SYSTEM, APPARATUS, AND COMPUTER-IMPLEMENTED METHOD FOR MONITORING PACKAGES IN TRANSIT THROUGH A LOGISTICS NETWORK - Patent application

A computer system using machine learning to predict and manage delivery risks in logistics networks addresses the challenge of delayed packages by automating the identification and adjustment of transportation plans, enhancing delivery reliability.

JP2024542369AActive Publication Date: 2024-11-15FEDEX CORP SERVICES INC
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
JP2024524463
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-11-30
Filing Date
2022-11-29
Publication Date
2024-11-15
Estimated Expiration
2042-11-29

AI Technical Summary

Technical Problem

Large logistics networks face challenges in identifying packages at risk of being delivered later than the specified commit time due to delays, which are not effectively addressed by traditional systems, requiring manual inspection and substantial personnel and time commitments.

Method used

A computer system that utilizes machine learning to analyze historical network data, train models, and apply thresholds to predict potential delays, enabling real-time monitoring and risk assessment of packages, and adjust transportation plans accordingly.

Benefits of technology

Enhances the ability to identify packages at risk of delayed delivery, reducing manual effort and improving delivery reliability by providing proactive risk management and route adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024542369000001_ABST
    Figure 2024542369000001_ABST
Patent Text Reader

Abstract

Methods, apparatus, and systems are described for monitoring in-transit parcels in a network to determine which parcels are at risk of being delivered after a specified committed time. In general, parcel fingerprints that define events expected to occur for in-transit parcels and define thresholds for those events are generated from historical data of events occurring in a logistics network for in-transit parcels. These thresholds may be applied in real-time to parcels as events occur for the parcel in transit to determine a risk level for any parcel having an event that occurs after the event thresholds. Attention can be given to parcels that are at a sufficient risk level.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present disclosure relates generally to systems, apparatus, computer readable media, and methods in the field of shipment management and logistics, and more particularly to various aspects relating to systems, apparatus, computer readable media, and methods for various enhanced and improved logistics monitoring capabilities to determine packages at risk of being delivered later than a specified committed time.

[0002] (Related Applications) This application claims priority to U.S. Provisional Application No. 63 / 284,615, entitled “SYSTEMS, APPARATUS, AND COMPUTER-IMPLEMENTED METHODS FOR MONITORING PACKAGES IN TRANSIT THROUGH A LOGISTICS NETWORK,” filed November 30, 2021, the entire contents of which are incorporated by reference herein. [Background technology]

[0003] A logistics network transports packages from an origin to a destination. Large logistics networks may operate on a massive scale, where millions of packages are transported from country to country around the world every day. Such large logistics networks may include hundreds of planes, tens of thousands of trucks / vans, and hundreds of thousands of team members. Furthermore, these logistics networks may be highly complex to operate, with day-specific packages, time-specific packages, different transit days, hundreds of service types, and packages moved via various modes of transportation, such as road, air, rail, etc. Network operations may vary over time and by day of the week. Additionally, networks may be affected by seasonality, where peak volumes are approximately 1.5 to 2 times higher than non-peak volumes. Furthermore, because new facilities and new transportation legs and modes may be added continuously, network plans must continually evolve to adapt to changing volumes and network configurations.

[0004] For a given package transported through the logistics network, the purchased service type may stipulate a committed time for delivery of the package from the origin to the destination. Thus, a customer may purchase the level of service required for a particular package. However, packages being transported through the logistics network may experience delays or other exceptions to the normal transport schedule for various reasons. Weather, traffic, equipment failures, internal network operational errors, etc. may affect the transport of the package. Thus, some packages may be delivered after the stated committed time.

[0005] While a slower than expected delivery may be harmless in some circumstances, in other circumstances a slower than expected delivery may have more serious consequences. It is therefore particularly important that awareness of late delivered packages is available. However, conventional computer systems in logistics networks lack the ability to provide such package-level insight to logistics network operator personnel or customers. Thus, such a determination may require manual review of potentially hundreds or thousands of in-transit packages at any given time for any large customer. Monitoring in-transit packages to determine which packages may be late to their destination is a huge and burdensome task requiring significant and expensive personnel and time efforts, which may not result in the identification of all packages of interest.

[0006] To address one or more of these issues, more capable and intelligent facility configurations are needed to learn from the historical behavior of the network and apply those learnings when monitoring the transportation of packages within a logistics network to identify those packages that are at risk of being delivered later than expected. Summary of the Invention

[0007] In the following description, it will become apparent that certain aspects and embodiments are generally directed to technical solutions for logistics operations, including providing a processing system with the ability to train a model of network operations that specifies thresholds for events that a package is expected to encounter while in transit between an origin and a destination based on an analysis of historical logistics network data. It will further become apparent that certain aspects and embodiments are generally directed to monitoring events for packages related to the specified thresholds to determine which packages are at risk of being delivered later than a specified committed time, so that the at-risk packages can be identified to personnel and / or customers. It should be understood that the aspects and embodiments, in their broadest sense, can be practiced without one or more features of these aspects and embodiments. It should be understood that these aspects and embodiments are merely exemplary.

[0008] In one aspect of the disclosure, a computer system for monitoring the transportation of packages within a logistics network is described. The computer system includes an interface to at least one logistics network data source. The computer system includes a first data storage module including historical network data coupled to the interface, the first data storage module constructing the historical network data from data feeds from the logistics network data source. The computer system includes a second data storage module including package transport model data. The computer system includes a first processing system that accesses the data storage and applies machine learning to the historical network data to train the package transport model data. The first processing system captures data-driven patterns including likely locations of historical events, likely types of historical events, and likely times of historical events. From the data-driven patterns, the first processing system creates a set of predicted events for a virtual package being transported from a first location to a final location through the logistics network, and determines thresholds for each of the predicted events for the virtual package to arrive at the final location by a specified time, each threshold associated with the likely time of a predicted event occurring for the virtual package, the predicted events associated with a likely location and a likely type. The computer system includes a second processing system coupled to the interface for receiving live data including data regarding actual loads in transit from the logistics network, the second processing system accessing the transportation model data from the second data storage device, applying the thresholds to the predicted events regarding the actual loads in transit, and evaluating whether the actual loads in transit will be delivered to the final location as scheduled for each predicted event based on the application of the thresholds.

[0009] In another aspect of the disclosure, a computer system for determining a level of risk that a package will be delivered to a final location after a specified time is described. The computer system includes a first processing system that determines a threshold value for each predicted event for the package between a first location and a final location. The computer system includes a second processing system that determines, for each predicted event, whether an event will occur during the package's transit according to a threshold value corresponding to each predicted event. At some point during the package's transit, the second processing system determines a level of risk that the package will be delivered after a specified time based on an amount of delay that exceeds the threshold value for a most recent predicted event.

[0010] Another aspect of the disclosure focuses on a computer-implemented method for determining a level of risk of a parcel being delivered to a final location after a specified time. The method includes determining a threshold value for each predicted event for the parcel between a first location and a final location. The method includes determining, for each predicted event, whether an event will occur during the parcel's transit according to the threshold value corresponding to each predicted event. Additionally, the method includes determining, at a point during the parcel's transit, a level of risk of the parcel being delivered after a specified time based on an amount of delay that exceeds the threshold value for a most recent predicted event.

[0011] Another aspect of the disclosure focuses on a computer-implemented method for monitoring transportation of packages in a logistics network. The method includes applying machine learning to historical network data for the logistics network to train package transportation model data. The machine learning trains package transportation model data by capturing data-driven patterns including likely locations of historical events, likely types of historical events, and likely times of historical events. The machine learning trains package transportation model data by creating a set of predicted events for a virtual package being transported through the logistics network from a first location to a final location from the data-driven patterns, and determining thresholds for each of the predicted events for the virtual package to arrive at the final location by a specified time, each threshold being associated with the likely time of a predicted event occurring for the virtual package, and the predicted events being associated with a likely location and a likely type.

[0012] Each of these aspects results in improvements in the art of logistics operations, including monitoring events of packages in transit through a logistics network to discover packages at risk of being delivered after a prescribed committed time. Additional advantages of this and other aspects of the disclosed embodiments and examples will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the invention. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. [Brief description of the drawings]

[0013] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments in accordance with one or more principles of the invention and, together with the description, serve to explain one or more principles of the invention.

[0014] [Figure 1]FIG. 1 is a diagram of an exemplary logistics network including parcel data collection that may be used for purposes of training a model of logistics network operations and monitoring parcels in real time by applying the model to live parcel data to determine the risk of a parcel in the logistics network being delivered after a specified time, in accordance with an embodiment of the present invention.

[0015] [Diagram 2] FIG. 1 illustrates an example computer system that may include one or more processing systems that implement one or more modules for training a model of network operations and one or more modules for monitoring packages in real time by applying the model to determine risk.

[0016] [Diagram 3] FIG. 1 illustrates three example aspects of package monitoring to detect packages at risk of being delivered after a specified time, including model training, model scoring, and configurable output.

