System and method for dynamically changing event allocation of islands

Abstract

The invention relates to a system and method for dynamically changing event allocation of islands. The present disclosure provides systems and methods that may advantageously apply machine learning to detect network interruptions and attribut network interruptions to particular components or nodes within the network. In one aspect, the disclosure provides a computer-implemented method comprising: mapping a network comprising a plurality of islands, the plurality of islands being dynamically alterable by splitting and / or merging of one or more islands, where the plurality of islands comprise a plurality of single components; and detecting and locating one or more local events at a single component level as well as an island level using the deaggregation model.

Description

[0001] This application is a divisional application of an application with a filing date of March 19, 2020, application number 202080038035.8, and invention name “System and method for dynamically changing event allocation of islands”. Cross-references

[0002] This application claims priority to U.S. Provisional Patent Application No. 62 / 822,300, filed on March 22, 2019, both of which are incorporated herein by reference in their entirety. Background Art

[0003] A network is a common system of interconnected devices or components. Even under ideal conditions, the operation and control of network systems can be quite complex, as network functions may need to perform certain tasks synchronously, sequentially, or in more complex patterns. An example of a complex network system is an electric distribution network. The control and maintenance of the power grid is a particularly complex problem given the ever-changing levels of supply and demand. The current surge in renewable energy sources and large-scale battery systems presents unique opportunities and challenges for the control of distribution networks, as they may provide greater resilience to the network, but their power output also tends to be intermittent. When unplanned events (such as power line failures) occur, the grid control system faces the challenge of responding correctly and minimizing the propagation of the fault to other parts of the grid.

[0004] There is a need for predictive maintenance applications for networks, such as electrical grids. Currently, most predictive maintenance applications can only assign outage events to specific areas of the grid, rather than to the specific components that caused the outage. Relatively simple logic can be used to predict certain outage events, such as the disconnection of one area of ​​the grid from other areas. However, as the complexity of the network increases, the logic of predictive maintenance becomes more difficult. Therefore, there is a need for improved predictive maintenance applications for complex networks to enhance the ability of control systems and system operators to respond to unexpected changes in network configuration. Summary of the invention

[0005] There is a need for a predictive maintenance application for networked systems that is capable of handling complex logic describing outage and reconnection events. Specifically, there is a need for a network maintenance application that is capable of detecting and attributing network outages to specific components or nodes within the network. Additionally, there is a need for a predictive maintenance application that can learn from a single network outage event and use the acquired data to predict and respond to future outage events.

[0006] The present disclosure addresses at least the need for enhanced predictive maintenance applications for networked systems by utilizing machine learning latent variable models. In some aspects, the present disclosure describes a disaggregated model that can be applied to predictive maintenance of networked systems such as power grids or telecommunications networks. The disaggregated model can provide further refinement of the network description, where outage events can be traced to specific nodes or components rather than broader regions. The described latent variable model may be able to handle more complex logical problems, such as islanding of power grid sections and reintegrating islands back into the network. Predictive maintenance applications may be able to use probabilistic modeling to assign outage events to specific components and use converged sets of probabilities to predict subsequent network behavior.

[0007] In some aspects, the present disclosure may describe a computer-implemented event allocation method that includes mapping a network comprising multiple islands, wherein the multiple islands can be dynamically changed by splitting and / or merging one or more islands, so that the multiple islands include multiple individual components, and predicting and / or detecting and locating one or more local events at the individual component level and at the island level can utilize a disaggregated model.

[0008] In some aspects, the disclosure may describe a computer-implemented event distribution method having a plurality of individual components including nodes and branches.

[0009] In some aspects, the disclosure may describe a computer-implemented event assignment method in which a disaggregated model is configured to detect one or more local events at an individual component level rather than at an island level.

[0010] In some aspects, the present disclosure may describe a computer-implemented event distribution method utilizing a disaggregated model configured to locate or isolate one or more local events at a single component level.

[0011] In some aspects, the disclosure may describe a computer-implemented event distribution method utilizing a disaggregated model configured to detect one or more local events at a single component level when one or more of the islands dynamically change.

[0012] In some aspects, the disclosure may describe a computer-implemented event distribution method in which dynamic changes to islands include multiple instances of splitting, merging, and / or re-merging of one or more islands.

[0013] In some aspects, the disclosure may describe a computer-implemented event distribution method in which dynamic alteration of an island includes connecting and / or splitting two or more of a plurality of nodes and branches.

[0014] In some aspects, the disclosure may describe a computer-implemented event distribution method in which dynamic changes to islands include splitting and / or merging of subsets of individual components.

[0015] In some aspects, the disclosure may describe a computer-implemented event distribution method in which subsets of individual components have the same size.

[0016] In some aspects, the disclosure may describe a computer-implemented event distribution method in which subsets of individual components have variable sizes.

[0017] In some aspects, the present disclosure may describe a computer-implemented event assignment method utilizing a disaggregated model as a machine learning latent variable model.

[0018] In some aspects, the disclosure may describe a computer-implemented event allocation method that utilizes a machine learning latent variable model implemented at the level of individual components.

[0019] In some aspects, the present disclosure may describe a computer-implemented event assignment method in which a machine learning latent variable model includes a plurality of latent variables corresponding to one or more local events at each of the individual components.

[0020] In some aspects, the disclosure may describe a computer-implemented event distribution method in which one or more local events occurring at an individual component level are associated with one or more island events occurring at an island level.

[0021] In some aspects, the present disclosure may describe a computer-implemented event assignment method in which a machine learning latent variable model is configured to receive observations of selected island events occurring on corresponding selected islands and process the observations based on an assumption that one of a plurality of latent variables has caused the selected island event to occur.

[0022] In some aspects, the disclosure may describe a computer-implemented method of assigning events in which selected island events are caused by a single component in or associated with a corresponding selected island.

[0023] In some aspects, the disclosure may describe a computer-implemented event assignment method in which a machine learning latent variable model includes a prior probability of each individual component causing each of one or more local events.

[0024] In some aspects, the disclosure may describe a computer-implemented event assignment method in which a machine learning latent variable model is configured to perform precise a posteriori inference based on prior probabilities to assign one or more local events to a single component.

[0025] In some aspects, the disclosure may describe a computer-implemented event assignment method wherein the assignment of the one or more local events to individual components includes determining a probability estimate for each local event occurring at each of the corresponding components.

[0026] In some aspects, the disclosure may describe a computer-implemented event assignment method in which a machine learning latent variable model is configured to iteratively update prior or previous probabilities using probability estimates until convergence.

[0027] In some aspects, the disclosure may describe a computer-implemented event assignment method in which a priori probabilities include initial known, actual, or estimated probabilities based on prior event data.

[0028] In some aspects, the disclosure may describe a computer-implemented event assignment method in which prior or a priori probabilities are iteratively updated using an expectation-maximization (EM) algorithm.

[0029] In some aspects, the disclosure may describe a computer-implemented event allocation method that also includes performing predictive maintenance at an individual component level based on one or more detected local events.

[0030] In some aspects, the disclosure may describe a computer-implemented method of event distribution in which a network includes an electric power distribution network.

[0031] In some aspects, the disclosure may describe a computer-implemented event distribution method in which an electric distribution network includes a plurality of distribution feeders.

[0032] In some aspects, the present disclosure may describe a computer-implemented event distribution method in which a plurality of nodes and branches are associated with a plurality of distribution feeders and connected nodes and branches within each feeder.

[0033] In some aspects, the plurality of individual components in the network are arranged in a geometric configuration selected from the group consisting of: a one-dimensional configuration, a two-dimensional configuration, and an irregular configuration.

[0034] In some aspects, a plurality of individual components in the network are arranged in a two-dimensional configuration, and wherein the two-dimensional configuration is a rectangular configuration, a radial configuration, or a hub-and-spoke configuration.

[0035] In some aspects, the one or more local events include non-technical losses associated with one or more of the plurality of individual components.

[0036] In some aspects, mapping the network includes identifying unexpected voltage levels in individual components of the network.

[0037] In some aspects, the present disclosure may describe a computer-implemented event distribution method, wherein one or more events include opening or closing one or more circuit breaker switches within a power distribution network, and the one or more circuit breaker switches are associated with one or more nodes and / or branches.

[0038] In some aspects, the present disclosure may describe a system for event distribution, comprising a server that communicates with a network comprising a plurality of islands; and a memory that stores instructions that, when executed by the server, cause the server to perform operations including mapping the network, such that the plurality of islands can be dynamically changed by splitting and / or merging one or more islands, and the plurality of islands include a plurality of individual components, and such that one or more local events are detected and located at the individual component level and at the island level using a disaggregated model.

