• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    Provided are methods and computerized tools for supply chain optimization. In one aspect, the present invention relates to computerized methods of optimizing a supply chain, the supply chain comprising a plurality of depots and at least one production facility, the method involving determining shipment volumes, storage volumes. and / or storage trigger level in said supply chain, and the method comprising the steps of a) forecasting demand using discrete event simulation and Monte Carlo simulations; b) adjust the replenishment strategy; c) propagate the information of the request backwards; d) planning production; e) evaluate the final strategy; and f) seek the best strategies.
    公开号:BE1023357B1
    申请号:E2016/5667
    申请日:2016-08-30
    公开日:2017-02-14
    发明作者:Benoit David；Sébastien COPPE；Sébastien MOUTHUY；Philippe Chevalier
    申请人:N-Side Sa；
    IPC主号:
    专利说明:

    METHODS FOR DETERMINING SHIPPING VOLUMES AND / OR STORAGE RELEASE LEVELS
    FIELD OF THE INVENTION
    Methods and computerized supply chain optimization tools are described herein. In one aspect, the present invention relates to computerized supply chain optimization methods, the supply chain comprising a plurality of depots and at least one production facility, the method involving the determination of shipping volumes, storage volumes. and / or storage trigger levels in said supply chain, and the method comprising the steps of: a) predicting the request using discrete event simulation and Monte Carlo simulations; b) adjust the replenishment strategy; (c) propagate backward the information in the application; d) plan production; e) evaluate the final strategy; and f) search for the best strategies.
    CONTEXT
    The logistics chains of large companies are complex systems that require advanced computer technologies for their effective management. In particular, the management of shipping volumes, the size of storage stocks and the minimum storage levels at which replenishment rounds are triggered is very difficult. This is particularly the case in complex logistics chains covering many companies and / or many different countries. Mathematical modeling and optimization are nowadays essential computerized techniques used to reduce the cost and to control the level of service of the supply chains.
    However, existing processes typically do not have the speed, detail, robustness, versatility, and adaptability needed in the day-to-day practice of supply management. As a result, there is a need for innovative processes for simulating and optimizing complex and changing supply chains, including the simulation and optimization of inventory levels, shipping volumes and / or trigger levels. replenishment.
    SUMMARY OF THE INVENTION
    The methods described herein overcome one or more of the disadvantages of the prior art processes. In particular, the methods described here provide a highly efficient and accurate method for integrating the three main themes of supply chain management: demand forecasting, inventory management, and production planning. As a result, the output from the production of goods, the storage of goods and / or the shipping of goods can be significantly improved. In addition, the methods described herein can deal with the particular case of a supply chain that never reaches a stationary state and where the variability of demand over time is a major complication. Such changing supply chains typically require a dynamic adaptation of the supply strategy based on an exact demand forecast and an adaptive replenishment algorithm.
    The inventors have found a way to take advantage of the combination of multiple advanced mathematical and computer-based techniques in a set of algorithms to accurately model the variability of demand and to infer from this variability the best inventory management strategy and the best planning of production both in terms of cost and risk (level of service). Shipping volumes, shipping intervals, storage volumes, and / or replenishment trigger levels can then be adapted accordingly.
    In some embodiments, the methods of the present invention provide a minimum cost curve for each given level of risk. A level of risk is measured as the probability of out of stock in the supply chain. From these curves, decision-makers can select the procurement strategy that offers the best compromise between cost and risk. Shipping volumes, shipping intervals, storage volumes, and / or replenishment trigger levels can then be adapted to coincide with this optimal tradeoff. As a result, the cost of logistics can be lowered while maintaining the risk of out of stock below a desired threshold.
    More particularly, the invention relates to computerized supply chain optimization methods, the supply chain comprising a plurality of depots and at least one production facility, the method involving the determination of shipping volumes, storage volumes and / or storage trigger levels in said supply chain, and the method comprising the steps of a) predicting the request using discrete event simulation and Monte Carlo simulations; b) adjust the replenishment strategy; (c) propagate backward the information in the application; d) plan production; e) evaluate a final strategy; and; f) search for the best strategies.
    In certain embodiments, the method further comprises the step of: g) according to one of said best strategies, performing one or more selected actions in the list comprising: • producing goods in said production facility (s) in a quantity predetermined and / or at a predetermined rate; • maintain the quantity of goods in one or more deposits included in said plurality of deposits in a predetermined range; • receive property in one or more deposits included in said plurality of deposits; • ship goods to and / or from the facility or production facilities, and / or the deposit (s) included in said plurality of depots; and or ; • ship goods between one or more depots.
    In some embodiments, the plurality of repositories includes a set of local repositories and / or the supply chain may further include a plurality of request locations, the request locations possibly including stores, and step a) involving the next sub-step: ai) modeling the supply chain demand using a set of agents, agents going to multiple demand locations, and agent behavior being simulated by modeling the way the agent consumes products.
    In some embodiments, the agent behavior modeling further includes the substep of: ai) taking into account multiple uncertainty parameters modeled by a probability curve, the multiple uncertainty parameters being selected from the list including the opening of new places or the closing of existing places of application, the affiliation of customers, a profile or a consumption stratum assigned to customers, the frequency of visits to a place of demand, the product consumed by the agent, and the probability of abandoning the agent.
    In some embodiments, the supply chain has a plurality of logistic endpoints and step a) involves a demand simulation on all logistic endpoints at the same time.
