WO2020121158A1 - Machine à auto-apprentissage pour superviser un système de gestion d'un stock de marchandises - Google Patents

Machine à auto-apprentissage pour superviser un système de gestion d'un stock de marchandises Download PDF

Info

Publication number
WO2020121158A1
WO2020121158A1 PCT/IB2019/060561 IB2019060561W WO2020121158A1 WO 2020121158 A1 WO2020121158 A1 WO 2020121158A1 IB 2019060561 W IB2019060561 W IB 2019060561W WO 2020121158 A1 WO2020121158 A1 WO 2020121158A1
Authority
WO
WIPO (PCT)
Prior art keywords
self
training
inventory
goods
optimization function
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/IB2019/060561
Other languages
English (en)
Inventor
Nicolas MAHLER
Thomas ORIOL
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Datapred SA
Original Assignee
Datapred SA
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 Datapred SA filed Critical Datapred SA
Publication of WO2020121158A1 publication Critical patent/WO2020121158A1/fr
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/087—Inventory or stock management, e.g. order filling, procurement or balancing against orders

Definitions

  • Self-learning machine supervising a system for managing an inventory of goods
  • the present invention concerns a self-learning machine
  • the present invention concerns a process for training the self-learning machine and a system comprising the trained self learning machine.
  • Self-learning machines are used in many industries and services for predicting or supervising systems. As an example, the management of inventory systems already uses self-learning machines for predicting future needs and availabilities of goods in the inventory.
  • Inventory costs are directly related to the quantity of goods in inventory. Storing too few goods could cause shortages leading to losses. But storing too many goods implies increased inventory costs, such as immobilized capital, higher rents, or additional personnel expenses.
  • EP3311348 describes an inventory management apparatus configured to notify a product provider (e.g. food and drinks) about store inventory levels on a daily basis, to avoid product shortages and profit losses.
  • a product provider e.g. food and drinks
  • US2016328724 describes a system and method for ordering products based on demand/sales forecasts provided by a set of dynamically weighted (using a sequential learning algorithm) predictive models.
  • WO2018/217954 describes an online marketplace system for online purchase of goods as well as for the efficient and expedient delivery of the goods purchased.
  • the system comprises an artificial intelligence module having one or more learning platforms including a learning engine and an analytics or interference platform, including an interference engine.
  • the interference engine may be configured for continuously running analytics on received data on a daily basis and/or with regard to one or more special or promotional events, such as prior to, during, and/or after the event, e.g. a shopping event, and functions with the purpose of improving the efficacity of the events results such as by improving the usefulness of producing, consuming, and/or delivery, of goods or services, being sold and purchased through the system.
  • Inventory systems often comprise large amounts of items whose availability needs to be monitored over long periods, often years or even decades, with a fine granularity, often days or even hours. As a
  • An aim of the invention is to provide a system capable of managing inventories of goods in a more cost-effective way than known systems.
  • Another aim of the invention is to provide a self-learning system for supervising an inventory that is more efficient when operational requirements are complex and/or datasets are large.
  • Another aim of the invention is to provide a solution for supervising an inventory that requires a smaller self-learning system and a smaller training dataset.
  • these aims are achieved by means of a method for training a self-learning machine supervising a system for managing an inventory of goods, wherein the self-learning machine comprises a first and second distinct self-learning units; the first self-learning unit being trained with a first input dataset including a reference time set, and the second self-learning unit being trained with a second input dataset including one or many outputs of the trained first self-learning unit.
  • this decomposition of the self-learning system in one first and one second unit may result in a partition of the required training dataset.
  • This partition is for example achieved when the first self-training unit is trained using a part of the training data set in order to achieve an approximation of a first optimization function, such as for example:
  • the second self training unit is trained using a part of the training data set in order to achieve another approximation of a second optimization function such as for example reducing the number of displacements.
  • This partition may also be achieved for example when the first self-training unit is trained according to a first time set (for example monthly variations of the inventory) while the second self-training unit is trained with the results of the first unit and according to a second time set (for example daily or hourly variations).
  • a first time set for example monthly variations of the inventory
  • a second time set for example daily or hourly variations
  • the first input dataset may include reference data representing observed amounts and/or observed variations of the inventory.
  • the first input dataset may include internal reference data representing observed human and/or mechanical events relevant to the inventory, such as for example changes in price, packaging, etc.
  • the reference time set may comprise reference times for: the goods reference data, the internal reference data, and/or the external reference data, for example times or dates of entries or exits from the inventory.
  • the first self-learning unit may infer one or a plurality of supplying orders in order to achieve a first given optimization function, for example a given cost function.
