METHOD AND DEVICE FOR PREDICTING LOCAL NETWORK CONNECTIVITY DEMAND
Local network agents with remotely trained machine learning models and cloud solutions improve the prediction and management of network connectivity demand, addressing the limitations of existing systems by providing accurate and adaptive network resource allocation.
Patent Information
- Application Number
- FR2023004169
- Authority / Receiving Office
- FR · FR
- Patent Type
- Utility models
- Current Assignee / Owner
- Filing Date
- 2023-04-25
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2033-04-25
AI Technical Summary
Current information technology infrastructures face challenges in predicting network connectivity demand accurately due to limited demand prediction capabilities and failure to consider the varying nature and seasonality of infrastructure use, leading to connectivity issues such as increased response times and disconnections.
Implementing local network agents with machine learning models that are trained remotely and deployed locally, allowing for long-term predictions and real-time adjustments, using algorithms like ExtraTreesRegressor or XGBoost, and integrating cloud solutions for governance and analytics to manage network resources effectively.
Enhances the accuracy of network connectivity demand prediction by adapting to changes in demand and external factors, enabling real-time resilient decisions and centralized global analysis, thereby optimizing IT infrastructure management.
Smart Images

Figure 00000033_0000 
Figure 00000034_0000 
Figure 00000035_0000
Abstract
Description
Title of the invention: METHOD AND DEVICE FOR PREDICTING LOCAL NETWORK CONNECTIVITY DEMAND Technical field of the invention
[0001] The present invention relates to a method for predicting demand for local network connectivity and a device for predicting demand for local network connectivity. It applies in particular to the field of information technology infrastructure management.
[0002] Background of the invention
[0003] The approaches described in this section are approaches that may be pursued, but not necessarily approaches that have been conceived or pursued before. Therefore, unless otherwise indicated, it should not be assumed that all approaches described in this section are considered prior art simply because they are included in this section.
[0004] The perpetual increase in the consumption of digital information puts information technology infrastructures under pressure, which can lead to connectivity issues (increased response times, disconnections, faulty information transfer).
[0005] The reasons that impact these infrastructures vary considerably in nature and seasonality.
[0006] In current systems, monitoring agents are integrated into network devices and configured to report, according to determined agent operational parameters, upon request from said network devices.
[0007] These infrastructures are limited in their ability to adapt to changes in demand, given their limited demand prediction capabilities.
[0008] These systems, limited to the analysis of network demand, also fail to take into account the reasons for the use of the infrastructure, which may have its own variation in nature and seasonality.
[0009] Therefore, there are currently no reliable tools for predicting demand for information technology infrastructures and, consequently, there are no optimized information technology infrastructures. Summary of the invention
[0010] The present invention aims to overcome the aforementioned drawbacks as well as other drawbacks which could be overcome, although not mentioned in the description below.
[0011] The concept of the present invention relates to the use of local network agents, associated with network devices, said agents having the ability to formulate long-term predictions based on the operation of a machine learning model trained by a central authority. Furthermore, between the initial prediction and the end of the prediction term, the machine learning model is retrained and reused to update the initial prediction, which leads to a more accurate prediction capability.
[0012] Such an invention differs from edge prediction computing, which requires significant computing power, because the model is not trained locally but remotely while the training outcome is used locally.
[0013] In particular embodiments, for low-resource physical network devices (FGPA or small microcontrollers), the trained machine learning model is compiled, or transformed, into a prediction algorithm, such as a decision tree for example. In such cases, local network agents are limited to exploiting a simple algorithm that has the benefits of being derived from a machine learning device.
[0014] In particular embodiments, for a more resource-intensive physical network device, an optimal algorithm may be used to leverage the benefits of machine learning performed by the central authority. Such an algorithm may be obtained by transforming the trained machine learning model, for example. For example, a central machine learning device may be leveraged as a classifier to associate tags for objects, which may then be executed locally to perform costless algorithms such as ExtraTreesRegressor or XGBoost, for example. This allows for local decision making without the need for a connection to the central authority, in real time, for locally storing these decisions and failures, and for sending this data to the central authority, at a decided frequency and / or as soon as a connection is available.Local storage must be able to store enough data before transmission is possible.
[0015] In such embodiments, the trained machine learning model may correspond to determining a "best of X" algorithm, wherein the X corresponds to a number of specific machine learning models.
[0016] Such an invention may take the form of a cloud solution (installed on a customer's server) providing a computing infrastructure providing governance metrics and analytics and interacting with a user to mitigate future risks to the infrastructure based on predictions from local agents.
[0017] Such an invention enables local, real-time, resilient decisions, as well as centralized global analysis to learn from all physical network devices, pool knowledge, and (re)deploy updated models at the edge.
[0018] The present invention is intended to remedy all or part of these disadvantages.
[0019] All embodiments disclosed and claimed herein are directed to computer-implemented programmed processes that interact with digital data to provide a practical application of computer technology to the problem of managing and maintaining network resources that can be used to increase the responsiveness of a society's information technology infrastructure. The disclosure is not intended to encompass techniques for organizing human activity, performing mental processes, or executing a mathematical concept, and any interpretation of the claims to encompass such techniques would be unreasonable based on the disclosure as a whole.
[0020] To this end, according to a first aspect, the present invention aims at a method for predicting demand for local connectivity to the network, comprising: - a step of predicting long-term local connectivity demand, by a machine learning model operated by a local agent associated with a physical network device belonging to a computer network, - in a duration equivalent to the duration of the long-term prediction, more than one short-term iteration of: - a step of aggregation of representative values, at least: - a network connectivity request comprising at least one physical network device, - a network usage purpose including at least one physical network device or - a long-term local connectivity demand prediction for the physical network device, - a step of training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction, - a step of deploying the machine learning model on at least one local agent, and - a step of updating the long-term local connectivity demand prediction by the deployed machine learning model, operated by a local agent.
[0021] With these provisions, between the initial prediction and the end of the prediction term, the machine learning model is retrained and reused to update the initial prediction, leading to a more accurate prediction capability.
[0022] In particular embodiments, the value aggregation step further aggregates values representative of an event external to the computer network.
[0023] Such embodiments make it possible to improve the predictions made by the local agents based on external events, such as changes in the weather, changes in a stock market price, the presence of holidays, or social movements for example.
[0024] In particular embodiments, the method object of the present invention comprises: - a step of training a machine learning device to produce a machine learning model configured to classify a local connectivity demand pattern, and - a step of exploiting the trained machine learning model to classify a local connectivity demand pattern according to at least one of the following elements: - a prediction difficulty for the physical network device and / or - an importance of the physical network device in the computer network as a hub in said network.
[0025] In particular embodiments, the aggregation step comprises a step of sampling a subset of information to form a training set, said training set being used during a step of training a machine learning device.