[0017] [Figure 4] FIG. 1 is a flow diagram illustrating an example method for training a model of logistics network operations including predicted events, thresholds associated with the predicted events, and risks associated with real-time delayed occurrence of the events.

[0018] [Diagram 5] FIG. 3 is a diagram of example modules that may be executed by a processing system such as that shown in FIG. 2 to train a model of logistics network operations.

[0019] [Figure 6] 6 is an exemplary histogram illustrating the distribution of marginal times of events as shown in FIGS. 4 and 5 and illustrating the determination of a threshold of events to be included in the model from the distribution of marginal times of events. FIG.

[0020] [Figure 7]FIG. 1 is a flow diagram illustrating an example method for determining event thresholds for current predicted events expected to occur for packages being transported within a logistics network.

[0021] [Figure 8A] FIG. 1 is a flow diagram illustrating an example method for determining the location and type of a next predicted event that is expected to occur for a package being transported within a logistics network after the occurrence of a corresponding current event. [Figure 8B] FIG. 1 is a flow diagram illustrating an example method for determining the location and type of a next predicted event that is expected to occur for a package being transported within a logistics network after the occurrence of a corresponding current event.

[0022] [Figure 9A] FIG. 13 is a flow diagram illustrating an example method for determining values ​​in a risk table that provide the risk associated with an amount of delay for each predicted event. [Figure 9B] FIG. 13 is a flow diagram illustrating an example method for determining values ​​in a risk table that provide the risk associated with an amount of delay for each predicted event.

[0023] [Figure 10] FIG. 1 is a flow diagram illustrating an example method of using a model to score risk of a package in relation to a predicted event, including both ex-post and ex-ante scoring of risk for a predicted event.

[0024] [Figure 11] FIG. 1 is a flow diagram illustrating an example method for using a model to provide a posteriori scoring of a package's risk in relation to the occurrence of a current predicted event.

[0025] [Figure 12] FIG. 13 is a flow diagram illustrating an example method of using a model to provide a proactive scoring of a package's risk in relation to the non-occurrence of a next predicted event following the occurrence of a current event.

[0026] [Figure 13] FIG. 13 is a flow diagram illustrating an example method of a risk aggregator aspect of a model for establishing a level of risk for a predictive event in terms of it being the next predictive event that, upon occurrence, becomes the current event.

[0027] [Figure 14] FIG. 1 is a flow diagram illustrating an example method for rerouting loads depending on a level of risk determined by a model.

[0028] [Figure 15] FIG. 1 illustrates an example timeline of events for a package being rerouted by level of risk.

[0029] [Figure 16] FIG. 1 is a flow diagram illustrating an example method for determining risk to a load from pre-shipment harsh condition advisories in conjunction with a forecasted transportation route provided by a model.

[0030] [Figure 17] 1 is a screen capture illustrating an exemplary user interface for presenting information about packages in transit within a logistics network, including those at risk of being delivered later than scheduled.

[0031] [Figure 18] 1 is a screen capture illustrating an example user interface for presenting information about a package in transit within a logistics network, including different levels of risk of being delivered later than a specified time.

[0032] [Figure 19]1 is a screen capture showing an exemplary user interface for presenting information about packages in transit within a logistics network, including where events are occurring that indicate the package is at risk of being delivered later than expected, including filters to control the nature of the information displayed. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0033] Reference will now be made in detail to various illustrative embodiments. Wherever possible, the same reference numbers are used in the drawings and the description to refer to the same or like parts. However, those skilled in the art will understand that different embodiments may implement particular parts in different manners depending on the needs of the intended deployment and operating environment of each embodiment.

[0034] Generally, various embodiments of systems, devices, computer readable media, and methods are described below that model logistics network operations in terms of the timing and location of events that a package encounters while in transit within the logistics network. Thresholds are determined for events that a package is expected to experience, and these thresholds provide a way to determine, for each event that a package is expected to encounter while in transit, whether the package is at risk of being delivered after a specified time, such as a commit time that the recipient expects to meet.

[0035] Those skilled in the art will also appreciate that the embodiments described herein provide improvements to certain technologies, such as systems for monitoring package shipments, logistics operations, and related infrastructure. The embodiments describe specific technical applications that leverage and apply specific embodiments that create a model of logistics network events and use the model to monitor packages in transit through a logistics network to identify packages that are at risk of being delivered after a defined committed time. Monitoring to identify packages at risk allows for relevant actions to be taken not only within the logistics network, but also at the customer level, and the specific technical applications improve or otherwise enhance such technology areas, as described and supported in the disclosure that follows.

[0036] FIG. 1 illustrates a portion of an example logistics network 100 used to transport packages between an origin and a final location, also referred to as a destination. The example logistics network 100 may include any number of facilities. Hubs 102, 104 are examples of facilities where packages arrive and are sorted for transport either to another hub or to a station 106, 108, 110, 112, 114, or 116 associated with the hub. Packages are transported from hubs 102, 104 to an associated station where they may be routed to their destination.

[0037] As packages are transported through the logistics network 100, they encounter various events from which information about the package is collected by the logistics network 100. This data is then collected at a computer system 118, such as one or more server computers that maintain one or more databases of package and event information. These events may be scan events, such as when a barcode, QR code, or similar readable code from a label or from an electronic circuit such as a radio frequency identification (RFID) tag placed on the package is scanned. These events may also include communications initiated by a transmitter on the package, such as an active RFID tag, a short-range transmitter, or a Bluetooth device, that transmits package information to a receiving device.

[0038] These events may occur at each milestone of the parcel during its journey. For example, an event may occur when the parcel is picked up at a remote location. Another event may occur when the parcel arrives at a station. Another event may occur when the parcel passes through a sorting or loading stage within a station. Another event may occur when the parcel departs from a station for a hub. Another event may occur when the parcel arrives at a hub. Another event may occur when the parcel is sorted at the hub, etc. Because each event identifies the parcel along with the time and location of the event, the location of the parcel at any time within the logistics network 100 is known.

[0039] The specific routes and timings that packages follow during transportation through the logistics network are defined by a plan. A package transportation planning system 120, such as one or more server computers, may determine an appropriate transportation plan for each package, and that data is provided to various facilities, including hubs and stations, so that packages identified at each event can be sorted and loaded to follow the routes described in the plan. The plan may also specify prescribed times by which packages should be delivered to their final destinations, for which routes may be selected. However, packages may experience delays and other deviations from the initial plan due to various factors.

[0040] 2 illustrates an exemplary computer system 200 that may be used to model a logistics network based on historical data collected by the parcel data collection system 118. The exemplary computer system 200 may then apply the model to real-time data streaming from the logistics network 100 to determine the level of risk of a parcel being delivered later than a specified time. The computer system 200 may include an interface 202 to data sources in the logistics network 100, including the data collection 118. The data streaming from the logistics network 100 to the computer system 200 via the parcel data interface 202 may be prepared and cleansed using typical data cleansing techniques. For example, the data interface may filter out events that may not be relevant to model training or scoring activities.

[0041] The parcel data storage module 204 collects historical network event data prepared for use in subsequent processes of both model training and scoring activities. For model training, the parcel data storage 204 may provide historical event data of the logistics network 100 to the first processing system 210. For model scoring, the parcel data storage 204 may provide live data that is streamed from the logistics network 100 to the second processing system 218. Although the first processing system 210 and the second processing system 218 are shown as separate items in FIG. 2, it will be understood that these processing systems may be provided by a single computer or a single collection of computers that implement separate functional modules for model training and scoring. Furthermore, when referring to computers, server computers, computer systems, and processing systems herein, it will be understood that these include one or more general purpose programmable processors, one or more application specific processors, wired digital logic, and / or various combinations of such devices.

[0042] The first processing system 210 may perform operations to train a model representing events occurring in the logistics network 100 from historical event data captured from the logistics network 100 over a predetermined period of time. Additionally, the first processing system 210 may repeatedly perform model training to account for new aspects of the logistics network 100 that may occur over time. For example, new events may be introduced into a collection of existing facilities, and these new events are captured in the historical data held in the parcel data storage 204 so that the first processing system 210 may continue to train the model to incorporate the new events. Additionally, the addition of a new facility may introduce new events into the historical data held in the parcel data storage so that the first processing system 210 may continue to train the model to incorporate new events corresponding to the new facility. In this manner, the computer system 200 maintains a model of the network 100 that is up to date, allowing the model to accommodate the logistics network 100, thereby allowing real-time application of the model for scoring to account for the evolving network 100.

[0043] Further, the computer system 200 is not limited to a single logistics network 100 or operator of the logistics network 100. The computer system, particularly the first processing system 210, may incorporate different networks when training the model, regardless of the operator, as long as historical event data from these additional networks is available. Additionally, the computer system 200, particularly the second processing system 218, may score the model for loads in different networks, as long as streaming live event data from the different networks is available.

