition probability from the current inventory status to the expected inventory status in the case where the predetermined price strategy is executed is determined.
  • the sales probability density function is determined based on the sales average and the sales standard deviation, and the difference between the current inventory quantity and the expected inventory quantity minus the predetermined value is the lower limit, and the difference between the current inventory quantity and the expected inventory quantity is the upper limit, and the sales probability density function is performed.
  • the integral calculation determines a state transition probability of the slow-moving product from the current inventory state to the expected inventory state in the case of executing the predetermined price policy; wherein, if the predetermined value is, for example, 1, the specific formula may be Among them, S is the current inventory, S' is the expected inventory, ⁇ is the sales average, and ⁇ is the sales standard deviation.
  • the historical sales volume of other similar commodities under the same category may be referred to, and the sales data of the current unsalable products may be filled to obtain sufficient sales data. Thereafter, the rule probabilities are fitted using steps 240-260.
  • the Markov decision model is constructed based on the state transition probability, the expected cumulative return, and the reward function with the expected goal as a reward function.
  • the expected cumulative return expression is determined based on the Markov decision model, and the predetermined price is used as a variable, and the enhanced learning algorithm is used for iterative calculation to dynamically determine the optimal price of the unsalable product.
  • the fitting MDP mainly reflects the dynamic conversion relationship between the inventory, price and sales volume of a given slow-moving product.
  • an optimal pricing strategy can be automatically calculated to maximize a given goal.
  • the optimal pricing strategy and expected goals of a certain slow-moving product under any given inventory level, price and clearance period can be determined, and the subjectivity and inaccuracy brought about by human decision-making can be avoided. Save a lot of inventory costs and avoid unnecessary losses.
  • the embodiment uses the classical price model to backfill the missing information, which improves the accuracy of the model fitting and solves the problem of unsalable pricing under the condition of lack of information.
  • the dynamic adjustment optimal strategy when the price is set artificially, when the system monitors the deviation, the dynamic adjustment optimal strategy can be supported, and the expected sales volume can be more accurately predicted, and one of the most recommended A good adjustment strategy to achieve the goal of supporting both manual intervention and dynamic adjustment.
  • step 260 using the maximum likelihood estimation algorithm to determine the sales volume of the unsalable item obeying the Naposon distribution; wherein S is the current inventory quantity, S' is the expected inventory quantity, and ⁇ is the sales average value.
  • the sales volume of the unsalable goods can be fitted to the distribution of the positive distribution, or the distribution of the Poisson can be made, or it can be fitted to other distributions, and the corresponding formula can be used to calculate the unsalable goods at the predetermined price.
  • the state transition probability from the current inventory status to the expected inventory status may be sufficient.
  • FIG. 3 is a schematic structural diagram of some embodiments of a commodity dynamic pricing data processing system according to the present disclosure.
  • the system includes a data acquisition unit 310, a probability determination unit 320, and an optimal price determination unit 330, wherein:
  • the data acquisition unit 310 is configured to acquire historical sales data of the commodity. For example, a sales record in which the commodity transaction price is between ⁇ -0.5 and ⁇ +0.5 within a predetermined time period may be extracted, where ⁇ is the price of the sale.
  • the probability determination unit 320 is configured to determine, based on the historical sales data, a state transition probability from the current inventory state to the expected inventory state in the case of executing the predetermined price policy, ie, determining an input of the Markov decision model.
  • the optimal price determining unit 330 is configured to construct a Markov decision model according to the state transition probability, the expected cumulative return, and the reward function with the expected target as a reward function, and determine an expected cumulative return expression based on the Markov decision model, at a predetermined price Variables, using the enhanced learning algorithm for iterative calculations, dynamically determine the optimal price of the commodity.
  • the expected target is the expected return from the current inventory status to the expected inventory status in the case of executing the predetermined price strategy, wherein the expected return may be the expected sales or the expected maximum gross profit.
  • Enhanced learning also known as reinforcement learning, is a machine learning field that emphasizes how to dynamically adjust strategies based on changes in the external environment to maximize expected benefits.
  • the optimal price of the commodity may be calculated using the formula, which may be referred to as the Bellman equation.
  • S is the current inventory quantity
  • S' is the expected inventory quantity
  • L is the remaining clearance time
  • ⁇ is the possible price of the commodity
  • P i is the optimal price of the commodity i day
  • P min is the minimum price of the commodity
  • P max For the maximum price of the commodity
  • T(S, ⁇ , S') is the state transition probability from the current inventory state S to the expected inventory state S' in the case where the price ⁇ is executed
  • T(S, P i , S') The state transition probability from the current stock quantity state S to the expected stock quantity state S' in the case where the optimal price P i is executed
  • F(S, i) is the cumulative profit when the remaining clearance time is i and the stock quantity is S
  • F(S', i-1) is the expected cumulative return when the remaining clearance period is i-1 and the stock quantity is S'
  • R(S, ⁇ , S') is the current stock status when the execution price is ⁇ .
  • R(S, P i , S') is the expected maximum return from the current stock quantity state S to the expected stock quantity state S' in the case of performing the optimal price P i , F(S, 0) and F(0, i) are initial values.
  • the historical sales data is used for data processing, and the optimal pricing strategy of the commodity is obtained, and the profit of the commodity is maximized under the premise of clearing the inventory.
  • the probability determining unit 320 is further configured to determine, according to the predetermined price, the first price interval, determine whether the quantity of historical sales data in the first price interval is greater than or equal to the threshold; if in the first price interval If the number of historical sales data is greater than or equal to the threshold, the average sales value of the commodity is determined based on the historical sales data of the daily price of the first price interval. For example, if the predetermined price is ⁇ , it may be determined that the first price interval is ⁇ -0.5 to ⁇ +0.5. Those skilled in the art will appreciate that different price ranges can be set depending on the value of the different commodities.
  • the maximum likelihood estimation algorithm can be used to determine that the sales volume of the commodity obeys the positive distribution. If the historical sales volume is expressed as (Q N , Q N-1 , ..., Q 1 ), where N is the number of days of sales.