    In some embodiments, step b) involves the substep of: bi) calculating replenishment trigger levels and replenishment quantities from the demand achievements at each supply chain endpoint ; the replenishment trigger levels corresponding to predetermined quantities of goods present in the plurality of demand locations, at or below which said demand locations are replenished.
    In some embodiments, the trigger levels are calculated such that the inventory remaining at the replenishment time is sufficient to cover the demand occurring during the replenishment lead time with a certain probability Ax.
    In some embodiments, step b) involves calculating shipping volumes to cover the average demand over a time horizon B1.
    In some embodiments, step c) involves the substep of: ci) determining trigger levels and shipping volumes for consecutive level nodes i in the supply chain, where the nodes include plurality of deposits, the plurality of demand locations, and the production facility or facilities, • trigger levels being calculated such that the remaining inventory in a level i node at the time of replenishment is sufficient to cover the demand occurring during the replenishment period with a certain probability At; and, • the shipping volumes being determined to cover the average demand at a level i node over a time horizon Bi.
    In some embodiments, step d) includes the substep of: di) determining the production planning to satisfy the consolidated demand with probability C.
    In some embodiments, step e) includes measuring the risk that a product is not available at an endpoint to meet a demand, and measuring the total cost of the supply chain.
    In some embodiments, the total cost of the supply chain includes the cost of materials and production, the cost of shipping, the cost of storage, and the potential cost of disposal.
    In some embodiments, the method involves the use of non-derivative optimization; preferably in which, for i-level nodes, with i = 1 to n, and n being the number of levels in the supply chain, Au Bu and C are determined by means of non-derivative optimization.
    In some embodiments, non-derivative optimization involves step-by-step introduction of new strategies for updating other strategies, and the distance between new strategies and other strategies is determined from the distance between the others. strategies.
    In some embodiments, the methods include the following steps: • Using the mean and variance of cost and risk, and the number of simulations that have been executed for each strategy, to calculate the probability that a strategy will outperform the performance of the strategies. other strategies, the central limit theorem and Bayesian inference being used in the calculation; • If the probability that a strategy outperforms one or more other strategies is less than a predetermined value, increase the number of simulations performed on that strategy until that probability is greater than or equal to that value. predetermined.
    BRIEF DESCRIPTION OF THE DRAWINGS
    The following description of the figures of particular embodiments of the invention is only illustrative in nature and is not intended to limit the present teachings, their application or their uses.
    Fig. Figure 1 shows a schematic representation of the trade-off between risk (X-axis) and cost (Y-axis) that is addressed in the methods of the present invention.
    In all the figures, the following numbering is adopted: 100 - inefficient strategies; 200 - optimal strategies; 210-high cost and low risk strategies; 220 - low cost and high risk strategies; 300 - impossible strategies.
    DETAILED DESCRIPTION
    The present invention will be described with respect to particular embodiments, but the invention is not limited thereto but only by the claims.
    As used herein, the singular forms "a", "an", and "the" include both singular and plural referents, unless the context obviously dictates otherwise.
    The terms "comprising", "comprises" and "consisting of" as used herein are synonymous with "comprising", "includes" or "containing", "contains", and are inclusive or unbounded and do not exclude additional process elements or steps. The terms "comprising", "comprises" and "consisting of", when referring to elements or process steps cited herein, also include embodiments that "consist of" said process elements or process steps right here.
    In addition, the terms first, second, third and the like as used herein are used to distinguish similar elements and not necessarily to describe a sequence or chronological order, unless specified. It should be understood that the terms so used are interchangeable under appropriate circumstances and that the embodiments of the invention described herein are likely to operate in other orders than those described or illustrated herein.
    The term "about" as used herein when referring to a measurable value such as a parameter, a quantity, a time duration, etc., is understood to encompass variations of at most +/- 10%, preferably not more than + 1-5%, more preferably not more than +/- 1%, and ideally not more than +/- 0.1% of and in relation to the specified value, in so far as It is appropriate to make such variations in the described invention. It should be understood that the value referred to by the "about" modifier is itself also, and preferably specifically, described.
    The specification of digital ranges by endpoints includes all numbers and fractions embedded within the respective ranges, as well as the specified endpoints. Unless otherwise defined, all terms used in the description of the invention, including technical and scientific terms, have the meanings commonly understood by a person normally skilled in the art to which this invention belongs. . As a further guide, the definitions of the terms used in the description are included to better understand the teaching of the present invention. The terms or definitions used herein are provided solely to assist in understanding the invention.
    In some aspects, methods for the simulation and optimization of complex and changing logistic chains are described herein. This includes determining shipping volumes and / or storage trigger levels. One of the main added values of these processes comes from their ability to handle a very large amount of data, and to represent the variability of demand. Another major advantage is that the methods described here have a good computational efficiency. In some embodiments, demand is propagated back and forth along the supply chain to optimize the inventory management strategy at different levels. In some embodiments, the different levels include local depots, regional depots, and central warehouses. In addition, the processes make it possible to optimize the production planning at the production sites.
    The terms "supply chain" and "supply network" are used interchangeably here.
    In some embodiments, the supply networks have four levels: level 1, level 2, level 3, and level 4. Preferably, level 1 includes demand locations or endpoints (eg, retail stores at retail, or hospitals in clinical trials) that supply the end customer. Preferably, level 2 includes regional or local repositories (typically supplying a country's request locations). Preferably, level 3 has a central warehouse (typically supplying the regional depots of a continent). Preferably, level 4 includes production sites (typically supplying central warehouses). Preferably, the central warehouses are located on production sites.