[0044] Focusing on the first processing system 210, the exemplary computer system 200 includes two subsystems that provide machine learning to train the model. The first subsystem 212 implements machine learning from the captured historical event data to perform a baggage fingerprint calculation to create a baggage fingerprint that defines the model. A baggage fingerprint describes the events that a baggage is expected to encounter during its transit from a specific first location, such as the baggage's origin location, to a specific final location that is the baggage's destination. The events captured from the network that make up the historical data and the events inherent in the baggage fingerprint are identified as having several characteristics. For example, an event may be identified as having three characteristics: the location where the event occurs, the type of event that occurs, and the time at which the event occurs.

[0045] The time included in the package fingerprint of the model specified for an event is the latest time that the event should occur for a virtual package to be delivered on time to its final location by the specified time. This time of the event in the model is therefore referred to as the package threshold. This package threshold is determined by machine learning from historical data, where events of a given location and a given type represented in the historical data typically occur over a specified range of times. An example of the machine learning operation of the first subsystem 212 that generates the threshold is described in more detail below with reference to Figures 4 through 8B.

[0046] The first processing system 210 stores the luggage fingerprints for all known luggage events determined by machine learning from historical data and model formation as luggage transportation model data in a luggage fingerprint storage module 216, which may be a database. The luggage fingerprint data can then be made available from the luggage fingerprint storage module 216 to other subsystems of the first processing system 210, including the second subsystem 214 for luggage risk table calculation. The luggage fingerprint data is used by the second subsystem 214 to determine the level of risk of the virtual luggage being delivered to a particular final location for different amounts of delay beyond the luggage threshold for each predicted event in the luggage fingerprint for the virtual luggage. The determination of the entry into the risk table in the second subsystem 214 is described in more detail below in connection with Figures 9A and 9B.

[0047] The risk table of the second subsystem 214 and the parcel fingerprints from the first subsystem 212 and the storage device 216 are made available to the second processing system 218 for model scoring purposes on the real-time parcel data. The second processing system 218 applies the parcel threshold when using model scoring to determine in real-time the amount of delay of a predicted event that exceeds the parcel threshold for a real parcel by also receiving live data streamed from the logistics network 100. Upon determining the amount of delay, if any, that a real parcel experienced for a particular predicted event, the second processing system 218 may determine a corresponding risk level for the parcel to be delivered to the final location after a prescribed time. The determination of the risk level for a real parcel in transit is described in more detail below in connection with Figures 10 to 13.

[0048] In addition to risk determination based on the risk table from the second subsystem 214, the second processing system 218 may also receive risk information regarding external factors that result in harsh conditions for transporting the package. An external source 206 of data related to such external factors may stream data to the computer system 200, more specifically to the external factors / risk storage 208 and the second processing system 218. For example, external factors may include weather, traffic, power outages, etc., which may result in predictable delays before the package is transported from one location to another, rather than being determined only in relation to the package threshold for the event. The second processing system 218 makes such pre-prediction of delay risk by having a predicted path from the package fingerprint, and then evaluates the predicted delay risk for each leg of the journey that the package is expected to make. The determination of risk from external sources is described in more detail below in relation to FIG. 16.

[0049] Once a level of risk has been assessed for a physical package in transit within the logistics network 100, information regarding the physical package and the level of risk is made available to downstream consumers. An alert may be generated immediately from the second processing system 218 to draw attention to specific situations regarding packages that have missed events or packages with significant levels of risk. Additionally, information regarding the package and the level of risk may be provided to another storage device 220 that maintains an up-to-date risk assessment for each physical package currently in transit within the logistics network 100. Additionally, one or more third processing systems 222, 224 may access the information and provide the information to one or more entities via user applications. For example, one system 222 may act as a web server and provide information to customers expecting to receive packages via a data feed to a dedicated application or a data feed to a website that the customers access via the Internet. Another system 224 may provide information to personnel of the shipping entities operating the logistics network 100.

[0050] A person viewing information provided by system 222 or 224 as a display on a device may be able to quickly see which packages are at risk of being delivered after a specified time. Thus, the person does not have to sift through information for all packages for which they wish to monitor on-time delivery status, as only packages at risk of being delivered after a specified time require attention. Additionally, the person may be provided with further options, including viewing package-specific information regarding packages at risk of being delivered after a specified time. Personnel at the shipping entity may be provided with additional features, including the ability to filter information regarding any package in the logistics network 100, such as by customer, and to identify specific facilities, days of the week, etc., within the logistics network 100 where a package is at risk, in order to further identify potential points of failure within the network. Screen captures of examples of such user interfaces are described in more detail below with reference to Figures 17-19.

[0051] In addition to displaying information about parcels at risk of being delivered after a predefined time for delivery, information about parcels at risk may be shared with a fourth processing system 226 that executes parcel transportation planning for the logistics network. The fourth processing system 226 may be a subsystem of the parcel transportation planning system 120, which determines the actual route through the logistics network 100 that the parcel will take to be delivered to its final location. Thus, at any given facility along the intended route for the parcel, the fourth processing system 226 may determine that a better route is available to increase the likelihood that the parcel will be delivered by the specified time. Once the parcel has taken the new route, the second processing system 218 may use events that occur along the new route to determine the level of risk, thereby providing a closed-loop system between the fourth processing system 226 for parcel transportation planning and the second processing system 218 for risk determination via model scoring.

[0052] For example, during the pre-shipment stage of a package, an external factor such as a weather forecast of adverse weather at a facility at the time the package is expected to arrive at or depart from the facility may indicate a risk for the package that triggers the fourth processing system 226 to create a new plan to route the package through a different facility. Similarly, a level of risk determined as a result of a package threshold being violated during transportation may trigger the fourth processing system 226 to create a new plan to route the package through a different facility to reduce the amount of delay associated with a subsequent event encountered by the package. In either case, the second processing system 218 evaluates the risk of the package being delivered to the final location, and that risk information is provided as feedback to the fourth processing system 226.

[0053] 3 is a diagram illustrating an overview 300 of the functionality of the computer system 200. Here, it can be seen that model training from historical data of events captured from the logistics network 100 is performed in overview step 302. This relates to the operation of the first processing system 210, where a package fingerprint containing predicted events and a corresponding package threshold for each event are determined along with associated risk levels for different amounts of delay to the package threshold. Model scoring is then performed in overview step 304. This relates to the operation of the second processing system 218, where a package fingerprint package threshold is compared to current events after they have occurred, and a package threshold is compared to next predicted events, while waiting for them to occur, proactively. The determined risk levels, and missed event exceptions as a whole, can then be provided to downstream consumers in overview step 306 via configurable outputs on a displayed user interface.

[0054] FIG. 4 is a flow diagram of machine learning steps 400 that may be performed for model training and executed by the first processing system 210. First, the first processing system 210 extracts historical data including event type, event time, and event location of parcels delivered over a certain time frame in step 402. The event type may specify whether the event is a pickup, arrival, departure, delivery, etc. The event time may specify the time and day of the week. The event location may specify the location where the event occurs, including any facility in the logistics network 100 or any other location where the event occurs. Next, a first module (module 1) implemented by the first processing system 210 calculates the slack time for all events in step 404 and categorizes the slack time into slack time bins. The slack time is the amount of time between the time of the event and a defined time that is the latest time that a delivery should occur in order to be considered on time. By categorizing the slack time of the events into slack time bins, a histogram is effectively constructed, an example of which is described in more detail below with reference to FIG. 5 and FIG. 6.

[0055] Once all the event slack times from the historical data have been classified, in step 406, the first part of the second module (module 2.1) determines a threshold value for each current predicted event of the virtual package. The threshold value for each current predicted event is determined by calculating a desired percentile of all slack times specific to on-time packages relative to the location of the current predicted event. For this current predicted event of the virtual package, the virtual package has a final destination, is shipped with a specific product type, and has a defined committed time for delivery on a specific day of the week. The product type is related to the level of service being offered to transport the virtual package, such as ground service, next day service, international service, etc. Thus, each threshold value determined in step 406 is associated with a current predicted event of a virtual package with some of these characteristics, which allows for the appropriate selection of the current predicted event for inclusion of a package fingerprint of a real package with the same characteristics as the virtual package. An example of a flow of sub-steps taken to determine a threshold value as in step 406 is described in more detail below with reference to FIG. 7.

[0056] Once the baggage thresholds have been determined for all possible current predictive events in the network, the second portion of the second module (module 2.2) determines the location and type of the next predictive event following the particular current event in step 408. To determine the location and type of the next predictive event following the particular current event, the thresholds of all possible next predictive events following the particular current event are determined using a histogram of the lead times of the possible next predictive events following the particular current event, again using a desired percentile to determine the potential next event threshold. The potential next event with the slowest threshold is selected as the location and type of the next predictive event for the baggage fingerprint. An example of a flow of sub-steps performed to determine the location and type of the next predictive event in step 408 is described in more detail below with reference to Figures 8A and 8B.