  • the mean ⁇ of the fitted positive distribution is as follows:
  • the probability determination unit 320 calculates the sales standard deviation ⁇ using the standard deviation formula based on the sales average value. E.g And based on the current inventory amount, the expected inventory amount, the sales average value, and the sales standard deviation, the state transition probability from the current inventory status to the expected inventory status in the case of executing the predetermined price policy is determined.
  • the optimal pricing strategy and the expected goal of a certain commodity under any given inventory level, price and clearance period can be determined, and the subjectivity and inaccuracy brought about by human decision-making can be avoided.
  • the company saves a lot of inventory costs and avoids unnecessary losses.
  • the technology supports manual adjustment of the results, and can quickly calculate the new optimal strategy based on the adjustment, greatly enhancing the flexibility.
  • the probability determining unit 320 is further configured to utilize the formula if the sales volume of the commodity is determined to obey the Poisson distribution using the maximum likelihood estimation algorithm.
  • the sales volume of the goods can be fitted to obey the distribution of positive and too, or the distribution of the Poisson can be made, or it can be fitted to other distributions, and the corresponding formula can be used to calculate the execution of the predetermined price strategy.
  • the state transition probability from the current inventory status to the expected inventory status is sufficient.
  • FIG. 4 is an application architecture diagram of some embodiments of a commodity dynamic pricing data processing system of the present disclosure.
  • An application platform 410, a machine learning platform 420, and a big data platform 430 may be included in the architectural diagram.
  • the present disclosure can achieve good results in the field of sales to slow-moving products.
  • the application platform 410 can implement functions such as information entry, model calling, result output, real-time monitoring, and manual adjustment.
  • the main input of information input includes the inventory quantity of the unsalable goods, the lowest price allowed, the current selling price of the goods, the length of the unsalable clearance, and the target (maximizing sales or maximizing gross profit). If the goal is to maximize the gross profit, the input also includes the cost price.
  • the model call refers to calling the model constructed by the machine learning platform 420 and the corresponding algorithm to derive the suggested price and predicted sales volume of a certain commodity.
  • the result of the output means that the output result is manually confirmed (optional), the commodity information is synchronized to the downstream slow-moving system, and the commodity price is modified to the suggested price.
  • Real-time monitoring and manual adjustment means that during the slow-moving clearance period, the system will capture the sales volume and inventory data of the goods in real time, and adjust the proposed optimal strategy and pricing to achieve the goal of slow-moving clearance, to a greater extent, in the presence of manual intervention.
  • the strategy of the model will be adjusted accordingly to calculate the optimal strategy for a given input.
  • the machine learning platform 420 is capable of data synchronization, calling volume modules, performing Markov decision model construction, and enhancing learning.
  • data synchronization means that after inputting the required information, the system will automatically synchronize the historical information of the commodity from the big data platform 430, mainly including sales volume and transaction price.
  • the volume price model, the Markov decision model and the machine learning specific implementation have been described in detail in the above embodiments, and will not be further elaborated here.
  • the big data platform 430 includes a product database, sales data, and an inventory database.
  • the slow-moving effect can be monitored in real time, and the price can be adjusted in time according to actual sales volume and real-time inventory, thereby improving the effect of slow sales clearance and increasing sales or gross profit.
  • the system also supports human intervention on the proposed price, which is to make a more accurate forecast of the expected sales volume when artificially setting a price, and propose an optimal adjustment strategy accordingly.
  • FIG. 5 is a schematic structural diagram of another embodiment of a commodity dynamic pricing data processing system according to the present disclosure.
  • the system includes a memory 510 and a processor 520, wherein:
  • Memory 510 can be a magnetic disk, flash memory, or any other non-volatile storage medium.
  • the memory is used to store the instructions in the embodiment corresponding to Figures 1-2.
  • the processor 520 is coupled to the memory 510 and can be implemented as one or more integrated circuits, such as a microprocessor or a microcontroller.
  • the processor 520 is configured to execute instructions stored in the memory, and can perform historical data processing using the historical sales data based on the Markov decision model and the enhanced learning algorithm to obtain an optimal pricing strategy for the commodity, and under the premise of clearing the inventory, the maximum is Increase the profitability of goods.
  • the system 600 can include a memory 610 and a processor 620.
  • Processor 620 is coupled to memory 610 via BUS bus 630.
  • the system 600 can also be coupled to the external storage device 650 via the storage interface 640 for invoking external data, and can also be connected to the network or another computer system (not shown) via the network interface 660, and will not be described in detail herein.
  • the data processing can be performed by using the historical sales data based on the Markov decision model and the enhanced learning algorithm to obtain the optimal pricing strategy of the commodity, and the inventory is cleared. Under the premise, maximize the profit of the goods.
  • a computer readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the steps of the method of the embodiment of FIGS. 1-2.
  • a processor implements the steps of the method of the embodiment of FIGS. 1-2.
  • embodiments of the present disclosure may be provided as a method, apparatus, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or a combination of software and hardware aspects. Moreover, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code. .
  • the computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture comprising the instruction device.
  • the apparatus implements the functions specified in one or more blocks of a flow or a flow and/or block diagram of the flowchart.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device such that a series of operational steps are performed on a computer or other programmable device to produce computer-implemented processing for execution on a computer or other programmable device.
  • the instructions provide steps for implementing the functions specified in one or more of the flow or in a block or blocks of a flow diagram.
  • the methods and apparatus of the present disclosure may be implemented in a number of ways.
  • the methods and apparatus of the present disclosure may be implemented in software, hardware, firmware or any combination of software, hardware, firmware.
  • the above-described sequence of steps for the method is for illustrative purposes only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless otherwise specifically stated.
  • the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine readable instructions for implementing a method in accordance with the present disclosure.
  • the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

