Method for characterizing flows between elements of a network
Patent Information
- Application Number
- EP2022839857
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2021-12-29
- Filing Date
- 2022-12-28
- Publication Date
- 2024-11-06
AI Technical Summary
Current systems for characterizing flows in dynamic networks face challenges in managing massive and evolving data, scalability during changes in network behavior, and handling unforeseen situations, particularly due to their unidirectional operation and reliance on historical learning databases.
A method and device for characterizing flows between network elements, involving a configuration phase with user-defined constraints, dynamic tracking, and grouping tools that utilize an expectation-maximization algorithm to adaptively group network elements based on flow data and user input, facilitating real-time analysis and alert generation.
Enables effective characterization and management of network flows by reducing cognitive load, improving scalability, and handling unforeseen situations through interactive human-centered and automated analysis, enhancing trust and reducing biases in decision-making.
Smart Images

Figure 1.1
Abstract
Description
[0001] DESCRIPTION
[0002] TITLE: Method for characterizing flows between elements of a network
[0003] The present invention relates to a method for characterizing flows between elements of a network. The present invention also relates to an associated electronic characterization device.
[0004] Faced with a situation that evolves dynamically over time, the operators in charge of flow control on a network have a role of supervision and analysis of unusual phenomena. Such unusual phenomena include the increase in demand or saturation for a given route, the decrease in transport demand for a station or for a given route, failures or accidents of mobility supports (trains, metros, buses, bicycles, etc.) or even passenger accidents or incidents. This role involves taking into account massive and evolving / dynamic data, as well as their analysis with regard to business knowledge on the structure of the network, the history of the behavior of agents (users) in this network, external knowledge on the phenomena influencing the network (location, attributes, one-off events, etc.), as well as taking into account possible means of action to resolve or optimize the situation.
[0005] This operational task is difficult to manage from simple passenger flow records, which are far too massive to be analyzed as such by operators. Therefore, tools have been developed to reduce the cognitive load of operators, by synthesizing the mass of information using analytical tools (i.e., which take into account the data and process it) and synthetic tools (i.e., which summarize the data and the results of the analysis).
[0006] To be usable, these tools are largely automated at the data retrieval and processing level. Most of the proposed systems operate in a unidirectional mode: data is grouped, analyzed, then a decision is made or provided in the form of a summary to an operator. The systems are therefore configured entirely upstream, or have a very high learning capacity, which is difficult to achieve in practice, particularly in the case of disruptive events (discontinuities), or rare events (no historical examples).
[0007] Expert systems are known from the state of the art, designed from human expertise formalized upstream of the system's operation. However, the design method of such systems raises questions of scalability during changes in behavior, availability of the expert and design effort, but also the question of bias introduced by the expert, insofar as the system only replicates what is described to it.
[0008] Other classical supervised approaches rely on learning from a learning base. While these systems can be effective in recognizing situations of interest already observed in the history of the network, and therefore “filter” events to reduce the operator’s load, they nevertheless present the need for a historical learning base, but also for references of normality or abnormality to learn from (labeling, tagging). We thus find the problem of managing the evolution of the system during fundamental changes in network behavior (for example, how to manage a new station created on the network), but also the ability to manage unforeseen or never-before-observed situations.
[0009] There is therefore a need for a means to facilitate the characterization of flows on a network.
[0010] To this end, the subject of the invention is a method for characterizing flows between elements of a network, the method being implemented by an electronic characterization device and comprising a configuration phase comprising the following steps: a. receiving data relating to flows between the elements of the network, called flow data, b. displaying, via a user interface, a representation of the flow data, c. receiving, via the user interface, at least one user constraint relating to the grouping of the elements of the network, each user constraint being chosen by a user as a function of the displayed representation of the flow data, and d.grouping, by a grouping tool, the elements of the network into groups according to the flow data and user constraints, the groups of elements obtained making it possible to characterize the flows between the elements of the network, the elements being assimilated, for the grouping, to nodes of a graph and the flows between the elements to edges of the graph, each group being such that the nodes corresponding to the elements of the group connect to each other in a similar way and connect in a dissimilar way to the nodes corresponding to the elements of the other groups. According to particular modes of implementation, the method comprises one or more of the following characteristics, taken in isolation or according to all technically possible combinations:.