  • the first self-learning unit may be trained with a first
  • optimization function that is a sub-optimal approximation of a given optimization function, for example a cost function. This approximation may for example be the approximation of a first inventory condition.
  • the optimization function may be determined over a first period, for example over successive months.
  • the second self-learning unit may be trained with a second optimization function that is another sub-optimal approximation of a given optimization function, for example another cost function.
  • approximation may for example be the approximation of a second inventory condition.
  • the optimization function may be determined over a second period, for example over successive hours or days.
  • these aims are achieved by means of a method for training a self-learning machine supervising a system for managing an inventory of goods, wherein the self learning machine comprises a first and second distinct self-learning units; the method comprising steps of:
  • A) Collecting an input dataset including, in any order:
  • a reference time set comprising reference times for: the goods reference data, the internal reference data, and/or the external reference data
  • step of training the self-learning machine comprises steps of:
  • the proposed solution provides more accurate and cost-effective inventory management because it is configured to infer re-supplying orders from collected data, not only according to an optimization function, which can be for example a given cost function, but also by considering at least one approximation thereof.
  • splitting the inferring operation into at least two distinct inferring operations is more efficient when operational
  • Fig. 1 shows a schematic view of a system managing one or more inventories of goods diversely located between a factory and final clients, according to the invention
  • Fig.2 and 3 show a self-learning machine for supervising the system of Figure 1, in self-learning mode (Fig. 2) and in operational mode (Fig.3);
  • Fig.4 shows a flow diagram of a method for training a self learning machine, according to the invention
  • Fig. 5 shows a self-learning machine comprising a plurality of self learning units with different learning periods and/or predictive horizons.
  • Figure 1 shows a system 1 for managing an inventory, notably a plurality of inventories 51, 52, 53, 54 of goods
  • the inventory can be the totality of the goods.
  • the inventory can be, or can comprise, a list or the aggregate value of the goods.
  • the list can comprise an identifier (e.g. code number), a quantity, and/or a value for each good.
  • the list can be physical and/or virtual (e.g. represented by digits and/or computer-readable data).
  • a good can be any physical object.
  • the goods can be any mass of material, such as raw materials, semi-finished or finished products, consumer goods, foodstuffs or consumables.
  • the goods can be rare materials, precious materials, and energy resources such as natural gas, electric energy, oil and refined products.
  • the goods can be (fixed or mobile) equipment or tools used for production or distribution, such as but not limited to: robots, machines, vehicles, containers, conveyors, engines.
  • the inventory can be located in a moveable storage unit, such as a moveable container, a shipping container, an intermodal container, the load bed of a vehicle (e.g. truck) and a trailer, or in a static storage unit (i.e. a storage unit not moving or not intended to be moved) such as a box, a safe or a bank vault.
  • a moveable storage unit such as a moveable container, a shipping container, an intermodal container, the load bed of a vehicle (e.g. truck) and a trailer, or in a static storage unit (i.e. a storage unit not moving or not intended to be moved) such as a box, a safe or a bank vault.
  • the inventory can also be located on an open or closed surface, such as a mine, a quarry, a plot of bare or cultivated land.
  • an asset can be, or can represent, any immaterial item that is/can be usable as a medium of exchange and/or as a medium of payment and/or that is convertible into a legal tender (e.g. cash) and/or into credit.
  • the assets can be credentials providing the ownership of (a given amount or quantity of) such an immaterial item, such as credit and debit credentials, currencies, cryptocurrencies, shares, bonds, or stock options.
  • the credentials can be physical and/or virtual (e.g. represented by digits and/or computer-readable data).
  • An asset can be, or can represent, any material item that is/can be usable as a medium of exchange and/or of payment, such as mediums of payment and legal tenders.
  • the inventory can thus be (alternatively or complementarily) located on a postal or bank account, in a portfolio of credentials (e.g.
  • each of the inventories is located in a dedicated storing entity 32, such as a warehouse 32, a hangar, a depot, a basement or cellar or a storage closet in a (trading or commercial) building 33 or in a creating entity 31, such as a factory 31, a fabrication site, a power plant or an oil well.
  • a dedicated storing entity 32 such as a warehouse 32, a hangar, a depot, a basement or cellar or a storage closet in a (trading or commercial) building 33 or in a creating entity 31, such as a factory 31, a fabrication site, a power plant or an oil well.
  • the storing entities 32 can be interconnected, or connected to one or more creating entities 31 by means of distribution channels 41-45.
  • the illustrated system 1 comprises a goods data module 24, an internal data module 25, an external data module 26, a timestamping module 27 and a self-learning machine 2.
  • the goods data module 24 is configured to collect goods data 811, 812, 813 representing observed amounts and/or observed variations of the inventory 51, 52, 54, notably of the goods thereof.
  • the observed amounts and/or variations can be related to the amounts of goods in the inventory.