[0057] Once all the next predicted events have been determined, the third part of the second module (module 2.3) combines the outputs of modules 2.1 and 2.2 in step 410. Thus, for a given current predicted event of a given virtual load with a determined threshold, module 2.3 combines the appropriate next predicted event location and type, and the next predicted event threshold determined in module 2.1. The threshold used with the next predicted event selected by module 2.2 is not the threshold determined by module 2.2, but the threshold calculated for the location and type of the next predicted event when that location and type is considered as the current predicted event in module 2.1. This removes the dependency on the current predicted event preceding the next predicted event when determining the threshold for that next predicted event, by using the output of module 2.1 as the threshold for each next predicted event. This is related to the posteriori scoring of the current event and the ex-ante scoring of the next event, which will be discussed below in connection with Figures 10-13. An example output of module 2.3, which combines the current predicted event from module 2.1 with the next predicted event from module 2.2, is shown in Table 1 immediately below. Table 1: [Table 1]

[0058] This luggage fingerprint of the virtual luggage shows that each predicted current event has a threshold T_1 (specified in minutes to the commit time) determined in step 406 of module 2.1. Each next predicted event following the current predicted event has a threshold T_2 (specified in minutes to the commit time). However, it is further seen that each next predicted event is followed by the same event, but this time as the current predicted event. In other words, when the next predicted event actually occurs, it becomes the current predicted event. Thus, T_2 of the next predicted event following one current predicted event will be equal to T_1 of the next current predicted event.

[0059] As a specific example of the package fingerprint shown in Table 1 above, for carrier entity FX, the first current predicted event occurs at facility ABEA of type PU (i.e., a pickup event), where the package is destined for a delivery address associated with facility CHIA. The service description code D indicates a domestic delivery service, the service code 3 indicates a two-day delivery, and the day of the week (dow) indicates a delivery date of Wednesday. The threshold T_1 of the first current predicted event is 2835, which means that the pickup is expected to occur 2835 minutes before the specified delivery time for the purchased service type. The next event expected to follow the current predicted event occurs at facility EWRHB as an arrival (AR) as determined by module 2.2, which searches for all possible next events that may follow the current event of the predicted ABEA PU. The threshold T_2 of this next predicted event is 1890 minutes before the specified delivery time, and was not found by module 2.2 due to this T_2 value. The current predicted event following the next predicted event is EWRHB arrival, whose threshold T_1 is 1890, exactly the same as the T_2 value of the preceding next predicted event. Module 2.1 required that the threshold T_1 be 1890 when evaluating the EWRHB arrival event as the current event. Module 2.3 used the T_1 value of the threshold of the current predicted event of EWRHB, 1890, as the T_2 value of the threshold of the next predicted event of EWRHB, eliminating the dependency of the T_2 EWRHB threshold on the previous ABEA PU event.

[0060] An additional module part 2.4 may be used in combination with the output of module 2.3 to append the destination time zone name, as shown in Table 1. This allows for subsequent adjustment of the time to a common frame of reference, such as Coordinated Universal Time (UTC), based on the time zone that applies to the location of each predicted event. This simplifies the application of thresholds as events may span from one time zone to the next.

[0061] Once the package fingerprint information is complete for all predicted events that may occur within the logistics network 100, a risk dimension table is created for all predicted events in step 412. This is done by calculating the risk of delivery delay for each time range in which the event may occur beyond the event threshold. For example, the range may be one hour, so that a risk level is calculated for each time the event occurs after the threshold from the package fingerprint. The risk is specific to the predicted event and the characteristics of the virtual package, including the event location, event type, product or service type, days until the commit time, day of the week for the commit time, etc.

[0062] FIG. 5 illustrates a set of components used in the flow of FIG. 4 to determine the current predicted event, the next predicted event, and respective thresholds such as event location, event type, product or service type, days to commit time, and day of the week for the commit time. Module 1, described above with respect to step 404 of FIG. 4, utilizes inputs 502 including event type, domestic vs. international service type, country code of the country the package is being shipped to, carrier code identifying the carrier of any particular package, and delivery date range to determine the slack such that the slack binned is the correct slack for a particular virtual package with the predetermined characteristics. Module 1 operation 504 accesses historical package event data 506 to calculate the slack distribution for each possible event in the logistics network 100. Module 1 output 508 includes the slack in bins to establish the slack distribution, which may be represented in histogram form 600 as shown in FIG. 6. In the example of FIG. 6, it can be seen that the histogram 600 includes slack binned into 15-minute intervals. Although other intervals may be selected, it has been found that the 15-minute interval provides sufficient resolution to determine the effective thresholds.

[0063] This output of Module 1 is then provided to Module 2 along with its input 512, which includes percentile numbers, minimum flow volume percent, and predefined time committed date range for delivery. Module 2 operation determines the package fingerprint events, including the current predicted event and the next predicted event, and thresholds for each event, as described in steps 406, 408, and 410 of Figure 4. The output of Module 2 is placed in storage 510 and includes a package fingerprint component that includes all predicted events and corresponding thresholds, similar to those shown in Table 1 for a given virtual package.

[0064] Returning to FIG. 6, histogram 600 further illustrates how module 2.1 determines a threshold for a given current event. In this example, the event is an outbound scan from a hub of a package destined for a particular final location. A threshold is then determined using a desired percentile of slack time. In this example, the desired percentile, indicated by reference numeral 606, is the 99th percentile, which has been shown to provide a useful threshold for determining the risk of package delay. In this example, the 99th percentile of observed on-time scans is 12:15 AM on the commit date for delivery. Thus, packages that leave this particular hub after 12:15 AM on the commit date and are destined for this particular final location are presumed to be at a level of risk of being delivered after the commit date. In this manner, packages with slack time in bins in area 602 are presumed to be on time with no level of risk, while packages with slack time in bins in area 604 are presumed to be at risk of delivery delay.

[0065] FIG. 7 is a flow diagram illustrating example steps 700 that may be followed when performing the functionality of module 2.1 to determine a current predictive event threshold for a package fingerprint. Module 2.1 begins at step 702, where module 2.1 reads an input parameter as the percentile number at which the event fingerprint threshold is selected. As described above for the example shown in FIG. 6, the percentile was 99. The current predictive event slack distribution provided to the histogram is read as input at step 704 from the database storing the histogram, e.g., a Hive table when Apache Hive is used. At step 706, the bin code columns are extracted for event location, ship-to location, event type, ship service type, ship commit day of week, and event slack bin number. At step 708, if the bin time code is negative, a flag is created. Next, at step 710, the total frequency of each of the event location, ship-to location, event type, ship service type, ship commit day of week, and event slack bin number where the bin time code is non-negative is calculated.

[0066] At this point, a set of steps 712 is performed for each unique combination of event location, ship-to location, event type, shipping service type, and shipping day of the week. In a first step 714 of the set of steps 712, the events are sorted in ascending order by bin time end. In step 716 of the set of steps 712, the cumulative frequency (CF) is calculated, and in step 718 of the set 712, the total frequency (TF) is calculated. Next, in step 720 of the set of steps 712, the cumulative percentage (CP) is calculated as CP=CF / TF. In step 722 of the set of steps 712, rows with associated CPs of at least 1-percentile number (e.g., 99) / 100 are selected. Next, in step 724, row numbers are assigned in ascending order of bin time end, and a row with row number equal to 1 is selected as the fingerprint threshold for the current predicted event. The current predicted event is now defined as a unique combination of event location, ship-to location, event type, shipping service type, and shipping commit day of the week. Next, in step 726, the current predicted events associated with each selected fingerprint threshold are written to a threshold data frame, completing module 2.1.

[0067] 8A-8B are flow diagrams illustrating an example 800 of steps that may be followed when performing the functionality of module 2.2 to determine the next predicted event location and type for a baggage fingerprint for a given current predicted event determined in module 2.1. First, in step 802, an input parameter is read as the desired percentile number of observed on-time scans. This may be the same percentile (such as 99) used in module 2.1, or it may be different since this threshold is used as a test for determining the next event in the most recent next predicted event after the given current event, but is not used as a threshold for the next predicted event in the baggage fingerprint. Next, in step 804, another input parameter is read as the desired minimum percentage of all historical next events that are required for the next predicted event to be a valid next predicted event. This minimum percentage represents a minimum flow volume for a potential predicted event so that next predicted events with very low occurrences can be filtered from the pool of potential next events for a given current event. A useful minimum percentage for such filtering has been found to be 10%, although other minimum percentages are also applicable.

[0068] Once the input parameters are read, in step 806, margin histogram data for each potential next event for a given current forecast event is read as input from a database that stores the histograms, e.g., a Hive table if Apache Hive is used. This allows the bin code columns to be extracted in step 808 into event location, ship to location, event type, ship service type, ship commit day, next event location, next event type, and next bin time code. In step 810, a flag is created if the bin time code is negative, otherwise a bin time end is created from the number in the next bin time code. Then, in step 812, the total frequency of each of the event location, ship to location, event type, ship service type, ship commit day, next event location, next event type, and next bin time code if the bin time code is not negative is calculated.