[0011] - the configuration phase includes, after the grouping step, a display step, via the user interface, of a representation of the groups of elements obtained;
[0012] - the method comprises a dynamic tracking phase comprising the following steps for each instant: a. receiving the flow data of the current instant, b. displaying, via a user interface, a representation of the flow data of the current instant, c. possibly updating, via the user interface, the user constraints so that the user constraints of the current instant are the updated user constraints or the user constraints of the previous instant otherwise, and d. possibly modifying, by the grouping tool, the groups of elements of the previous instant according to the flow data of the current instant and the user constraints of the current instant;
[0013] - the dynamic tracking phase also includes for each instant a display step, via the user interface, of a representation of the groups of elements obtained for the current instant;
[0014] - during each stage of displaying a representation of the groups of elements obtained, an indicator is also displayed relating to the consideration of user constraints in the groupings carried out;
[0015] - during the stage of displaying a representation of the groups of elements obtained for the current moment, any modifications to the groups of elements are highlighted on the displayed representation so as to facilitate the management of flows between the elements of the network;
[0016] - the dynamic monitoring phase includes a step of generating an alert and / or implementing an action when the groups of elements of the current instant are different from the groups of elements of the previous instant;
[0017] - the or each user constraint is chosen from: a. a constraint relating to an imposed grouping between elements of the network, and b. a constraint relating to a prohibited grouping between elements of the network; - the grouping tool implements an expectation-maximization type algorithm.
[0018] The present invention also relates to an electronic device for characterizing flows between elements of a network, the device being capable of implementing a method as described previously and comprising: a. a user interface, b. a grouping tool, and c. a tool for receiving flow data and communicating a representation of the flow data, on the one hand, to the user interface and, on the other hand, to the grouping tool.
[0019] Other features and advantages of the invention will become apparent upon reading the following description of embodiments of the invention, given by way of example only and with reference to the drawings which are:
[0020] [Fig 1], Figure 1, a schematic representation of an electronic device for characterizing flows between elements of a network,
[0021] [Fig 2] Figure 2, a flowchart of an example of implementation of a method for characterizing flows between elements of a network, and
[0022] [Fig 3] Figure 3, a schematic representation of user constraints communicated via a user interface on which elements of a network are represented.
[0023] An electronic device 10 for characterizing flux between elements E of a network R is illustrated in FIG. 1.
[0024] The network R groups together all the elements E capable of exchanging flows between them. The elements E are physical entities (objects, installations, etc.) or individuals or animals.
[0025] For example, network R is a transport network and elements E are stations of network R. In this case, flows are the movements of the transport considered between stations. The transport network is, for example, a rail network (train), a road network (car, bus, bicycle) or an urban network (pedestrians, scooters). In one example, the stations are self-service rental stations (bicycles, cars, scooters), for which the operator's objective is to replenish the stations according to traffic, or to detect anomalies in the rental profiles, for example in the event of a breakdown.
[0026] In another example, the network R is a communication network or a social network. The elements E are then the users of the network R and the flows between the elements E are, for example, the number of interactions between pairs of users over a given time interval.
[0027] The device 10 is configured to implement a method for characterizing flow between the elements E of a network R, which will be described in the remainder of the description.
[0028] As illustrated in FIG. 1, the device 10 comprises a user interface 12, a grouping tool 14, and a tool 16 for receiving the flow data.
[0029] The user interface 12 is capable of displaying information transmitted by the grouping tool 14 and by the receiving tool 16. The user interface 12 is also capable of receiving information from the user for transmission at least to the grouping tool 14.