Abstract

本公开提供了一种商品动态定价数据处理方法和系统，涉及数据处理领域。该方法包括：获取商品的历史销售数据；根据历史销售数据，确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法迭代计算，确定商品的最优价格；其中，预期目标为在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益。本公开基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略，能够最大限度的提高商品的收益。

Description

商品动态定价数据处理方法和系统
相关申请的交叉引用
本申请是以CN申请号为201710510912.7，申请日为2017年6月29的申请为基础，并主张其优先权，该CN申请的公开内容在此作为整体引入本申请中。
技术领域
本公开涉及数据处理领域，尤其涉及一种商品动态定价数据处理方法和系统。
背景技术
随着电子商务的发展，大型电子商务公司通常经营数百万的商品，其中大部分商品是自采自销，并存储在自有的仓库中。在数量众多的商品中，难免会有很多商品由于各种原因产生了滞销，滞销产品一方面占用了公司的大量流动资金，另一方面也占用了宝贵的库存资源。相较于传统零售行业，电子商务公司对库存效率的要求更高，标准更严格。库存成本的精细化控制对于整个公司的运营有着很大的影响，因此对于滞销产品的有效处理对于公司的发展有着重大的影响。
发明内容
发明人认识到，相关技术中，可以利用算法拟合出每个商品的价格弹性参数，并据此预测出某个商品在给定价格条件下的销量，然后，结合现有的库存水平以及期望清仓的时长自动推荐一个最合理的定价，但相关算法无法对建议价格进行动态调整。价格弹性参数在确定后，在整个滞销清仓期间都不会改变，因而缺乏根据最新数据进行自我更新的能力。由于商品在正常销售期间和滞销清仓期间的量价关系很有可能是不同的，而相关算法采用的是正常销售期间的价格弹性参数，因此，这对定价准确度会有一定的影响。另外，因为任何预测都难免会出现误差，在预测销量和实际销量出现偏差时，现有静态模型无法得到反馈并做出相应的调整，所以即使商品的量价关系在滞销清仓期间变化不大，采用静态的量价关系也会影响定价的准确度。再者，相关算法不支持对建议价格进行人为干预。在算法给出建议价格后，出于种种原因，用户有可能对价格进行调整，从而导致相关算法对接下来的清仓效果评估产生偏差，同时，相关算法无法根据当前价格调整未来的定价策略。
本公开要解决的一个技术问题是：提供一种商品动态定价数据处理方法和系统，在清库存的前提下，最大限度的提高商品的收益。
根据本公开的一些实施例，提出一种商品动态定价数据处理方法，包括：获取商品的历史销售数据；根据历史销售数据，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，确定商品的最优价格；其中，预期目标为商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益。
可选地，确定商品的销量均值；基于当前库存量、预期库存量和销量均值，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
可选地，根据预定价格确定第一价格区间；判断在第一价格区间的历史销售数据的数量是否大于等于阈值；若在第一价格区间的历史销售数据的数量大于等于阈值，则根据第一价格区间的每天的历史销量数据确定商品的销量均值；若在第一价格区间的历史销售数据的数量小于阈值，则将第一价格区间逐渐扩大到第二价格区间，以使第二价格区间的历史销售数据的数量大于等于阈值，利用量价模型确定商品在第二价格区间的价格弹性参数，根据价格弹性参数确定商品的销量均值。
可选地，利用最大似然估计算法若确定商品的销量服从正太分布，则基于销量均值和销量标准差确定销量概率密度函数，以当前库存量与预期库存量之差减预定值为下限，以当前库存量与预期库存量之差为上限，对所述销量概率密度函数进行积分计算，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，所示销量标准差根据销量均值利用标准差公式确定。
可选地，利用最大似然估计算法若确定商品的销量服从拿泊松分布，则利用公式
Figure PCTCN2018084382-appb-000001
确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，S为当前库存量，S'为预期库存量，μ为销量均值。
可选地，根据以下公式确定商品的最优价格；F(S,0)＝0；F(0,i)＝0,1≤i≤L；
Figure PCTCN2018084382-appb-000002
Figure PCTCN2018084382-appb-000003
其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格，T(S,α,S')为执行价格α的情况 下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，当前库存量为S时的累计收益，F(S',i-1)为剩余清仓时长为i-1，预期库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
根据本公开的另一些实施例，还提出一种商品动态定价数据处理系统，包括：数据获取单元，用于获取商品的历史销售数据；概率确定单元，用于根据历史销售数据，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；最优价格确定单元，用于以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，确定商品的最优价格；其中，预期目标为商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益。
可选地，概率确定单元用于确定商品的销量均值；基于当前库存量、预期库存量和销量均值，确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
可选地，概率确定单元还用于根据预定价格确定第一价格区间；判断在第一价格区间的历史销售数据的数量是否大于等于阈值；若在第一价格区间的历史销售数据的数量大于等于阈值，则根据第一价格区间的每天的历史销量数据确定商品的销量均值；若在第一价格区间的历史销售数据的数量小于阈值，则将第一价格区间逐渐扩大到第二价格区间，以使第二价格区间的历史销售数据的数量大于等于阈值，利用量价模型确定商品在第二价格区间的价格弹性参数，根据价格弹性参数确定商品的销量均值。
可选地，概率确定单元还用于利用最大似然估计算法若确定商品的销量服从正太分布，则基于销量均值和销量标准差确定销量概率密度函数，以当前库存量与预期库存量之差减预定值为下限，以当前库存量与预期库存量之差为上限，对销量概率密度函数进行积分计算，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，所示销量标准差根据销量均值利用标准差公式确定。
可选地，概率确定单元还用于利用最大似然估计算法若确定商品的销量服从拿泊 松分布，则利用公式
Figure PCTCN2018084382-appb-000004