  • the observed amounts and/or variations are preferably observed, registered and/or determined at a specific time or within a specific time interval.
  • the internal data module 25 is configured to collect internal data 821, 822, 823, 824 representing observed human and/or mechanical events on the goods of the inventory.
  • the observed human and/or mechanical events can notably relate to:
  • the cost of the inventory such as production, acquisition, managing, and/or sales costs of the goods
  • the external data module 26 is configured to collect external data 831 representing human, natural, and/or environmental events affecting a given geographical region 61 having a relationship with and/or a connection to the inventory.
  • the geographical region 61 can relate to:
  • the geographical region can be designed by a governmental and/or commercial organization. Alternatively or complementarily, the geographical region can be defined in relation to the inventory.
  • the human events can relate to human activities such as concerts, sports competitions, school and academic overtures and/or closures, holidays in the corresponding geographic region. Human activities can further concern the financial state and/or health of the geographical region 61, the purchasing power of residents and/or potential clients 62.
  • the environmental events can relate to temperature, weather, and/or season of the geographical region.
  • the natural events can relate to weather conditions, to a weather hazard (e.g. wind, thunderstorms, rain, snow, heat wave and frost hazards), natural disasters (e.g. floods) affecting the geographical region.
  • a weather hazard e.g. wind, thunderstorms, rain, snow, heat wave and frost hazards
  • natural disasters e.g. floods
  • the system further comprises a timestamping module 27 configured to provide a consistent (i.e. unvarying, unmodifiable) time reference for the goods data, the external data, and/or to the internal data.
  • Each module can be a dedicated electronic circuit, a set of instructions executed on a (shared and/or common) processor of a device, and/or a combination thereof.
  • the self-learning machine 2 of the system 1 is configured to infer a supplying order 911, 912, 913 using a given optimization function 90 assigned to the inventory, notably by a manager or an operator of the system 1 in order to provide a more cost-effective solution for the management of an inventory.
  • the self-learning machine 2 is trained to provide the supplying order 91 in response to an input dataset 89 comprising:
  • a supplying order 911, 912, 913 can be any (physical or virtual) order causing a modification of the goods of the inventory, notably by an increase or a reduction of the inventory.
  • the modification can relate to (at least a parameter of) the amount and/or value of the inventory, such as a quantity and a mass of goods of the inventory, notably by adding new goods to the inventory and/or by delivering (parts of) the goods of the inventory.
  • the modification can be immediate or short-term, i.e. within a short time interval from the execution of the inferred supplying order.
  • the short interval can be a second, a minute, an hour, or a day.
  • the modification can take effect, or being configured to take effect within a time interval that is delayed with respect to the execution of the inferred supplying order.
  • the supplying order 911, 912, 913 can also be an order potentially causing a modification of the inventory, in case the order is configured and/or destined to stabilize the inventory (e.g. amount and/or the value thereof), e.g. following a variation of:
  • the supplying order can be based (notably by a logical or mathematical function) on an inferred:
  • the given optimization function 90 can include any element and/or combination of elements contributing to the cost of the inventory, such as related effort, material resources, time and utilities consumed, risk incurred, and opportunity forgone. Examples are: an economic inventory cost, immobilized capital, rents for storing the inventory, transportation cost and times, return-on-investment time, loss of sales and/or clients due to sold-outs, (economic) penalty.
  • the given optimization function 90 can also include any element and/or combination of elements generating an income related to the inventory, such as a profit, an excess of revenue, an opportunity. Examples are: a return-on-investment, an increase in sales and/or clients.
  • the given optimization function 90 can also include any element and/or combination of elements related to client satisfaction and promotions affecting the inventory.
  • the given optimization function can be related to a specific time horizon (i.e. a fixed point of time in the future at which point the
  • the given optimization function can be also related to (evaluated within) a specific time period (e.g. defined in one or more second(s), minute(s), hour(s), day(s), week(s), month(s), year(s), or by a combination thereof).
  • a specific time period e.g. defined in one or more second(s), minute(s), hour(s), day(s), week(s), month(s), year(s), or by a combination thereof).
  • the self-learning machine 2 of the system 1 comprises a plurality of self-learning units 21, 22.
  • Each of these self-learning units is configured (i.e. trained) to infer a specific output from a dataset according to a specific optimization function.
  • the self-learning machine 2 can comprise two or more self learning units.
  • the self-learning machine 2 comprises a first and a second self- learning units 21, 22.
  • a plurality of self-learning units allows to split the inferring operation of the self-learning machine 2, into a plurality (i.e. equal number) of distinct inferring operations.
  • the self-learning machine 2 can thus be efficiently trained by collecting an input dataset including, in any order:
  • the step of collecting an input dataset also includes a step of providing a reference time set 80 comprising reference times for: the goods reference data, and/or the internal reference data, and/or the external reference data.