[0030] The user interface 12 is a human-machine interface. The user interface 12 comprises, for example, a display such as a touch screen or not, and possibly a keyboard and / or a mouse.
[0031] The grouping tool 14 is suitable for determining groupings of the elements E of the network R, as will be described in the remainder of the description.
[0032] The reception tool 16 is capable of receiving and communicating a representation of the flow data, on the one hand, to the user interface 12 and, on the other hand, to the grouping tool 14, as will be described in the remainder of the description.
[0033] The grouping tool 14 and the receiving tool 16 are, for example, software modules, such as computer program products comprising program instructions for causing the respective operation of the grouping tool 14 and the receiving tool 16 when such computer programs are executed by a computer.
[0034] For example, the user interface 12, the grouping tool 14 and the reception tool 16 are integrated in the same computer. In this case, the computer programs are for example stored in memories of the computer.
[0035] Alternatively, each of the grouping tool 14 and the reception tool 16 are integrated into a calculator of their own, and thus form two separate entities.
[0036] As a further variant, the computer program products corresponding to the grouping tool 14 and the receiving tool 16 are stored on the same information medium or on their own information mediums. The or each readable information medium is a medium suitable for storing electronic instructions and capable of being coupled to a bus of a computer system. For example, the or each information medium is a USB key, a floppy disk or flexible disk (also known as a "Floppy disk"), an optical disk, a CD-ROM, a magneto-optical disk, a ROM memory, a RAM memory, an EPROM memory, an EEPROM memory, a magnetic card or an optical card.
[0037] A method for characterizing flow between the elements E of a network R implemented by the device 10 will now be described with reference to the flowchart of figure 2 and to the examples of figures 3 and 4 which illustrate certain steps of the method.
[0038] The characterization method comprises a configuration phase 100 and, optionally, a dynamic tracking phase 200.
[0039] Configuration phase 100 corresponds to the first reception of user constraints as will be described below. Configuration phase 100 thus corresponds to a single instant. Configuration phase 100 alone corresponds to a flow characterization at a given instant.
[0040] When implemented, the dynamic tracking phase 200 follows the configuration phase 100 and comprises steps repeated over time. The dynamic tracking phase 200 is thus implemented in real time.
[0041] The configuration phase 100 comprises a step 110 of receiving data relating to flows between the elements E of the network R, called flow data. Step 110 is, for example, implemented by the reception tool 16.
[0042] Flow data are, for example, derived from measurements made on the network R by sensors. Flow data are, for example, a count of interactions or movements between the elements E of the network R.
[0043] In step 110, the received flow data is optionally processed and communicated, if necessary in the form of an adapted representation, to each of the user interface 12 and the grouping tool 14. For example, the representation communicated to the user interface 12 has been adapted to be understandable by a user (see FIG. 3 for example) while the grouping tool 14 receives the raw flow data.
[0044] The configuration phase 100 comprises a step 120 of displaying, via the user interface 12, a representation of the flow data. The displayed representation is thus suitable for being viewed by a user of the device 10.
[0045] The configuration phase 100 comprises a step 130 of receiving, via the user interface 12, at least one user constraint relating to the grouping of the elements E of the network R. Each user constraint is chosen by a user according to the displayed representation of the flow data. Each user constraint is communicated to the grouping tool 14.
[0046] Preferably, the or each user constraint relates to similarities or dissimilarities between the elements E of the network R. Preferably, the or each user constraint is chosen from:
[0047] - a constraint relating to a grouping imposed between elements E of the network R, and
[0048] - a constraint relating to a prohibited grouping between elements E of the network R.
[0049] Preferably, user constraints are chosen by the user based on his own knowledge about the functioning of the network R, but also based on available external knowledge, such as incident reports, particular events that would lead to saturations or closures of certain nodes of the network R.
[0050] In an example implementation, user constraints are already predefined, and the user adapts these constraints (for example via a slider) specifically for the R network considered.