    The terms "client" and "agent" as used herein represent the same entity. A customer is the natural person generating a request. In some embodiments, the customer is a buyer and / or a consumer. The term "agent" generally refers to the way a real client is represented in the algorithm. In general, an agent is a more generic concept than a client. In very general and abstract terms, an "agent" typically represents the simulation of an entity that generates a certain behavior that influences the simulation results at particular moments.
    In some embodiments, the request is captured by modeling all possible "clients" with agents. The method then simulates the possible behaviors of the agents over time. In addition, a simulation of how these behaviors generate demand is included. The behavior of an agent is typically governed by certain random variables such as the frequency of its visits to a logistic endpoint generating a request. Another random variable is typically used to represent the product type (s) for which the request is generated during said visit. Generally, in order to represent the possible behaviors of the agents, multiple simulations with different and independent implementations of these random variables may be useful. In addition, assumptions can be made concerning the parameters of these random variables (for example the minimum and maximum values of a uniform distribution, the mean value of a Poisson distribution, the mean and the standard deviation for a distribution. Gaussian, ...). These assumptions can be inputs to the process or can be refined using historical data (see below).
    An endpoint in a supply chain is a terminal node of the supply network where the customer can acquire the product, that is, where the final product demand occurs. The term "place of application" is also used as a synonym for the term "endpoint in a supply chain".
    In particular embodiments, methods described herein can be used to process a very large and complex supply chain with a number of "clients" up to 20000, a number of time steps to simulate in the horizon up to 4000 (no daily time over a 10-year planning period), a number of demand locations up to 2000, and / or a total number of deliverables of up to 10000000. Generally, the present processes intervene the introduction of many random variables to model many possible behaviors. As a result, a very large number of independent simulations of the same process may be required to cover the range of possible outcomes with statistically sufficient accuracy. In particular embodiments, the Monte Carlo simulation technique is used for this purpose. In particular embodiments, the present methods can process up to 100,000 simulations.
    To optimize very complex logistic chains, processes with very high efficiency in terms of calculation time are necessary because of both the dimensions of the problem and the number of simulations to be performed. A discrete event simulation method has this advantage by simulating agent behaviors only at certain time steps, thus avoiding iterating sequentially over all time steps in the simulation period. The method maintains for each agent the next time step where this agent is to be considered in the simulation. Calculations are then made only at the required time steps, thus significantly reducing the number of operations to be performed. This method is particularly appropriate when the actions of the simulated agent take place only on certain particular (discrete) instants in the planning horizon. In other words, if, for a specific client, endpoint visits generating a request do not occur at each time step but only at certain times with a long period of inactivity between these times.
    The methods described herein can also start from an existing situation. An existing situation may include current inventory levels and / or current status of all the agents (customers) generating the demand.
    In particular embodiments, and from an existing initial point, the method may also benefit from historical data recorded about actual (observed) customer behavior. For example, if they are available in any database, the actual timing of client visits to endpoints and the actual product (s) consumed during these visits can be used to refine initial assumptions about random variables. For this purpose, the Bayesian estimation method can be used to balance the observed data with the initial hypothesis. The Bayesian method provides a mathematical formulation to weight the estimated variable parameter from data observed with the initial hypothesis. Basically, the formula increases the weight of the observed data based on the number of observations. This method is particularly suitable for automatically updating the random variables when the number of observations that can be collected for each agent can be very different.
    The methods according to the present invention are capable of taking into account many complex factors of a supply chain, such as factors selected from the list comprising: - a multilevel supply network (production sites, central warehouses, regional warehouses, local deposits); - global reach of multi-language supply chains (in some embodiments, multiple languages are considered as labels); - long and variable delays; - expiry of the products; - various shipping and storage costs depending on the region; - high uncertainty on demand; and, - loss of loading.
    Moreover, the present methods allow high-performance computer implementations. As a result, CPU and / or memory requirements are minimized, resulting in reduced power usage and / or more efficient use of the IT infrastructure.
    In a first aspect, the present invention relates to a computerized supply chain optimization method, the logistics chain comprising a plurality of depots and at least one production facility, the method involving the determination of shipping volumes, volumes of storage and / or storage trigger levels in said supply chain, and the method comprising the steps of a) predicting the demand, preferably using discrete event simulation and Monte Carlo simulations; b) adjusting the replenishment strategy; (c) backward propagation of demand intelligence; d) production planning; e) final evaluation; and, f) search for the best strategies.
    As a result, the aggregate cost of storage and shipping costs can be efficiently balanced with the risk of supply disruptions. Preferably, step a) involves calculating the demand and its variability at the endpoints of the supply chain using Monte Carlo simulations. Endpoints in the supply chain may include local depots or stores. Furthermore, step a) preferably involves the use of a discrete event simulation. As a result, demand can be simulated with high efficiency at the end points of the supply chain.
    Preferably, step b) involves calculating replenishment trigger levels and replenishment quantities at the end-points of the supply chain, using the demand and its variability, and using and Bx as tuning parameters. to control risk and cost. This contributes to improving the computational efficiency of the present methods. The interpretation of parameters A1 and B1 is explained below.