[0039] In some aspects, the present disclosure may describe a non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform an event distribution method, the method comprising mapping a network comprising a plurality of islands, the plurality of islands being capable of dynamically changing by splitting and / or merging one or more islands, such that the plurality of islands comprises a plurality of single components, and such that one or more local events are detected and located at a single component level and at an island level using a disaggregated model. Join by reference

[0040] All publications, patents, and patent applications mentioned in this specification are incorporated herein by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference. To the extent the publications and patents or patent applications incorporated by reference contradict the disclosure contained in the specification, the specification is intended to supersede and / or take precedence over any such conflicting material. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] FIG. 1 is a schematic diagram of an example predictive maintenance system.

[0042] 2 is a diagram of an example power distribution network.

[0043] 3 is a flow diagram of an example process for detecting and locating events occurring in a distributed system using a disaggregated model.

[0044] 4 is a flow chart of an example process for training a latent variable model.

[0045] Figure 5a depicts the splitting, merging, and re-merging of nodes in a one-dimensional distributed system.

[0046] FIG5b is a graph showing the hypothetical probability of an event occurring at each node in FIG5a.

[0047] FIG. 6 illustrates an example of a computer system that is programmed or otherwise configured to implement the methods provided herein. DETAILED DESCRIPTION

[0048] The present disclosure describes systems and methods for event distribution during the splitting and merging of network systems. Networks (such as power grids) can be very complex dynamic systems that require equally complex control systems to maintain normal function. Although the control of the system may be relatively simple during normal operation, the logic of controlling the network may become more complex when unexpected changes (such as component removal or connection failure) occur. The present disclosure provides systems and methods for improving the control of dynamically changing networks, which can be applied to the control of power grids, telecommunications networks, supply chain distribution networks, computer networks, social networks, oil and gas production networks, etc.

[0049] A network, such as an electrical grid, may be comprised of a variety of components in a variety of configurations. For example, in an electrical grid, the network may be comprised of power plants, transmission lines, transformers, substations, and local energy sources such as rooftop solar panels. A network may be highly connected, such as a grid in an urban area, or highly linear and less interconnected, such as a grid in a rural area. Generally speaking, more isolated components in a network system are more vulnerable to attack due to the lack of a fallback connection if another connection fails. Conversely, a highly connected network may be prone to cascading effects, such as power surges associated with power outages, if there are intricate connections between the components. Current challenges and opportunities in the control of network systems are to develop methods to understand the past behavior of the network and apply that understanding to predict the future behavior of the network.

[0050] In a complex network, such as an electrical grid, specific areas of the network with limited connectivity may become isolated if one or more connections fail. Because a subset of network components is isolated from the rest of the network, such events may be referred to as "islands." In highly complex networks, islands may frequently form and then rejoin the network as connectivity is restored. As an extended example, a storm may cause multiple grid failures due to failed power lines, resulting in the formation of isolated islands of multiple outage customers. As power lines are restored, the islands will rejoin, changing the way power must be distributed. In some cases, power line failures may be random, but in some cases, the failures may be associated with particularly vulnerable grid components. The ability to accurately predict possible failures, locate failures at the island / component level, and proactively anticipate changes in power demand as the grid is restored helps improve reliability and reduce grid operating costs.

[0051] The present disclosure describes the use of one or more machine learning algorithms to analyze the behavior of a complex network during an islanding event and predict how the network will react when an island forms and re-merges with the network. The machine learning algorithm can be trained with a data set over time. As the machine learning algorithm becomes better trained, it gains a better ability to predict which specific component within the island system may have failed. Using the predictive results, control of the rest of the network can be adjusted accordingly to take into account the eventual re-merger of the island based on past performance. Relative to conventional network control systems, the present disclosure provides a faster and more efficient way to handle control of a complex network.

[0052] FIG. 1 is a diagram of an example predictive maintenance system 100 configured to assign island-level events to specific components in a network and predict future component-level events. System 100 includes a network 110. Network 110 includes nodes, such as nodes 1 to 10, and branches connecting the nodes to each other, such as branches 21, 22, and 23 that connect node 1 to nodes 2, 6, and 7, respectively. For convenience, only some of the branches in network 110 are identified. In addition, the nodes and branches depicted in network 110 are illustrative only. In general, a network can have any useful combination, as well as any useful number of nodes and branches. For example, a network can have any of the properties, characteristics, or features described in the following paragraphs:

[0053] A network as described herein may include a system of multiple interconnected components. In some aspects, a network may include at least 2 components, at least 3 components, at least 4 components, at least 5 components, at least 10 components, at least 25 components, at least 50 components, at least 100 components, at least 150 components, at least 200 components, at least 250 components, at least 500 components, at least 1000 components, at least 5000 components, at least 10000 components, at least 25000 components, at least 50000 components, at least 100000 components, at least 500000 components, at least 1000000 components, or more. In some aspects, each component may include a source, device, connector, node, switch, or receiver. In some aspects, a component in a networked system may include a dynamic component (e.g., a power plant) so that it actively affects the behavior of the network, or it may be a passive component (e.g., a light bulb) that reacts to the behavior of the network. A network may consist of interconnected components of similar size or behavior, such as a computer network, or it may consist of interconnected components of highly variable size, such as household appliances.

[0054] A network as described herein may be connected directly or indirectly. In some aspects, a network may include a network connection, the network connection including physical links between components, such as power lines in an electrical grid. In other aspects, a network may include remote communication links between components, such as a wireless computer network. In some aspects, a network connection may be constant, static, or permanent. In other aspects, a network connection may be variable, changeable, or reconfigurable. In some aspects, a network connection may have physical or other properties that affect network behavior. In some aspects, a network connection may include network components, such as power lines, transformers, or wireless routers.

[0055] The network referred to herein can also be described as a system of multiple interconnected nodes. In some respects, the network can include at least 2 nodes, at least 3 nodes, at least 4 nodes, at least 5 nodes, at least 10 nodes, at least 25 nodes, at least 50 nodes, at least 100 nodes, at least 150 nodes, at least 200 nodes, at least 250 nodes, at least 500 nodes, at least 1000 nodes, at least 5000 nodes, at least 10000 nodes, at least 25000 nodes, at least 50000 nodes, at least 100000 nodes, at least 500000 nodes, at least 1000000 nodes or more. The node can include one or more components. In some respects, the node can be a single device, connector, node, switch or receiver. In some respects, the node can include a plurality of components grouped together, for example, including the house of multiple electrical appliances and other electrical appliances. In some respects, the node in the network can have one or more connections to other nodes. In some respects, a node may not have the connection to other nodes.

[0056] Networks as described herein can be arranged in a variety of geometric configurations. In some aspects, a network can include a one-dimensional connection of nodes, for example, nodes can be classified as terminals, terminal nodes can have only a connection to one other node, central nodes can have two connections to other nodes, or isolated nodes have no connection to other nodes. In some aspects, the connection between two nodes can be cut off, the central node can become a terminal node, or the terminal node can become an isolated node.

[0057] In some aspects, the network may include a two-dimensional network. In some aspects, a node in a two-dimensional network may include one or more connections to other nodes. In other aspects, an isolated node in a two-dimensional network may not have connections to other nodes. In some aspects, a connection in a two-dimensional network may include a bottleneck if there are no other connections at the end of a selected connection that allow a path between two nodes.

[0058] In some aspects, a network may include a two-dimensional grid that may be spatially represented in a rectangular manner. In some aspects, a rectangular grid may include nodes that have operable connections only to their nearest neighbors. In some aspects, a network may include a two-dimensional grid that may be spatially represented in a radial manner. In some aspects, a radial grid may include multiple linear series of nodes radiating outward from one or more central nodes. In some aspects, a network may include a hub-and-spoke geometry in which connections radiate outward from a central node to satellite nodes. In some aspects, satellite nodes of a hub-and-spoke network may have operable connections. In some aspects, a spoke and hub model may include several nested layers of satellite nodes, whereby concentrically connected rings of satellites may be formed around a hub node.

[0059] In some aspects, the network can include a hybrid geometry of linearly interconnected, rectangularly interconnected, or radially interconnected nodes. In some aspects, the network can include irregular geometries with complex or asymmetric patterns of connectivity. In some aspects, the network can include networks of more than two dimensions, whereby additional dimensions can include different characteristics or physical properties that distinguish aspects of the network, such as a single-phase grid connected to a multi-phase grid.

[0060] A network as described herein may include many types of dynamic systems. The methods of the present disclosure may be applied to any system that satisfies the basic description of a network, such as those described above. The systems described subsequently are exemplary to help describe the types of systems for which the event allocation algorithm may be beneficial.

[0061] A network may be defined by its physical connections. In some aspects, a network may include an electrical grid system whereby electrical power may be distributed from various generation sources to various receivers.