[0069] At this point, a set of steps 814 is performed for each unique combination of event location, ship-to location, event type, shipping service type, shipping commit day, next event location, and next event type. In a first step 816 of the set of steps 814, the events are sorted in ascending order by bin time end. In a step 818 of the set of steps 814, a cumulative frequency (CF) is calculated, and in a step 820 of the set 814, a first total frequency (TF1) is calculated. Next, in a step 822 of the set of steps 814, a cumulative percentage (CP) is calculated as CP=CF / TF1. In a step 824 of the set of steps 814, rows with an associated CP of at least 1-percentile number (e.g., 99) / 100 are selected. Next, in a step 826, row numbers are assigned in ascending order of bin time end, and rows with row number equal to 1 are selected.

[0070] Once a row equal to 1 is selected for each unique combination, a series of steps 828 is performed for each unique combination of event location, ship-to location, event type, ship service type, and ship commit day. In step 830, the first of the series of steps 828, a second total frequency (TF2) is calculated. Then, a series of steps 832 nested within the series of steps 828 is performed for each unique combination of event location, ship-to location, event type, ship service type, ship commit day, next event location, and next event type. In step 834, a distribution percentage (DP) is calculated as DP=TF1 / TF2. In step 836 of the set 832, rows with associated DPs at least a minimum percentage (e.g., 10) / 100 are selected, thereby retaining only potential next events that meet this minimum flow volume. Next, in step 838, row numbers are assigned in ascending order of bin time end, and a row with row number equal to 1 is selected as the next predicted event type and location. Here, the next predicted event for each current predicted event is defined as the unique combination of the event location, ship-to location, event type, shipping service type, and shipping commit day corresponding to the associated previous current predicted event. In step 840, this next predicted event type and location are written as output to the next event data frame. As mentioned above, the parcel threshold used in the parcel fingerprint of the next predicted event is determined separately by module 2.1, where the next predicted event is considered as the current predicted event, and module 2.3 combines the outputs of modules 2.1 and 2.2 to generate the parcel fingerprint as shown in Table 1. Here, the threshold T_2 of the next predicted event is the same as T_1 for the current predicted event following the predicted event, because the next predicted event (T_2) is the same predicted event monitored during pre-event scoring and the next current event (T_1) is the same predicted event monitored during post-event scoring, respectively.

[0071] A specific example of how module 2.2 determines the next predicted event location and type is as follows: A historical current event that occurred at a DENR event location of departure type has three potential next locations and event types in the historical data. For each of the three potential next locations and event types, a 99th percentile threshold time and sample size are determined. For example, an arrival event type at the ORDR event location occurs 5% of the time, with a 99th percentile threshold for this pair of 10 / 1 / 19 5:30. An arrival event type at the INDH event location and an arrival event type at the MEMH event location occur 45% and 50% of the time, respectively, with a 99th percentile threshold of 10 / 1 / 19 1:45 and 10 / 1 / 19 00:00, respectively. Given this information, the arrival event type at the INDH event location is selected as the next predicted event type and location for the current event of departure from DENR. Because that next predicted event location and type has a sample size greater than 10%, its 99th percentile threshold is the most recent among those with a sample size greater than 10%.

[0072] 9A and 9B are flow diagrams illustrating example steps 900 that may be followed when performing the functions of the parcel risk table calculation subsystem 214 of the first processing system 210. In a first step 902, input parameters are read as the shipping start date and shipping end date of the parcel of the historical data. Next, in step 904, historical event data for parcels with shipping dates within the shipping start date and shipping end date are read as further input. In step 906, the service type, commit day, and service result (delivered on time or delayed by a certain amount of time) are extracted from the historical data of each parcel. In step 908, an event array is expanded for the extracted information. Next, in step 910, the data is filtered to keep only fingerprint eligible events. For example, all events that are not related to the predicted events of the logistics network 100 used in the parcel fingerprint are removed.

[0073] Once the historical event data has been filtered, it is ready to be applied to determine a risk level for parcels being delivered after a prescribed time based on the length of delay over a threshold for each predictive event in the parcel fingerprint. In step 912, for each event for each parcel from the historical data, the event type, event location, and number of days from event time to commit date are determined, with maximum values ​​specific to each service type. For example, an express service type has a maximum of 3 days, and a ground service type has a maximum of 7 days. In step 914, for each event for each historical parcel, the corresponding parcel fingerprint threshold is obtained. It is then determined in step 916 whether any of the parcel's historical events violate the parcel fingerprint threshold.

[0074] At this point, the amount of delay for each historical event for each package determined in step 916 can be calculated in step 918 by determining the difference between the actual event time and the fingerprint threshold up to a desired maximum value. For example, the maximum delay may be limited to a level such as 10 hours where the risk level of a package being delayed is already considered to be very high. In this example, the amount of delay is calculated in hours, but it will be understood that other units of time can also be the basis for the determination, depending on the desired level of granularity. In step 920, for each historical event with a fingerprint threshold violation, the fingerprint violation counter assigned to the set of matching events to which the event with the threshold violation belongs is incremented by one. Thus, the result of step 920 is an aggregation of threshold violations for each particular set of matching events.

[0075] In step 922, for each historical event that has a fingerprint threshold violation and a delivery that is delayed because it occurred after a specified time, such as the commit time, a late delivery counter assigned to the set of matching events to which the event with the threshold violation and late delivery belongs is incremented by one. Thus, the result of step 922 is a tally of late deliveries for each of the matching events in a particular set of matching events. For each of these sets of matching events, a level of risk may be calculated by dividing the value of the late counter by the value of the violation counter. Thus, the output of step 924 is a value between 0 and 1 to set the level of risk. If one wishes to express the output as a percentage, this value may be multiplied by 100. The output is then written to a risk table in step 926.

[0076] An example risk table is shown below in Table 2 for an arrival (AR) event type at event location MEMH. This shows that there may be delay amounts not shown in the historical event data. For example, delays of 1, 2, 3, and 6 hours are not represented in the historical data. However, these delay amounts may occur in the future. Therefore, it may be appropriate to interpolate to determine the risk amount for such missed delay amounts. A risk table including such an interpolation is shown below in Table 3. As historical data continues to be captured, the process of creating the risk table may be continually performed by the first processing system 210 so that the interpolated delay amounts may eventually be replaced with the delay amounts occurring in the historical data. Table 2 (without interpolation): [Table 2] Table 3 (after interpolation): [Table 3]

[0077] The operation of the first processing system 210 to create a model defined by the package fingerprint and associated risk table is described above, and the second processing system 218 may score the model against real-time data of a real package in transit through the logistics network 100 to determine a level of risk for each predicted event, both ex post and ex post. FIG. 10 is a flow diagram of an exemplary method 1000 for scoring a model against real-time event data of a real package in transit. In step 1002, real-time live event data streaming for a real live package in transit is received by the second processing system 218. At this point, the ex post scoring for the next predicted event from the package fingerprint is cancelled by the occurrence of the current predicted event corresponding to the event data just received in step 1002. In step 1006, the new current event is scored against the model by comparing the current predicted event threshold that matches the current event that actually occurred for the real package against the new current event that just occurred for the real package to determine the amount of delay (if any) for the occurrence of the event that exceeds the threshold. This amount of delay may then be determined for a predicted event in the risk table that matches the current event to reveal a corresponding level of risk specified in the risk table, which may then be made available to downstream processes.

[0078] Once the level of risk of the current event to the actual package has been determined, a pre-type scoring may be scheduled for the next predictive event defined for the current event in the package fingerprint in step 1008. However, if the current event is of delivery type, the process may end since delivery has occurred. If the current event is not of delivery type, i.e., the package is still in transit, then a pre-type scoring is scheduled and begins to occur at the time period specified by the schedule in step 1008. For example, the pre-type scoring may be scheduled to run every 15 minutes. It will be appreciated that the schedule need not be every 15 minutes, in which case the amount of time may be selected as a matter of balance between the resolution of the pre-type score and the amount of processing required by the second processing system 218, given the potentially large scale of the logistics network 100 and the sheer number of actual packages in transit at any given time to be scored against the model. Once the threshold of the next predicted event to be scored has passed, a level of risk is assigned. The scheduled pre-type scoring may then continue to periodically update the risk level based on the increase in delay beyond the threshold. If a new occurrence of an event occurs for the package, the flow of the example method 1000 returns to step 1002 .

[0079] Because proactive scoring is looking for parcels for which the next predicted event has not yet occurred, it is not yet possible to score that event as a current event. This allows proactive scoring at step 1008 to determine risk for the next predicted event before it actually occurs in the network 100, rather than determining risk only after the event has occurred via reactive scoring at step 1006. When proactive risk assessment is performed at a scheduled time, risk aggregation may be used in conjunction with the latest proactive score and the current proactive score to determine whether the risk from the latest proactive score or the risk from the existing proactive score is used as the risk level for the parcel at that time. However, once the next predicted event actually occurs and becomes the current event being proactively scored, the risk from that proactive score becomes the risk level for the parcel. An example of the steps of reactive scoring is described below in conjunction with FIG. 11, and an example of the steps of proactive scoring is described below in conjunction with FIG. 12.