    Preferably, step c) involves performing new simulations including the endpoint replenishment strategy to generate demand realizations at the higher nodes of the supply chain. This process is preferably repeated up to the point (s) of entry of the supply chain. The entry points of the supply chain typically include production sites. This contributes to improving the computational efficiency of the present methods.
    Preferably, step d) involves calculating the production planning for a given value of probability C of demand coverage. This contributes to improving the computational efficiency of the present methods.
    Preferably, step e) involves measuring and / or simulating the risk of supply disruptions, and evaluating the cost of the entire value chain. This contributes to improving the computational efficiency of the present methods.
    Preferably, step f) involves an optimization without derivatives. This contributes to improving the computational efficiency of the present methods.
    This process is explained in more detail below.
    In some embodiments, the method is applied to determining shipping volumes, storage volumes, and / or storage trigger levels in retail supply chains. Processes according to the present invention can be very valuable in the management of supply chains in the retail sector, since the retail supply chains are becoming more and more complex, such as the logistics chains of the retail sector. clinical tests. In particular, the importance of fashion makes goods more and more perishable; if goods are not sold quickly, they will be less desirable to consumers, and will have to be sold at a loss. Like clinical trial supply chains, retail supply chains typically do not reach a steady state due to seasonality, ad-hoc marketing campaigns, short product life cycles, and so on. ...
    In particular embodiments, the method further comprises the step of: g) according to one of said best strategies, performing one or more actions selected from the list consisting of: producing goods in said one or more production facilities in a predetermined quantity and / or at a predetermined rate; maintaining the quantity of goods in one or more deposits included in said plurality of deposits in a predetermined range; receiving goods in one or more deposits included in said plurality of deposits; ship goods to and / or from the facility or production facilities, and / or deposit (s) included in said plurality of depots; and / or ship goods between one or more depots.
    As a result, real world supply chains can be made more efficient by the methods described herein.
    The terms "goods" and "products" as used herein can be used interchangeably.
    In some embodiments, supply chain demand is modeled using a set of agents. An agent represents a customer or a set of customers whose behavior generates demand at a certain time and place. In particular, the precise behavior of an agent is preferably simulated by modeling its multiple visits to demand locations (eg stores), and modeling the way agents consume products. As a result, demand in supply chains can be modeled efficiently.
    In some embodiments, simulations are performed using the discrete event simulation (DES) technique. The DES technique is well suited to the processing of complex and changing logistic chains with many events occurring at irregular time steps. It allows both short calculation times and low memory allocation.
    In some embodiments, step a) involves modeling agent behavior by taking into account multiple uncertainty parameters. Multiple parameters of uncertainty may be selected from the list including the opening of new places or the closing of existing places of demand, the affiliation of customers, a profile or consumption stratum assigned to customers (eg male / woman, age level, ...), the frequency of client visits to the places of demand, the products consumed by clients, the probability of abandonment of the client, etc. Preferably, random generators having an appropriate probability curve are used to represent a possible outcome for each random parameter. An example of an appropriate probability curve is the Poisson distribution. As a result, the variability of agent behavior in logistics chains can be accounted for efficiently.
    In some embodiments, step a) involves the use of Monte Carlo simulations, preferably Monte Carlo simulations for modeling the demand and its variability at the endpoints of the supply chain. Endpoints in the supply chain may include local depots or stores. In particular, a simulation over a certain period represents a possible realization of the request. By calculating a very large number of realizations, a fairly satisfactory representation of the space of the possible demand curves can be obtained.
    For highly complex logistic chains with high variability, up to 100000 implementations (ie individual simulation runs included in the Monte Carlo simulation) may be necessary to cover the variability of demand with sufficient accuracy. The combination of Monte-Carlo and DES simulations can be particularly advantageous in this case, because this approach makes it possible to calculate the large amount of simulations in a limited time.
    In some embodiments, step a) involves the simulation of the demand on all endpoints of the supply chain at the same time. This allows the simulation to take into account a possible correlation between the demand in different locations. Such correlations could occur when, for example, two stores in the same region compete for the same customer base. It should be noted that, as mentioned above, the terms "place of demand" and "end point of supply chain" can be used interchangeably.
    In some embodiments, step b) involves calculating replenishment trigger levels and replenishment quantities from the demand realizations at each endpoint. The replenishment trigger levels correspond to predetermined quantities of goods present in the plurality of demand locations, at or below which said demand locations are replenished. The calculation of replenishment trigger levels and replenishment quantities from the demand achievements at each endpoint can enhance the applicability and versatility of the present methods; since the demand may have a high variability over time, replenishment trigger levels and replenishment quantities may change over time.
    In some embodiments, step b) involves the calculation of trigger levels. Preferably, the trigger levels are calculated such that the stock remaining at the replenishment time is sufficient to cover the demand occurring during the replenishment lead time with a certain probability. Preferably, Monte-Carlo simulations are used to generate many realizations of the demand, which then makes it possible to characterize the statistical distribution of the demand. By using quantiles, this statistical distribution can then be used to measure the trigger level for a replenishment for any given value of the probability Ax. Typically, the ^ parameter depends on the desired level of service (ie, the probability / risk of supply disruptions). As a result, Ax plays an important role in the trade-off between cost and risk. In particular, a high value of i.e. a value close to 100%, ensures a low risk of out of stock but gives rise to high security stocks that can generate high storage costs and, in the case of perishable products, significant waste. Typical values of A are 99% to 99.9%. Preferably, the value of A is 99.9 in clinical trial logistic chains, where the risk of out of stock must be kept very low.