  • the method for training the self-learning machine 2 further comprises a step of building a training dataset 88 by pairing the goods reference data 81 and/or the internal reference data 82 and/or the external reference data 83 with the reference time set 80. [0067] The self-learning machine 2 can then be trained with the built training dataset 88 so as to infer the supplying order 91 optimizing the given optimization function 90.
  • the training of the self-learning machine 2 thus comprises steps of:
  • the training database used for training the first self-learning unit 21 can correspond to the entire, or to a (selected) portion, of the training database 88 provided to the self-learning machine 2.
  • the training can thus comprise a step of training another self learning unit (e.g. the second self-learning unit 22) using an output 211 of the (already trained) first self-learning unit 21, i.e. the training dataset used for training the second self-learning unit 22 comprises this output 211.
  • This training dataset can further comprise the entire, or a (selected) portion, of the training database 88 provided to the self-learning machine 2.
  • the training of the second self- learning unit 22 can be executed, as shown in Figure 3, according to a second optimization function 902 which may be a (another) sub-optimal approximation of the given optimization function 90.
  • the training dataset used for training the second self-learning unit 22 can comprise the entire, or a (selected) portion, of the training database 88 provided to the self- learning machine 2.
  • the supplying order 91 can be based (e.g. inferred, calculated, mathematically or logically derived) from the output 211 of the first self-learning unit 21 and from the output 221 of the second self-learning unit 22, e.g. by means of an aggregating unit 23 and/or by means of one or more additional self-learning unit(s) of the self learning machine 2.
  • the supplying order 91 can be obtained by aggregating the output 211 of the first self-learning unit 21 and the output 221 of the second self-learning unit 22 by means of an aggregating unit 23 being trained or configured to aggregate these results to optimize the given optimization function of the self-learning machine 2.
  • the supplying order 91 can be obtained by providing one or both of the outputs 211, 221 of the first and second self-learning units 21, 22 to one or multiple other self-learning units of the self-learning machine 2, each unit being trained to optimize another (sub-optimal) approximation of the given optimization function and/or the given optimization function.
  • the supplying order 91 can thus be obtained by training a (final) self-learning unit or by aggregating a set of outputs provided by two or more self-learning units by means of an aggregating unit, according to the given optimization function.
  • the second self-learning unit 22 can be alternatively trained according to the given function of the self-learning machine 2.
  • the second self-learning unit 22 acts as a final self-learning unit providing the supplying order 91.
  • a (sub-optimal) approximation of the given optimization function can be any optimization function:
  • approximation of the given optimization function can be any optimization function related to:
  • each self-learning unit of the self-learning machine 2 comprises a set of distinct models 212- 214, 222- 224, each being individually trained according to the optimization function assigned to the self-learning unit, or according to any relevant and pre-specified
  • the output of the self-learning unit is provided by aggregating the (set of) outputs provided by the set of distinct models 215-217, 225-227.
  • the aggregation can be provided by an aggregating unit 218, 228 being trained or configured to linearly combine with or without weighting the (set of) outputs to optimize the optimization function assigned to the self-learning unit.
  • the training of a self-learning unit 21, 22 of the self-learning machine 2 can thus comprise the steps of:
  • the method can further include, after the determination of the aggregation, the steps of:
  • the method can comprise the steps of:
  • the set of models can comprise one or more of the following models:
  • a simple lag model e.g. providing (a part of) the input dataset delayed by a given time interval (e.g. a day, a week or a year);
  • Figure 4 shows a flow diagram of an example method for configuring and training the self-learning machine 2 for supervising cost- optimally or cost effectively a system 1 managing an inventory 51, 52, 53, 54 of goods and/or assets.
  • the first step (S 1 ) concerns the definition of one or more approximated optimization functions for the given optimization function.
  • the approximated optimization function can be derived, extracted or identifying by
  • Each approximated optimization function is then assigned to a distinct self-learning unit.
  • the self-learning machine comprises (at least) the same number of self-learning units as the number of defined approximated optimization functions.
  • the aim of the following step (S2) is the creation of a flow graph, which expresses the dependencies between the inputs and/or outputs of the self-learning units of the self-learning machine 2.
  • the input and the output are defined, notably according to the output provided by other self-learning units.
  • the flow graph can comprise a final self-learning unit providing the supplying order 91 from an output or a set of outputs provided by (an)other self-learning unit(s).
  • the independent self-learning unit i.e. the self learning unit whose inputs are devoid of outputs from other self-learning units.
  • the dependent self-learning node can be (individually) trained (S5, S6).
  • the training then comprises a (final) aggregation or training (S7) for optimizing the given optimization function.
  • This step involves the optimization of an aggregating unit processing a set of outputs or a training of the final self-learning unit.
