This invention relates to agents, and more particularly to an inventory early warning agent used in a supply chain management system.
Today's companies need to adapt to many competitive pressures. For example, financial markets are increasingly demanding that companies use capital more efficiently; other businesses are seeking global playing fields to maintain growth and diversify risk; customers are demanding service as markets of one and forcing companies toward mass customization; and innovation cycles are continually accelerating.
These pressures on businesses are driving changes that have enormous implications for supply networks. For some companies, shrinking capital availability is forcing companies to streamline manufacturing and supply operations and build efficiencies, which are critical to the supply network. For other companies, information ubiquity is driving and facilitating globalization, which shrinks distances to markets and resources. The information ubiquity also requires levels of supply network visibility and collaboration that were not essential in traditional supply chains. Customers are armed with information about the real value of products, which is shrinking customer loyalty and requiring customer-service levels too expensive for companies that are unable to manage supply chain efficiencies. Finally, shrinkages in the time available to build and launch products are forcing companies to innovate at velocities far greater than before.
Ultimately, competitive pressures push profit margins lower. Product manufacturers must find ways to improve efficiency, thereby reducing costs, to survive in highly competitive markets. Supply chain efficiency plays a key role in improving margins and can be a determining factor in the success of manufacturers.
A supply chain is a network of facilities and distribution options that performs the functions of procuring materials, transforming the materials into semi-finished and finished products, and distributing the finished products to customers. Supply chain management (“SCM”) is a business policy that aims to improve all activities along the supply chain. SCM results in improved integration and visibility within individual companies, as well as flexibility across supply and demand environments. As a result, a company's competitive position is greatly enhanced by building supply networks that are more responsive than the current sequential supply chains.
SAP AG and SAP America, Inc. provide SCM solutions for product manufacturers to help them reach their goals. Some of the SCM solutions are based on the mySAP.com e-business platform. One of the building blocks of the e-business platform is the SAP R/3 component that provides enterprise resource planning functionality. The SAP R/3 product includes a Web Application Server (“Web AS”), an R/3 core, and various R/3 extensions. The SCM Extensions of R/3 provide various planning, coordination, execution, and optimization solutions that are associated with a supply chain.
In one general aspect, an inventory agent operating on software includes instructions operable to cause a programmable processor to receive inventory data relating to stock in an inventory, apply the inventory data to a conditional probabilistic predictive statistical algorithm, calculate a predicted inventory level, and use the calculated predicted inventory level to determine whether to order additional stock for the inventory. The statistical algorithm uses a conditional probabilistic model to process the data.
Embodiments of the inventory agent may include one or more of the following features. For example, the inventory data may be generated using a radio-frequency identification device and/or upon the sale of the stock in the inventory. The inventory data may be generated upon the transfer of the stock in the inventory.
The inventory agent may be implemented in a supply chain management system.
The conditional probabilistic model may use a conditional Gaussian approximation. The conditional probabilistic model may use historical inventory data and current inventory data. The conditional probabilistic model may further calculate an upside 10% confidence bound and a downside 10% confidence bound. The inventory agent may further include instructions to use the calculated upside 10% confidence bound and the downside 10% confidence bound in determining whether to order additional stock for the inventory.
The inventory data may further include actual cumulative replenishment, forecast cumulative replenishment, actual cumulative consumption, and forecast cumulative consumption data. The inventory agent may further include instructions operable to cause the agent to order a replenishment of the inventory.
The inventory agent may further include instructions operable to record data related to inventory consumption and replenishment. The inventory agent may still further include instructions operable to use the recorded data to calculate cumulative forecast consumption, cumulative forecast replenishment, cumulative actual consumption, and cumulative actual replenishment. The inventory agent may still further include instructions operable to use the inventory data with the calculated cumulative forecast consumption, cumulative forecast replenishment, cumulative actual consumption, and cumulative actual replenishment to calculate a second predicted inventory level.
In another general aspect a system includes one or more computer systems and an inventory agent computer coupled to the computer systems over a network. The inventory agent computer is operable to receive inventory data relating to stock in an inventory, apply the inventory data to a conditional probabilistic predictive statistical algorithm, calculate an expected inventory level, and use the calculated expected inventory level to determine whether to order additional stock for the inventory. The statistical algorithm uses a conditional probabilistic model to process the data.
The inventory agent computer may be implemented in a supply chain management system.