    In some embodiments, step b) involves calculating shipping volumes to cover the average demand over a time horizon B1. Bx is a parameter that can significantly improve the cost-risk tradeoff. In particular, a high value of Bx corresponds to large shipping volumes and low shipping frequencies. Low shipping frequencies reduce shipping costs but may decrease flexibility, so unnecessary materials are shipped.
    In some embodiments, step c) of propagating back the request information is performed as follows. For a given value of the parameters Ax and Bx, the replenishment strategy of each endpoint can be determined and can be simulated using the endpoint request as an input. Typically, the end points of the logistics chains include local depots and / or stores. New (unseen) implementations of the endpoint request could thus be generated using the process described for step (a), and can be used to simulate the replenishment strategy, which in turn generates new requests for the depot supplying this endpoint. End-point requests and replenishment strategies can therefore be consolidated into demand-side outcomes for the penultimate level of supply chain nodes (eg for regional repositories). Using the same principle as that described above for step b), replenishment strategies of the penultimate level of supply chain nodes can be determined, again using two general setting parameters A2 and B2 acting on trigger levels and shipping volumes. The above procedure can be repeated to successively calculate the demand and replenishment strategies of the upstream levels i (i being a natural number> 1) until reaching the production sites. The index i in At and Bi refers to level i of a supply network. At a given level i, there may be many nodes. Defining the parameters Ai and Bi for an integer level makes it possible to model a large variety of nodes, for example thousands of nodes, at level i by only two parameters: At and Bi. This can greatly increase the computational efficiency and reduce the memory requirements of the present methods.
    The At and Bt parameters can be used at each level to calibrate the strategy.
    Accordingly, in some embodiments, step c) involves the substep of determining trigger levels and shipping volumes for consecutive level nodes i in the supply chain. The nodes of the consecutive levels may comprise a plurality of depots, a plurality of request locations and / or at least one production facility. In particular, trigger levels may be calculated such that the remaining inventory at a level i node at the replenishment time is sufficient to cover the demand occurring during the replenishment time with a certain probability At. the shipping volumes can be calculated to cover the average demand at a level i node over a time horizon B ^ Step c) can be used to determine the demand in all logistic chain nodes, including at the level of d. production facilities. The above embodiments make it possible to improve the efficiency with which step c) is performed.
    In some embodiments, step d) includes the substep of determining the production planning to satisfy the consolidated demand with probability C. The term "consolidated demand" refers to the demand at the level of a production site. In some embodiments, the determination of the parameter C is made taking into account constraints on the size of the batches, the expiration of the products and the production time as boundary conditions. As a result, production planning can be determined efficiently while taking into account up-to-date real-world information.
    In some embodiments, meta-parameters A 1, Bi and C can be used. Generally, the parameters Ai and Bi are used per level of the supply network, and a parameter C is used for the production site (s). Parameters Ai, Bi, and C are used to drive the actual supply rules of all nodes in the supply network and production quantities. Preferably, at each level i, for a given value of the pair of parameters Ai and Bi, the same value for all the nodes of the level in question, a different replenishment rule is still elaborated for each node of the level in question, in depending on the specific variability of the demand observed on the node in question. These different replenishment rules are calculated to achieve the desired level of risk (Ai) and demand horizon (Bi) for all nodes but typically correspond to different trigger levels and replenishment amounts on each node, since the shape of the request and its variability can be different on each node.
    In some embodiments, the parameters A 1, B 2 and C can be seen as the setting button with which one can play to develop the actual replenishment rule on each node and that can be used to optimize risk and cost. Due to the limited number of these tuning variables, the search space for the optimization algorithm remains very limited. The optimization of the replenishment strategies of all the nodes of the network is thus simplified by the optimization of a very small number of variables Ai, Bi and C of decision.
    In some embodiments, step e) includes the step of measuring the risk that a product will not be available at an endpoint to satisfy a request. Preferably, step e) also comprises the step of measuring the total cost of the supply chain. As a result, the performance of particular supply chain strategies can be evaluated efficiently.
    In some embodiments, the total cost of the supply chain includes the cost of materials and production, the cost of shipping, the cost of storage, and the potential cost of waste disposal. This makes it possible to take realistic conditions into account in an efficient manner.
    We have detailed how the risk and cost can be estimated for a set of Ait BÎt and / or C meta-parameters for a level i, where i = 1 to n, where n is the number of levels in a network supply chain (typically n = 2). When used, these meta-parameters strongly influence the performance of procurement strategies. However, even if the number of meta-parameters to be set is small, the large number of possible values that can be explored for these parameters Ait Bi, and / or C can make it difficult to identify their best possible combinations, which are located on the Pareto border (see Fig. 1).
    We now discuss how step f), the search for the best strategies, can be performed according to some embodiments of the methods described herein.
    In order to efficiently characterize the Pareto border of the best possible strategies in the risk / cost space, a neighborhood optimization algorithm can be used, the neighborhood optimization algorithm exploring a series of values for the set of values. meta-parameters Ait Bi, and / or C. An optimization without derivatives can be used by the algorithm in order to evaluate the least possible strategies. Derivative optimization techniques are generally well suited for use in the methods described herein, because the search space is rather smooth and because the risk and cost assessment of a set of values is generally very expensive (typically involving Monte-Carlo simulations, each simulation involving discrete-event simulation of a complex random process).