[0080] FIG. 11 is a flow diagram illustrating example steps 1100 that may be followed when performing a reactive scoring function of the second processing system 218. In step 1102, the flow begins by reading streamed events from event locations in the logistics network 100 that are directed to the second processing system 218. The streamed events may identify the package, the origin, the final location, a predefined time for delivery known as the service commit, the product / service type, and the delivery day of the week. In step 1104, the events are organized into the specific data processing format required and cleansed to filter out events that are not relevant for scoring or are otherwise out of scope. In step 1106, any proactive scoring activity scheduled for this package is deactivated because a reactive scorable event has occurred.

[0081] In step 1108, a determination is made as to whether the time of the event has already passed the service commit. If yes, then the package is already late, and therefore the risk of the package being late is 1 or 100%, and the post-hoc scoring ends. If the time of the current event has not yet passed the service commit, then in step 1112, a current predicted event fingerprint threshold corresponding to the actual current event streamed from the logistics network 100 is obtained and applied to the time of the event. The amount of delay of the actual event relative to the current predicted event cutoff time indicated by the threshold is determined, and in step 1114, a determination is made as to whether there is a delay relative to the threshold cutoff time.

[0082] If there is no delay exceeding the threshold, the package is on track for delivery by service commitment. Thus, a risk score of 0 is assigned in step 116. If there is an amount of delay between the occurrence of the actual current event versus the threshold of the matching current predicted event from the package fingerprint, then in step 1118, the level of risk is retrieved from the risk table and this level of risk is assigned as the risk score for the package. After risk scoring is completed in either step 1116 or 1118, in step 1120, proactive scoring is scheduled to activate for the next predicted event for the package fingerprint. The reactive scoring process is then completed for this current event and reactive scoring goes into an idle state until the next event is streamed from the location in the logistics network 100.

[0083] One scenario that may occur during reactive scoring is that the actual current event that occurred in the logistics network 100 may not match the current predicted event from the luggage fingerprint. In other words, the transportation plan may have sent the luggage to a different facility than predicted by the luggage fingerprint. This may be the result of normal business operations such as load balancing along the route of the logistics network and / or the application of a pre-shipment risk analysis that caused an unexpected route through the logistics network to be selected. However, when a current event occurs in the network, the reactive scoring obtains a threshold value associated with the actual current event from the luggage fingerprint information available for the logistics network 100, and then proceeds from that current event so that the reactive and proactive scoring can continue based on the next predicted event and the current predicted event from the actual current event until the final destination is reached. Since the next predicted event may later differ from the next predicted event that was based on the location of the current predicted event based on the actual current location, the fingerprint may differ from what was initially predicted by the luggage fingerprint, but the process continues with the reactive and proactive scoring based on the predicted event and threshold value of the new luggage fingerprint.

[0084] FIG. 12 is a flow diagram illustrating example steps 1200 that may be followed when executing a proactive scoring function of the second processing system 218 scheduled during the reactive scoring method of FIG. 11. The method starts periodically according to a schedule and begins at step 1202 where the next predicted event of the package fingerprint is evaluated. This evaluation compares the current time to the service commit time and a threshold value specified by the package fingerprint for this next predicted event to determine the amount of delay, if any, that exceeds the threshold. Decision step 1204 determines whether the current time is past the service commit time. If yes, the package is already late, so the risk of the package being late is assigned as 1 or 100% and this instance of proactive scoring for the next predicted event ends. If the time of the current event has not yet passed the service commit time, the amount of delay of the actual event relative to the cutoff time of the next predicted event as indicated by the previously determined threshold value is considered, and in step 1208 it is determined whether there is a delay relative to the threshold cutoff time.

[0085] If there is no delay, the package is currently no higher risk at this time than when the most recent reactive scoring was performed. Thus, the risk score is not changed and this instance of proactive scoring for the next predictive event is terminated. If there is indeed a delay to the threshold, in this example, in step 1210, the scheduled proactive event is cleared for this package since a new proactive schedule is created instead. Then, in step 1212, the risk level of this next predictive event at this amount of delay exceeding the threshold is found in the risk table and assigned to the package. Then, a new proactive scoring action for the next predictive event may be scheduled to start at a new time in the future. For example, a new schedule of proactive scoring for the package may be scheduled to start one hour later, which corresponds to an increment in the amount of delay in the risk table, e.g., a one-hour increment. The amount of time before the next proactive scoring is initiated is a balance of the amount of processing power required for proactive scheduling versus the value of the particular level of granularity for the proactive scoring update. Then, this instance of proactive scoring is terminated.

[0086] FIG. 13 is a flow diagram illustrating an example 1300 of steps that may be followed when performing the risk aggregation function of the second processing system 218. In some embodiments, it may be desirable to use the pre-emptive scoring only for the purpose of determining whether the risk of a package being delivered after a specified time is worse than the risk from the most recent post-emptive scoring, and record that worse risk for the package. Thus, if the pre-emptive scoring of the next predicted event results in a lower risk score than that assigned by the most pre-emptive scoring, the higher risk in the most pre-emptive scoring is maintained as the risk level of the package. Thus, in such an embodiment, only the actual occurrence of an event in the network that is post-emptively scored may reduce the risk level assigned to the package. Example 1300 illustrates this implementation.

[0087] The example 1300 begins at step 1302 where a proactive scoring of the next predicted event of the package fingerprint is performed to determine a proactive scored risk level. This proactive scored risk level, i.e., the new risk, is compared to the most recent reactive scored risk level for the package, i.e., the old risk, and a determination is made as to whether the new risk is greater than the old risk. If the new risk is not greater than the old risk, the old risk found from the most recent reactive scoring is maintained as the level of risk for the package at step 1308. If the new risk is greater than the old risk, the new risk found from the most recent reactive scoring is maintained as the level of risk for the package at step 1306. The example risk aggregator 1300 then ends until the next proactive scoring for the package occurs.

[0088] FIG. 14 is a flow diagram illustrating example steps 1400 that the fourth processing system 226 may follow to reroute a parcel via changes to the transportation plan implemented by the logistics network 100 in response to the level of risk determined by the second processing system 218. The second processing system 218 may output live parcel risk data in transit as a stream directed to and received by the fourth processing system 226 in step 1402. The fourth processing system 226 may then determine an alternative transportation plan from the parcel's current location to its final destination with a risk estimated to be lower than the current risk level in step 1404. For example, this step may include implementing an alternative transportation plan in step 1406 in which the parcel is directed to an atypical next event location that is different from the next predicted event location of the fingerprint threshold, which then triggers a new assessment of the parcel's risk that is likely different from the previous one. This can be even more effective when used in combination with pre-shipment risk determinations from external risk analysis, such as whether weather or traffic to a selected atypical next location is predicted to be a lower risk than the risk posed by weather or traffic to a more typical next location. Transportation planning and external risk analysis are described below in connection with Figures 15 and 16.

[0089] FIG. 15 illustrates an example 1500 of actual package events 1502 and associated risk levels 1504 of a package being delivered after a prescribed time that occur during the package's transportation. A package arrives at a location INDH, generating a risk score of 0. However, within a station at the INDH location, an exception occurs due to weather risk, which results in a 73% chance of the package being delivered after the prescribed committed time. However, due to the high level of risk, the package's transportation plan is changed and the package is rerouted, resulting in an updated risk prediction of 0 since the new route does not encounter weather risk. Additionally, the package fingerprint thresholds that apply to packages departing from an INDH location and bound for a MICA final destination at a given committed time and service level are not violated at the time of the departure event from INDH.

[0090] Along the route, the departure event from the MEMH location is 9 minutes later than the fingerprint threshold for this event (5:09AM vs. 5:00AM), which results in a post-event risk score of 34% based on the risk table. However, the risk score of 34% is considered low risk because the package is more likely than not to reach its final location by the prescribed commit time of 10:30AM. The package then arrives at the next event location MSPA 56 minutes later than the fingerprint threshold for this event (7:41AM vs. 6:45AM), but the risk table only indicates a risk score of 7%, which places it in the no risk category because the package is very likely to reach its final destination by the prescribed commit time of 10:30AM. This causes the risk to continue to fluctuate, but does not exceed the low risk category, and the package is delivered by the prescribed commit time of 10:30AM.

[0091] FIG. 16 is a flow diagram illustrating an example 1600 of steps that the fourth processing system 226 may follow to consider pre-shipment analysis using external factors and associated risks. In step 1602, the start location or origin of the package, start time, final destination, and product / service type are obtained. Then, in step 1604, the expected path and timing of the product moving through the network 100 along the expected path is calculated using the predicted events in the package fingerprint to move such package to the final location. In step 1606, based on the predicted location and time calculated from the package fingerprint, external factors and associated risk information 1608 is obtained so that it can be identified whether the package will pass through harsh conditions such as weather or traffic, and this information is then converted into an ambient risk score.