Embodiments of the system may include one or more of the following features. For example, the system may further include instructions to generate the inventory data upon receipt from a radio-frequency identification device and instructions to generate the inventory data upon the sale of the stock in the inventory. The system also may further include instructions to generate the inventory data upon the transfer of the stock in the inventory.
The conditional probabilistic model may use a conditional Gaussian approximation. The system may further include instructions to use historical inventory data and current inventory data in the conditional probabilistic model.
The system may further include instructions to further calculate an upside 10% confidence bound and a downside 10% confidence bound in the conditional probabilistic model. The system may still further include instructions to use the calculated upside 10% confidence bound and the downside 10% confidence bound in determining whether to order additional stock for the inventory.
The inventory data may include actual cumulative replenishment, forecast cumulative replenishment, actual cumulative consumption, and forecast cumulative consumption data. The system may further include instructions operable to cause the inventory agent computer to order a replenishment of the inventory.
The system may further include instructions operable to record data related to inventory consumption and replenishment. The system may still further include instructions operable to use the recorded data to calculate cumulative forecast consumption, cumulative forecast replenishment, cumulative actual consumption, and cumulative actual replenishment. The system may still further include instructions operable to use the inventory data with the calculated cumulative forecast consumption, cumulative forecast replenishment, cumulative actual consumption, and cumulative actual replenishment to calculate a second predicted inventory level.
The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description, the drawings, and the claims.
Like reference symbols in the various drawings indicate like elements.
An agent is an active program that performs some high-level business functions, such as monitoring the inventory for a particular SKU for potential stock out situations and sending alerts when various thresholds are reached. For example, an agent can be programmed as an inventory early warning agent (“IEWA”) that monitors one or more inventories in a factory, warehouse, store, or shelf within a store. The IEWA is programmed to monitor the inventory based on real time data that is obtained, for example, from cash registers or radio-frequency identification devices that scan information relating to an item being purchased or transferred and use that information to update the IEWA as to the level of stock on a particular shelf.
As described in more detail below, the IEWA is a predictive and adaptive inventory management application that can be used to monitor and predict future inventory levels by modeling variability in both demand and supply related supply chain activities. The IEWA uses learning techniques that can estimate potential variation in inventory levels in the near future in order to identify potentially risky situations early enough to allow for corrective measures. To provide increased effectiveness the IEWA is operated with minimal human intervention, parameter adjustment, and manual calibration. These characteristics are achieved with machine-learning techniques to recognize patterns of behavior from historical data around consumption and replenishment and around the performance of resources and supply chain partners. The IEWA can use two types of modes of operation for its predictive function: in a first mode use the IEWA predictive models built from detailed activities; and in a second mode use the IEWA predictive models built from the aggregation of multiple activities.
In the detailed mode of operation, potential variation in inventory is estimated for each planned replenishment or consumption activity. For example, given historical performance data, the IEWA estimates whether a planned truckload of 12 ounce bottles of Bob's bottles will arrive any time within four hours prior to and six hours after the planned delivery time with a quantity that is between 95% and 100% of the requested quantity. A second stage of processing then combines the estimates of potential variation for individual activities into an estimate of the potential variation of the projected inventory levels.
In the aggregate mode of operation, potential variation of the projected inventory level is estimated directly from projected cumulative replenishment and consumption. This estimate is calculated as the aggregation of replenishment and consumption activities.
Both modes of operation provide advantages. For example, the detailed mode of operation is advantageous when low volumes of detailed data are available and when the degree of variation in activities is highly dependent on the particular resource or partner involved. The aggregate mode of operation is advantageous when it is necessary to deal with high volumes of data, or when it is difficult to track the outcome of any particular planned activity. Although either mode of operation can be applied to the IEWA, only the aggregate model is discussed in detail below.
Referring to
The IEWA 115 also can be used when the shelf is replenished. For example, when the shelf 105 is replenished, the shelf-monitoring program 120 sends a message to the execution module 140 along the message path 135, which sends a message to the distribution center 165 along the message path 170. These messages are used to update the distribution center of the amount of stock on the shelf 105. The distribution center can use the same IEWA or a separate IEWA to apply predictive models to the distribution center to estimate whether and when there will be an undesirable variation in inventory levels.
At another level, a store can use an IEWA to monitor the levels of an item on the shelf and in the inventory for one or more items to estimate potential undesirable variations in inventory levels. When items are sold, for example, by being scanned at a cash register, the IEWA takes that sales data and uses algorithms to determine whether there will be an undesirable variation in the inventory levels and when to send an order to replenish the shelf and/or order more of that item from a warehouse or distribution center.