[0051] An example of user constraints for a transport network R (metro) is illustrated in Figure 3. In this example, the user interface 12 allows the user to visually indicate the grouping constraints for the algorithm. In this example, the operator can easily propose groupings of stations with identical behavior (stations surrounded by solid lines), or conversely indicate a desire for separation (stations surrounded by dotted lines).
[0052] The configuration phase 100 comprises a step 140 of grouping, by the grouping tool 14, the elements E of the network R into groups according to the flow data and the user constraints. The groups of elements E obtained make it possible to characterize the flows between the elements E of the network R.
[0053] In particular, the elements E are assimilated, for grouping, to nodes of a graph and the flows between the elements E to edges of the graph. The graph is dynamic, that is to say that its edges evolve over time, from the evolution of the flow data. Such a graph is also discretized in time, to be able to aggregate individual transactions in the form of an interval count. Such a graph is potentially bipartite, in the case where the nodes of the network R are divided into two differentiated groups (for example, a group of entry stations, and a group of exit stations).
[0054] Each group is such that the nodes corresponding to the elements E of the group connect to each other in a similar way and connect in a dissimilar way to the nodes corresponding to the elements E of the other groups. By the term "similar" it is understood that the connections are close but not necessarily strictly identical. In an example implementation, this discretized dynamic graph is represented in the form of a sequence of weighted adjacency matrices. An adjacency matrix for a finite graph with n vertices is an nxn matrix whose non-diagonal element ay is the number of edges connecting vertex i to vertex j. The diagonal element an is the number of loops at vertex i (for simple graphs, this number is therefore equal to 0 or 1).Thus, in this example, the inputs to clustering tool 14 are adjacency matrices and the outputs of clustering tool 14 are reordered matrices whose rows and columns have been permuted with respect to the input adjacency matrices so as to reveal groups.
[0055] In this example, the nodes of the dynamic graph are the nodes of the network R, and the edges of the dynamic graph at each instant correspond to an aggregation of the interactions between two nodes over a given period of time. In a transport network R, for example, the nodes of the dynamic graph are the stations of the network R, and the edges at a given instant correspond to the number of trips made between two given stations over a fixed period of time.
[0056] For example, the clustering tool 14 implements a parametric mathematical model describing how flows are generated in the network R as well as an associated learning algorithm. These parameters (e.g., level distributions, connectivities, transition dynamics, etc.) are learned by the algorithm from the data.
[0057] For example, clustering tool 14 implements an expectation-maximization algorithm, which allows for user constraints to be taken into account. An expectation-maximization algorithm (often abbreviated EM) is an iterative algorithm that allows for finding the maximum likelihood parameters of a probabilistic model when the latter depends on unobservable latent variables.
[0058] In one example, user constraints are assigned an importance indicator. The user constraints are then taken into account for grouping according to the importance indicator. For example, the grouping tool 14 will ensure that a constraint with a high importance indicator is respected more than a constraint with a low importance indicator.
[0059] Preferably, the configuration phase 100 comprises a step 150 of displaying, via the user interface 12, a representation of the groups of elements E obtained. The displayed representation is thus suitable for being viewed by a user of the device 10. The displayed representation makes it possible in particular to assist in the management of flows between the elements E, for example to detect abnormal situations. Preferably, during each display step 150, an indicator relating to the consideration of user constraints in the groupings carried out is also displayed. Thus, if the constraint(s) are not respected, this makes it possible to indicate possible anomalies in the flow data.
[0060] Thus, the user has two sources of information:
[0061] The actual flows, synthesized in the form of a real-time view of the R network, and
[0062] The synthesis elements E resulting from the automatic analysis: the partition of the nodes of the network R at each instant of the study period of the graph and, where applicable, the parameters of the model estimated by the algorithm on the data.
[0063] As previously indicated, the dynamic tracking phase 200 comprises different steps repeated for each instant (preferably real time).