[0092] This risk score may then be used by the transportation planning system 226 to make a pre-shipment decision about the transportation plan and whether a different plan has a lower risk. An example was described above in connection with FIG. 15, where a risk level of 73% was caused by weather. Similarly, this example of a pre-shipment determination of risk 1600 may be applied by the second processing system 218 in real time at each location visited by the physical load along the route to determine the surrounding risk that the transportation planning system 226 may consider when deciding to re-route the load from the visited locations.

[0093] The risk information that has been determined for each package in transit within logistics network 100 can be extremely useful to both the entities operating logistics network 100 and the entities expecting to receive the shipment. This is especially true for packages that have special handling requirements, such as temperature control, such as certain vaccine and other medical related packages, and / or packages that have extreme time sensitivity, such as certain vaccine or medical related packages. Thus, the risk information may be provided in a display format to various stakeholders to enable them to monitor packages of interest. Furthermore, this allows personnel of the shipping entities to monitor for failure points in the network that cause delays that impact their ability to deliver packages by the prescribed time.

[0094] FIG. 17 illustrates an example 1700 of a screen capture of such a user display provided by the application 222, 224 of FIG. 2. The example 1700 allows the user to apply various filters to control the information displayed, such as a filter 1702 for payer versus shipper, a filter 1704 for selecting the payer or shipper account selected in the filter 1702, and the service type of the package, such as ground service or express service. Depending on the filter selected, the example 1700 further displays the total number of packages in transit in field 1708, the total number of packages that are at a sufficient level of risk of interest in field 1710, and the number of packages that have experienced an exception, such as a missed event due to weather or other atypical factors in field 1712. The example 1700 further includes a map 1714 showing the number of packages at risk of being delivered after a specified time to locations within a given geographic area, such as within each state in the United States.

[0095] Providing this information in example 1700 allows a person interested in monitoring parcels to focus on parcels with a sufficient level of risk that requires attention, for example, greater than 50% risk. Thus, rather than monitoring all 446 parcels in transit, the person may choose to focus on the 283 that have a sufficient level of risk. These parcels may be identified by the user selecting field 1710 to view all 283 or by selecting a geographic area from map 1714 and focusing on parcels at risk for that selected area. Additionally, the user may focus on any parcels at risk by selecting to view more information, and in response to receiving the user's selection, the third processing system may then display a list of parcels at risk and the user may select a parcel of interest from the list. Once the third processing system receives a selection of a particular parcel from the list, the third processing system may display parcel-specific information.

[0096] It will be understood that packages for a given customer, such as a shipper or recipient, have an account for which the shipper has. Each package may be tied to that account such that the third processing system provides information to the shipper account to display only risk information related to packages for that shipper account. Additionally, packages for a given shipper account may have different package-specific final locations, and events that occur to the packages during transit may be package-specific events. Thus, in the example of FIG. 17, the 446 packages for the shipper account may be destined for hundreds of different final locations. Additionally, these packages may be traveling independently, so even two packages with the same origin and final destination following the same transit route may be at different points along the transit route and therefore may experience separate package-specific events that result in separate determinations of delay and risk. Example 1700 shows a screen capture taken at a first display time. The user may then choose to view the display at a second display time, where the numbers may have changed because additional scoring has occurred for these packages. For example, the number of packages at risk may be greater or less than the 283 packages at risk at the first display time shown in FIG.

[0097] Figure 18 shows another example 1800 of a screen capture of a user display provided by the applications 222, 224 of Figure 2. This example 1800 provides information 1802 including total number of packages / shipments for a given criteria, such as a particular customer or a particular location, number of packages at risk of delay, number without a commit date, number on schedule, number near the commit date, and number past the commit date. This allows personnel at the shipping entity to quickly view a snapshot of a given customer.

[0098] The bar graph in example 1800 also allows for a quick view of a snapshot of the status of shipments in chart 1804, including the number of packages that are late 1806. A lateness status chart 1808 shows the number of packages in each classification of risk that are delivered past their committed time, including the number of packages at a sufficient level of risk 1810, such as at least 80%, that are considered likely to be late, the number of packages at low risk, such as less than 40%, and the number of packages at medium risk, such as between 40% and 80%, that are considered 1814. Other information is also shown, such as a chart 1816 showing the last known location of the packages. Thus, personnel at the shipping entity can gather a snapshot of information that provides insight into the operation of the network and the status of shipments for a given customer.

[0099] 19 shows another example 1900 of a screen capture of a user display provided by the application 222, 224 of FIG. 2. This example 1900 is particularly suited for personnel of a shipping entity attempting to find fault points in the logistics network 100 that cause delays that pose a risk of a package being delivered after the prescribed committed time. The example 1900 includes a heat map 1902 that represents locations in the logistics network 100 where delays are occurring based on the number of violated fingerprint thresholds found during post-mortem and pre-mortem scoring. The delays may also be shown distributed within days of the week as in chart 1904.

[0100] A set of filters 1906 may also be available for selection to allow the user to filter the information as needed and drill down into the data in various ways, taking into account that the risk data is stored in association with all other parameters of the package including customer, origin, current location, final location, service type, prescribed commit time, etc. Filters may be applied individually or in combination. A first field 1908 provides for selection of a customer name (i.e., shipper) to filter the information provided for display. A second field 1910 provides for selection of an EAN code to filter the information provided for display. A third field 1912 provides for selection of a facility name to filter the information provided for display. A fourth field 1916 provides for selection of a facility name to filter the information provided for display. A fifth field 1918 provides for selection of a delivery date to filter the information provided for display. A sixth field 1920 provides for selection of a number of service delay days to filter the information provided for display. A seventh field 1922 provides for selection of the scan type that resulted in the event data to filter the information provided for display, or the event type described above. An eighth field 1924 provides a facility type selection for filtering the provided information for display. A ninth field 1926 provides an origin location selection for filtering the provided information for display, and a tenth field 1928 provides a destination location selection for filtering the provided information for display, with both fields 1926 and 1928 allowing the user to specify an origin / destination combination.

[0101] In addition to the heat map 1902 and chart 1904, a table of information 1930 is displayed to provide specific details to the user. For example, the shipments resulting from the application of the filters appearing in table 1930 show the event type in column 1932, the event timestamp in column 1934, and the corresponding fingerprint threshold timestamp in column 1936. In this manner, it is possible to view the raw data that generates the risk of delay and filter them as needed by facility, customer, etc. Other charts, such as donut charts 1938, 1940, and 1942, may be taught to further show the relative contribution to delay by scan type, facility type, origin / destination pair, etc. Learning points of failure by facility, facility type, scan type, customer, origin / destination pair, etc., allows the user to consider corrective actions to improve the operation of the logistics network 100.

[0102] Various examples and embodiments are described above for creating and applying network models in the form of package fingerprints to provide the ability to determine late delivery risks and exceptions for packages in transit, and to enable monitoring of such packages at risk and the occurrence of late delivery risks and exceptions. Improvements to computer systems related to logistics networks include the ability to create these thresholds from historical data and then apply these thresholds in real time to enable real-time monitoring of late delivery risks.

[0103] In summary, it will be appreciated by those skilled in the art that the sequences of operations for carrying out any of the methods and method variations described in the embodiments herein are merely exemplary, and it is emphasized that various sequences of operations may be followed as legally possible and in accordance with the principles of the invention.

[0104] At least some portions of the exemplary embodiments outlined above may be used in conjunction with other exemplary embodiments to enhance and improve logistics operations and related monitoring of packages at risk of delivery delay. For example, risk based on reactive and proactive scoring may be used in conjunction with pre-shipment risk analysis. Similarly, either or both of these risk assessments may be used solely for monitoring and intelligence personnel or may be used to create a closed loop with package transportation planning. Furthermore, at least some of the exemplary embodiments disclosed herein may be used independently and / or in combination with each other and may have application in apparatus and methods not disclosed herein. However, those skilled in the art will appreciate that the exemplary model training, model scoring, and related outputs described above provide enhancements and improvements to techniques used in logistics and shipment management.

[0105] Those skilled in the art will appreciate that an embodiment may provide one or more advantages, and not all embodiments necessarily provide all or more of the specific advantages described herein. Moreover, it will be apparent to those skilled in the art that various modifications and variations can be made to the structures and methods described herein. It is therefore to be understood that the invention is not limited to what has been described herein. Rather, the invention encompasses modifications and variations, as set forth in the following claims.