  • the training of the self-learning unit and/or the definition of the approximated optimization functions can be preceded by a step of data loading and/or pre-processing, e.g. stationarization, de-seasonalization, denoising, and/or outlier filtering.
  • pre-processing e.g. stationarization, de-seasonalization, denoising, and/or outlier filtering.
  • the step of training the self-learning units or of aggregating can comprise a step of modifying, cancelling and/or adding dependencies between self-learning units following observed training and/or
  • a self-learning machine 2 comprising a plurality of self-learning units, at least one configured/trained with a shorter horizon and another with a longer time horizon, allows to aggregate and/or infer the supplying order 91 based on the short-term, mid-term, and long-term trends inferred by the self-learning units.
  • Fig. 5 shows a self-learning machine trained according to a given optimization function, e.g. for providing a buying order for a steel inventory optimizing a financial profit and loss (PnL) function, at a given time horizon, e.g. 1-day time horizon.
  • the self-learning machine comprises a plurality (a set) of self learning units 28, 28', 28", 28'"
  • a sub-set of the self-learning units 28-28" is arranged as independent units, i.e. units non-depending on outputs provided by others units. Each of these units is configured to optimize the given optimization function (or another approximated optimization function) at:
  • a shorter time horizon e.g. 12-hours
  • time horizon e.g. 1-day time horizon; or a longer time horizon, e.g. 3-day, 5-day, 10-day, 15-day and 20- day time horizons.
  • the self-learning machine 2 then comprises a final self-learning unit 28'" arranged to (sequentially) aggregate the outputs 281-281"' provided by the independent units 28-28" (as input 280"') to provide the supplying order 91 according to the given optimization function (e.g. the PnL function) at the given time horizon (1-day time horizon).
  • the given optimization function e.g. the PnL function
  • the self-learning unit with a shorter time horizon provides an inferring of short-term up to mid-term trends permitting the self-learning machine to rapidly react to trend reversals.
  • the self-learning units with a longer time horizon provide an inferring of mid-term up to long-term trends to avoid overfitting effects.
  • acts, events, or functions of any of the algorithms described herein can be performed in a different sequence, can be added, merged, or left out altogether (for example, not all described acts or events are necessary for the practice of the methods).
  • acts or events can be performed concurrently, for instance, through multi-threaded processing, interrupted processing, or on multiple processors or on processor cores or on other parallel architectures, rather than sequentially.
  • different tasks or processes can be performed by different machines or computing systems that can function together.
  • a hardware processor can include electrical circuitry or digital logic circuitry configured to process computer-executable instructions.
  • a processor includes an FPGA or other programmable device that performs logic operations without processing computer-executable instructions.
  • a processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
  • a processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
  • computing environment can include any type of computer system, including, but not limited to, a computer system based on a microprocessor, a mainframe computer, a digital signal processor, a portable computing device, a device controller, or a computational engine within an appliance, to name a few.
  • a computer system based on a microprocessor, a mainframe computer, a digital signal processor, a portable computing device, a device controller, or a computational engine within an appliance, to name a few.
  • the steps of a method, process, or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module stored in one or more memory devices and executed by one or more processors, or in a combination of the two.
  • a software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of non-transitory computer- readable storage medium, media, or physical computer storage.
  • An example storage medium can be coupled to the processor such that the processor can read information from, and write information to, the storage medium.
  • the storage medium can be integral to the processor.
  • the storage medium can be volatile or nonvolatile.
  • Self-learning units designate any computer system used to perform a task without using explicit instructions, relying on patterns and inference instead.
  • the self-learning unit may be supervised or unsupervised. It may be implemented for example as a neural network. It may be implemented as a hardware, software or mixed solution. Reducing the size of a self-learning unit may include reducing the number of neurons, reducing the number of layers of the network, reducing the number of inputs, reducing the number of interconnexions, or any combination of those reductions.
  • a cost or optimization function may be implemented as a table or as a software module or function for representing a cost depending on certain input data.
  • a cost or optimization function may indicate the volume that is required for storing a number of items given as input.
  • Conditional language used herein such as, among others, “can,” “might,” “may,” “e.g.,” and the like, unless specifically stated otherwise, or otherwise understood within the context as used, is generally intended to convey that certain embodiments include, while other embodiments do not include, certain features, elements or states. Thus, such conditional language is not generally intended to imply that features, elements or states are in any way required for one or more embodiments or that one or more embodiments necessarily include logic for deciding, with or without author input or prompting, whether these features, elements or states are included or are to be performed in any particular embodiment.