    In some embodiments, the search for the best strategies involves starting from a set of initial strategies, i.e. an initial set of meta-parameters. It is likely that this set is suboptimal. In other words, it is likely that the initial set of meta-parameters is located in an area of the cost-risk space that is not on the Pareto border. This set of initial parameters can then be evaluated to determine costs and risks. Then, a new strategy can be generated, and its performance can be evaluated. In some embodiments, the new strategy is combined in pairs with older strategies, and the new strategy is added if it outperforms any other strategy. Preferably, this is repeated until the search no longer finds any strategy leading to improvement during tmax iterations. At this point, the set of strategies is considered a good representation of the Pareto front. The tmax parameter is specific to the problem and depends on the accuracy required by the user.
    In some embodiments, non-derivative optimization techniques are used in the search for better strategies. Derivative-free optimization techniques, such as Nelder-Mead's multi-directional search or method, help to speed up the optimization phase by quickly orienting the algorithm towards new good strategies. A strategy s is a set of values of the parameters Ait BÎt and / or C, for the level t, with t = 1 to n. We denote the value of the parameter A, in the strategy s by s (Ai). Preferably, these methods select several Sv. ^ Sk strategies in the current set and compose them to obtain a set of new strategies, whose performance can then be evaluated. Preferably, the more remote Sv. ^ Sk policies, the more remote is snew from the original strategies. This is thought to allow the DFO algorithm to begin by diversifying the search and finding entirely new strategies at the beginning (when the set of strategies is suboptimal with many different solutions). This quickly leads the DFO algorithm to solutions close to the Pareto front. When the set contains similar strategies, it is likely to represent the Pareto front, and the algorithm intensifies the search around that front. The distance between two strategies is the sum on all the adjustment parameters of the absolute difference of the values affected by these two strategies: Σρερ Isi (p) - s2 (p) where l P is the set of all the parameters Ait Bi, and vs.
    For the optimization of stochastic logistic chains, an enormous number of executions (simulations) is generally necessary in order to use the optimization techniques without derivatives with precision. Running large numbers of simulations is usually expensive. In order to reduce the need for large numbers of simulations, the Applicant has developed methods for evaluating strategies with varying accuracy. This could speed up the execution of the methods described here.
    Accordingly, methods are provided for evaluating strategies with high accuracy only when it is necessary to compare them. Preferably, the suboptimal strategies are evaluated at a level of precision that makes it possible to compare them to the best ones. In fact, it is generally unnecessary to achieve a high degree of accuracy in evaluating suboptimal strategies because it is clear that they are worse than the best ones. Basically, the level of precision and therefore the number of simulations needed to evaluate each strategy is usually adjusted to the distance to the other strategies to which it is compared. As the algorithm gradually approaches the Pareto boundary, the distance between strategies typically decreases and the number of simulations required to discriminate between their performance increases.
    In some embodiments, non-derivative optimization involves comparing two solutions. Preferably, the comparison involves using the average and variance of the cost and risk of the solutions, the number of simulations performed for each strategy, and the use of the central limit theorem and Bayesian inference. A simulation run typically corresponds to a series of Monte Carlo simulations of demand curves at each demand location, with each Monte Carlo simulation involving discrete event simulation of the demand generating process. This process is modeled using agents representing the end customer consuming the product that is the subject of the supply chain.
    When the performances of the two simulation runs are close, and if a small number of simulations have been executed so far for both solutions, additional simulations can be performed on both solutions. This is preferably done because the probability that one solution outperforms the others is low, and additional simulations are needed to determine if this is the case or not. On the other hand, when the two performances are very different, a simulation can be eliminated after only a small number of simulations. A small number of simulations can be, for example, 100, 1000, or 10000 simulations. Accordingly, if the probability that a strategy outperforms one or more other strategies is less than a predetermined value, then the number of simulations executed on the strategy in question is increased until said probability is greater than or equal to at said predetermined value. Therefore, computational resources can be efficiently allocated to the detailed evaluation of strategies that actually need to be evaluated in detail.
    In some embodiments, the method is applied to determining shipping volumes, storage volumes, and / or storage trigger levels in clinical trial supply management. The scope of clinical trial supply chains (which is often global and typically with three levels - central warehouse, local repository, investigative sites - for a Phase III trial), highly variable and difficult to anticipate demand (resulting patient recruitment at the investigative sites and patient visits during treatment), expiry of products, high cost of packaging and shipments (eg in temperature-controlled transport), are some of the reasons for which the calculation approach described here results in a better compromise between the cost of clinical trials and the risk of supply disruptions.
    The processes described here can be deeply integrated into the global process of supply management of clinical trials. In particular, they can be used first in pre-study planning and budgeting to help assess the impact of different study designs on the cost and risk of the supply chain. They can then be used at the start of a study to launch the initial production plant and to implement the supply chain strategy, which includes determining the amount of safety stocks and replenishment volumes. Finally, the methods described herein can be used on a regular basis in clinical trials. Typically, the methods can be used every month. As a result, production planning and supply strategies can be re-evaluated on the basis of actual operation and on the basis of observed deviations from the original plan. To this end, the collection of actual data regarding the status of patients in the treatment assigned to them and the inventory levels at each node of the supply chain is necessary. These data are generally available in information systems typically used for the operational management of supply chains, for example in the operational management of clinical trial logistic chains, called IVRS (Interactive Voice Response System) or IWR (Response System). interactive web-based). Our process can exploit data provided by these IVRS / IWR systems for the re-evaluation of the supply strategy of an ongoing study.