[0062] A network as described herein may be defined by its data connections. In some aspects, a network may include a computer network. A computer network may include components such as a processor, a hard drive, an Ethernet card, an Ethernet cable, a wireless receiver, or a wireless router. In some aspects, a computer network may include a large computer or server system.

[0063] A network as described herein may be defined by spatial connections.In some aspects, a network may include an air traffic control system whereby individual aircraft are operably associated with a particular control station and switch associations when they enter the control area of ​​a different station.

[0064] A network as described herein may be defined by a variety of connection modes. In some aspects, a network may include a manufacturing system. A manufacturing system may include electrical parts, computer parts, conveying devices, material supply systems, parts handling systems, robotic devices, or other interconnected devices.

[0065] A network as described herein may include one or more computers for controlling the network. In some aspects, a computer control system may automatically control the functions of one or more components of a network. In some aspects, a computer control system may require user input as part of its decision-making process. In some aspects, a computer control system may include a reaction system that redistributes and balances network tasks or behaviors when changes occur. In some aspects, a computer control system may include a predictive system that anticipates and preemptively redistributes and balances tasks or behaviors before changes occur. In some aspects, a predictive control system may include a machine learning algorithm.

[0066] A network as described herein may experience an event that changes the behavior or characteristics of the network. An event may include a change, deletion, upgrade, failure or other change to one or more components or nodes within the network. In some aspects, an event may include a failure within the network. A failure may include a cut link, a hardware failure, a data transmission interruption or any other event that inhibits network performance. An event may include an upgrade of the system, such as adding a new component or node, or a transient fluctuation in the output of a component or node with a variable operation. In some aspects, an event may form a bottleneck when the flow between two components or nodes is limited due to a lack of connection. In some aspects, the cutting off of a bottleneck in a network may include an island. By reestablishing a connection between two entities, an island may be re-merged with the main network.

[0067] Returning to Figure 1, network 110 also includes sensors 112, 114 and 116. Sensor 112 collects data at node 1, and sensors 114 and 116 collect data at branches 24 and 25, respectively. Sensors 112-116 can collect different data depending on the network type of network 110. For example, if network 110 is a power distribution system, sensors 112-116 can be voltage sensors that collect voltage data. If network 110 is changed to a supply chain network, sensors 112-116 can be, for example, vehicle sensors or traffic sensors. Sensors 112-116 can communicate with event distribution server 120 via a wired or wireless network 130.

[0068] System 100 also includes an event distribution server 120 that implements disaggregation model 125. Event distribution server 120 may be located at or near any node or branch of network 110, or it may be located remotely. Although referred to as a single server in this disclosure, event distribution server 120 may actually be multiple servers in multiple locations.

[0069] The disaggregation model 125 may be implemented in software, hardware, or a combination of both on the event distribution server 120. The disaggregation model 125 is configured, i.e., trained, to detect, isolate, and locate island-level events occurring in the network 110 using data from the sensors 112-116. That is, the disaggregation model 125 is configured to determine the node or branch in the network 110 that is responsible for or is the source of the event. For example, using the data collected by the sensors 112-116, the trained disaggregation model 125 may determine that the source of the event 140 is the node 3 of the network 110. The disaggregation model 125 may additionally or alternatively use data from an external source (e.g., weather data) to determine the source of the event. Events will be described in more detail later.

[0070] Without the disaggregation model 125, the event distribution server 120 would not be able to localize events to a specific node or branch in the network 110. This is because sensor data is not available at every node and branch. The disaggregation model 125 will be described in more detail with reference to FIGS.

[0071] The event distribution server 125 also includes a training module 127 that helps train the disaggregation model 125. The training module 127 may be a computer program that implements the training process of the disaggregation model 125. The training of the disaggregation model 125 will be described in more detail with reference to FIG.

[0072] FIG. 2 is a diagram of an example distribution network 200. The distribution network 200 includes a power plant 210, a transmission substation 220, a high-voltage transmission line 230, switches 240a-c, a substation 250, a distribution line 260, transformers 270a-d, and consumers 280a-d. The transmission substation 220 can convert the power generated by the power plant 210 into high-voltage power. The high-voltage transmission line 230 can transmit high-voltage power over long distances. The switches 240a-c can isolate parts of the distribution network. The substation 250 can convert the high-voltage power from the high-voltage power line 230 into power that can be safely transmitted by the distribution line 260. And the transformers 270a-d can convert the power from the distribution line into power with a voltage suitable for consumers. Although not depicted, the distribution network may include distribution feeders. The distribution feeder is a connection between the output terminal of the substation 250 and the input terminal of the distribution line 260. The nodes and branches in the network described herein can be associated with distribution feeders and the connected nodes and branches within each distribution feeder. Any of the components described above can be equipped with or integrated with sensors that collect data about the components. Such data can be used by the algorithms described herein. The distribution network is described in more detail in the following paragraphs:

[0073] A power distribution network as described herein may include any system for transmitting electrical energy from a source to a receiver. A power distribution network may include an electrical grid system. In some aspects, a power distribution network may include an interconnected system of electrically connected components whereby electrical current is used to drive or actuate multiple components, such as a factory assembly line. A power distribution network may include an interconnected system of power sources, transmission lines, nodes, switches, and power receivers. A power distribution network may include one or more computers for purposes such as control, monitoring, maintenance, optimization, or any other necessary network function.

[0074] The distribution network as described herein may include one or more power sources. The power source may include any device, component, source or device that feeds electrical energy into the distribution network. The power source may include a large-scale power generation facility, such as a combustion-driven power plant, a gasifier power plant, a nuclear power plant, a hydroelectric power plant, a geothermal power plant, a solar thermal power plant, a photovoltaic power plant, a wind farm or a tidal generator complex. Large-scale power sources may be base loads, peak loads or load-following sources. Large-scale combustion and gasification systems may utilize a variety of fuels, including coal, petroleum, biomass, oil shale and municipal solid waste. The power source may include a small-scale power generation source, such as a rooftop photovoltaic panel, a micro-turbine, a gas generator, a diesel generator or a windmill. The power source may include an electrochemical device, such as a hydrogen fuel cell, a solid-state fuel cell or a battery. The battery system may include an alkaline battery, an aluminum-air battery, a lithium metal battery, a molten salt battery, a nickel-hydrogen battery, a solid-state battery, an aluminum ion battery, a lead-acid battery, a nickel-metal hydride battery, a lithium cobalt battery, a lithium manganese battery, a lithium ion polymer battery, a lithium phosphate battery, a lithium sulfur battery, a magnesium ion battery and a nickel-cadmium battery.

[0075] The power supply can be powered in a variety of ways. The power supply can include a DC power supply. The power supply device can include an AC power supply. In some aspects, the AC power supply can include a single-phase or three-phase power supply. The power supply can include a specific voltage. The DC power supply can include 72V or higher, 48V or higher, 36V or higher, 24V or higher, 18V or higher, 12V or higher, 5V or higher, 1.5V or higher, 1V or higher, 0.1V or higher, or 0.01V or higher. Alternatively, the DC power supply can include at most 72V, at most 48V, at most 36V, at most 24V, at most 18V, at most 12V, at most 5V, at most 1.5V, at most 1V, at most 0.1V or at most 0.01V. The AC power supply can include a specific voltage. The AC power source can include 765 kV or more, 500 kV or more, 230 kV or more, 115 kV or more, 69 kV or more, 240 V or more, 230 V or more, 220 V or more, 120 V or more, 110 V or more, 10 V or more, 1 V or more, or 0.01 V or more. In other cases, the AC power source can include at most 765 kV, at most 500 kV, at most 230 kV, at most 115 kV, at most 69 kV, at most 240 V, at most 230 V, at most 220 V, at most 120 V, at most 110 V, at most 10 V, at most 1 V, or at most 0.01 V. In some aspects, the AC power source can operate at 50 Hz or 60 Hz.

[0076] A power distribution network as described herein may include a plurality of transmission elements that operably connect components or nodes within the network. A transmission element may include components or nodes having properties that may affect the behavior of the network. A transmission element may include any object or device capable of transmitting an electric charge. A transmission element may include a wire, a cable, a conduit, or a wire. A transmission element may include a bundle or collection of multiple individual wires or cables. A transmission element may include a metallic conductor, a superconductor, or a semiconductor. In some aspects, a transmission element may include a power line. In some aspects, a power line may include a high-voltage power line. In other aspects, a transmission element may include a sub-transmission power line, a feeder, or a distribution line. In some aspects, a transmission element may be above the ground. In other aspects, a transmission element may be buried or submerged. In some aspects, a transmission element may be permanently fixed. In other aspects, a transmission element may be detachable or separable.