Landscapes

  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Quality & Reliability (AREA)
  • Finance (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Human Resources & Organizations (AREA)
  • Accounting & Taxation (AREA)
  • Operations Research (AREA)
  • Development Economics (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

L'invention concerne un procédé d'entraînement d'une machine à auto-apprentissage (2) qui supervise un système (1) de gestion d'un stock (51, 52, 53, 54) de marchandises de manière à inférer une commande de fourniture (91, 911, 912, 913) qui optimise une fonction d'optimisation donnée (90). L'entraînement de la machine à auto-apprentissage (2) comprend une étape d'entraînement d'une première unité d'auto-apprentissage (21) de la machine à auto-apprentissage (5) selon une première fonction d'optimisation (90) étant une approximation sous-optimale de ladite fonction d'optimisation donnée. L'entraînement comprend également une étape d'entraînement d'une seconde unité d'auto-apprentissage (22) à l'aide d'une sortie de la première unité d'auto-apprentissage entraînée (21) et/ou en fonction d'une seconde fonction d'optimisation qui est une approximation sous-optimale de ladite fonction d'optimisation donnée (90).
PCT/IB2019/060561 2018-12-10 2019-12-09 Machine à auto-apprentissage pour superviser un système de gestion d'un stock de marchandises Ceased WO2020121158A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CH15222018 2018-12-10
CH01522/18 2018-12-10