确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，S为当前库存量，S'为预期库存量，μ为销量均值。
可选地，最优价格确定单元用于根据以下公式确定商品的最优价格；F(S,0)＝0；F(0,i)＝0,1≤i≤L；
Figure PCTCN2018084382-appb-000005
Figure PCTCN2018084382-appb-000006
其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格，T(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，当前库存量为S'时的累计收益，F(S',i-1)为剩余清仓时长为i-1，预期库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
根据本公开的另一些实施例，还提出一种商品动态定价数据处理系统，包括：存储器；以及耦接至存储器的处理器，处理器被配置为基于存储在存储器的指令执行如上述的方法。
根据本公开的另一些实施例，还提出一种计算机可读存储介质，其上存储有计算机程序指令，该指令被处理器执行时实现上述的方法的步骤。
与相关技术相比，本公开实施例基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略，能够最大限度的提高商品的收益。
通过以下参照附图对本公开的示例性实施例的详细描述，本公开的其它特征及其优点将会变得清楚。
附图说明
构成说明书的一部分的附图描述了本公开的实施例，并且连同说明书一起用于解释本公开的原理。
参照附图，根据下面的详细描述，可以更加清楚地理解本公开，其中：
图1为本公开商品动态定价数据处理方法的一些实施例的流程示意图。
图2为本公开商品动态定价数据处理方法的另一些实施例的流程示意图。
图3为本公开商品动态定价数据处理系统的一些实施例的结构示意图。
图4为本公开商品动态定价数据处理系统的一些实施例的应用架构图。
图5为本公开商品动态定价数据处理系统的另一些实施例的结构示意图。
图6为本公开商品动态定价数据处理系统的再一些实施例的结构示意图。
具体实施方式
现在将参照附图来详细描述本公开的各种示例性实施例。应注意到：除非另外具体说明，否则在这些实施例中阐述的部件和步骤的相对布置、数字表达式和数值不限制本公开的范围。
同时，应当明白，为了便于描述，附图中所示出的各个部分的尺寸并不是按照实际的比例关系绘制的。
以下对至少一个示例性实施例的描述实际上仅仅是说明性的，决不作为对本公开及其应用或使用的任何限制。
对于相关领域普通技术人员已知的技术、方法和设备可能不作详细讨论，但在适当情况下，所述技术、方法和设备应当被视为授权说明书的一部分。
在这里示出和讨论的所有示例中，任何具体值应被解释为仅仅是示例性的，而不是作为限制。因此，示例性实施例的其它示例可以具有不同的值。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步讨论。
为使本公开的目的、技术方案和优点更加清楚明白，以下结合具体实施例，并参照附图，对本公开进一步详细说明。
图1为本公开商品动态定价数据处理方法的一些实施例的流程示意图。该方法包括以下步骤110-140。
在步骤110，获取商品的历史销售数据。例如，可以提取预定时间段内该商品成交价在α-0.5到α+0.5之间的销售记录，其中，α为商品销售的价格，销售数据包括销量数据、售价等。
在步骤120，根据历史销售数据，确定商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
在步骤130，以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型。预期目标为商品在执行预定价格策略的情况下从当前库 存量状态到预期库存量状态的预期收益，其中，预期收益可以是预期销售额也可以是预期最大毛利。马尔科夫决策过程(Markov Decision Process，简称MDP)是一种数学模型，用于处理部分随机、部分有决策者控制系统的最优化调度问题。马尔科夫决策过程可以表示为一个如下的元组：
M＝(Sta,Act,T,R)
其中，Sta为一个状态集合；Act是一个包含所有决策的集合；T对所有形如(S,α,S')的状态转移规则指定一个合适的概率，即给定一个状态S、一个决策α和一个后续状态S'，T(S,α,S')表示在执行决策α的情况下，从S转移到S'的概率；R为奖励函数，即执行某个规则的时候，系统能够获得的奖励。具体到滞销模型中，状态为某商品所有可能的库存量，决策为该商品所有可能的定价。奖励函数R的设定取决于给定的目标，如目标为最大化销售额，则规则(S,α,S')对应的奖励为(S-S')×α，其中(S-S')表示库存量的变化，即当天该商品销售的数量，α为商品在当天的售价。如目标设定为最大化毛利，则对应的奖励应为(S-S')×(α-P cost)，其中P cost为该商品的成本价。
该马尔科夫决策模型的主要任务是确定状态转移概率T(S,α,S')，即每条规则(S,α,S')对应的概率，也即在价格为α的情况下，一天销售(S-S')的概率。
在步骤140，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，动态确定商品的最优价格。增强学习又称为强化学习，属于机器学习领域，强调如何基于外部环境的变化动态调整策略，以最大化预期的收益。例如，以价格α为变量，以为剩余清仓时长为i，库存量为S时的累计收益F(S,i)为目标函数进行迭代计算，使得F(S,i)取得最大值，当F(S,i)取得最大值时的α的取值为商品的最优价格。
在一些实施例中，可以利用如下公式计算商品的最优价格，该公式可以称为Bellman(最短路径算法)等式。
F(S,0)＝0；
F(0,i)＝0,1≤i≤L；
Figure PCTCN2018084382-appb-000007
Figure PCTCN2018084382-appb-000008
其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格， T(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，库存量为S时的累计收益，F(S',i-1)为剩余清仓时长为i-1，库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
在该实施例中，基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略，在针对滞销品时，能够在清库存的前提下，最大限度的提高商品的收益。
图2为本公开商品动态定价数据处理方法的另一些实施例的流程示意图。本公开在滞销品销售环节能够起到良好效果，下面将以滞销品为例进行说明，该方法包括以下步骤210-290。
在步骤210，获取滞销品的历史销售数据。
在步骤220，根据预定价格确定第一价格区间。例如，预定价格为α，则可以确定第一价格区间为α-0.5到α+0.5。本领域的技术人员应当理解，根据不同滞销品价值的不同，可以设定不同的价格区间。
在步骤230，判断在第一价格区间的历史销售数据的数量是否大于等于阈值，若大于阈值，则执行步骤240，否则，执行步骤250。例如，若销售记录的数量超过30天，则认为数据可用，否则，认为数据过少。