[0026] Such embodiments allow adjustment of the dataset sampling and prediction horizon, according to a specific use case. These settings are located in the database linked to a particular service (use case) and are used to prepare the dataset before enrichment and machine learning. These settings are part of the prediction, which is saved in the database, so that it is possible to "replay" the prediction with the exact original settings, whenever the settings are changed for future predictions.
[0027] In particular embodiments, the aggregation step comprises a step of defining a prediction duration, said prediction duration being used during a step of training a machine learning device.
[0028] The prediction duration is the prediction horizon or the distance to the future prediction. Example: a data set in ten-minute increments. For this particular use case, a one-hour forecast horizon is the best compromise. For this data set in ten-minute increments, predicting at a one-hour horizon means that it is necessary to predict the next six points in the time series. In particular embodiments, the step enriching a training set, said enriched training set being used during a training step of a machine learning device.
[0029] The purpose of enrichers is to build additional information from raw metrics. Example: the orchestrator gives information that can be considered as the IP destination for each volume of traffic sent by an application via a device. A part of this traffic is then directed to the cloud (external destination), and the rest of the traffic is directed to another physical network device in the global enterprise infrastructure, this destination physical network device having the role of data center or hub.
[0030] By aggregating this information and performing specific calculations, it is possible to calculate for each physical network device during a period (10-minute slice) the total incoming traffic volume (entering the physical network device from another physical network device) and the detail by application, for example incoming traffic from other physical network devices or from productivity-tagged applications.
[0031] This produces two types of metrics: - the volume of incoming traffic (in bytes and hours), and - the part of this incoming traffic which is associated with applications marked productivity.
[0032] In particular embodiments, the method object of the present invention comprises a step of adjusting local connectivity to the network based on a prediction of long-term local connectivity demand.
[0033] Such embodiments allow dynamic reconfiguration of the infrastructure based on the provided predictions, to avoid major connectivity issues.
[0034] In particular embodiments, a value representative of the connectivity request corresponds to a response time of a remote device.
[0035] In particular embodiments, the step of training a machine learning device comprises: - training an extreme random forest device, - training a support vector machine regressor device and - training a gradient amplification regressor device.
[0036] In particular embodiments, the method object of the present invention comprises a step of associating a prediction reliability indicator with at least one physical network device, such a reliability indicator being aggregated during the step of aggregation, provision of a long-term prediction and / or updating of the long-term prediction.
[0037] In particular embodiments, the method object of the present invention comprises a step of associating a network importance indicator with at least one physical network device, such a network importance indicator being aggregated during the step of aggregating, providing a long-term prediction and / or updating the long-term prediction.
[0038] According to a second aspect, the present invention aims at a device for predicting demand for local connectivity to the network, comprising: - a local agent associated with a physical network device that belongs to a computer network, comprising means for operating a machine learning model to provide a long-term local connectivity demand prediction, - means for aggregating representative values, at least: - a network connectivity request comprising at least one physical network device, - a network usage purpose including at least one physical network device, - a long-term local connectivity demand prediction for the physical network device, - a means for training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction, - a means of deploying the machine learning model on at least one local agent, and - a means for updating the long-term local connectivity demand prediction by the deployed machine learning model operated by a local agent, the aggregation, training and deployment means being configured to operate iteratively, more than once, for a duration equivalent to the duration of the long-term prediction.
[0039] The device object of the present invention has the same advantages as the method object of the present invention. Brief description of the drawings
[0040] Other advantages, aims and particular characteristics of the invention will emerge from the following non-exhaustive description of at least one particular object, method or device of the present invention, in relation to the drawings annexed thereto, in which:
[0041] [Fig-1] represents, schematically and in the form of an organizational chart, a particular succession of steps of a method which is the subject of the present invention,
[0042] [Fig.2] schematically represents a particular embodiment of a system computer which can be used to implement a method which is the subject of the present invention,
[0043] [Fig.3] schematically represents a particular embodiment of the device object of the present invention, and
[0044] [Fig.4] schematically represents a long-term prediction sequence and training and short-term prediction updating enabled by the use of the present invention. Detailed description
[0045] This description is not exhaustive, since each feature of one embodiment may be combined with any other feature of any other embodiment in an advantageous manner.
[0046] Various inventive concepts may be embodied in one or more methods, an example of which has been provided. The acts performed as part of the method may be ordered in any suitable manner. Accordingly, it is possible to design embodiments in which acts are performed in a different order than illustrated, which may include performing certain acts simultaneously, even if they are shown as sequential acts in illustrative embodiments.
[0047] The expression "and / or" as used herein in the specification and in the claims, is to be understood to mean "either or both" of the elements so conjugated, i.e., elements which are present conjunctively in some cases and disjunctively in other cases. Multiple elements listed with "and / or" are to be interpreted in the same manner, i.e., as "one or more" of the elements so conjugated. Other elements other than the elements specifically identified by the "and / or" clause, may optionally be present, whether or not related to those specifically identified elements.Thus, by way of non-limiting example, a reference to "A and / or B", when used in conjunction with open language such as "comprising", may refer, in one embodiment, to A only (optionally including elements other than B); in another embodiment, to B only (optionally including elements other than A); in yet another embodiment, to both A and B (optionally including other elements); etc.
[0048] As used herein in the specification and claims, "or" is to be understood as inclusive.
[0049] As used herein in the specification and claims, the term "at least one," with reference to a list of one or more elements, shall be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each element specifically listed in the list of elements and not excluding any combination of elements in the list of elements. This definition also allows that elements, other than the elements specifically identified in the list of elements to which the phrase "at least one" refers, may optionally be present, whether or not related to those specifically identified elements.Thus, by way of non-limiting example, "at least one of A and B" (or, equivalently, "at least one of A or B" or equivalently, "at least one of A and / or B") may refer, in one embodiment, to at least one, optionally including more than one, A, without B present (and optionally including elements other than B); in another embodiment, to at least one, optionally including more than one, B, without A present (and optionally including elements other than A); in yet another embodiment, to at least one, optionally including more than one, A, and to at least one, optionally including more than one, B (and optionally including other elements); etc.
[0050] In the claims, as well as in the above specification, all transitional expressions such as "comprising", "including", "carrying", "having", "containing", "involving", "holding", "composed of" and the like are to be understood as being open, i.e., including, but not limited to. Only the transitional expressions "consisting of" and "consisting essentially of" are to be understood as closed or semi-closed transitional expressions, respectively.
[0051] It should be noted that the figures are not to scale.