[0077] A power distribution network as described herein may include a plurality of nodes or switches. A node may include a connection point between two or more transmission elements. A node may include a transmission element that is divided into two or more transmission elements. A node may include two or more transmission elements that are merged to produce a smaller number of transmission elements. In some aspects, a node may include a transformer. In some aspects, a node may include a substation. In some aspects, a node may include a terminal or a connection board. A power distribution network may include a plurality of switches or circuit breakers. A switch may include any element capable of interrupting current or switching current from one transmission element to a different transmission element. A switch may include a manually or mechanically controlled device. A switch may be electronically controlled. A switch may be controlled by the action of a computer. A switch may include a relay, such as a mechanical relay or a solid-state relay. A power distribution system may include a plurality of circuit breakers. A circuit breaker may include any device or component that temporarily or permanently interrupts current. A circuit breaker may include a fuse or a disconnect switch. A circuit breaker may be a passive or active element in a power distribution system.

[0078] Electrical energy may be consumed or dissipated at a variety of receivers. An electrical radiator may include any device that converts electrical energy into mechanical, thermal, or electromagnetic energy. Electrical receivers may include common devices including, but not limited to, computers, lights, washers, dryers, refrigerators, pumps, compressors, microwave ovens, electric stoves, ranges, fans, winches, motors, tools, and electrical radiators. In some aspects, a battery system may act as a source during discharge and as a sink during charge. In some aspects, a sink may include a ground or neutral connection.

[0079] A power distribution network as described herein may include other components. The electrical elements may be passive or active. The electrical components may include resistors, diodes, rectifiers, transistors, light emitting diodes, amplifiers, inductors, fuses, potentiometers, capacitors, or solenoids.

[0080] A power distribution network as described herein may include one or more computers for controlling the network. In some aspects, a computer control system may automatically control the functions of the power distribution network. In some aspects, a computer control system may require user input as part of its decision-making process. In some aspects, a computer control system may include a reactive system that redistributes and balances electrical loads when changes occur. In some aspects, a computer control system may include a predictive system that anticipates and preemptively redistributes and balances electrical loads before changes occur. In some aspects, a predictive control system may include a machine learning algorithm.

[0081] The distribution network as described herein may include a conventional power grid system. The power grid may include the interconnection of power sources, transmission devices, and power receivers in a local area, region, national, or international power grid. The power grid may include one or more large-scale power sources that provide a continuous base load. In some aspects, the power grid may include a network of small-scale devices (e.g., photovoltaic cells) that feed electrical energy to the power grid in a regular or irregular pattern. The power grid may include a synchronous power grid, whereby alternating current appears in the same phase and frequency throughout the power grid. The power grid may include more than one voltage, and the more than one voltage variation is regulated by a device such as a transformer.

[0082] A distribution network as described herein may include a microgrid. A microgrid may include a small-scale or isolated power grid. A microgrid may include conventional energy sources, such as a gas generator. A microgrid may include renewable energy sources, such as a wind turbine. In some aspects, a microgrid may operate independently of a conventional power grid. In other aspects, a microgrid may be connected to a large-scale power grid. A microgrid may include one or more computers that control the functions of the microgrid. In the event that one or more connections between a microgrid and a conventional power grid are severed, the microgrid may be able to self-sustain operation without supply or control from a main grid system.

[0083] The power grid as described herein can be arranged in several geometric shapes. The distribution network can include a radial network. The radial network can include a tree network whereby a single radial branch is subdivided into smaller branches, which in turn can be subdivided into even smaller branches. The distribution network can include a mesh network. The mesh network can include a loop system or a ring system. The distribution network can include more complex geometries that mix aspects of radial and mesh networks.

[0084] A distribution network as described herein may experience events that change the distribution and flow of electrical energy. An event may include a change, deletion, upgrade, failure, or other change to one or more components or nodes within the power network. In some aspects, an event may include a failure within the network. A failure may include a transmission line cut, a power outage, a power station emergency, or a cascading failure. An event may include an upgrade to the system, such as adding a new power source or a transient fluctuation in the power output from a power source such as a wind turbine or photovoltaic cell. In some aspects, an event may form a bottleneck when the flow between two components or nodes is limited due to a lack of connection. In some aspects, the cutting of a bottleneck in a distribution network may include an island. By reestablishing a connection between two regions, an electrical island may be reunited with the main grid.

[0085] Events in the distribution network may include loss events. Loss events may include expected or unexpected power load drops in one or more regions of the power grid. Loss events may include technical losses. Technical losses may include constant sources of losses, such as resistive losses or transformer losses in transmission lines. In some aspects, the power grid control system may consider technical losses in its algorithms. Loss events may include non-technical losses (NTL). NTL may include any unexpected source of loss. In some aspects, NTL may include theft of power service, tampering with meters, meter failures, control system errors, or acts of vandalism of electrical equipment. In some aspects, NTL may not be considered or not well considered in the power grid control system. In other aspects, NTL may be considered by the power grid control system applying machine learning algorithms to detect unexpected losses.

[0086] In some cases, the components described herein may include a power sensing meter. The power sensing meter may be attached to a power supply unit in a building. In some cases, the power sensing meter may be located in a substation, on a utility pole serving a building, or in a meter box adjacent to or in a building. The building may be a residential building or a commercial building. The building may be a residence, or any type of structure that may generate and / or consume electricity.

[0087] A power sensing meter as described herein may include a smart meter and / or a solar monitor.A power distribution network as described herein may include a power monitoring module.

[0088] A smart meter may include one or more sensors configured to measure and record the power consumption of a building in real time or near real time. Electrical appliances and fixtures in a building may consume electricity, and their power consumption may be measured and recorded by a smart meter. For example, a smart meter may be configured to record power consumption at time intervals and provide data to a power monitoring module at a predetermined frequency. The range of the predetermined frequency may be every minute, 5 minutes, 10 minutes, 15 minutes, 30 minutes, 1 hour, 12 hours, or 24 hours. Any frequency range may be envisioned. In some cases, the range of the predetermined frequency may be less than every 1 hour, 30 minutes, 1 minute, 45 seconds, 30 seconds, or 15 seconds. Therefore, power consumption data may include a measure of power consumption as a function of time. For example, the measured power consumption may be timestamped. In some cases, power consumption data may be recorded and stored at a remote database or server.

[0089] In some cases, smart meters can be configured to collect power consumption data for remote reporting to a utility provider for monitoring and / or billing by the utility provider. Smart meters can also be configured to send outage notifications to the utility provider and monitor the quality of power delivered to the building (e.g., power delivery rate, power continuity, voltage peaks, etc.). Smart meters can be connected to the power grid and can be used to determine the power load of the building on the grid.

[0090] Any description of a smart meter may also apply to any type of device used to measure electricity consumption. For example, this may include a built-in meter at a building that sends electricity usage data to a utility company. Any description of a smart meter may apply to a meter that includes "smart" features, such as two-way real-time or near real-time signal communication between the meter and the utility.

[0091] A solar monitor as described herein may include one or more sensors configured to measure and record solar-based power generation in real time, near real time, or intermittently. A solar energy system located in a building may be used to generate electricity. A solar energy system may include one or more solar panels configured to convert sunlight into electrical energy. In some embodiments, the solar energy system may be a photovoltaic system (also referred to as a solar PV power generation system or PV system). A photovoltaic system is an electrical system intended to supply available solar energy with the aid of photovoltaic power generation. A photovoltaic system may consist of several parts, including solar panels that absorb sunlight and convert it into electrical energy, solar inverters that convert current from DC to AC, and installation, wiring, and other electrical accessories for establishing a working system. Photovoltaic systems can generally be divided into three different market segments: utility-scale systems installed on residential rooftops, commercial rooftops, and ground. Their capacity may range from a few kilowatts to hundreds of megawatts. A typical residential system may be around 10 kilowatts and installed on a sloping roof, while a commercial system may reach megawatts and is typically installed on a low-slope or even flat roof.

[0092] The solar monitor described herein can be configured to record electrical energy generation at time intervals and provide data to the power monitoring module at a predetermined frequency. The range of the predetermined frequency can be every minute, 5 minutes, 10 minutes, 15 minutes, 30 minutes, 1 hour, 12 hours or 24 hours. Any frequency range can be envisioned. In some cases, the range of the predetermined frequency can be less than every 1 hour, 30 minutes, 1 minute, 45 seconds, 30 seconds or 15 seconds. Therefore, the power generation data can include a measure of power generation as a function of time.

[0093] In some cases, a solar monitor may be configured to collect solar power generation data. A solar monitor may include any device or system capable of monitoring solar power production data. Solar power generation data may be obtained through one or more inverters located in a solar power system of a building. The inverter may convert the variable direct current (DC) output of a photovoltaic (PV) solar cell panel in a solar power system into industrial frequency alternating current (AC) electricity to power ordinary AC powered equipment. Measurement of the converted current may indicate solar power generation data. Solar power generation data may be sent to a solar power monitoring system. A solar power monitoring system may be configured to store, update, and monitor solar power production data.