    Plots representing simulations that can be generated by the present methods can be visualized in such a way that supply chain managers can easily see which costs are related to which service levels, and in such a way that supply chain managers can choose appropriate compromises. Once this choice is made by a supply chain manager, the actual replenishment parameters (trigger levels and shipping quantities) for each supply chain node can be returned together with the production planning.
    EXAMPLES 1. Risk and Cost
    In a first example, reference is made to FIG. 1. FIG. 1 shows a schematic representation of the trade-off between risk and cost which is addressed in the methods of the present invention. In particular, the risk is identified on the x axis and the cost is identified on the y axis. A particular point in the risk-cost space is called strategy. Three types of strategies are shown in FIG. 1: inefficient strategies (100), optimal strategies (200) and impossible strategies (300).
    A strategy A, occupying a position in the risk-cost space (x ^ yj, is an inefficient strategy (100) if and only if there is at least one strategy B, occupying a position in the risk-cost space ( xB, yB), for which: (f> Xb) AND (yA> yB)) OR ((xA> xB) AND (yA> yB)). (1)
    In other words, the cost of an inefficient strategy can be reduced without increasing the risk and / or the risk of an inefficient strategy can be reduced without increasing the cost. Or, again differently, a strategy is said to be inefficient if there is (if it is possible to find) another strategy offering the same level of risk at a lower cost or a lower risk level for the same cost.
    The optimal strategies (200) are the strategies A for which no strategy B can be found such that the equation (1) is verified. In other words, the cost of an optimal strategy (200) can only be lowered by increasing its risk, and the risk of an optimal strategy (200) can only be lowered by increasing its cost. Optimal strategies (200) include both high cost and low risk strategies (210) and low cost, high risk strategies (220). On the other hand, optimal strategies (200) form the boundary between inefficient strategies (100) and impossible strategies (300). This border is called the Pareto border.
    Impossible strategies (300) are strategies that can not be accomplished in practice. In other words, impossible strategies (300) correspond to positions in the risk-cost space that lie outside the realm of possible strategies. Possible strategies consist of the union of inefficient strategies (100) and optimal strategies (200). 2. Logistics chain
    In this example, reference is made to a particular supply chain with four levels; namely demand locations, regional warehouses, central warehouses and production sites. Each level has several nodes. For example, demand locations include several individual locations (network end points, stores), local depots have multiple individual depots, central warehouses have multiple individual warehouses, and production sites have multiple individual production sites. 3. Supply Network
    In this example, reference is made to a typical supply network with 10 production sites, 100 depots (level 2), and 1000 demand locations (level 1). This network comprises 1100 nodes (level 2 and level 1) for which replenishment strategies must be determined and for which 10 production schedules must be established. For each node, the replenishment strategy is a trigger level and a shipping volume. In other words, for each node, two quantities (or more if they are made dependent on time) must be specified. Assuming a single production lot, we have 10 quantities to be determined at the production sites. In total, this makes a minimum of 2210 variables to optimize. The present methods make it possible to replace these 2210 variables by, in this example, 5 variables A1, B1, A2, B2 and C from which can be directly drawn the 2210 variables to which we are ultimately interested. This greatly simplifies the optimization problem. Because of this dramatic simplification, the present methods make it possible to solve similar problems even in very complex logistic chains.
    权利要求:
    Claims (15)
    [1]
    A computerized supply chain optimization method, the supply chain comprising a plurality of depots and at least one production facility, the method involving the determination of shipping volumes, storage volumes and / or trigger levels storage in said supply chain, and the method comprising the steps of a) predicting the demand using discrete event simulation and Monte Carlo simulations; b) adjust the replenishment strategy; (c) propagate backward the information in the application; d) plan production; e) evaluate a final strategy; and; f) search for the best strategies.
    [2]
    The method according to claim 1, further comprising the step of: g) according to one of said best strategies, performing one or more actions selected from the list comprising: • producing goods in said production facility (s) in a quantity predetermined and / or at a predetermined rate; • maintain the quantity of goods in one or more deposits included in said plurality of deposits in a predetermined range; • receive property in one or more deposits included in said plurality of deposits; • ship goods to and / or from the facility or production facilities, and / or the deposit (s) included in said plurality of depots; and or ; • ship goods between one or more depots.
    [3]
    3. The method of claim 1 or 2, the plurality of repositories comprising a set of local depots and / or the supply chain may further comprise a plurality of demand locations, the request locations possibly comprising stores, and the step a) involving the following substep: ai) modeling the supply chain demand using a set of agents, agents going to multiple demand locations, and the behavior of an agent being simulated by modeling the way the agent consumes products.
    [4]
    4. Method according to claim 3, the modeling of the behavior of the agents further comprising the sub-step consisting of: aii) taking into account multiple uncertainty parameters modeled by a probability curve, the multiple uncertainty parameters being chosen in the list including the opening of new places or the closing of existing places of demand, the affiliation of customers, a profile or a layer of consumption assigned to customers, the frequency of visits to a place of demand, the product consumed by the agent, and the probability of abandoning the agent.
    [5]
    The method of any one of claims 1 to 4, the supply chain having a plurality of logistic endpoints and the step a) involving a simulation of the request on all chain endpoints. logistics at the same time.