[0052] According to one embodiment, the techniques described herein are implemented by at least one computing device. The techniques may be implemented in whole or in part using a combination of at least one server computer and / or other computing devices that are coupled using a network, such as a packet data network. The computing devices may be hardwired to perform the techniques or may include digital electronic devices such as at least one application-specific integrated circuit (ASIC) or field-programmable gate array (FPGA) that is persistently programmed to perform the techniques or may include at least one general-purpose hardware processor programmed to perform the techniques in accordance with program instructions in firmware, memory, another storage unit, or a combination.Such computing devices may also combine custom hard-wired logic, ASICs, or FPGAs with per-programming. customized to perform the described techniques. The computing devices may be server computers, workstations, personal computers, portable computing systems, handheld devices, mobile computing devices, wearable devices, body-mounted or implantable devices, smartphones, smart appliances, networking devices, autonomous or semi-autonomous devices such as robots or unmanned ground or aerial vehicles, any other electronic device that incorporates hard-wired and / or program logic to implement the described techniques, one or more virtual computing machines or instances in a data center and / or a network of server computers and / or personal computers.
[0053] [Fig. 2] depicts a block diagram that illustrates an exemplary computer system 200 with which an embodiment may be implemented. In the example of [Fig. 2], a computer system 205 and instructions for implementing the disclosed technologies in hardware, software, or a combination of hardware and software, are depicted schematically, e.g., as boxes and circles, at the same level of detail commonly used by those of ordinary skill in the art to which this disclosure relates to communicate about computer architecture and computer system implementations.
[0054] The computer system 205 includes an input / output (I / O) subsystem 220 that may include a bus and / or one or more other communication mechanisms for communicating information and / or instructions between components of the computer system 205 over electronic signal paths. The I / O subsystem 220 may include an I / O controller, a memory controller, and at least one I / O port. The electronic signal paths are shown schematically in the drawings, for example, as lines, one-way arrows, or two-way arrows.
[0055] At least one hardware processor 210 is coupled to the I / O subsystem 220 for processing information and instructions. The hardware processor 210 may include, for example, a general-purpose microprocessor or microcontroller and / or a special-purpose microprocessor such as an embedded system or Graphics Processing Unit (GPU) or a digital signal processor or an ARM processor. The processor 210 may have an integrated Arithmetic Logic Unit (ALU) or may be coupled to a separate ALU.
[0056] The computer system 205 includes one or more memory units 225, such as a main memory, which is coupled to the I / O subsystem 220 for storage digital electronic storage of data and instructions to be executed by the processor 210. The memory 225 may include volatile memory such as various forms of random access memory (RAM) or other dynamic storage devices. The memory 225 may also be used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processor 210. Such instructions, when stored on non-transitory computer-readable storage media accessible to the processor 210, may transform the computer system 205 into a specialized machine that is customized to perform the operations specified in the instructions.
[0057] The computer system 205 further includes non-volatile memory such as read-only memory (ROM) 230 or other static storage device coupled to the I / O subsystem 220 for storing information and instructions for the processor 210. The ROM 230 may include various forms of programmable ROM (PROM) such as erasable PROM (EPROM) or electrically erasable PROM (EEPROM). A persistent storage unit 215 may include various forms of non-volatile RAM (NVRAM), such as FLASH memory, or a solid state storage unit, magnetic disk, or optical disk such as CD-ROM or DVD-ROM, and may be coupled to the I / O subsystem 220 for storing information and instructions.Storage 215 is an example of a non-transitory computer-readable medium that may be used to store instructions and data that, when executed by processor 210, cause computer-implemented methods to be performed to perform the present techniques.
[0058] The instructions in memory 225, ROM 230, or storage unit 215 may comprise one or more sets of instructions organized into modules, methods, objects, functions, routines, or calls. The instructions may be organized into one or more computer programs, operating system services, or application programs, including mobile applications.The instructions may include an operating system and / or system software; one or more libraries to support multimedia, programming, or other functions; data protocol instructions or stacks to implement TCP / IP, HTTP, or other communications protocols; file format processing instructions to parse or render HTML, XML, JPEG, MPEG, or PNG encoded files; user interface instructions to render or interpret commands for a graphical user interface (GUI), command-line interface, or text-based user interface; application software such as an office suite, Internet access applications, design and manufacturing applications, graphics applications, audio applications, software engineering applications, educational applications, . games or various applications. Instructions may implement a web server, a web application server, or a web client. Instructions may be organized as a presentation layer, an application layer, and a data storage layer, such as a relational database system using Structured Query Language (SQL) or without SQL, an object store, a graph database, a flat file system, or another data storage unit.
[0059] The computer system 205 may be coupled via the I / O subsystem 220 to at least one output device 235. In one embodiment, the output device 235 is a digital computer display. Examples of a display that may be used in various embodiments include a touchscreen display or a light-emitting diode (LED) display or a liquid crystal display (LCD) or an e-paper display. The computer system 205 may include one or more other types of output devices 235, alternatively or in addition to a display device. Other output devices 235 include, for example, printers, ticket printers, plotters, projectors, sound or video cards, speakers, buzzers or piezoelectric devices or other audible devices, LED or LCD lamps or indicators, haptic devices, actuators, or servos.
[0060] At least one input device 240 is coupled to the I / O subsystem 220 for communicating signals, data, command selections, or gestures to the processor 210. Examples of input devices 240 include touchscreens, microphones, digital still and video cameras, alphanumeric and other keys, keypads, keyboards, graphics tablets, image scanners, joysticks, clocks, switches, buttons, dials, slides.
[0061] Another type of input device is a control device 245, which may perform cursor control or other automated control functions such as navigating a graphical interface on a display screen, as an alternative to or in addition to input functions. The control device 245 may be a touchpad, mouse, trackball, or cursor direction keys to communicate direction information and control selections to the processor 210 and to control movement of the cursor on the display 235. The input device may have at least two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), which allows the device to specify positions in a plane.Another type of input device is a wired, wireless, or optical control device such as a joystick, stylus, console, steering wheel, pedal, gear shift mechanism, or other type of control device. An input device 240 may include a com . combination of several different input devices, such as a video camera and a depth sensor.
[0062] In another embodiment, the computer system 205 may include an Internet of Things (IoT) device in which one or more of the output device 235, the input device 240, and the control device 245 are omitted. Or, in such an embodiment, the input device 240 may include one or more cameras, motion detectors, thermometers, microphones, seismic detectors, other sensors or detectors, measuring devices, or encoders, and the output device 235 may include a special display such as a single-line LED or LCD display, one or more indicators, a display panel, a meter, a valve, a solenoid valve, an actuator, or a servo.