[0094] The predictive maintenance system described herein can be used to predict component-level events in networks other than power distribution networks. For example, the predictive maintenance system can be used to predict events in telecommunications networks, supply chain networks, computer networks, oil and gas production networks, or social networks.

[0095] A telecommunications network as described herein may include multiple devices for transmitting data within the network. A telecommunications network may include a telephone network, a computer network, a financial transaction network, a satellite network, and any other interconnected network based on data transmission. A telecommunications network may include multiple data transmission and data acquisition devices that are in direct or indirect operational communication with each other. A telecommunications network may also include devices for data processing and control. In some aspects, a telecommunications network may include a series of interconnected components capable of sending and receiving data, such as an Aircraft Communications Addressing and Reporting System (ACARS) system for communication between an aircraft and a ground control station. In some aspects, a telecommunications network may include a series of interconnected devices that transmit data only to a central acquisition center, such as a network for remote weather sensing.

[0096] A component or node in a telecommunications network may include a data transmission or acquisition device. A component or node in a telecommunications network may be associated with an address that identifies the location of the device within the network. In some aspects, the data transmission or acquisition device may be associated with a fixed address. In other aspects, the data transmission or acquisition device may have a variable address that is reconfigured at random or regular intervals. A component or node may include a data transmission device, such as a sensor or a radio transmitter. A component or node may include a data receiving device, such as a pager, a television, a radio, a printer, or an answering machine. A component or node may include a device that can send and receive data, such as a fax machine, a switch, an Ethernet hub, a wireless router, a computer, a landline phone, a wireless phone, a communication satellite, or a modem.

[0097] Telecommunication network can realize the operable connection of data transmission via electric or electromagnetic signal.In some respects, the connection in telecommunication network can be realized via hard-wired connection (such as Ethernet cable, coaxial cable, fiber optic cable or telephone line).In some respects, the connection in telecommunication network can utilize electromagnetic transmission equipment for data transmission, such as transmitter, receiver, dish type satellite antenna, waveguide, antenna and repeater.In some respects, data transmission can occur via the transmission of packet.In other respects, data can be transmitted in streaming or continuous mode.In some respects, data transmission can be managed by communication protocol (such as TCP / IP, IPX / SPX, X.25 or AX.25).

[0098] A telecommunications network may experience events that create a topological rearrangement of components or nodes and connections within the network. In some aspects, an event may include adding or removing a device from the network. In other aspects, an event may include a change or reallocation of addresses of network components. In some aspects, an event may include a physical or non-physical fault within the network. A physical fault may include a power line outage or an outage due to the electromagnetic effects of a solar storm. A non-physical fault may include an erroneous or faulty transmission. An event in a telecommunications network may be temporary, such as an outage, or permanent, such as removing a device from the network.

[0099] The telecommunications network may be a mesh network. A mesh network is a network in which nodes of the network are directly, dynamically, and non-hierarchically connected to other nodes in the network. The non-hierarchical arrangement of the nodes prevents the failure of one node from affecting a large portion of the network, and allows the nodes to cooperate with each other to efficiently route data to or from clients. Mesh networks may be "self-healing" mesh networks in that when specific nodes fail, such as when an event occurs at the node, they may automatically reconfigure to avoid transmitting data through the node. Data about how the self-healing mesh network automatically reconfigures itself may be used to locate the event.

[0100] Some telecommunication networks may include computer networks. A computer network may include multiple computer devices and devices under the control of one or more computers. A computer network may be a closed system or an open system. An open computer network may include one or more connections to other computer networks. Computer network systems may include personal area networks, local area networks, wireless local area networks, wide area networks, system area networks, enterprise private networks, and virtual private networks.

[0101] A computer network may include multiple components or nodes including interconnected devices. A computer network may include components such as desktop computers, laptop computers, handheld or tablet computers, wireless telephone devices, modems, routers, hubs, servers, printers, monitors, touch screens, and scanners. A computer network may also include hardware components such as robots or programmable logic controllers (PLC) controllers that perform tasks under the control of the network. The components of a computer network may involve a variety of tasks, which involve data transmission, data acquisition, control, signaling, calculation, processing, and display. Components in a computer network, such as display screens, may play a passive role. Components in a computer system, such as process controllers, may play an active role.

[0102] Computer networks may have operable connections via hardwired or wireless connections. Hardwired connections may include Ethernet cables, USB cables, flash drives, pin connectors, and any other form of connection that permits data to be transferred between computer network components. Computer networks may be connected via wireless networks. Wireless connections may include hubs, routers, and any other devices required to transmit electromagnetic signals between computer devices.

[0103] A computer network may experience events that create a topological rearrangement of components or nodes and connections within the network. In some aspects, an event may include adding or removing a computer device from the network. In other aspects, an event may include a change or reallocation of component addresses within the network. In some aspects, an event may include physical or non-physical failures within the network. A physical failure may include any event associated with a hardware outage or disconnection, such as a printer jam. A non-physical failure may include any event associated with a data transmission, acquisition, processing, or output failure. A non-physical failure may include a transmission data failure, such as a server or router outage. Non-physical failures may also include events associated with malicious activity, such as, during a distributed denial of service attack, removing a networked computer through software, forcing isolation of a computer device infected with malicious code, or overloading one or more components through high-volume data transmission. Events in telecommunications networks may be temporary, such as an outage, or permanent, such as removing a device from the network.

[0104] A social network as described herein may include multiple users connected via a relationship framework. A social network may construct a relationship framework based on factors such as geography, occupation, nationality, political preference, family relationships, friendships, or any other factor that distinguishes network components. A social network may have a classification system that includes various levels of classification, including institutions, families, individuals, or a combination thereof.

[0105] In some aspects, components of a social network may include individual users of the social network. In other aspects, components of a social network may include a collection of individual users grouped together within a unified classification, such as a common employer, alumni status, family, hobbies, or any other classification basis. In some aspects, individual users within a social network may be connected to other users via multiple forms of classification.

[0106] Connections in a social network may be self-defined or system-defined. In some aspects, users of the social networking platform may choose how they are categorized and may establish connections with other users within the network accordingly. In other aspects, computer algorithms may form connections between individual users or components based on acquired data or user-supplied data. Connections between components in a social network may be permanent or severable by the user or by a management computer algorithm.

[0107] Events in a social network can include any change in connections within the network. For individual users, events may be associated with life events such as birth, marriage, death, divorce, school enrollment, graduation, employment, promotion, dismissal or layoff, illness, or any other personal event that changes the relationships between individual users in the social network. At the institutional level, connections may change due to events such as mergers, acquisitions, departmental or subsidiary relationships, turnover, leadership changes, reorganizations, or throughput (e.g., class changes at school). Event occurrences in social networks can be highly random, such as the formation of a new friendship, or very predictable, such as the planned availability of an employee for work.

[0108] The oil and gas production network as described herein may include drilling assets, refining assets, and pipeline assets. Drilling assets may include platforms, drilling rigs, drill bits, casing, pumps, etc. Drilling assets may be onshore or offshore assets. Refining assets may include any machine or equipment commonly found in refineries, including distillation units (e.g., crude distillation units and vacuum distillation units), catalytic reformers, hydrocrackers, processors (e.g., amine processors, merox processors, hydroprocessors, etc.), etc. Pipeline assets may include pipelines, pumps, compressors, heat exchangers, valves, etc.

[0109] Nodes in an oil and gas production network may be drilling and refining assets. Nodes may be connected by pipeline assets and other supply chain infrastructure, such as roads, railroads, waterways, and vehicles that use these roads, railroads, and waterways. Events in an oil and gas production network may be mechanical or electrical failures. For example, an event may be a failure of a drilling, refining, or pipeline asset. Additionally or alternatively, an event may be a weather event, such as weather that prevents drilling or causes a supply chain disruption. Additionally or alternatively.

[0110] Drilling, refining, and pipeline assets may be equipped with sensors that collect data about such assets. The sensors may be thermometers, pressure gauges, flow meters, accelerometers, magnetometers, etc. The data collected by such sensors may be used by the algorithms described herein to predict and locate events in the oil and gas production network.

[0111] A supply chain network as described herein includes multiple physical locations, routes between the physical locations, vehicles to navigate the routes, and personnel to work at the physical locations and operate the vehicles. Supply chain networks typically rely on coordination between multiple entities (e.g., business organizations operating over thousands or millions of square miles in many countries). Supply chain networks are configured to efficiently transport goods from one or more sources to multiple destinations.