Publications (1)

Publication Number Publication Date
WO2020121158A1 true WO2020121158A1 (fr) 2020-06-18

Family

ID=65811985

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/IB2019/060561 Ceased WO2020121158A1 (fr) 2018-12-10 2019-12-09 Machine à auto-apprentissage pour superviser un système de gestion d'un stock de marchandises

Country Status (1)

Country Link
WO (1) WO2020121158A1 (fr)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160328724A1 (en) 2015-05-06 2016-11-10 Wal-Mart Stores, Inc. System and method for forecasting with sparse time panel series using dynamic linear models
EP3311348A1 (fr) 2015-06-17 2018-04-25 Panasonic Intellectual Property Management Co., Ltd. Appareil de gestion des stocks, procédé et système
WO2018217954A1 (fr) 2017-05-23 2018-11-29 Mercato, Inc. Systèmes et procédés permettant d'attribuer et de répartir un inventaire

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160328724A1 (en) 2015-05-06 2016-11-10 Wal-Mart Stores, Inc. System and method for forecasting with sparse time panel series using dynamic linear models
EP3311348A1 (fr) 2015-06-17 2018-04-25 Panasonic Intellectual Property Management Co., Ltd. Appareil de gestion des stocks, procédé et système
WO2018217954A1 (fr) 2017-05-23 2018-11-29 Mercato, Inc. Systèmes et procédés permettant d'attribuer et de répartir un inventaire

Similar Documents

Publication Publication Date Title
Albertzeth et al. Mitigating transportation disruptions in a supply chain: a cost-effective strategy
Violi et al. The inventory routing problem under uncertainty with perishable products: an application in the agri-food supply chain
Iyer et al. Quick response in manufacturer-retailer channels
Martel et al. Designing value-creating supply chain networks
Sui et al. A reinforcement learning approach for inventory replenishment in vendor-managed inventory systems with consignment inventory
KR102326878B1 (ko) Ｓｃｍ 기반 제３자 물류 서비스 제공 시스템
Ahmadi et al. A bi-objective location-inventory model with capacitated transportation and lateral transshipments
CN104115167A (zh) 用于集中采购的方法和装置
Martel et al. Supply chains: Issues and opportunities
JP2024536940A (ja) 最適化されたツリー・アンサンブル・ベースの需要モデル
Pakhira et al. Two-level supply chain of a seasonal deteriorating item with time, price, and promotional cost dependent demand under finite time horizon
Oguntola et al. On the value of shipment consolidation and machine learning techniques for the optimal design of a multimodal logistics network
Rogulin A model for optimizing plans for procurement of raw materials from regions of Russia in a timber-processing enterprise
Shukla et al. An inventory model for continuously deteriorating agri–fresh produce: an artificial immune system–based solution approach
Toe Teoh et al. AI in supply chain management
WO2020121158A1 (fr) Machine à auto-apprentissage pour superviser un système de gestion d'un stock de marchandises
Caro et al. Coordination of inventory distribution and price markdowns for clearance sales at zara
Guigues et al. Robust management and pricing of liquefied natural gas contracts with cancelation options
Troyer et al. Improving asset management and order fulfillment at Deere & Company's C&CE division
Rahbari et al. Wheat supply chain network design: Lesson for resilience and sustainability in a situation of war and crisis
Chen et al. Dynamic programming model for attended delivery time slot management
Lutsenko et al. A METHOD TO FORM CONTROL OVER QUEUING SYSTEMS TAKING INTO CONSIDERATION THE PROBABILISTIC CHARACTER OF DEMAND.
Namazian et al. Predict and optimize: a smart inventory management model for beauty retail: Z. Namazian et al.
Sengar et al. Review on trends in machine learning applied to demand & sales forecasting
Bordusenko AUTOMATION OF FINANCIAL FLOWS IN INDUSTRY: INTELLIGENT ALGORITHMS FOR PROCUREMENT AND INVENTORY MANAGEMENT

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: 19835485

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: 19835485

Country of ref document: EP

Kind code of ref document: A1