At an even higher level, a warehouse or distribution center can use an IEWA to monitor the levels of an item within the warehouse, such as on shelves, on pallets, in quarantine, or at another location within the warehouse, to determine whether there will be an undesirable variation in the inventory levels and when to send an order to replenish. Customers of the warehouse, such as a retailer or a factory, order the item from the warehouse. For example, a consumer product goods (“CPG”) retailer may order a pallet load of an item which the warehouse operator loads onto a delivery truck of either the warehouse, the retailer, or a third party logistics supplier. When the pallet is loaded on the truck, the warehouse operator may use a wireless communications device to notify the inventory management software that a pallet-load of a particular item has been transferred from the warehouse. Either the wireless communications device or the inventory management software may be programmed to notify the IEWA that a pallet-load of the particular item has been transferred from the warehouse. The IEWA takes that transfer data and analyzes it using algorithms to determine whether there is likely to be an undesirable variation in inventory levels and when to order addition stock of that item. One example of a framework of the IEWAs, an example of a messaging system used by the IEWAs, and the algorithms used by the IEWAs are described in more detail below.
Referring to
The MTS 220 is a service that is provided by a particular agent framework and allows agents to send messages, such as an AclMessage, to other agents. An AclMessage encapsulates a communication between two agents and has some characteristics in common with an email message, such as, for example, specifying a sender and recipient, having a subject, and providing message content.
The DF 225 is a service provided by a particular agent framework and provides the framework's agents with access to the central directory service. The directory service is a central service that provides identity (white page) and capability (yellow page) search facilities across an agent community. There is one directory service for one agent community. A directory service might be federated with the directory service for another agent community. An agent community is a collection of agent frameworks (and therefore agents) that collaborate to provide various business functions. A community may consist of several agent frameworks, and each framework may in turn contain several agents.
The AMS 230 is a service provided by a particular agent framework and provides agent lifecycle management facilities within the framework. The facilities allow remote management of a community of agent frameworks by essentially providing an external interface to the AMS in each agent framework. For example, the AMS allows administrators and control agents to manage the execution of other agents by stopping, starting and suspending agents. The agent framework architecture 200 also includes an administrative user interface (“AUI”) that allows a system administrator to manage an agent community. The AUI uses the directory service to identify currently running agent frameworks. It then uses each framework's AMS to manage the agents within that framework.
The post office 235 is a service provided by a particular agent framework. The post office 235 receives and maintains AclMessages addressed to the agents within that framework.
The directory facilitator 225 (“DF”) in each agent framework 217 updates the directory service 240, registering new agents as they are started up and deregistering agents as they are shut down. The directory service 240 then allows agents to search for other agents, both by identity (e.g., using a service that is analogous to a white page service) and capability (e.g., using a service that is analogous to a yellow page service). The DF 225 provides access to this functionality within each agent framework 217.
In the scenario of
As described briefly above, the agents 205 may perform analysis of the inventory data that they receive to determine if there is likely to be a potential variation in inventory levels and, if so, when to request a replenishment of the inventory from a warehouse or the production of goods by a manufacturer to replenish the inventory in a warehouse or at a store. The analysis may be a simple process involving the comparison of inventory on the shelf or in a warehouse to a threshold value. If the inventory level is above the threshold value, the agent does not request replenishment but if the inventory level is at or below the threshold value the agent requests replenishment. Such an analysis does not take into account the likelihood that the stock on the shelf or in the warehouse will be sold quickly or slowly or that there may be some variability in the rate at which the stock is sold. Moreover, such an analysis does not take into account the likelihood that the warehouse or manufacturer will have sufficient stock on hand to supply the store, much less the likelihood that the warehouse or manufacturer can make a timely delivery. To ensure that the store, warehouse, or manufacturer has adequate quantities of stock to meet its needs while also minimizing excessive inventory levels, the agent 205 uses a conditional probabilistic predictive statistical analysis to predict an expected inventory level, an upside confidence bound, and a downside confidence bound. For many companies, an expected inventory level that is near or below zero is undesirable. The company can determine the minimum low level, compare the expected inventory level to the company's minimum low level, and order more stock if these expected levels are below the minimum low level.