[0063] The computer system 205 may implement the techniques described herein using custom hardwired logic, at least one ASIC or FPGA, firmware, and / or program instructions or logic that, when loaded and used or executed in combination with the computer system, operates or programs the computer system as a specialized machine. In one embodiment, the techniques herein are performed by the computer system 205 in response to the processor 210 executing at least one sequence of at least one instruction contained in the main memory 225. Such instructions may be read into the main memory 225 from another storage medium, such as the storage unit 215. Execution of the sequences of instructions contained in the main memory 225 causes the processor 210 to perform the process steps described herein.In alternative embodiments, hard-wired circuits may be used instead of or in combination with software instructions.
[0064] The term "storage medium" as used herein means any non-transitory medium that stores data and / or instructions that cause a machine to operate in a specific manner. These storage media may include non-volatile and / or volatile media. Non-volatile media include, for example, optical or magnetic disks, such as storage unit 215. Volatile media include dynamic memory, such as memory 225. Common forms of storage media include, for example, a hard disk drive, a solid-state disk, a USB flash drive, a magnetic data storage medium, any optical or physical data storage medium, a memory chip, or the like.
[0065] The storage medium is separate from but may be used in conjunction with a transmission medium. The transmission medium participates in the transfer of information between the storage media. For example, the transmission medium includes coaxial cables, copper wires, and optical fibers, including the wires that make up an I / O subsystem bus 220. Transmission media may also take the form of acoustic or light waves, such as those generated during radio and infrared data communications.
[0066] Various forms of means may be involved in transporting at least one sequence of at least one instruction to the processor 210 for execution. For example, the instructions may first be carried on a magnetic disk or a solid-state disk of a remote computer. The remote computer may load the instructions into its dynamic memory and send the instructions over a communications link such as a fiber optic or coaxial cable or a telephone line using a modem. A modem or router local to the computer system 205 may receive the data over the communications link and convert the data into a format that can be read by the computer system 205.For example, a receiver such as a radio frequency antenna or an infrared detector may receive the data carried in a wireless or optical signal and suitable circuitry may provide the data to an I / O subsystem 220, e.g., placing the data on a bus. The I / O subsystem 220 transports the data to memory 225, from which the processor 210 retrieves and executes the instructions. The instructions received by the memory 225 may optionally be stored on the storage unit 215 before or after execution by the processor 210.
[0067] The computer system 205 also includes a communication interface 260 coupled to the bus 220. The communication interface 260 provides a bidirectional data communication coupling to the network link or links 265 which is (are) directly or indirectly connected to at least one communication network, such as a network 270 or a public or private cloud on the Internet. For example, the communication interface 260 may be an Ethernet network interface, an Integrated Services Digital Network (ISDN) card, a cable modem, a satellite modem, or a modem for providing a data communication connection to a corresponding type of communication line, for example, an Ethernet cable or a metallic cable of any kind or a fiber optic line or a telephone line.The network 270 broadly represents a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a campus network, an internetwork, or any combination thereof. The communication interface 260 may include a LAN card to provide a data communication connection to a compatible local area network, or a wired cellular radiotelephone interface to send or receive cellular data according to cellular radiotelephone wireless network standards, or a wired satellite radio interface to send or receive digital data according to satellite wireless network standards. In such an implementation. In implementation, the communication interface 260 sends and receives electrical, electromagnetic, or optical signals via signal paths that carry digital data streams representing various types of information.
[0068] The network link 265 generally provides electrical, electromagnetic, or optical data communication directly or via at least one network to other data devices, using, for example, satellite, cellular, Wi-Fi, or BLUETOOTH technology. For example, the network link 265 may provide a connection via a network 270 to a host computer 250.
[0069] Further, network link 265 may provide a connection via network 270 or to other computing devices via internetworking devices and / or computers that are operated by an Internet Service Provider (ISP) 275. ISP 275 provides data communication services via a global packet data communication network represented by Internet 280. A server computer 255 may be coupled to Internet 280. Server 255 broadly represents any computer, data center, virtual machine or virtual computing instance with or without a hypervisor, or computer running a containerized program system such as DOCKER or KU-BERNETES.The server 255 may represent an electronic digital service that is implemented using multiple computers or instances and is accessed and used by transmitting web service requests, URL (Uniform Resource Locator) strings with parameters in HTTP payloads, API calls, application service calls, or other service calls. The computer system 205 and the server 255 may constitute elements of a distributed computing system that includes other computers, a processing cluster, a server farm, or other organization of computers that cooperate to perform tasks or run applications or services. The server 255 may include one or more sets of instructions organized into modules, methods, objects, functions, routines, or calls.The instructions may be organized into one or more computer programs, operating system services, or application programs, including mobile applications. The instructions may include an operating system and / or system software; one or more libraries to support multimedia, programming, or other functions; data protocol instructions or stacks to implement TCP / IP, HTTP, or other communications protocols; file format processing instructions to parse or render HTML, XML, JPEG, MPEG, or PNG encoded files; user interface instructions to render or interpret commands for a graphical user interface (GUI), command-line interface, or text-based user interface; . application software such as an office suite, Internet access applications, design and manufacturing applications, graphics applications, audio applications, software engineering applications, educational applications, games, or miscellaneous applications. The server 255 may include a web application server that hosts a presentation layer, an application layer, and a data storage layer such as a relational database system using Structured Query Language (SQL) or no SQL, an object store, a graph database, a flat file system, or another data storage unit.
[0070] The computer system 205 may send messages and receive data and instructions, including program code, via the network(s), network link 265, and communications interface 260. In the Internet example, a server 255 may transmit requested code for an application program via the Internet 280, ISP 275, local area network 270, and communications interface 260. The received code may be executed by the processor 210 when received and / or stored in the storage unit 215, or another non-volatile storage unit for later execution.
[0071] The execution of instructions as described in this section may implement a process as an instance of a computer program that is executed and consisting of program code and its ongoing activity. Depending on the operating system (OS), a process may be composed of multiple threads that concurrently execute instructions. In this context, a computer program is a passive set of instructions, while a process may be the actual execution of those instructions. Multiple processes may be associated with the same program; for example, opening multiple instances of the same program often means that multiple processes are being executed. Multitasking may be implemented to allow multiple processes to share the processor 210.While each processor 210 or processor core executes a single task at a time, the computer system 205 may be programmed to implement multitasking to allow each processor to switch between currently executing tasks without having to wait for each task to complete. In one embodiment, switches may be leveraged when tasks perform input / output operations, when a task indicates that a switch is possible, or upon hardware interrupts. Time sharing may be implemented to enable rapid response for interactive user applications by rapidly performing context switches to provide the appearance of concurrent execution of multiple processes simultaneously. In one embodiment, for security and reliability reasons, a system . exploitation can prevent direct communication between independent processes, providing strictly mediated and controlled inter-process communication functionality.
[0072] In the description below, the term "physical network device" means any piece of hardware connected over a computer network from which said piece derives connectivity to other physical and / or virtual network devices.