Claims

1. 1. A computer system for monitoring the transportation of packages in a logistics network, comprising: an interface to at least one logistics network data source; a first data storage module including historical network data coupled to the interface, the first data storage module constructing the historical network data from data feeds from the logistics network data sources; a second data storage module containing cargo transport model data; a first processing system that accesses a data storage device and applies machine learning to the historical network data to train the cargo transportation model data; Capturing data-driven patterns including likely locations of historical events, likely types of historical events, and likely times of historical events; and generating from the data-driven pattern a set of predicted events for a virtual package being transported through the logistics network from a first location to a final location; and determining thresholds for each of the predicted events such that the virtual package will arrive at the final location by a specified time, each threshold being associated with the expected time of a predicted event occurring for the virtual package, and the predicted events being associated with an expected location and an expected type. a first processing system for performing a second processing system coupled to the interface for receiving live data from the logistics network, the second processing system accessing the transportation model data from the second data storage module, applying the thresholds to the predicted events for the actual shipments, and evaluating whether the actual shipments will be delivered as scheduled to the final location by the specified time for each predicted event based on the application of the thresholds; A computer system comprising:

2. The computer system of claim 1 , wherein the first processing system further calculates a table of risk amounts associated with delay amounts exceeding the threshold in any of the predicted events for the virtual load to reach the final location by the specified time.

3. The computer system of claim 2, wherein the second processing system determines a level of risk that the actual package will not be delivered to the final location by the specified time by accessing a table of risk amounts to determine the risk amount associated with an amount of delay exceeding the threshold for a current event detected by the second processing system.

4. 4. The computer system of claim 3, wherein the second processing system further determines a level of risk that the actual package will not be delivered to the final location by the specified time by accessing a table of risk amounts to determine the risk amount associated with the amount of delay exceeding the threshold for a next predicted event that has not yet been detected by the second processing system.

5. 4. The computer system of claim 3, further comprising a third processing system that receives the level of risk associated with the physical shipment and, upon receiving a request, generates a display indicating the level of risk for the physical shipment.

6. a fourth processing module for receiving the level of risk associated with the physical shipment; 4. The computer system of claim 3, wherein when the level of risk that the actual load will not be delivered to the final location by the specified time reaches a rerouting threshold, the fourth processing module determines a next event for the actual load that is different from a next event planned during transportation to the final location that is expected to reduce the level of risk.

7. 4. The computer system of claim 3, wherein the second processing system further receives external factor information, and the second processing system determines a future risk to the actual baggage reaching the location by the specified time based on the external factor information related to the predicted event expected to occur for the actual baggage during transportation.

8. The computer system of claim 7 , wherein the external factor information includes weather and traffic information related to the predicted event.

9. The computer system of claim 3 , wherein the current event includes the physical package being scanned at a specified location within the logistics network.

10. 10. The computer system of claim 9, wherein the current event further comprises an arrival scan at the specified location in the logistics network.

11. The computer system of claim 9 , wherein the current event further comprises a departure scan from the specified location within the logistics network.

12. the first processing system determines the threshold for each predicted event by determining, for each predicted event that is a current predicted event, a time that a desired percentile of packages historically arrive on time; The computer system of claim 1 , wherein the threshold is set based on the time for reactive monitoring of the current predictive event for the physical load.

13. 13. The computer system of claim 12, wherein the desired percentile of historically on-time arriving packages for the current forecast event is a 99th percentile.

14. 13. The computer system of claim 12, wherein the first processing system further determines the threshold for each predicted event by determining, for each current predicted event, a next predicted event that is likely to follow the current predicted event by selecting the next predicted event location having a most recent threshold from a set of potential next event locations, wherein the threshold for each potential next event location for determining the most recent threshold is determined by determining, for each potential next event, a time that a desired percentile of packages moving from the current event location to the potential next event location would historically arrive on time, and wherein the most recent threshold identifies a location and type of the next predicted event for purposes of proactively monitoring the next predicted event for the real package.

15. For a first shipper account having a plurality of packages being transported within the logistics network, at least a portion of the plurality of packages having a first shipper location and a package-specific final location, while each package is being transported within the logistics network, a second processing system captures event data for each package at each package-specific current event, retroactively compares the threshold value from a corresponding current predicted event for each package with an occurrence time of the package-specific current event to determine from a table of risks a level of risk that the package will not be delivered to the final location by a package-specific specified time, and proactively compares the threshold value from a corresponding next predicted event for each package with a current time, 3. The computer system of claim 2, further comprising a third processing system that receives output from the second processing system, the third processing system providing information for display to the first shipper account at the first display time including a total number of packages among the plurality of packages that are at least at a particular risk level of not reaching the package-specific final location by the package-specific specified time as of the first display time.

16. While each package is being transported within the logistics network, the second processing system continuously captures event data for each package at each package-specific current event, retroactively compares the threshold value from the corresponding current predicted event for each package with the occurrence time of the package-specific current event to determine from the risk table the level of risk of not being delivered to the final location by the package-specific specified time, and proactively compares the threshold value from the corresponding next predicted event for each package with the current time; The computer system of claim 15, wherein the third processing system provides, for the first shipper account, information for display at a second display time following the first display time, including a total number of packages among the plurality of packages that are at least at the risk level of not reaching the package-specific final location by the package-specific specified time.

17. For the first shipper account, the third processing system provides information for display including information regarding packages at least at the specified risk level that will not reach the package-specific final location by the package-specific specified time; 16. The computer system of claim 15, wherein the third processing system receives a selection of one of the packages being displayed and provides information for display specific to the selected package.

18. 3. The computer system of claim 2, further comprising a third processing system that receives output from the second processing system, the third processing system providing information for display at a first display time including aggregate information regarding packages that are at least at a particular risk level of not reaching a package-specific final location.

19. 20. The computer system of claim 18, wherein the logistics network includes a plurality of facilities, and the third processing system provides for selection of a particular facility to filter information for a provided display.

20. 20. The computer system of claim 18, wherein the third processing system provides for selection of specific event types to filter information for a provided display.

21. Each facility has a specific type, 20. The computer system of claim 19, wherein the third processing system provides for selection of a particular facility type to filter information for a provided display.

22. 20. The computer system of claim 18, wherein the third processing system provides a selection of package origin to filter information provided for display.

23. 20. The computer system of claim 18, wherein the third processing system provides a selection of a package's last location to filter information for display provided.

24. 20. The computer system of claim 18, wherein the third processing system provides for selection of the specified time of delivery of the package to filter information for display provided.

25. 20. The computer system of claim 18, wherein the third processing system provides a selection of a package's last location to filter information for display provided.

26. 20. The computer system of claim 18, wherein the third processing system provides a package sender selection for filtering information provided for display.

27. The computer system of claim 18, wherein the third processing system provides information for display including the number of actual packages at least at the particular level of risk grouped by each day of the week on which a predictive event threshold underlying the particular level of risk for the actual package occurs.

28. 1. A computer-implemented method for monitoring transportation of packages in a logistics network, comprising: applying machine learning to historical network data for the logistics network to train parcel transportation model data; Capturing data-driven patterns including likely locations of historical events, likely types of historical events, and likely times of historical events; and generating from the data-driven pattern a set of predicted events for a virtual package being transported through the logistics network from a first location to a final location; and determining thresholds for each of the predicted events for the virtual package to arrive at the final location by a specified time, each threshold being associated with the expected time of a predicted event occurring for the virtual package, the predicted events being associated with an expected location and an expected type. and applying the threshold to the predicted events for a physical shipment; evaluating, for each predicted event, whether the actual baggage will arrive at the final location as scheduled by the specified time based on application of the threshold; further calculating a table of risk amounts associated with delay amounts exceeding the threshold in any of the predicted events for the virtual load to reach the final location by the specified time; determining a level of risk that the physical package will not be delivered to the final location by the specified time by accessing a table of risk amounts to determine the risk amount associated with a delay amount exceeding the threshold for the current event detected; determining a level of risk that the physical package will not be delivered to the final location by the specified time by accessing a table of risk amounts to determine the risk amount associated with the amount of delay exceeding the threshold for a next predicted event that has not yet been detected; A computer-implemented method comprising:

29. For a first shipper account having a plurality of packages being transported within the logistics network, at least a portion of the plurality of packages having a first shipper location and a package-specific final location, while each package is being transported within the logistics network, capturing event data for each package at each package-specific current event, retroactively comparing the threshold value from a corresponding current predicted event for each package with an occurrence time of the package-specific current event to determine from the table of risk amounts the level of risk of not being delivered to the final location by the package-specific specified time, and proactively comparing the threshold value from a corresponding next predicted event for each package with a current time to further determine from the table of risk the level of risk of not being delivered to the final location by the package-specific specified time; The computer-implemented method comprises: providing information for display at a first display time including a total number of packages among the plurality of packages that are at least at a particular risk level of not reaching the package-specific final location by a package-specific defined time at the first display time.

30. The computer-implemented method of claim 28, further comprising:

30. 30. The computer-implemented method of claim 29, providing display information for the first shipper account including information regarding package(s) at least at the particular risk level that will not reach the package-specific final location by the package-specific specified time, receiving a selection of one of the displayed packages, and providing display information specific to the selected package.

Citation Information

Patent Citations

  • Delivery control method and device therefor, and delivery information service method

    JP2002042005A

  • Traffic information providing apparatus and traffic information providing method

    JP2006039978A

  • Continuous delivery

    JP2020098573A

  • Systems and methods for estimating arrival times

    JP2020502601A

  • Mobile device-based system and method for self-directed flexible delivery task allocation

    JP2021513130A