Referring to
Any suitable statistical method can be used to determine the inventory level and confidence bounds. For example, the statistical analysis used by the agent 205 can be implemented with a predictive statistical model that can predict the variability at any point along the planned inventory trajectory, can be started with very little historical inventory data, and that improves as the data is generated over time. The plot of planned and actual inventory trajectories illustrated in
One such statistical model that can be used is a probabilistic inference model based on a conditional Gaussian approximation. This model, its derivation, and its application to the inventory early warning agent are explained in more detail below.
To derive the model used for explanation purposes, there is an assumption of a known, planned forecast for inventory levels over time. The planned forecast is a piecewise constant function specified by its breakpoints. As illustrated in
To actually develop a prediction algorithm, assume that inventory level plans are generated in regular fixed intervals, Δ (e.g., every day at midnight), and set the origin of time to be the time at which there is the earliest historical plan in the dataset. This origin of time may be recent and provide a very limited number of data points. Thus, plans are generated at times nΔ for nεZ+, where n is the index of the most recent planning period. Further, assume that all plans are generated out to a horizon of τ(>Δ) and indicate the inventory levels of plans generated at time nΔ by Inp(nΔ+δ), where 0≦δ≦τ is a continuous quantity. Because plans are generated at midnight, δ=0 in the future. This inventory plan is piecewise constant and specified by its breakpoints (Inp(b),δnp(b)) for δnp(b)≧0 and bε[1,K,Bn] where Bnp is the number of breakpoints defining the nth plan. For conciseness define Inp(•) and δnp(•) to be the Bnp vectors with components [Inp(b)K Inp(Bnp)]T and [δnp(b)K δnp(Bnp)]T. Furthermore, let Ia(t) be the actual inventory at time t>0. The inventory is not measured at all times, but can be interpolated from a piecewise constant function. The breakpoints defining the actual inventory are given by (Ia(b),ta(b)) for bε[1,Ba] where Ba is the number of breakpoints defining the actual inventory and all t(b)≧0. Ia(•) and ta(•)$ are the Ba vectors [Ia(b)K Ia(Ba)]T and [ta(b)K ta(Ba)]T, respectively.
Based on the definitions above, for any time t there are a number of plans that predict the inventory at that time. For example, the most recent plan was formed at
Based on the above notation it is straightforward to develop or build the model and specify the data from which to build the model. Specifically, if n is the index of the most recent planning period, the times at which the model must predict the inventory levels are given by tβ=nΔ−δβ. By defining the q vector δ≡[δ1,K,δq]T, the model can be used to determine P(I|n,δ,I2p(•1),δ2p(•1),Ia(•),ta(•)), where I2p(•1)=[I1p(•)L INp(•)]T and δ•2p(•1)=[δ1p(•)L δNp(•)]T. To understand how the algorithm works it is helpful to understand how order streams are converted to the inventory levels that serve as the input to the aggregate predictor. Since the algorithm works entirely with inventory levels (whether planned or actual), consumption and replenishment order streams need to be converted into inventory levels. Conceptually, this is straightforward: replenishment orders increase the inventory while consumption orders decrease inventory. The only complication is that consumption and replenishment are only measured at certain times such that there is not data describing what happened to the inventory between those times at which the inventory was measured. Additionally, the times at which the measurements are taken may be irregularly spaced. Both of these problems are easily solved with a model of inventory levels between measurements.
The model of inventory levels is based on scenarios to which the inventory is actually subjected. As such, there are a variety of occurrences that might happen to the inventory levels between measurements. The simplest case is that the inventory was unchanging between measurements. This is termed the piecewise constant case since it is possible to interpolate inventory levels between measurements by using a piecewise constant function. Alternatively, if successive inventories are measured as (I1, t1) and (I2, I2) it is possible to linearly interpolate the inventory level at time t (where t1≦t≦t2) as I(t)=I1+(I2−I1)(t−t−t1)/(t2−t1). The model is based on the assumption of a piecewise constant rather than piecewise linear interpolation, although either choice or other interpolation mechanism is acceptable for data preparation since the algorithm is independent as to the choice. All the algorithm requires is that the inventory levels be measured at equally spaced intervals, Δ, and this can always be accomplished by interpolation by either method.