在步骤240，根据第一价格区间的每天的历史销量数据确定滞销品的销量均值。其中，在价格变动不大的情况下，可以假设销量符合正太分布，这种假设在大部分滞销品销售中能够得到验证。其中，可以利用最大似然估计算法确定滞销品的销量服从正太分布。若历史销量表示为(Q N,Q N-1,...,Q 1)，其中，N为销售的天数。拟合的正太分布的均值μ如下所示：
Figure PCTCN2018084382-appb-000009
在步骤250，将第一价格区间逐渐扩大到第二价格区间，以使第二价格区间的历史销售数据的数量大于等于阈值。例如，当在给定的价格区间销售记录的数量少于30，无法拟合出可信的分布时，则逐步增大价格区间直到有足够数量的历史记录。
在步骤251，利用量价模型确定滞销品在第二价格区间的价格弹性参数。量价模型是一种经济模型，用于衡量滞销品对于价格变动的敏感度。量价模型具体如下所示：
Q＝a×P b
其中，Q为销量参数、P为价格参数、a和b为价格弹性参数，该公式需要通过历史信息，利用最小二乘法拟合。以滞销品A为例，详细步骤如下：
将公式Q＝a×P b两边取自然对数，转换为logQ＝b×logP+loga。取滞销品A前N天的历史销量和历史成交价，分别记为(P N,P N-1,...P 1)和(Q N,Q N-1,...,Q 1)。将历史销量和历史成交价分别取自然对数，取得以下数据(P' N,P' N-1,...,P 1')和(Q' N,Q' N-1,...,Q 1')，其中，P′ i＝logP i,Q′ i＝logQ i,1≤i≤N。再利用最小二乘法拟合系数a'，b'使得
Figure PCTCN2018084382-appb-000010
的取值最小，最后，使a＝e a′，b＝e b′。
在步骤252，根据价格弹性参数确定滞销品的销量均值。利用公式μ＝a×α b，确定销量均值。
在步骤260，根据销量均值利用标准差公式计算销量标准差δ。例如：
Figure PCTCN2018084382-appb-000011
在步骤270，基于当前库存量、预期库存量、销量均值和销量标准差，确定滞销品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
例如，基于销量均值和销量标准差确定销量概率密度函数，以当前库存量与预期库存量之差减预定值为下限，以当前库存量与预期库存量之差为上限，对销量概率密度函数进行积分计算，确定滞销品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，预定值例如为1，则具体公式可以为
Figure PCTCN2018084382-appb-000012
其中，S为当前库存量，S'为预期库存量，μ为销量均值，δ为销量标准差。
在一些实施例中，若经过步骤240-260不能取得满意的拟合结果时，可以参考同一品类下的其他相似商品的历史销量，用以填充当前滞销品的销量数据，在获得足够销量的数据后，重新利用步骤240-260对规则概率进行拟合。
在步骤280，以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型。
在步骤290，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，动态确定滞销品的最优价格。
其中，拟合MDP主要体现的是给定滞销品库存量、价格、销量之间的动态转换关系。利用增强学习技术，可以自动计算出一个最优的定价策略，使得给定的目标最大化。
通过上述方法，可以确定出某滞销品在任意给定库存水平、价格和清仓周期条件下的最优定价策略和预期达到的目标，避免人为决策带来的主观性和不准确性，能够为企业节省大量的库存成本，也避免了不必要的损失。
另外，该实施例在历史信息缺失的情况下，利用经典量价模型对缺失的信息进行回填，提升了模型拟合的准确度，解决信息缺失情况下的滞销定价问题。
在上述实施例中，当在人为设定一个价格的情况下，在系统监控到出现偏差时，可以支持动态调整最优策略，对预期的销量进行更为准确的预测，并据此建议一个最佳的调整策略，从而达到同时支持人工干预和动态调整的目的。
在本公开的另一些实施例中，利用最大似然估计算法若确定滞销品的销量服从拿泊松分布，则在步骤260中，利用公式
Figure PCTCN2018084382-appb-000013
确定滞销品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，S为当前库存量，S'为预期库存量，μ为销量均值。
本领域的技术人员应当理解，在滞销品的销量可以拟合出服从正太分布，也可以使拿泊松分布，还可以是拟合为其他分布，利用相应的公式计算出滞销品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率即可。
图3为本公开商品动态定价数据处理系统的一些实施例的结构示意图。该系统包括数据获取单元310、概率确定单元320和最优价格确定单元330，其中：
数据获取单元310用于获取商品的历史销售数据。例如，可以提取预定时间段内该商品成交价在α-0.5到α+0.5之间的销售记录，其中，α为销售的价格。
概率确定单元320用于根据历史销售数据，确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率，即确定马尔科夫决策模型的输入。状态转移概率T(S,α,S')，即每条规则(S,α,S')对应的概率，也即在价格为α的情况下，一天销售(S-S')的概率。
最优价格确定单元330用于以预期目标为奖励函数，根据状态转移概率、预期累 计收益和奖励函数构建马尔科夫决策模型，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，动态确定商品的最优价格。其中，预期目标为在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益，其中，预期收益可以是预期销售额也可以是预期最大毛利。增强学习又称为强化学习，属于机器学习领域，强调如何基于外部环境的变化动态调整策略，以最大化预期的收益。
在一些实施例中，可以利用如下公式计算商品的最优价格，该公式可以称为Bellman等式。
F(S,0)＝0；
F(0,i)＝0,1≤i≤L；
Figure PCTCN2018084382-appb-000014
Figure PCTCN2018084382-appb-000015
其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格，T(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，库存量为S时的累计收益，F(S',i-1)为剩余清仓时长为i-1，库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
在该实施例中，基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略，在清库存的前提下，最大限度的提高商品的收益。