[0073] In the description below, the terms "computer network" refer to any embodiment of the network 270 disclosed in connection with [Fig. 2].
[0074] In the description below, the terms "connectivity demand" may refer to any metric generally used to monitor the demand of a communication link, such that a value representative of the connectivity demand corresponds to: quality: response time from a remote device, time to reach a DNS server, packet loss, latency of a packet to the destination, time to establish an SSL connection and / or capacity: use of bandwidth in transmission or reception for example.
[0075] In the description below, the terms “local agent” designate a hardware and / or software component paired with a physical network device and configured to, at least: monitoring and reporting the connectivity request of said physical network device, meaning that the local agent is configured to monitor the request of at least one layer of the Open Systems Information (OSI) stack and issue data to a computing device and leverage at least one trained machine learning model configured to associate a demand prediction with a given term (or for a given duration) for a given set of inputs, such as a time series representative of past connectivity demand for example.
[0076] Such a physical network device, or physical network device, can correspond to any physical network device in a computer network. An edge network device is a physical network device located at the edge of a network that serves as an entry or exit point for data traffic. These devices are responsible for routing traffic between different networks and can perform various functions such as filtering, encryption, and quality of service (QoS) management. Edge network devices can include routers, switches, firewalls, load balancers, and other network devices that are placed at the edge of a network. These devices are often used to provide connectivity and security between different networks, such as connecting a local area network (LAN) to a wide area network (WAN) or the Internet.
[0077] The general principle of the present invention can be seen in [Fig.4].
[0078] [Fig.4] represents, schematically, a series of predictions, 405, 406 and 407, by a single local agent in relation to a single physical network device.
[0079] Each prediction, associated with the same prediction time (or horizon), is associated with a prediction training set, 410, 411 and 411, comprising data used to train a machine learning device. These training sets, 410, 411 and 411, correspond to data of the same nature measured up to a different time. These different times are either: the initial sampling cutoff time 410 before training before the initial prediction 405 for a determined time is any time, 411 and 412, in the interval from the initial sampling cutoff time 410 to the time determined for the prediction 405, 406 and 407.
[0080] Before each prediction, associated with the same prediction time (or horizon), the training set, 410, 411 or 411, is used as input for a singular machine learning device 415 configured to provide data associated with a machine learning model of a similar nature to the training sets, 410, 411 and 411, to a prediction at a determined prediction time.
[0081] The result of these training steps is the provision of trained machine learning models, 420, 421 and 422, deployed on local agents and operated to provide adjusted predictions, 406 and 407, at the time of the initial prediction 405.
[0082] [Fig.l] schematically represents a particular succession of steps of the method 100 which is the subject of the present invention. This method for predicting demand for local connectivity to the network 100 comprises: a step 105 of predicting long-term local connectivity demand, by a machine learning model operated by a local agent associated with a physical network device which belongs to a computer network, in a duration equivalent to the duration of the long-term prediction, more than one short-term iteration of: a step 110 of aggregation of values representative of at least: a network connectivity request including at least one physical network device, a network usage purpose comprising at least one physical network device, a long-term local connectivity demand prediction for the physical network device, a step 115 of training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction, a step 120 of deploying the machine learning model on at least one local agent, and a step 125 of updating the long-term local connectivity demand prediction by the deployed machine learning model operated by a local agent.
[0083] The long-term local connectivity demand prediction step 105 may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said long-term local connectivity demand prediction step 105. Such a set of instructions corresponds, at least, to the operation of a trained machine learning model configured to associate a prediction, for a determined duration in the future, of a connectivity status of the physical network device in relation to at least one other network device in the computer network. Such a prediction time may, for example, be one hour in the future.
[0084] In the case where the long-term local connectivity demand prediction is linked to a capacity prediction: multiple parameters can be used, such as traffic metrics, quality metrics, demand metrics, business factors, which can be provided by multiple sources, if necessary, to predict the maximum bandwidth percentage that will be used during the next hour by a local agent, such a prediction can correspond to the maximum, minimum, average and estimates when the maximum peak will occur in the next hour, - such a prediction can be based on: - this prediction (% bandwidth), - knowledge of peripheral network devices such as the different links available with their bandwidth, - if an additional backup connection is available, not already activated, - if this request for backup connection is authorized, - a quota associated with this additional link, and the actual demand for this quota and / or - a specific business factor giving a coefficient to be applied to action thresholds; - the prediction can lead to an action: if the predicted bandwidth demand reaches 75% of the actual total available bandwidth for this edge network device or physical network device, and all checks are good (authorization, quota, availability...), then the decision is made to activate the additional link to increase the bandwidth - this action can be triggered by a local or remote actuation device, configured to change the value of at least one connectivity parameter based on the prediction and / or - different thresholds can be used to generate perspectives, such as notifications when 70% of bandwidth will be needed, when a link's monthly quota is about to be reached, when the prediction % has led to the activation of a backup link.
[0085] In the case where the long-term local connectivity demand prediction is linked to a failure prediction, the following process may take place: - infrastructure usage over the past 30 days can be used as a sample, and events can be identified and labeled as “outages”, based on a calculation such as a rapid decrease in connectivity of more than 10% over a short period of time, the physical network devices that are primarily responsible for this outage can also be identified, - these events are used as targets to input into a supervised machine learning device using an enriched dataset (up to 40 features) consisting of connectivity metrics, predictions for each physical network device that was responsible for at least one outage event, and that was classified with high hub importance, such as: - a loss ratio variation representative of the change in the average packet loss ratio for a physical network device over the next hour, - a risk of saturation representing a demand for bandwidth greater than the network capacity over the next hour, - three algorithms (such as an extra random forest regressor, a support vector regressor or a boosting gradient regressor) executed to predict future points (24 hours) when a physical network device will be responsible for a global outage, and optionally uses the ExtraTreesRegressor algorithm to provide the importance of features, - the output of the predictions is: - physical network device responsible for identifying the predicted failure, site, labels, - risk level measured between 50% and 100%, lower values being ignored, and the time stamp of each failure.
[0086] This information can be recorded every 10 minutes to enable further analysis and provide unique statistics on edge network devices or the behavior and responsibilities of physical network devices. And this information is aggregated on a map to show sites at risk in the next 24 hours, with a risk level, in what is called a heat map.
[0087] All predictions such as capacity, quality of experience, link quality, application demand are stored in a database, together with the settings used for the machine learning model and the context made up of business factors. They are reused for other predictions (such as failure prediction) and can be replayed to test “what if” scenarios that could improve prediction accuracy.
[0088] All predictions are analyzed every night to calculate the actual accuracy, comparing the predictions to what actually happened.
[0089] When the accuracy of a prediction service decreases, it can launch an in-depth analysis with “What if” scenarios to calibrate the machine learning model settings.