    [6]
    The method of any one of claims 1 to 5, wherein step b) includes the substep of: bi) calculating replenishment trigger levels and replenishment quantities from the realizations of the request to each end point of the supply chain; the replenishment trigger levels corresponding to predetermined quantities of goods present in the plurality of demand locations, at or below which said demand locations are replenished.
    [7]
    7. The method of claim 6, the trigger levels being calculated such that the stock remaining at the replenishment time is sufficient to cover the demand occurring during the replenishment time with a certain probability Ax.
    [8]
    8. The method of claim 6 or 7, wherein step b) involves the calculation of shipping volumes to cover the average demand over a time horizon B1.
    [9]
    The method of claim 8, wherein step c) comprises the substep of: ci) determining trigger levels and shipping volumes for consecutive level nodes i in the supply chain, the nodes comprising the plurality of depots, the plurality of request locations, and the production facility or facilities, • the trigger levels being calculated such that the remaining inventory in a level i node at the replenishment time is sufficient to cover the demand occurring during the replenishment period with a certain probability At; and, • the shipping volumes being determined to cover the average demand at a level i node over a time horizon Bi.
    [10]
    The method of any one of claims 1 to 9, wherein step d) comprises the substep of: di) determining the production planning to satisfy the consolidated demand with a probability C.
    [11]
    The method according to any one of claims 1 to 10, wherein step e) comprises measuring the risk that a product is not available at an endpoint to satisfy a request, and measuring the total cost of the supply chain.
    [12]
    12. The method of claim 11, the total cost of the supply chain including the cost of materials and production, the cost of shipping, the cost of storage and the possible cost of disposal of waste.
    [13]
    The method of any one of claims 1 to 12, involving the use of non-derivative optimization; preferably in which, for nodes of level t, with i = 1 to n, and n being the number of levels in the supply chain, Ait Bi, and C are determined by means of optimization without derivatives.
    [14]
    A method according to any one of claims 13, the non-derivative optimization involving step-by-step introduction of new strategies for updating other strategies, and the distance between new strategies and other strategies being determined from after the distance between the other strategies.
    [15]
    The method of claim 14, comprising the steps of: • using the mean and variance of the cost and risk, and the number of simulations that have been executed for each strategy, to calculate the probability that a strategy outperforms the performance other strategies, the central limit theorem and Bayesian inference being used in the calculation; • If the probability that a strategy outperforms one or more other strategies is less than a predetermined value, increase the number of simulations performed on that strategy until that probability is greater than or equal to that value. predetermined.
    类似技术:
    公开号 | 公开日 | 专利标题
    Aydin et al.2016|Coordination of the closed-loop supply chain for product line design with consideration of remanufactured products
    US11250499B2|2022-02-15|Generating optimal strategy for providing offers
    Negahban et al.2014|Managing production level in new product diffusion: an agent-based simulation approach
    US20170147955A1|2017-05-25|Enterprise resource management tools
    Chyrun et al.2019|Heterogeneous Data with Agreed Content Aggregation System Development.
    Yelland et al.2014|Forecasting demand for fashion goods: a hierarchical Bayesian approach
    Canetta et al.2013|Modelling hybrid demand | evolution: A scenario planning approach
    BE1023357B1|2017-02-14|METHODS FOR DETERMINING SHIPPING VOLUMES AND / OR STORAGE RELEASE LEVELS
    Sun et al.2015|Estimating the long-term cost to serve new customers in joint distribution
    Singhal et al.2019|Understanding multi-stage diffusion process in presence of attrition of potential market and related pricing policy
    US20170293269A1|2017-10-12|Control system with machine learning time-series modeling
    KR20140072427A|2014-06-13|Method and apparatus to recommend products reflecting characteristics of users
    Hernandez-Garcia et al.2017|A real option based model for the valuation of patent protected technological innovation projects
    Asai et al.2020|Realized stochastic volatility models with generalized Gegenbauer long memory
    US8374904B2|2013-02-12|Market forecasting
    KR20210116435A|2021-09-27|Systems and methods for inventory management and optimization
    Tilson et al.2014|Monopoly production and pricing of finitely durable goods with strategic consumers׳ fluctuating willingness to pay
    Niknejad et al.2017|Analysis of impact of uncertainty in global production networks’ parameters
    Li et al.2018|Pricing strategies with promotion time limitation in online social networks
    Li et al.2021|Lifecycle forecast for consumer technology products with limited sales data
    Predić et al.2020|Time series analysis: Forecasting sales periods in wholesale systems
    TW201837814A|2018-10-16|Method and system for forecasting product sales on model-free prediction basis
    US20150066596A1|2015-03-05|Behavior pricing analytics
    US20200143391A1|2020-05-07|METHODS, SYSTEMS, APPARATUS AND ARTICLES OF MANUFACTURE TO MODEL eCOMMERCE SALES
    US11176495B1|2021-11-16|Machine learning model ensemble for computing likelihood of an entity failing to meet a target parameter
    同族专利:
    公开号 | 公开日
    EP3193288A1|2017-07-19|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    ES2877924A1|2020-06-30|2021-11-17|Inst Tecnologico De Aragon Itainnova|Digital Twin Simulation of a Supply Chain in a Physical Internet Framework |
    法律状态:
    优先权:
    申请号 | 申请日 | 专利标题
    EP16151686.9A|EP3193288A1|2016-01-18|2016-01-18|Methods for optimisation of a supply chain by determining shipment volumes and/ or storage levels|
    EP16151686.9|2016-01-18|
    [返回顶部]