[0064] The dynamic tracking phase 200 comprises a step 210 of receiving the flow data of the current instant. The reception step 210 is, for example, implemented in a similar manner to the reception step 110.
[0065] The dynamic tracking phase 200 comprises a step 220 of displaying, via the user interface 12, a representation of the flow data of the current instant. The display step 220 is, for example, implemented in a similar manner to the display step 210.
[0066] The dynamic tracking phase 200 comprises a step 230 of possibly updating, via the user interface 12, the user constraints so that the user constraints of the current instant are the updated user constraints or the user constraints of the previous instant otherwise. The update step 230 is, for example, implemented in a similar manner to the reception step 130.
[0067] The update is, for example, due to developments in the R network brought to the user's attention or is linked to the grouping of entities carried out at the previous moment.
[0068] The dynamic tracking phase 200 comprises a step 240 of possible modification, by the grouping tool 14, of the groups of elements E of the previous instant as a function of the flow data of the current instant and the user constraints of the current instant. The modification step 240 is, for example, implemented in a similar manner to the grouping step 140.
[0069] Changes are, for example, due to changes in the data flow and / or changes in user constraints.
[0070] The dynamic tracking phase 200 comprises a step 250 of displaying, via the user interface 12, a representation of the groups of elements E obtained for the current moment. The display step 250 is, for example, implemented in a similar manner to the display step 150.
[0071] Thus, preferably, during each display step 250, an indicator is also displayed relating to the consideration of user constraints in the groupings carried out.
[0072] Preferably, during the display step 250, any modifications to the groups of elements E are highlighted on the displayed representation so as to facilitate the management of flows between the elements E of the network R.
[0073] Optionally, the dynamic tracking phase 200 comprises a step 260 of generating an alert and / or implementing an action when the groups of elements E of the current instant are different from the groups of elements E of the previous instant. Step 260 is, for example, implemented by the user or by the device 10.
[0074] Thus, the present method implements an interaction loop between a human-centered system (user interface) and an automated analysis system (clustering tool), each contributing to the analysis of the other through a common knowledge representation mode, all in a dynamic manner. The consideration of user constraints by the clustering tool 14 thus makes it possible to facilitate the characterization of flows on a network R.
[0075] In particular, the automated analysis subsystem makes it possible to synthesize the flows on the network R, and where appropriate their evolution over time. This synthesis consists in part of a partitioning (clustering) of the nodes of the graph, based on a mathematical modeling of the way in which the flows are generated, as well as on knowledge presented by the operator aimed at making certain partitions of the nodes of the graph less probable than others. The partitioning being dynamic, the groups sought do not necessarily consist of the same nodes over time and one interest of the analysis is to capture the dynamics of these groups over time. The grouping tool 14 then transmits to the user interface 12 a summary of these results.
[0076] Symmetrically, the operator-centric part performs an expert analysis from a graphical representation of the flows synthesized by the automated analysis subsystem, possible at different granularities. In particular, dynamic changes in the composition of the groups are the main indices that can be used by the operator to trigger actions, raise alerts or simply produce reports. For example, it may be desirable to raise an alert when a significant number of nodes in the network R change groups between two successive moments, indicating a significant change in the structure of the flows at a given moment. This phenomenon, however, must be analyzed in relation to the previously observed dynamics of evolution of the nodes in the groups.The results provided by the automatic system to the operator are therefore a series of partitions of the graph nodes for each instant of the dynamic graph as well as the parameters learned by the model, describing in a synthetic way the way in which the groups of nodes interact and evolve over time. These parameters allow a better understanding of the results for the operator.
[0077] Thus, the expert extracts from his analysis indications on current behaviors, and can, if necessary, transmit them in turn to the automaton in the form of possible modifications of the user constraints. In particular, the user constraints describe, for example, spatial, structural or conjunctural similarities between nodes of the network R and are likely to evolve over time depending on the operational situation.