[0090] In a time interval corresponding to the interval between the first operation of the machine learning model and the time (date) of the prediction, several iterations of the training set assembly set (aggregation 110), training 115, deployment 120 and update 125 of the predictions occur.
[0091] For example, one of these iterations may require ten minutes and a prediction may be made for a duration of one hour in advance, meaning that four iterations may occur and gradually adjust the prediction.
[0092] The aggregation prediction step 110 may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said aggregation step 110.
[0093] During the aggregation step 110, data of different natures can be associated to generate a training set for a machine learning device.
[0094] During the aggregation step 110, a long-term prediction may be part of the training set, said long-term prediction being representative of the same connectivity demand parameter or a different connectivity demand parameter. For example, a capacity demand prediction may be used in a failure prediction.
[0095] At least part of this data may correspond, for example, to connectivity request data from local agents.
[0096] At least a portion of this data may correspond, for example, to a label or identifier defining the purpose of the connectivity request data from local agents. Such a purpose may be, for example, emails, video conferencing, video streaming, or instant messaging.
[0097] At least part of this data may correspond, for example, to connectivity demand predictions from local agents.
[0098] At least part of this data may correspond, for example, to values representative of an event external to the computer network. Such an event may correspond, for example, to weather information, stock market information, calendar information (holidays) or information relating to the news.
[0099] The aggregated data may be associated with a time value, corresponding to an indicator of duration or seasonality (such as time of day, day, month, day of week, season).
[0100] Aggregated data may correspond to derived data, such as averages over a given period, for example.
[0101] As understood, this data may correspond, for example: - to the evolution of connectivity / latency / loss packets... in the last hour, - to the average of the latency of the links for a physical network device during a slot (10 min), and - the percentage of “productivity” traffic (from applications that are associated with a label representative of productivity) compared to the total traffic passing through the physical network device in a 10-minute slot, weighted by the importance of the physical network device in the network.
[0102] In particular embodiments, such as that shown in [Fig.l], the aggregation step 110 comprises a step 140 of sampling a subset of information to form a training set, said training set being used during a step, 115 and / or 130, of training a machine learning device.
[0103] Depending on the data source (e.g. SD-WAN Orchestrator, Partner Agents, Telcos or CDR files), data values (points) can be collected or streamed at different frequencies, which can vary. Sometimes information arrives every minute, sometimes every ten minutes, sometimes several times per second. The frequency depends on the specific use case being implemented.
[0104] This raw data can be stored upon arrival, but depending on the data demand (the use case), the data can be resampled to align all sources to a desired sample rate.
[0105] For example, a partner delivers some streaming data, which can be at any frequency, such as control plane data from local agents, and an orchestrator API is used to obtain ten-minute snapshots of aggregated data. This data can be transferred in bulk at given time intervals.
[0106] For capacity predictions, this data may be aggregated into ten-minute slices or blocks, including, if applicable, business factors provided by a customer responsible for the local device, to obtain a regular time series data set with a point every ten minutes. Each column is aggregated using the appropriate formula, for example min, max, mean, sum and standard deviation.
[0107] The final dataset is constructed and sampled on the fly if necessary, or a use-case specific database table is created with sampled data, in order to accelerate machine learning operations.
[0108] For some use cases, such as route quality prediction (gateway and hop), raw data can be used, and this data can be sampled in several versions correlated to the desired prediction horizon. Example: Sampling 1 minute, then 2 minutes, to predict in the following 10 minutes. Then sampling 5 minutes to predict a horizon of 30 minutes. Then sampling 10 minutes to predict at 1 hour and 4 hours.
[0109] This allows to provide multi-time interval predictions, which will be the short-term quality risk, medium and long-term predictions. The route change decision can take into account all time intervals to assess the urgency of the action.
[0110] In particular embodiments, such as that shown in [Fig. 1], the aggregation step 110 comprises a step 145 of defining a prediction duration, said prediction duration being used during a step, 115 and / or 130, of training a machine learning device.
[0111] Such a step 145 of defining a prediction duration may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said step 145 of defining a prediction duration.
[0112] During such a prediction step 145, an input subsystem of a computing device may be operated to automatically or manually establish a prediction time for local agent predictions. Such a prediction time may correspond to an offset from the prediction time (e.g. +1 hour) or to determined times of the day (e.g. 11 a.m.).
[0113] In particular embodiments, such as that shown in [Fig. 1], step 150 of enriching a training set, said enriched training set being used during a step, 115 and / or 130, of training a machine learning device.
[0114] Such a step 150 of enriching a training set may be carried out, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said step 150 of enriching a training set.
[0115] During this enrichment step 150, the objective is to add context to each point in a data set. For example, if there is a situation by particular in an airport, given by the person in charge of a local device as a business factor, and associated with an adjustment coefficient, starting at a timestamp, all points in the data set starting from this timestamp and until the "alert" is removed are enriched by the information (boolean coefficient +).
[0116] Weekly in-depth analysis of the behavior of physical network devices can add tags to the physical network device or local devices, which can be used for building the dataset.
[0117] Thus, a specific weather alert situation can be mentioned in the dataset during the defined period, for example.
[0118] The final dataset used by the machine learning device is the result of resampling the raw data, enriched with contextual data (calculated or provided as a business factor).
[0119] Predictions made on this dataset can provide more information that is added to another even richer dataset, adding connectivity metrics and prediction results (e.g.: failure predictions).
[0120] Step 115 of training a machine learning device may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said step 115 of training a machine learning device.
[0121] During this training step 115, the data aggregated during the step 110 aggregation can be used as input to a machine learning device of a type suitable for the particular use case.
[0122] Such a machine learning device can use pre-processing algorithms such as LSTM, Transformers, TSNE-XGBOOST (to pre-train models for subsequent real-time demand with prediction).
[0123] Such a machine learning device may use post-processing algorithms such as ExtraTreesRegressor, Support Vector Regressor, GradientBoost Regressor, HistGradientBoost, TSNE-XGBOOST for pre-trained models.
[0124] In particular embodiments, such as that shown in [Fig.l], step 115 of training a machine learning device comprises: - training 160 of an extreme random forest device, - the drive 165 of a support vector machine regressor device and - training 170 of a gradient amplification regressor device.
[0125] For example, using a thirty-day data set that has been resampled by 10-minute points, enriched with aggregated metrics and business factors, seasonality information can be added, such as the day of the week, weekend, time of day.
[0126] This example assumes that the physical network device associated with the local agents has been classified and labeled by an asynchronous process.
[0127] The dataset for all physical network devices of a customer or a tenant of a customer, as needed, is obtained.