[0112] Nodes in a supply chain network can include: supplier manufacturing plants, assembly plants, regional distribution centers, local distribution centers, and destinations, such as residential or commercial customers. Branches in a supply chain network can include roads, railroads, waterways, and flight paths that connect nodes in the supply chain network together.

[0113] The supply chain network also includes the vehicles that move goods along the supply chain network routes, as well as the people who operate the vehicles and work in the factories and distribution centers. The vehicles in the supply chain network can include trucks, trains, ships, airplanes, etc. Typically, each vehicle is operated by one or more people. Manufacturing plants and distribution centers also usually employ many people to facilitate the manufacturing process, organize the goods, and load the goods into the vehicles.

[0114] Operators of supply chain networks can collect data about the supply chain network from a variety of sources. For example, they can collect weather data from Doppler radar sensors or public sources; traffic data from vehicles in the supply chain network or from public sources such as Google Maps or Waze; vehicle data, such as mileage or maintenance records of vehicles in the chain network; and manufacturing equipment data. Because supply chain networks can span hundreds or thousands of miles and multiple countries, it is often not feasible to collect data from every node, branch, and vehicle in the supply chain network.

[0115] An event in a supply chain network, i.e., a failure, can disrupt the supply chain network and cause costly shipping delays. For a business customer, a disruption in the supply chain network can result in a loss of production, e.g., because a component of a product is unavailable. Therefore, predicting and preventing events before they occur can be beneficial.

[0116] Events in the supply chain network can include weather events or similar natural disasters, such as hurricanes, blizzards, floods, forest fires, etc. Weather events can make routes impassable, thereby breaking the supply chain network into isolated islands. However, the operator of the supply chain may not know the precise location of the impossible roads. For example, the operator may know that the blizzard affected a specific city or region, but may not know which specific roads are impassable. The operator may also know the roads that were affected by blizzards in the past. Based on this limited data, and using the algorithms disclosed in this article, the operator can predict specific roads that are impassable, and adjust the transportation routes or send snow removal equipment to these roads accordingly.

[0117] Events in a supply chain network can also include vehicle failures. For example, a truck may break down or run out of fuel. Using limited prior data, such as vehicle maintenance data and data about which vehicles have broken down in the past, and using the algorithms described herein, operators of the supply chain network can predict which vehicles will break down and perform preventive maintenance on those vehicles. Similarly, supply chain operators can predict equipment failures at manufacturing plants.

[0118] Events in a supply chain network may also include traffic incidents. In some cases, an incident may render an entire road impassable. In other cases, an incident may only affect a specific mode of transport. Examples of traffic incidents include traffic accidents, construction, and large public events. Traffic incidents can delay the transportation of goods from source to destination by increasing transit time.

[0119] Events in the supply chain network can also include worker shortages or strikes. Lack of workers can cause increased shipping times.

[0120] Events in a supply chain network can be localized to a specific branch, such as a road, waterway, rail or flight path, a specific vehicle, a specific plant or distribution center, or a specific supplier.

[0121] Events in a supply chain network, like events in a distribution network, can cause parts of the supply chain network to become isolated from each other. Such isolated parts are referred to as "islands" in this specification. As routes are restored, the islands reassemble. The ability to accurately predict possible failures in a supply chain network and locate the failures at the node / route level helps improve the reliability and timeliness of the supply chain network.

[0122] FIG3 is a flow chart of an example process 300 for detecting and locating events occurring in a network using a disaggregated model. Process 300 may be performed by one or more appropriately programmed computers configured to collect data about the network, i.e., event data. For example, process 300 may be performed by the event distribution server 120 depicted in FIG1 . For convenience, process 300 will be described as being performed by a single server. Process 300 may be adapted or modified for detecting and locating events in any of the network types discussed above (including power distribution networks, telecommunications networks, computer networks, supply chain power distribution networks, social networks, oil and gas production networks, etc.).

[0123] The server observes island level events in a network (310). The network may be a power distribution network, a telecommunications network, a computing network, a supply chain network, a social network, an oil and gas production network, or any other type of network.

[0124] Island events are events that occur in a network and can be directly traced back to an island, but not to a component within the island. An island is an isolated partition of a network. Islands are formed when connectivity in a network is disrupted (e.g., when a distribution line in a power distribution network is disrupted, a component in a telecommunications network or computing network fails, or a route in a supply chain network becomes inaccessible). Similarly, two or more islands can merge when connectivity in a network is restored, for example, when a power company repairs a failed distribution line in a power distribution network, when a failed component in a telecommunications network or computing network is repaired, or when an inaccessible route is cleared. Thus, the islands that make up a network can change dynamically by splitting or merging with each other.

[0125] The dynamic change of an island may involve multiple instances of island splitting, merging, or re-merging. Generally speaking, the splitting of an island involves splitting two or more nodes or branches in the island, while merging two or more islands involves merging nodes or branches from each of the two or more islands. In other words, the dynamic change of an island may involve the splitting and / or merging of subsets of individual components of the island. The subsets of individual components may be the same size, or the subsets of individual components may be of variable size.

[0126] Each island may include one or more individual components. Generally speaking, individual components in an island may be represented as nodes and one or more branches to or from these nodes. More specifically, for example, an island in a distribution network may include one or more of the following: a power plant, a transmission substation, a high-voltage transmission line, a switch, a circuit breaker, a substation, a feeder or distribution line, a local transformer, and a consumer. An island in a computing network may include a subset of servers, processors, interconnections, wireless routers, and wireless receivers in a distributed computing network. And an island in an automation equipment system in a manufacturing plant may include a subset of such automation equipment. In some embodiments, an island is a smaller partition, for example, a partition that includes only subcomponents of the components listed above. For example, an island in a distribution network may include only a subset of generators in a power plant, or a subset of transformers in a transmission substation. Any level of abstraction is possible.

[0127] Generally speaking, an event is a change in the operational state of one or more individual components in a network. For example, an event can be a failure, deletion, upgrade, change, or other similar change of a component. Examples of failures include branch outages, hardware failures, data transmission path interruptions, and other similar events that inhibit network performance. Removal may involve intentionally removing a component temporarily or permanently from a network, for example, temporarily removing a component for maintenance. An upgrade may be the addition of a new component to a network or a transient fluctuation in component output that occurs naturally with variable operation. And a change may involve replacing one or more other components with one or more components. In some embodiments, an event may form a bottleneck when the flow between two components or nodes becomes restricted due to a lack of connection.

[0128] An example of an event in a power distribution network is the opening or closing of a circuit breaker switch within the power distribution network, where the circuit breaker switch is associated with one or more of the nodes and / or branches of the power distribution network. Other examples of events in a power distribution network include generator failures, transmission line and distribution line failures, transformer failures, circuit breaker overloads, etc. Examples of events in a distributed computing system include hard drive failures, memory overloads, race conditions, etc.

[0129] An event can lead to other events, or even a series of cascading events. For example, a failed transmission line in a distribution network can cause a circuit breaker in another part of the distribution network to overload. In some cases, a single event can cause an entire island or the entire distribution network to fail.

[0130] The server can "observe" the event by collecting data that defines the event. Typically, the server cannot immediately identify the specific component in the network that caused or was the source of the event. This is because the server does not receive data from every node and branch in the network. For example, it is impractical and costly to equip the entire distribution network with necessary sensors, which may include thousands of electrical equipment, hundreds of miles of transmission and distribution lines, and thousands of consumers. Similarly, it is impractical to equip the entire supply chain network with sensors, which may cover thousands or millions of square miles of land, sea, and air space across multiple countries. However, the server can determine the general area in the network where the event occurred based on the data collected by a subset of nodes in the network. For example, the server can receive transmission line and distribution line data from a subset of transmission and distribution lines. The data may include line temperature and vibration, instantaneous and average current and voltage measurements, etc. Similarly, a subset of transformers, switches, circuit breakers, and meters in the distribution network can be equipped with sensors that transmit data to the server. The sensor can communicate with the server via a wired or wireless network.

[0131] By comparing data from different nodes in the network, the server can generally attribute or assign an event to a particular island. The server may also use the data to map the network, particularly when the server does not have a pre-existing map of the network, or when an event changes the network configuration. Mapping the network may involve identifying isolated partitions of the network, i.e., islands, and their boundaries. Islands may be identifiable and distinguishable from one another based on one or more characteristics or attributes of the data received by the server. For example, in the context of an electrical distribution network, islands may be identified and distinguished based on unexpected voltage levels in a particular portion of the network.

[0132] The server provides data describing island-level events to a disaggregated model (220). Generally speaking, a disaggregated model is a model that divides a system into several parts. In the context of this disclosure, these parts are the individual parts that make up the island.