在本公开的另一些实施例中，概率确定单元320还用于根据预定价格确定第一价格区间，判断在第一价格区间的历史销售数据的数量是否大于等于阈值；若在第一价格区间的历史销售数据的数量大于等于阈值，则根据第一价格区间的每天的历史销量数据确定商品的销量均值。例如，预定价格为α，则可以确定第一价格区间为α-0.5到α+0.5。本领域的技术人员应当理解，根据不同商品价值的不同，可以设定不同的价格区间。在价格变动不大的情况下，可以假设销量符合正太分布，这种假设在大部分商品销售中能够得到验证。其中，可以利用最大似然估计算法确定商品的销量服从正 太分布。若历史销量表示为(Q N,Q N-1,...,Q 1)，其中，N为销售的天数。拟合的正太分布的均值μ如下所示：
Figure PCTCN2018084382-appb-000016
概率确定单元320还用于若在第一价格区间的历史销售数据的数量小于阈值，则将第一价格区间逐渐扩大到第二价格区间，以使第二价格区间的历史销售数据的数量大于等于阈值；利用量价模型确定商品在第二价格区间的价格弹性参数；根据价格弹性参数确定商品的销量均值。例如，当在给定的价格区间销售记录的数量少于30，无法拟合出可信的分布时，则逐步增大价格区间直到有足够数量的历史记录。具体可以利用公式μ＝a×α b，确定销量均值，其中，a和b为价格弹性参数。
概率确定单元320根据销量均值利用标准差公式计算销量标准差δ。例如
Figure PCTCN2018084382-appb-000017
并基于当前库存量、预期库存量、销量均值和销量标准差，确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
例如，根据公式
Figure PCTCN2018084382-appb-000018
确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率T(S,α,S')；其中，S为当前库存量，S'为预期库存量，μ为销量均值，δ为销量标准差。
在上述实施例中，可以确定出某商品在任意给定库存水平、价格和清仓周期条件下的最优定价策略和预期达到的目标，避免人为决策带来的主观性和不准确性，能够为企业节省大量的库存成本，也避免了不必要的损失。
另外，由于信息缺失等各种原因，模型构建过程中也会出现偏差，给出用户认为不合理的定价。这种情况下，该技术支持对结果进行人工调整，并能在调整的基础上，重新快速计算出新的最优策略，极大的增强了灵活性。
在本公开的另一些实施例中，利用最大似然估计算法若确定商品的销量服从拿泊松分布，则概率确定单元320还用于利用公式
Figure PCTCN2018084382-appb-000019
确定在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，S为当前库存量，S'为预期库存量，μ为销量均值。
本领域的技术人员应当理解，在商品的销量可以拟合出服从正太分布，也可以使拿泊松分布，还可以是拟合为其他分布，利用相应的公式计算出在执行预定价格策略 的情况下从当前库存量状态到预期库存量状态的状态转移概率即可。
图4为本公开商品动态定价数据处理系统的一些实施例的应用架构图。架构图中可以包括应用平台410、机器学习平台420和大数据平台430。本公开在应用到滞销品的销售领域，能够取得良好效果。
应用平台410能够实现信息录入、模型调用、结果输出、实时监控以及人工调整等功能。其中，信息录入主要输入包括滞销商品的库存数量、允许的最低价格、商品当前的售价、滞销清仓的时长、目标(最大化销售额或最大化毛利)。如果目标选择的是最大化毛利，输入还要包括成本价。模型调用指调用机器学习平台420所构造的模型和相应算法，得出某个商品的建议价格和预测销量。结果输出指输出的结果经人工确认后(可选)，同步商品信息到下游滞销系统，同时把商品价格修改为建议价格。实时监控和人工调整指在滞销清仓期间，系统会实时抓取商品的销量和库存数据，并对建议的最优策略和定价进行调整，以更大程度实现滞销清仓的目标，在有人工干预存在的情况下，模型的策略也会相应的调整，从而计算出在给定输入下的最优策略。
机器学习平台420能够实现数据同步、调用量价模块、进行马尔科夫决策模型构造和增强学习的功能。其中，数据同步指在输入需要的信息后，系统会自动从大数据平台430同步该商品的历史信息，主要包括销量和成交价等。量价模型、马尔科夫决策模型和机器学习具体实现方式已在上述实施例中进行了详细介绍，此处不再进一步阐述。
大数据平台430中包括商品数据库、销量数据以及库存数据库等。
本公开的实施例中，能够实时监控滞销效果，并根据实际销量、实时库存及时调整价格，从而提升滞销清仓的效果，提高销售额或者毛利等。另一方面，该系统还支持对建议价格进行人为干预，即在人为设定一个价格的情况下，对预期的销量进行更为准确的预测，并据此建议一个最佳的调整策略。
图5为本公开商品动态定价数据处理系统的另一些实施例的结构示意图。该系统包括存储器510和处理器520，其中：
存储器510可以是磁盘、闪存或其它任何非易失性存储介质。存储器用于存储图1-2所对应实施例中的指令。
处理器520耦接至存储器510，可以作为一个或多个集成电路来实施，例如微处理器或微控制器。该处理器520用于执行存储器中存储的指令，能够基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略， 在清库存的前提下，最大限度的提高商品的收益。
在一些实施例中，还可以如图6所示，该系统600包括存储器610和处理器620。处理器620通过BUS总线630耦合至存储器610。该系统600还可以通过存储接口640连接至外部存储装置650以便调用外部数据，还可以通过网络接口660连接至网络或者另外一台计算机系统(未标出)，此处不再进行详细介绍。
在该实施例中，通过存储器存储数据指令，再通过处理器处理上述指令，能够基于马尔科夫决策模型和增强学习算法利用历史销售数据进行数据处理，得到商品的最优定价策略，在清库存的前提下，最大限度的提高商品的收益。
在另一些实施例中，一种计算机可读存储介质，其上存储有计算机程序指令，该指令被处理器执行时实现图1-2所对应实施例中的方法的步骤。本领域内的技术人员应明白，本公开的实施例可提供为方法、装置、或计算机程序产品。因此，本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用非瞬时性存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
至此，已经详细描述了本公开。为了避免遮蔽本公开的构思，没有描述本领域所公知的一些细节。本领域技术人员根据上面的描述，完全可以明白如何实施这里公开的技术方案。
可能以许多方式来实现本公开的方法以及装置。例如，可通过软件、硬件、固件或者软件、硬件、固件的任何组合来实现本公开的方法以及装置。用于所述方法的步骤的上述顺序仅是为了进行说明，本公开的方法的步骤不限于以上具体描述的顺序，除非以其它方式特别说明。此外，在一些实施例中，还可将本公开实施为记录在记录介质中的程序，这些程序包括用于实现根据本公开的方法的机器可读指令。因而，本公开还覆盖存储用于执行根据本公开的方法的程序的记录介质。
虽然已经通过示例对本公开的一些特定实施例进行了详细说明，但是本领域的技术人员应该理解，以上示例仅是为了进行说明，而不是为了限制本公开的范围。本领域的技术人员应该理解，可在不脱离本公开的范围和精神的情况下，对以上实施例进行修改。本公开的范围由所附权利要求来限定。