[0128] Then: - a service that runs the pre-trained model (TSNE-XGBOOST) is called for each physical network device that is classified as “HIGH” (prediction difficulty level). This process takes 1 or 2 seconds to process about 50 physical network devices, and it returns a list of HIGH physical network devices with the resulting prediction (using the same dataset), - the overall dataset is split by physical network device, to calculate physical network device predictions by physical network device and - physical network devices are classified into three levels: LOW, MEDIUM, HIGH by a classification algorithm, such as TSNE XGboost - a different strategy being applied for each class of physical network device: - LOW: An ExtraTreesRegressor algorithm can be run, which works very well for these "easy to predict" physical network devices, and takes a fraction of a second per physical network device, - MEDIUM: These physical network devices can sometimes exhibit traffic spikes that are not easy to predict with a single algorithm: - an “Outlier Detector” algorithm can be run to normalize the dataset using the 5-95 percentiles, - holes in the dataset can be filled, - the behavior of the physical network device over the past 30 days is analyzed to detect whether traffic sometimes bypasses 50% of the bandwidth capacity, or whether the physical network device is used extensively as a "recipient" hub for other physical network devices, - two types of physical network devices MEDIUMS can be determined, and two distinct strategies can be applied: - for the first type (no specific behavior), three algorithms can be run on the physical network device (ExtraTreesRegressor, SVR, Gra-dientBoostRegressor), then the one with the best confidence is elected, and the "explainability" of the ExtraTreesRegressor is preferably kept - the prediction values of the result of the "best" algorithm are kept and the prediction includes a table of all predicted values within the prediction horizon (example: prediction actual + 6 points of 10 minutes for a horizon of 1 hour) and - for the second type (kind of data centers, or hubs with a strong relational hub position), a HistGradientBoosRegressor algorithm is executed because it better captures abrupt traffic variations, with less latency and an “amplified” prediction, - HIGH: for this class of physical network devices, the calculation has already been done in the first step, so we consider this prediction result.
[0129] This allows to obtain a prediction for each physical network device, including the maximum peak during the next hour, which is the main information needed.
[0130] Preferably, each prediction is made with additional descriptive information (site, physical network device, link mapping, available & authorized backup links, quotas, etc.).
[0131] From there, a decision is made to activate (prediction >= 75%) or deactivate (prediction < 50%) an additional link, if applicable. The decision process uses fallback / quota / authorization / prediction.
[0132] If the action is ACTIVATE or DISABLE a backup link and everything is good, then it is possible to communicate with the physical network device to modify the internal policies to adjust the backup link demand (it can also be changing traffic priorities to focus for example on video conferencing applications and a prohibited request during a critical period).
[0133] Depending on the available perspectives and customer subscriptions, it is possible, during the dressing process, to notify a perspective for any action performed, or prediction threshold or quota request.
[0134] The final prediction (in j son format) can be saved in a database table for future request, statistics, smart metrics, API gateway and for nightly accuracy control.
[0135] In particular embodiments, the method 100 which is the subject of the present invention comprises a step 101 of associating a prediction reliability indicator with at least one physical network device, such a reliability indicator being aggregated during the step 110 of aggregation, provision 105 of a long-term prediction and / or updating 125 of the long-term prediction.
[0136] Such an association step 101 can be performed manually, via input means of a computer device, or automatically.
[0137] In embodiments in which the association step 101 is performed automatically, a set of instructions representative of an algorithm may be executed by a computing device in order, from a series of long-term predictions past term and corresponding actual events that the predictions accurately or inaccurately predicted, to determine a reliability indicator. The more accurate the predictions were in the past, the higher the reliability indicator for a specific physical network device.
[0138] Such an association step 101 may use a classifier algorithm, such as a trained machine learning classifier model, to associate an indicator representative of a reliability class (“HIGH”, “MEDIUM”, “LOW”, for example) for physical network devices.
[0139] In particular embodiments, the method 100 which is the subject of the present invention comprises a step 102 of associating a network importance indicator with at least one physical network device, such a network importance indicator being aggregated during the step 110 of aggregation, provision 105 of a long-term prediction and / or updating 125 of the long-term prediction.
[0140] Such an association step 102 can be performed manually, via input means of a computer device, or automatically.
[0141] In embodiments in which the association step 102 is performed automatically, a set of instructions representative of an algorithm may be executed by a computing device in order, from a series of trading volume and / or a number of physical network devices communicating with a targeted physical network device, to determine a network importance indicator of said targeted physical network device. The more a targeted physical network device is associated with a high trading volume and / or a high number of physical network devices communicating with said targeted physical network device, the higher the network importance indicator.
[0142] Such an association step 102 may use a classifier algorithm, such as a trained machine learning classifier model, to associate an indicator representative of a reliability class (“HIGH”, “MEDIUM”, “LOW”, for example) for physical network devices.
[0143] Step 120 of deploying a machine learning model on at least one local agent may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said deployment step 120.
[0144] During this deployment step 120, a trained machine learning model can be downloaded by a local agent and stored in a memory awaiting exploitation of this trained machine learning model.
[0145] Step 125 of updating the long-term local connectivity demand prediction may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said update step 125.
[0146] During this update step 125, the trained machine learning model is used to provide a connectivity demand prediction for the time (date) of the initial prediction.
[0147] In particular embodiments, such as that shown in [Fig.l], the method 100 which is the subject of the present invention comprises: - a step 130 of training a machine learning device to produce a machine learning model configured to classify a local connectivity request pattern, and - a step 135 of using the trained machine learning model to classify a local connectivity request pattern according to at least one of the following elements: - a prediction difficulty for the physical network device and / or - an importance of the physical network device in the computer network as a hub in said network.
[0148] The prediction difficulty (“LOW”, “MEDIUM” or “HIGH”) may be based on previous predictions and the behavior of the physical network device over a specific time interval (the last six months), as well as the function of the physical network device in the infrastructure graph.
[0149] This classification can be used to choose the machine learning strategy for capacity prediction, for example.
[0150] The importance of the physical network device can be based on traffic interactions (IP destination is on one physical network device in the network, egress to another physical network device in the network, etc.) and the number of physical network devices that are connected to that physical network device. This results in three classes per importance level: "HUBS", "HUB.1" and "HUB.2". This classification allows for selecting "significant" physical network devices on the network, for example to evaluate whether a failure prediction is appropriate for that physical network device, as it has a significant impact on the overall enterprise network.
[0151] In particular embodiments, such as that shown in [Fig.l], the method 100 which is the subject of the present invention comprises a step 155 of adjusting local network connectivity based on a prediction of long-term local connectivity demand.
[0152] The step 155 of adjusting local connectivity of the network may be performed, for example, by at least one processor of a computer system 200 configured to execute a set of instructions corresponding to said step 155 of adjusting local connectivity of the network.