[0133] The disaggregation model is configured, i.e., trained, to use data to detect, isolate, and locate events at the level of individual components. That is, the disaggregation model is configured to determine the individual components in the island that are responsible for the event or the source of the event. For example, if a transmission line fails, e.g., bends under the weight of heavy snow during a snowstorm, a particular distribution line in the distribution network may be the source of the event. Using island-level data describing the event, the disaggregation model is configured to identify the distribution line as the source of the event. As another example, if a server memory in a distributed computing network exceeds capacity and interrupts a data path in the distributed computing network, the server memory may be the source of the event. Using island-level data describing the event, e.g., data identifying interrupted data paths in a distributed computing system, the disaggregation model is configured to identify the memory as the source of the event. The training of the disaggregation will be described in more detail with reference to FIG. 4 .

[0134] The disaggregation model is configured to detect one or more local events at the level of individual components when one or more of the islands dynamically change. It is for this reason that it is important to assign or attribute events to individual components—because islands can split, merge, and re-merge, it is difficult to keep track of which events belong to which islands.

[0135] In some embodiments, the disaggregated model is a machine learning latent variable model. A latent variable is a variable that is not directly observed, but is inferred from other observed variables. A classic example of a latent variable is general intelligence. General intelligence cannot be directly observed and measured, but it can be informed by other observed variables (such as standardized test scores, high school or college grade point average, subjective assessments, or a combination thereof). Another example of a latent variable is overall health. Similarly, although overall health cannot be measured directly, it can be understood through many other observed variables, including body mass index, resting heart rate, blood pressure, cholesterol levels, etc.

[0136] In the context of the networks described in this article, the components responsible for a particular event or the source of a particular event are typically unobservable and are therefore latent variables that can be modeled. Although it is theoretically possible to observe the individual components that cause an event, doing so is impractical for the reasons described previously. However, (i) the components responsible for a particular event and (ii) the probability that a particular component is the source of a given event can be inferred from other observable data organized into a latent variable model.

[0137] A machine learning latent variable model is a statistical model that (i) models latent variables similar to those described above and (ii) is refined over time through training, typically in an unsupervised learning setting. In the context of the networks described in the present disclosure, a machine learning latent variable model may be implemented at the component level. That is, the machine learning latent variable model may include latent variables, each of which corresponds to the probability of a local event occurring at each of the individual components.

[0138] For convenience, the machine learning latent variable model will be referred to as the latent variable model. The training of the latent variable model will be described in more detail with reference to FIG.

[0139] Latent variable models typically operate under the assumption that the observable variables are only the result of their location on the latent variable and that after controlling for the latent variable, the observable variables have nothing in common. This is called local independence. For example, a latent variable model for an electric distribution network may operate under the assumption that voltage differences are only the result of being on the specific island where the component causing the event is located.

[0140] An example of a latent variable model is a mixture model. A mixture model is a probabilistic model used to represent the presence or absence of a subpopulation within a larger population. That is, a mixture model can be used to make statistical inferences about the properties of a subpopulation, given only observations of the entire population, without information about the identity of the subpopulation. The model does not require observational data to identify the subpopulation to which a single observation belongs. In the context of the networks described herein, an observed island event may be the large population of the mixture model, and the individual components responsible for the event may be the current but unobserved subpopulation.

[0141] In a classic example, mixture models are used to model housing prices. Generally, prices of different types of houses in different neighborhoods vary widely, but, for example, prices of a particular type of house in a particular neighborhood tend to cluster together. One possible model for such prices is one that assumes that (i) the prices are exactly described by a mixture model with k different components, (ii) each component has a normal distribution, and (iii) each component specifies a particular combination of house type / neighborhood. Fitting this model to observed prices, for example using an expectation-maximization algorithm, will tend to cluster prices by house type / neighborhood and reveal the distribution of prices within each house type / neighborhood.

[0142] This same approach applies to the network described in this paper. A possible model for events in the network is to assume that: (i) island-level events can be traced back to one of k different components within the island, (ii) each component has a normal distribution, and (iii) each component of the model specifies a physical component within the island. Fitting this model to observed island-level events, for example using an expectation-maximization algorithm, will tend to cluster the observed events by the component responsible for the event and reveal the likelihood that a given event originated from a specific component.

[0143] Another example of a latent variable model is a Bayesian model. Generally speaking, a Bayesian model is a probabilistic graphical model that represents a set of variables or features and their conditional dependencies. A Bayesian model can use a directed acyclic graph (DAG) to represent variables or features. Each variable or feature can be represented by a node in the Bayesian model. In one example, if a node is affected by or depends on another node, there can be a connection, link, or edge between a first node and a second node. If a first node is affected by or depends on a second node, the connection can be directed from the second node to the first node. If a second node is affected by or depends on a first node, the connection can be directed from the first node to the second node.

[0144] The Bayesian model can represent the probabilistic relationship between island-level events. In one example, when limited data about island-level events are available, the Bayesian model can be used to determine, infer or predict the individual components responsible for the island-level events.

[0145] The inference of the single component responsible for the event can be based on at least some data provided to the Bayesian model. For example, some island-level event data can be provided to Bayes. In some implementations, the single component responsible for the event in the inference island includes applying the maximum a posteriori estimation process to the Bayesian model to obtain multiple energy consumption values ​​of multiple energy consumption sources associated with the user. Generally speaking, with respect to the Bayesian network model, the maximum a posteriori estimation process can correspond to the mode of the posterior distribution. The maximum a posteriori estimation process can be used to obtain a point estimate of an unobserved quantity based on empirical data (e.g., island-level data from previous events), such as the probability that a specific component is responsible for the event. In one example, one or more of the probabilities of a single component causing a given event in the island are unknown. The maximum a posteriori estimation process can determine, infer or estimate the most likely value of one or more unknown probabilities based on available information (such as limited data island event data). Therefore, the maximum a posteriori estimation process can be used to determine, infer or estimate the probability of disaggregation. It should be understood that many variations are possible. In addition, in some embodiments, the inference of the probability of multiple single components can be performed without installing additional sensors in the power distribution network.

[0146] Referring again to FIG. 3 , the server receives detections of local events at the component level and island level from the disaggregated model (which may be any of the models described above) ( 330 ). The detections may be used to perform predictive maintenance on the components. In some implementations, process 300 is used to predict component level events before they occur and perform maintenance on the corresponding components.

[0147] FIG. 4 is a flow chart of an example process 400 for training a latent variable model, such as one or more of the latent variable models described with reference to FIG. 3 . Specifically, process 400 corresponds to an expectation-maximization algorithm. The expectation-maximization algorithm is an unsupervised learning algorithm because no labeled training examples are required. In fact, because the latent variable model models unobserved variables, it is often impractical or impossible to provide labels for the training examples. Process 400 may be performed by the training module 127 depicted in FIG. 1 .

[0148] Originally, the expectation-maximization algorithm is an iterative method for finding the maximum likelihood or maximum a posteriori estimates of parameters in a statistical model where the model depends on unobserved latent variables. In the context of the networks described in this article, the unobserved latent variables are the individual components that cause island events, and the parameters are the probabilities that an individual component is the source of a given island event.

[0149] The maximum a posteriori probability estimate is an estimate of an unknown quantity that is equal to the posterior distribution mode of the unknown quantity. The maximum a posteriori probability can be used to obtain point estimates of unobserved quantities based on empirical data (such as island-level data). It is closely related to the maximum likelihood estimation method, but uses an enhanced optimization objective that incorporates a prior distribution of the quantity to be estimated (quantifying the additional information available through prior knowledge of related events). Therefore, the maximum a posteriori estimate can be seen as a regularization of the maximum likelihood estimate.

[0150] The expectation-maximization iterations alternate between performing an expectation step, which creates a function for the expectation of the log-likelihood evaluated using the current estimates of the parameters (e.g., component probabilities), and a maximization step, which computes the parameters found in the expectation step that maximize the expected log-likelihood. These parameter estimates are then used to determine the distribution of the latent variables in the next expectation step.

[0151] In other words, one uses prior knowledge of related events to model the probability distribution of future events. In the context of the network described in this article, this means that the probability that each component in the island is responsible for the island-level event or is the source of the island-level event is pre-computed or pre-estimated. The pre-computed probabilities are actually probability distributions. When an event is observed, the model infers the probability that a specific component causes the event, and updates the pre-computed probability distribution based on the inference and the new event.

[0152] The training module estimates the prior probability of each component on the island causing the event (410). Generally, the prior probability is the probability distribution of the observed variable. The prior probability can be an initially known, actual or estimated probability based on prior event data.

[0153] The training module implemented on the server observes events on the island (420). As previously described, observing events involves collecting island-level data that defines the events. However, the data does not include the components to which the events should be assigned, i.e., the latent variables being modeled.