Claims (14)

  1. 一种商品动态定价数据处理方法，包括：
    获取商品的历史销售数据；
    根据所述历史销售数据，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；
    以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型；
    基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，确定所述商品的最优价格；
    其中，所述预期目标为所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益。
  2. 根据权利要求1所述的方法，其中，
    确定所述商品的销量均值；
    基于当前库存量、预期库存量和销量均值，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
  3. 根据权利要求2所述的方法，其中，
    根据预定价格确定第一价格区间；
    判断在所述第一价格区间的历史销售数据的数量是否大于等于阈值；
    若在所述第一价格区间的历史销售数据的数量大于等于阈值，则根据所述第一价格区间的每天的历史销量数据确定所述商品的销量均值；
    若在所述第一价格区间的历史销售数据的数量小于阈值，则将所述第一价格区间逐渐扩大到第二价格区间，以使所述第二价格区间的历史销售数据的数量大于等于阈值，利用量价模型确定所述商品在所述第二价格区间的价格弹性参数，根据价格弹性参数确定所述商品的销量均值。
  4. 根据权利要求2或3所述的方法，其中，
    利用最大似然估计算法若确定所述商品的销量服从正太分布，则基于销量均值和 销量标准差确定销量概率密度函数，以当前库存量与预期库存量之差减预定值为下限，以当前库存量与预期库存量之差为上限，对所述销量概率密度函数进行积分计算，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；
    其中，所示销量标准差根据销量均值利用标准差公式确定。
  5. 根据权利2或3所述的方法，其中，
    利用最大似然估计算法若确定所述商品的销量服从拿泊松分布，则利用公式
    Figure PCTCN2018084382-appb-100001
    确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；
    其中，S为当前库存量，S'为预期库存量，μ为销量均值。
  6. 根据权利要求1所述的方法，其中，根据以下公式确定商品的最优价格：
    F(S,0)＝0；
    F(0,i)＝0,1≤i≤L；
    Figure PCTCN2018084382-appb-100002
    Figure PCTCN2018084382-appb-100003
    其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格，T(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，当前库存量为S时的累计收益，F(S',i-1)为剩余清仓时长为i-1，预期库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
  7. 一种商品动态定价数据处理系统，包括：
    数据获取单元，用于获取商品的历史销售数据；
    概率确定单元，用于根据所述历史销售数据，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；
    最优价格确定单元，用于以预期目标为奖励函数，根据状态转移概率、预期累计收益和奖励函数构建马尔科夫决策模型，基于马尔科夫决策模型确定预期累计收益表达式，以预定价格为变量，利用增强学习算法进行迭代计算，确定所述商品的最优价格；
    其中，所述预期目标为所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的预期收益。
  8. 根据权利要求7所述的系统，其中，所述概率确定单元用于确定所述商品的销量均值；基于当前库存量、预期库存量和销量均值，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率。
  9. 根据权利要求8所述的系统，其中，所述概率确定单元还用于根据预定价格确定第一价格区间；判断在所述第一价格区间的历史销售数据的数量是否大于等于阈值；若在所述第一价格区间的历史销售数据的数量大于等于阈值，则根据所述第一价格区间的每天的历史销量数据确定所述商品的销量均值；若在所述第一价格区间的历史销售数据的数量小于阈值，则将所述第一价格区间逐渐扩大到第二价格区间，以使所述第二价格区间的历史销售数据的数量大于等于阈值，利用量价模型确定所述商品在所述第二价格区间的价格弹性参数，根据价格弹性参数确定所述商品的销量均值。
  10. 根据权利要求8或9所述的系统，其中，所述概率确定单元还用于利用最大似然估计算法若确定所述商品的销量服从正太分布，则基于销量均值和销量标准差确定销量概率密度函数，以当前库存量与预期库存量之差减预定值为下限，以当前库存量与预期库存量之差为上限，对所述销量概率密度函数进行积分计算，确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，所示销量标准差根据销量均值利用标准差公式确定。
  11. 根据权利要求8或9所述的系统，其中，所述概率确定单元还用于利用最大 似然估计算法若确定所述商品的销量服从拿泊松分布，则利用公式
    Figure PCTCN2018084382-appb-100004
    确定所述商品在执行预定价格策略的情况下从当前库存量状态到预期库存量状态的状态转移概率；其中，S为当前库存量，S'为预期库存量，μ为销量均值。
  12. 根据权利要求7所述的系统，其中，所述最优价格确定单元用于根据以下公式确定商品的最优价格；
    F(S,0)＝0；
    F(0,i)＝0,1≤i≤L；
    Figure PCTCN2018084382-appb-100005
    Figure PCTCN2018084382-appb-100006
    其中，S为当前库存量，S'为预期库存量，L为剩余清仓时长，α为商品的可能价格，P i为商品第i天的最优价格，P min为商品的最小价格，P max为商品的最大价格，T(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，T(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的状态转移概率，F(S,i)为剩余清仓时长为i，当前库存量为S时的累计收益，F(S',i-1)为剩余清仓时长为i-1，预期库存量为S'时的预期累计收益，R(S,α,S')为执行价格α的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，R(S,P i,S')为执行最优价格P i的情况下从当前库存量状态S到预期库存量状态S'的预期最大收益，F(S,0)和F(0,i)为初始值。
  13. 一种商品动态定价数据处理系统，包括：
    存储器；以及
    耦接至所述存储器的处理器，所述处理器被配置为基于存储在所述存储器的指令执行如权利要求1至6任一项所述的方法。
  14. 一种计算机可读存储介质，其上存储有计算机程序指令，该指令被处理器执行时实现权利要求1至6任一项所述的方法的步骤。
PCT/CN2018/084382 2017-06-29 2018-04-25 商品动态定价数据处理方法和系统 WO2019001120A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710510912.7A CN107123004A (zh) 2017-06-29 2017-06-29 商品动态定价数据处理方法和系统
CN201710510912.7 2017-06-29