[0078] Thus, this coupling between the grouping tool 14 and the user via the user interface 12, allows: a reduction in biases of the actors (cognitive for the operator, learning for the automaton), and a rapid convergence towards a satisfactory common analysis; a strengthening of the bond of trust of the human towards the automatic analysis; a greater independence from operating conditions and a priori than for a supervised analysis (from a base of examples).
[0079] Those skilled in the art will understand that although the invention has been described in particular in the context of transport networks, it applies to any graph structure evolving over time, representable for example by a series of adjacency matrices, for which the objective is to trace changes in the behavior of the nodes (and relationships between nodes) over time.
[0080] Those skilled in the art will understand that the embodiments and variants previously described may be combined provided that they are technically compatible.
Claims
CLAIMS process for characterizing flows between elements (E) of a network (R), the process being implemented by an electronic characterization device (10) and comprising a configuration phase including the following steps: a. receiving data relating to flows between the elements (E) of the network (R), called flow data, b. displaying, via a user interface (12), a representation of the flow data, c. receiving, via the user interface (12), at least one user constraint relating to the grouping of the elements (E) of the network (R), each user constraint being chosen by a user according to the displayed representation of the flow data, and d.The grouping, by a grouping tool (14), of the elements (E) of the network (R) into groups based on flow data and user constraints, the resulting groups of elements (E) characterizing the flows between the elements (E) of the network (R), the elements (E) being considered, for the purposes of grouping, as nodes of a graph and the flows between the elements (E) as edges of the graph, each group being such that the nodes corresponding to the elements (E) of the group connect to each other in a similar manner and connect dissimilarly to the nodes corresponding to the elements (E) of the other groups. The procedure according to claim 1, wherein the configuration phase comprises, after the grouping step, a display step, via the user interface (12), of a representation of the resulting groups of elements (E).proceeded according to claim 1 or 2, wherein the method comprises a dynamic tracking phase including the following steps for each instant: a. receiving the flow data of the current instant, b. displaying, via a user interface (12), a representation of the flow data of the current instant, c. possibly updating, via the user interface (12), the user constraints so that the user constraints of the current instant are the updated user constraints or the user constraints of the previous instant otherwise, and. d. the possible modification, by the grouping tool (14), of the groups of elements (E) of the previous instant according to the flow data of the current instant and the user constraints of the current instant.
4. Method according to claim 3, wherein the dynamic tracking phase also includes for each instant a display step, via the user interface (12), of a representation of the groups of elements (E) obtained for the current instant.
5. Method according to claim 2 or 4, wherein at each display step of a representation of the groups of elements (E) obtained, an indicator relating to the consideration of user constraints in the groupings made is also displayed.
6. A method according to claim 4 or claim 5 in its dependence on claim 4, wherein during the display step of a representation of the groups of elements (E) obtained for the current time, any modifications of the groups of elements (E) are highlighted on the displayed representation so as to facilitate the management of flows between the elements (E) of the network (R).
7. A method according to any one of claims 1 to 6, wherein the dynamic monitoring phase includes a step of generating an alert and / or implementing an action when the element groups (E) of the current time are different from the element groups (E) of the previous time.
8. A method according to any one of claims 1 to 7, wherein the user constraint or constraints are chosen from: a. a constraint relating to an imposed grouping between elements (E) of the network (R), and b. a constraint relating to a prohibited grouping between elements (E) of the network (R).
9. A method according to any one of claims 1 to 8, wherein the grouping tool (14) implements an expectation-maximization type algorithm. 14 Electronic device (10) for characterizing flows between elements (E) of a network (R), the device (10) being suitable for implementing a method according to any one of claims 1 to 9 and comprising: a. a user interface (12), b. a grouping tool (14), and c. a tool (16) for receiving flow data and communicating a representation of the flow data, on the one hand, to the user interface (12) and, on the other hand, to the grouping tool (14).