The particular approach used with the IEWA uses the essential idea of exploiting correlations in errors between planned inventories and the actual inventories themselves. These correlations are used to estimate a probability density for a number of errors and then condition this density on the available data at the time a prediction needs to be made to find the likely errors at the time of prediction. The most likely correction to the planned inventory can then be applied to estimate the most likely actual inventory. The full probability can be used to define confidence intervals around this most likely prediction. Examples of correlations in errors include seasonal errors in inventory, such as season changes in inventory related to holidays or other seasonal events.
The estimation of the actual inventories uses an estimation of the systematic errors made in the planning process. Consequently, the model uses an error function defined so that Ia(t)=Inp(t−nΔ)+ƒn(t−nΔ) where nΔ is the time at which the plan was generated and which must satisfy t−nΔ≦τ. For prediction into the future the model needs only an estimate of ƒn(t) since I(nΔ+δβ)=Ia(nΔ+δβ)=Imp((n−m)Δ+δβ)+ƒm((n−m)Δ+δβ) for all q choices of δ. In the model, fn is defined as fn=[ƒn(nΔ+δ1)L ƒn(nΔ+δq)]T, the vector of errors made by the nth plan as it forecasts the future. Knowing the probability density for fn is all that is needed since the density for Ia simply shifts the mean.
The next step is to estimate P(fn) given the historical data by exploiting two types of correlations that likely exist in the data. First, the elements of fn are likely correlated because if inventory is high (or low) at time nΔ+δ1 then it is also likely to be high (or low) at the later time nΔ+δ2 so long as δ is not much greater than δ1. Moreover, it is reasonable to assume that the plans formed at subsequent planning periods will be correlated since the more recent plan is probably an update of an older plan. To capture such relationships in the historical data, the algorithm uses a model of the joint density P(fn,fn−1,L,fn−p). A reasonable choice of p is p=τ, although a good choice of p is that value determined by an autoregression algorithm. A suitable model for the joint density is that of a Gaussian joint density, as follows:
where f% is the vector of length (p+1)q given by
As described in more detail below, this Gaussian model can be used for predicting inventory levels and calculating the mean and covariance.
If predicting forward from time nΔ, P(fn) gives the distribution over the possibilities from which it is possible to infer expected values and confidence intervals. In general, this is obtained from P(fn%) by marginalizing over the previous errors fn−1,L,fn−p. However, the inventory situation is not this simple because it is desirable to condition the estimate on previously observed errors and not simply use a static estimate. In addition, many of the elements of the previous errors fn−l are for times greater than nΔ and consequently are unknown for the purposes of prediction. The unknown elements from fn−i are those for which (n−i)Δ+δΔ>nΔ or more simply those for which δα>iΔ.
As a solution, define
and
The density that then must be determined from P(fn%) is
which is easily determined by integration and an application of Bayes rule.
The first step is integrating to determine
as illustrated below:
where
is given by Eq. (0.1). The integration results in another Gaussian distribution with mean μ≦ and covariance Σ≦. If N> is a subset of {1,2,L,(p+1)q} giving the indices in
and N≦={1,2,L,(p+1)q}\N> then
μ≦=μ%(N≦) and Σ≦Σ%(N≦,N≦). (0.3)
The notation a(Nr) indicates the |Nr|-vector formed from a by keeping the elements in the ordered index set Nr. The elements are ordered according to the order in N.
Similarly A(Nr,Nc) is the |Nr|×|Nc| matrix formed from A by keeping the rows in Nr and the columns in Nc in their respective orders. This description uses the common notation that |S| gives the number of elements in the set or vector S.
Applying Bayes rule results in
A standard calculation yields this conditional probability as Gaussian having mean and covariance given by
where
The above formula assumes that Σ2 is invertible. In many situations this may not be the case (e.g., if the planned and actual activities are always in complete agreement). As such, Eq. (0.4) is modified to be:
μ(fn%≦)=μ1+Σ1,2(Σ2+λI)−1(fn%≦−μ2) and Σ(fn%≦)=Σ1−Σ1,2(Σ2+λI)−1Σ2,1 (0.5)
where I is the identity matrix having the same size as Σ2 and λ is a small positive quantity (e.g., 10−5). In Eq. (0.5) the μ vectors are given by
μ1=μ≦(N1), μ2=μ≦(N2) (0.6)
and the Σ matrices are defined by
Σ1=Σ≦(N1,N1), Σ1,2=Σ≦(N1,N2), Σ2,1=Σ≦(N2,N1), and Σ2=Σ≦(N2,N2) (0.7)
where the index sets are N1={1,2,L,q} and N2={q+1,q+2,L,|fn%≦|+q}.