Publications (1)

Publication Number Publication Date
WO2019001120A1 true WO2019001120A1 (zh) 2019-01-03

Family

ID=59718884

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/084382 WO2019001120A1 (zh) 2017-06-29 2018-04-25 商品动态定价数据处理方法和系统

Country Status (2)

Country Link
CN (1) CN107123004A (zh)
WO (1) WO2019001120A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113343577A (zh) * 2021-06-23 2021-09-03 平安国际融资租赁有限公司 一种参数优化方法、装置、计算机设备及可读存储介质

Families Citing this family (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107123004A (zh) * 2017-06-29 2017-09-01 北京京东尚科信息技术有限公司 商品动态定价数据处理方法和系统
CN109978428A (zh) * 2017-12-28 2019-07-05 北京京东尚科信息技术有限公司 数据估算方法和装置以及计算机可读存储介质
CN108428153A (zh) * 2018-03-12 2018-08-21 上海哔哩哔哩科技有限公司 预售商品的策略调整方法、服务器及策略调整系统
CN110516895B (zh) * 2018-06-20 2021-09-03 北京京东尚科信息技术有限公司 人机仓的选品决策方法和装置、计算机可读存储介质
CN110634007A (zh) * 2018-06-22 2019-12-31 北京京东尚科信息技术有限公司 数据处理方法、装置、系统、计算机可读介质
CN108960923A (zh) * 2018-07-09 2018-12-07 北京百悟科技有限公司 一种定价的方法、装置及计算机存储介质
CN110827047A (zh) * 2018-08-07 2020-02-21 北京京东尚科信息技术有限公司 动态定价方法和装置
CN109272405B (zh) * 2018-09-30 2020-09-29 大唐碳资产有限公司 碳资产交易方法及系统
CN109816459A (zh) * 2019-03-21 2019-05-28 深圳市丰巢科技有限公司 一种动态调价方法、装置、终端及存储介质
CN110298689A (zh) * 2019-05-28 2019-10-01 合肥工业大学 用于更新商品信息的方法、系统及存储介质
CN110570245A (zh) * 2019-09-05 2019-12-13 上海摩库数据技术有限公司 可配置、多维度的实时计价引擎
CN111126917A (zh) * 2019-12-24 2020-05-08 北京每日优鲜电子商务有限公司 单位转移数值调整方法、装置、计算机设备及存储介质
CN110930201A (zh) * 2020-01-02 2020-03-27 山东浪潮通软信息科技有限公司 一种基于动态建模的计价方法及系统
CN111311362B (zh) * 2020-02-12 2023-12-01 安超云软件有限公司 产品属性更新方法以及系统
CN111460363A (zh) * 2020-04-01 2020-07-28 丰车(上海)信息技术有限公司 一种二手车供应链选址与库存水平管理算法
CN113762993A (zh) * 2020-06-03 2021-12-07 阿里巴巴集团控股有限公司 数据处理方法、系统、装置、设备和存储介质
CN112036985A (zh) * 2020-09-04 2020-12-04 烟台冰兔网络科技有限公司 一种电商经营的营销系统
CN112232953A (zh) * 2020-10-14 2021-01-15 深圳壹账通智能科技有限公司 一种债券交易预付款方法、装置、设备及存储介质
CN112819379A (zh) * 2021-02-28 2021-05-18 广东电力交易中心有限责任公司 风险偏好信息的获取方法、系统、电子装置及存储介质
CN114662933B (zh) * 2022-03-24 2023-04-07 杭州良时智能科技有限公司 营销运营风险管控方法及系统
CN115358427A (zh) * 2022-08-31 2022-11-18 成都九洲电子信息系统股份有限公司 一种零部件配置建议采集计算方法
CN117236649B (zh) * 2023-11-10 2024-01-26 天津麦旺生物技术有限公司 一种用于宠物饲料加工原料需求量的调度方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140039979A1 (en) * 2012-08-01 2014-02-06 Opera Solutions, Llc System and Method for Demand Forecasting
CN103593770A (zh) * 2013-10-24 2014-02-19 清华大学 基于马尔科夫模型的亚马逊弹性计算云竞价方法
CN105205701A (zh) * 2015-09-22 2015-12-30 创点客（北京）科技有限公司 一种网络动态定价方法和系统
CN107123004A (zh) * 2017-06-29 2017-09-01 北京京东尚科信息技术有限公司 商品动态定价数据处理方法和系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140039979A1 (en) * 2012-08-01 2014-02-06 Opera Solutions, Llc System and Method for Demand Forecasting
CN103593770A (zh) * 2013-10-24 2014-02-19 清华大学 基于马尔科夫模型的亚马逊弹性计算云竞价方法
CN105205701A (zh) * 2015-09-22 2015-12-30 创点客（北京）科技有限公司 一种网络动态定价方法和系统
CN107123004A (zh) * 2017-06-29 2017-09-01 北京京东尚科信息技术有限公司 商品动态定价数据处理方法和系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113343577A (zh) * 2021-06-23 2021-09-03 平安国际融资租赁有限公司 一种参数优化方法、装置、计算机设备及可读存储介质
CN113343577B (zh) * 2021-06-23 2023-09-26 平安国际融资租赁有限公司 一种基于机器学习的参数优化方法、装置、设备及介质

Also Published As

Publication number Publication date
CN107123004A (zh) 2017-09-01

Similar Documents

Publication Publication Date Title
WO2019001120A1 (zh) 商品动态定价数据处理方法和系统
US11281969B1 (en) Artificial intelligence system combining state space models and neural networks for time series forecasting
Xia et al. A seasonal discrete grey forecasting model for fashion retailing
TWI690865B (zh) 交易量的預測方法及裝置
JP6514342B2 (ja) オンライン予測システムおよび方法
US20140310059A1 (en) System , method and computer program forecasting energy price
CN110059896A (zh) 一种基于强化学习的股票预测方法及系统
US11854022B2 (en) Proactively predicting transaction dates based on sparse transaction data
EP3371764A1 (en) Systems and methods for pricing optimization with competitive influence effects
CN114549046A (zh) 基于融合模型的销售预测方法、系统、设备及存储介质
CN109544233A (zh) 一种面向电商业务的时尚品需求预测方法
WO2019214455A1 (zh) 一种数据序列预测方法及计算设备
CN109767032A (zh) 一种基于数据分析的企业财务运营数字化管理优化系统
WO2020024718A1 (zh) 外汇交易量预测方法和装置
Li et al. A belief-rule-based inventory control method under nonstationary and uncertain demand
CN112950262A (zh) 销量预测方法、装置、计算机设备和存储介质
Saha et al. Demand forecasting of a multinational retail company using deep learning frameworks
CN116188061A (zh) 商品销量预测方法、装置、电子设备及存储介质
JP5251217B2 (ja) 販売数予測システム、販売数予測システムの動作方法および販売数予測プログラム
CN112308293A (zh) 违约概率预测方法及装置
CN115759751A (zh) 一种企业风险预测方法、装置、存储介质、电子设备及产品
CN106537442A (zh) 收集量调节辅助装置，收集量调节辅助方法和计算机可读记录介质
CN106920000B (zh) 基于es预测技术的煤炭供应链牛鞭效应抑制方法
JP7438008B2 (ja) 商品の販売数予測方法、装置及びプログラム、発注量決定方法、装置及びプログラム
CN117011063B (zh) 客户交易风险预测处理方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18824006

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18824006

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 21.04.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18824006

Country of ref document: EP

Kind code of ref document: A1