[0153] During this step 155, there may be two types of adjustments. The first type of adjustments is network adjustments which are policies that can be passed to network components to impact connectivity between physical and / or virtual devices in the networks. For example, particular protocols or services may be blocked or connections between two devices may be prohibited. The second type of adjustments is application adjustment to impact sending, receiving data via the networks or moving an application from the local system to another computer.
[0154] Such a step 155 can be achieved by mapping the links for the physical network device, to check if there is a backup link, if it is in use or on standby, this link can be operated remotely, and if there is a quota limitation for this link (start date in the month and monthly allocation, plus threshold and action decisions).
[0155] Example: A customer can for example authorize the request for a backup link, but establish a quota (10 GB from the 5th of the month) and give 2 quota actions: perspective when the monthly quota request is at 70%, and perspective and blocking of the link request once the quota is used at 99%.
[0156] Then, depending on the prediction (% bandwidth) and the settings for the thresholds, if the prediction for the next max hour is greater than or equal to 75%, and there is a link that can be used, an order to the orchestrator can be sent or directly to the physical network device to activate this backup link, which adds its bandwidth to the overall bandwidth of the physical network device.
[0157] This step is described in the detailed capacity prediction process. In a separate timeframe (e.g. once a week, or once a day), the classification algorithms are executed to train a model (heavy power demand). The resulting model is saved for reuse with fresh data added, during the capacity prediction process, but reusing a pre-trained model with added data is extremely fast compared to computing the original model.
[0158] For other algorithms in the capacity process, for example, the models are created, trained, tested, evaluated and finally applied to the entire dataset, every 10 minutes, for each physical network device, to adjust the predictions. This is possible because the algorithms and adjustments of the present invention are very fast.
[0159] If the predicted capacity requirement is less than 50% and there is an activated backup link, or the activated backup link is blocked for quota reasons, then the backup link is deactivated, which returns to standby mode.
[0160] Thus, for each prediction, the environment of the physical network device (link availability...) and prediction (%) are obtained.
[0161] This information is used to create a smart metric: if the required connectivity percentage in the next hour is more than 75%, but there is no available, authorized, quota-free link to provide additional bandwidth, there is a risk of outage.
[0162] [Fig. 3] schematically represents a particular embodiment of the device 300 which is the subject of the present invention. This device for predicting demand for local connectivity to the network 300 comprises: - a local agent 305 associated with a physical network device 306 which belongs to a computer network 307, comprising a means 310 for exploiting a machine learning model to provide a long-term local connectivity demand prediction, - a means 315 of aggregating values representative of at least: - a network connectivity request including at least one physical network device, - a network usage purpose including at least one physical network device, - a long-term local connectivity demand prediction for the physical network device, - a means 320 for training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction, - a means 325 of deploying the machine learning model on at least one local agent, and - a means 330 of updating the long-term local connectivity demand prediction by the deployed machine learning model operated by a local agent,
[0163] the aggregation, training and deployment means being configured to operate iteratively, more than once, for a duration equivalent to the duration of the long-term prediction.
[0164] Particular embodiments of the means of this device 300 are disclosed in relation to the method 100 which is the subject of the present invention.
Claims
Claims
1. Method for predicting local connectivity demand to the network (100), characterized in that it comprises: - a step (105) of predicting long-term local connectivity demand, by a machine learning model operated by a local agent associated with a physical network device which belongs to a computer network, in a duration equivalent to the duration of the long-term prediction, more than one short-term iteration of the following steps: - a step (110) of aggregating representative values, at least of a connectivity demand of the network comprising at least one physical network device, of a purpose of use of the network comprising at least one physical network device or of a prediction of long-term local connectivity demand for the physical network device,- a step (115) of training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction, - a step (120) of deploying the machine learning model on at least one local agent, and - a step (125) of updating the long-term local connectivity demand prediction by the deployed machine learning model operated by a local agent.,
2. The method (100) of claim 1, wherein the step (110) of aggregating values further aggregates values representative of an event external to the computer network.
3. A method (100) according to any one of claims 1 or 2, which comprises: - a step (130) of training a machine learning device to produce a machine learning model configured to classify a local connectivity demand pattern, and - a step (135) of operating the trained machine learning model to classify a local connectivity demand pattern according to at least one of the following: - a prediction difficulty for the physical network device and / or - an importance of the physical network device in the network in- computer as a hub in said network.
4. Method (100) according to any one of claims 1 to 3, wherein the aggregation step (110) comprises a step (140) of sampling a subset of information to form a training set, said training set being used during a step (115, 130) of training a machine learning device.
5. Method (100) according to any one of claims 1 to 4, wherein the aggregation step (110) comprises a step (145) of defining a prediction duration, said prediction duration being used during a step (115, 130) of training a machine learning device.
6. Method (100) according to any one of claims 1 to 5, which comprises a step (150) of enriching a training set, said enriched enrichment set being used during a step (115, 130) of training a machine learning device.
7. A method (100) according to any one of claims 1 to 6, which comprises a step (155) of adjusting local network connectivity based on a long-term local connectivity demand prediction.
8. The method (100) of any one of claims 1 to 7, wherein a value representative of the connectivity request corresponds to a response time of a remote device.
9. The method (100) of any one of claims 1 to 8, wherein the step (115) of training a machine learning device comprises: - training (160) an extreme random forest device, - training (165) a support vector machine regressor device, and - training (170) a gradient boosting regressor device.
10. Method (100) according to any one of claims 1 to 9, which comprises a step (101) of associating a prediction reliability indicator with at least one physical network device, such a reliability indicator being aggregated during the step (110) of aggregating, providing (105) a long-term prediction and / or updating (125) the long-term prediction.
11. A method (100) according to any one of claims 1 to 10, which comprises a step (102) of associating a network importance indicator with at least one physical network device, such a network importance indicator being aggregated during the step (110) of aggregating, providing (105) a long-term prediction and / or updating (125) the long-term prediction.
12. A device for predicting local connectivity demand to the network (300), characterized in that it comprises: - a local agent (305) associated with a physical network device (306) which belongs to a computer network (307), comprising means (310) for operating a machine learning model to provide a long-term local connectivity demand prediction, - means (315) for aggregating values representative of at least: - a connectivity demand of the network comprising at least one physical network device, - a purpose of use of the network comprising at least one physical network device or - a long-term local connectivity demand prediction for the physical network device, - means (320) for training a machine learning device to produce a machine learning model configured to provide a long-term local connectivity demand prediction,- means (325) for deploying the machine learning model on at least one local agent, and - means (330) for updating the long-term local connectivity demand prediction by the deployed machine learning model operated by a local agent, the aggregation, training and deployment means being configured to operate iteratively, more than once, for a duration equivalent to the duration of the long-term prediction.,