Using the above results, the complete algorithm can be used to make the prediction as follows: (1) assuming a value of p (or having the algorithm determine p), construct an estimate of μ% and Σ% from the historical data; (2) construct the index set N≦and form μ≦and Σ≦ from μ% and Σ% according to Eq. (0.3); (3) using historical error values construct the vectors of past errors
(4) build μ(fn%≦) and Σ(fn%≦) according to Eqs. (0.5), (0.6) and (0.7); and (5) return [Ip(nΔ+δ1)L Ip(nΔ+δq)]T+μ(fn%≦) as the predicted actual inventory levels and return Σ(fn%≦) as the covariance on this estimate.
The first two steps can be done offline in a discrete calculation since they have no dependence on the time for which the prediction is being made. To accomplish the first step and determine the parameters of the Gaussian, there a number of approaches or models that can be used. A first model is based on frequency counts while a second and a third method use an autoregressive model of the multivariate time series {fl}. The second and third models differ in the underlying simplifications made, with the second model being more accurate but more complex.
If there is sufficient historical inventory data, the simplest possible way to determine the parameters of the Gaussian is through unbiased frequency count estimates. From the historical record it is relatively easily to build up the following estimates:
As additional inventory data is generated, these estimates are easily updated online. As such, the model is adaptive to and learns from changes in conditions.
Because subsequent plans are likely to be related to each other (e.g., the latter plan is probably a modification of the former plan) it is a reasonable assumption to model the time series {fi} as an autoregressive process of order p:
where the errors, εn, are modeled as i.i.d. zero-mean Gaussian variables described by N(0,Cp). There are many software packages which will efficiently estimate the parameters by maximizing the Gaussian likelihood (e.g., least squares) so the model uses the assumption that wp, Cp, and all Alp are all available. Some packages will attempt to determine the best value of p using a Bayesian information criterion. However, a natural choice in the present setting is p=τ. The proposed predictive algorithm makes use of a number of p.
This probability can be determined as follows:
P(fn,fn−1,L,fn−p)=P(fn|fn−1,L,fn−p)P(fn−1,L,fn−p) where
The model uses a simplification in order to tractably model P(fn−1,L,fn−p), which can be written as
The remaining probability P(f) is assumed to be Gaussian, therefore, P(f)=exp[−(f−μ)TΣ(f−μ)/2]/√{square root over ((2π)τdet Σ)} where μ and Σ can be estimated by simple frequency counts from the historical data:
The joint probability is then
where f%=[fn fn−1 L fn−p]T and the parameters of the Gaussian are
Σ−1 can be inverted using the identity that if
In the present case take B=Σ−1, V1=−Cp−1/2, and Vl=AlTCp−1/2 for iε[2,L,p] where Cp−1/2 is the Cholesky decomposition of C−1 to find
and, by multiplying, the mean is found to be:
As noted above, P(fn−1,L,fn−p) was simplified to the form
This can be improved by writing
As noted in the above equations, each conditional probability can be modeled as an autoregressive process of successively smaller orders. The parameters of each process can be determined from the historical times series. For the final unconditional probability the model uses the same Gaussian assumption as before. Each conditional probability for the AR(p−i) process is written as
and the unconditional probability for fn−p is the same as before having mean μ and covariance Σ. Combining all these results shows that the joint density P(fn,L,fn−p) is Gaussian with precision (inverse covariance) given by Σ%
and Ap is the q×pq matrix given by Ap=[A1p A2p L App]. Given this particular form for the precision matrix, its Cholesky decomposition can immediately be written down as
In this form the inverse is easier to calculate. If an inverse of the form below is necessary:
then the Q matrices must satisfy:
where A0i=Q0l≡−I. The iterative solution satisfying the requirement is given by
starting from Q0p−j=−I.
Thus the covariance matrix, Σ%, is equal to
This matrix can be easily implemented in software as executable instructions to provide the covariance.
Having determined the covariance matrix, the mean is determined by completing the square:
with Σ% as above. Again this matrix and its square are easily implemented in software as executable instructions.
Although complex on paper, the algorithm is computationally very efficient. During the training phase the covariance matrix and its mean can be estimated with a single pass through the data by recording frequency counts. If there are D data points then the complexity of accomplishing both tasks is of the order O(D(p+1)q((p+1)q+1)). Prediction is not much more complicated. The dominant factor is the computation of Σ2−1, which is of order O(|N2|3). Moreover, this matrix inverse can be stored and need not be recalculated with each prediction.
To be used as an online algorithm, the above algorithm must be modified so that it can operate in an online mode in which all parameters of the algorithm (i.e., the mean and covariance) are continually updated as new data arrives so that it is adaptive. This is straightforward because updates to the mean and covariance are simple to implement in software. The algorithm also can be modified to discount data from the past by a factor 0≦λ≦1. This allows the algorithm to learn from recent events and adapt to changing conditions. Thus, if the data f1,L,fT−2,fT−1,fT is available at time T, the estimate of the mean is
and the estimate of the covariance at time T as
where the denominator is given by DT(λ)=1+λ+L+λT−1=(1−λT)/(1−λ). With these definitions the algorithm can be implemented in software to provide running estimates of the mean and covariance and update them as
The discount factor λ should be set to match the time scale for variation of the problem. If behavior is stationary then λ=1 otherwise λ should be set from the data. One quick and basic manner to accomplish this is to match some characteristic time of variation,
Referring to
A number of embodiments of the invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, other embodiments are within the scope of the following claims.
This application claims priority based on U.S. Provisional Application Ser. No. 60/336,227 filed Nov. 14, 2001, for SCM Supply Network Planning, and U.S. Provisional Application Ser. No. 60/384,638, filed May 31, 2002, for Inventory Early Warning Agent in a Supply Chain Management System, the disclosures of which applications are incorporated here by reference in their entirety.
Number | Name | Date | Kind |
---|---|---|---|
4688026 | Scribner et al. | Aug 1987 | A |
5168445 | Kawashima et al. | Dec 1992 | A |
5382784 | Eberhardt | Jan 1995 | A |
5548110 | Storch et al. | Aug 1996 | A |
5564890 | Knudsen, Jr. | Oct 1996 | A |
5582497 | Noguchi | Dec 1996 | A |
5595356 | Kewin | Jan 1997 | A |
5640002 | Ruppert et al. | Jun 1997 | A |
5819232 | Shipman | Oct 1998 | A |
5838253 | Wurz et al. | Nov 1998 | A |
5887176 | Griffith et al. | Mar 1999 | A |
5913210 | Call | Jun 1999 | A |
5953234 | Singer et al. | Sep 1999 | A |
5963134 | Bowers et al. | Oct 1999 | A |
6006196 | Feigin et al. | Dec 1999 | A |
6032125 | Ando | Feb 2000 | A |
6047290 | Kennedy et al. | Apr 2000 | A |
6144945 | Garg et al. | Nov 2000 | A |
6182070 | Megiddo et al. | Jan 2001 | B1 |
6205431 | Willemain et al. | Mar 2001 | B1 |
6232876 | Maloney | May 2001 | B1 |
6243613 | Desiraju et al. | Jun 2001 | B1 |
6341266 | Braun | Jan 2002 | B1 |
6341271 | Salvo et al. | Jan 2002 | B1 |
6357662 | Helton et al. | Mar 2002 | B1 |
6415978 | McAllister | Jul 2002 | B1 |
6557758 | Monico | May 2003 | B1 |
6611726 | Crosswhite | Aug 2003 | B1 |
6693539 | Bowers et al. | Feb 2004 | B2 |
6758403 | Keys et al. | Jul 2004 | B1 |
6763996 | Rakers et al. | Jul 2004 | B2 |
6834266 | Kumar et al. | Dec 2004 | B2 |
7249068 | Kakouros et al. | Jul 2007 | B1 |
20020138336 | Bakes et al. | Sep 2002 | A1 |
20020178127 | Byde et al. | Nov 2002 | A1 |
20020188529 | Krever | Dec 2002 | A1 |
20030093307 | Renz et al. | May 2003 | A1 |
20030126103 | Chen et al. | Jul 2003 | A1 |
20030225635 | Renz et al. | Dec 2003 | A1 |
20040186794 | Renz et al. | Sep 2004 | A1 |
20040215551 | Eder | Oct 2004 | A1 |
20050075949 | Uhrig et al. | Apr 2005 | A1 |
Number | Date | Country | |
---|---|---|---|
20030225635 A1 | Dec 2003 | US |
Number | Date | Country | |
---|---|---|---|
60384638 | May 2002 | US | |
60336227 | Nov 2001 | US |