[0154] Using the available island-level event data and the prior probability distribution, the training module performs an exact a posteriori analysis to determine the probability that an observed event is caused by each component on the island (430). A posteriori inference is an estimate of an unknown quantity that is equal to the mode of the posterior distribution, which is the conditional probability assigned to a latent variable (e.g., a single component in the island that is responsible for an island-level event after taking into account relevant evidence or context).

[0155] The training module updates the prior probabilities based on the new event data and the inferences (440).

[0156] Steps 420 to 440 are repeated until convergence, ie, until the latent variable distribution changes by less than some threshold amount when a new event is observed ( 450 ).

[0157] Figure 5a depicts a one-dimensional example of island splitting and merging in a network. At time T = 1, the network has three islands A, B, and C. Island A consists of nodes 1 and 2, node 3, and island C consists of nodes 4 and 5.

[0158] At time T=2, due to the event between nodes 1 and 2 at time T=1, node 2 splits from node 1 and merges with node 3. The result is a new island D consisting of only node 1 and a new island E consisting of nodes 2 and 3. Node C remains unchanged.

[0159] At time T=3, due to the event between nodes 2 and 3 at time T=2, node 2 re-merges with node 1, and node 3 merges with nodes 4 and 5. The result is that island A is re-formed, and nodes 3, 4, and 5 form a new island F.

[0160] At time T=4, due to the event at node 4 at time T=3, node 3 splits from node 4, causing all original islands A, B, and C to reform.

[0161] Figure 5b plots the hypothetical probability of an event occurring at each node at each time.

[0162] The present disclosure provides a computer system programmed to implement the method of the present disclosure.

[0163] FIG. 6 illustrates a computer system 601 that is programmed or otherwise configured to implement the methods provided herein.

[0164] Computer system 601 can adjust various aspects of the present disclosure, for example, obtain an inventory data set including multiple inventory variables, apply a trained algorithm to the inventory data set to generate a forecast for a variable with future uncertainty, and apply an optimization algorithm to the inventory data set to optimize the multiple inventory variables. Computer system 601 can be a user's electronic device or a computer system located remotely relative to the electronic device. The electronic device can be a mobile electronic device.

[0165] The computer system 601 includes a central processing unit (CPU, also referred to herein as a "processor" and "computer processor") 605, which may be a single-core or multi-core processor, or multiple processors processing in parallel. The computer system 601 also includes a memory or memory location 610 (e.g., random access memory, read-only memory, flash memory), an electronic storage unit 615 (e.g., a hard disk), a communication interface 620 (e.g., a network adapter) for communicating with one or more other systems, and peripherals 625, such as cache, other memory, data storage, and / or an electronic display adapter. The memory 610, the storage unit 615, the interface 620, and the peripherals 625 communicate with the CPU 605 via a communication bus (solid lines) such as a motherboard. The storage unit 615 may be a data storage unit (or data warehouse) for storing data. The computer system 601 may be operatively coupled to a computer network ("network") 630 with the help of the communication interface 620. The network 630 may be the Internet, an intranet and / or an extranet, or an intranet and / or an extranet in communication with the Internet.

[0166] In some cases, network 630 is a telecommunications and / or data network. Network 630 may include one or more computer servers that may implement distributed computing, such as cloud computing. For example, one or more computer servers may enable cloud computing on network 630 ("cloud") to perform various aspects of the analysis, calculation, and generation of the present disclosure, such as obtaining an inventory data set including multiple inventory variables, applying a trained algorithm to the inventory data set to generate predictions for variables with future uncertainty, and applying an optimization algorithm to the inventory data set to optimize multiple inventory variables. Such cloud computing may be provided by cloud computing platforms, such as Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform, and IBM Cloud. In some cases, with the help of computer system 601, network 630 may implement a peer-to-peer network that enables devices coupled to computer system 601 to act as clients or servers.

[0167] The CPU 605 may execute a series of machine-readable instructions, which may be embodied in a program or software. The instructions may be stored in a memory location, such as the memory 610. The instructions may be directed to the CPU 605, which may then be programmed or otherwise configured to implement the methods of the present disclosure. Examples of operations performed by the CPU 605 may include fetching, decoding, executing, and writing back.

[0168] CPU 605 may be part of a circuit, such as an integrated circuit. One or more other components of system 601 may be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0169] Storage unit 615 can store files, such as drivers, libraries, and saved programs. Storage unit 615 can store user data, such as user preferences and user programs. In some cases, computer system 601 may include one or more additional data storage units located external to computer system 601, such as on a remote server that communicates with computer system 601 via an intranet or the Internet.

[0170] Computer system 601 can communicate with one or more remote computer systems via network 630. For example, computer system 601 can communicate with a user's remote computer system. Examples of remote computer systems include personal computers (e.g., portable PCs), tablet computers or tablet PCs (e.g., Apple® iPad, Samsung® Galaxy Tab), phones, smart phones (e.g., Apple® iPhone, Android-enabled devices, Blackberry®), or personal digital assistants. Users can access computer system 601 via network 630.

[0171] The methods described herein may be implemented by machine (e.g., computer processor) executable code stored in an electronic storage location of the computer system 601, such as stored in the memory 610 or electronic storage unit 615. The machine executable or machine readable code may be provided in the form of software. During use, the code may be executed by the processor 605. In some cases, the code may be retrieved from the storage unit 615 and stored in the memory 610 for access by the processor 605 at any time. In some cases, the electronic storage unit 615 may be excluded, and the machine executable instructions are stored on the memory 610.

[0172] The code may be precompiled and configured for use with a machine having a processor suitable for executing the code, or may be compiled during runtime. The code may be provided in a programming language, which may be selected to enable the code to be executed in a precompiled or compiled form.

[0173] Aspects of the systems and methods provided herein, such as computer system 601, can be embodied in programming. Various aspects of the technology can be considered as "products" or "articles of manufacture", which are generally present in or contained in a machine-readable medium in the form of machine (or processor) executable code and / or related data. The machine executable code can be stored on an electronic storage unit, such as a memory (e.g., read-only memory, random access memory, flash memory) or a hard disk. "Storage" type media can include any or all tangible memories of a computer, processor, etc. or its related modules, such as various semiconductor memories, tape drives, disk drives, etc., which can provide non-transitory storage for software programming at any time. All or part of the software can sometimes be communicated through the Internet or various other telecommunications networks. For example, such communication can enable software to be loaded from one computer or processor to another computer or processor, such as from a management server or host computer to a computer platform of an application server. Therefore, another type of media that can carry software elements includes optical waves, radio waves, and electromagnetic waves, such as across physical interfaces between local devices, through wired and optical landline networks, and through various air links. The physical elements that carry such waves, such as wired or wireless links, optical links, etc., can also be considered to be the medium that carries the software. As used herein, unless limited to non-transitory, tangible "storage" media, terms such as computer or machine "readable media" refer to any medium that participates in providing instructions to a processor for execution.

[0174] Thus, machine-readable media, such as computer executable code, can take a variety of forms, including but not limited to tangible storage media, carrier media, or physical transmission media. Non-volatile storage media include, for example, optical or magnetic disks, such as any storage device in any computer, etc., such as can be used to implement the database shown in the figure, etc. Volatile storage media include dynamic memory, such as the main memory of such a computer platform. Tangible transmission media include coaxial cables; copper wire and optical fiber, including the wires that make up the bus within a computer system. Carrier transmission media can take the form of electrical or electromagnetic signals, or acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Thus, common forms of computer-readable media include, for example: floppy disks, diskettes, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, DVDs or DVD-ROMs, any other optical media, punched card tapes, any other physical storage media with patterns of holes, RAM, ROM, PROM and EPROM, FLASH-EPROM, any other memory chip or cassette, carrier waves that transmit data or instructions, cables or links that transmit such carriers, or any other medium from which a computer can read programming code and / or data. Many of these forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0175] The computer system 601 may include or communicate with an electronic display 635 that includes a user interface (UI) 640. Examples of user interfaces (UIs) include, but are not limited to, graphical user interfaces (GUIs) and web-based user interfaces. For example, the computer system may include a web-based dashboard (e.g., GUI) configured to display, for example, a BOM to a user.

[0176] The methods and systems of the present disclosure may be implemented by one or more algorithms. The algorithms may be implemented by software when executed by the central processing unit 605. The algorithms may, for example, obtain an inventory data set including a plurality of inventory variables, apply a trained algorithm to the inventory data set to generate a forecast for a variable with future uncertainty, and apply an optimization algorithm to the inventory data set to optimize the plurality of inventory variables.

[0177] Although the description has been made with respect to specific embodiments thereof, these specific embodiments are merely illustrative and not restrictive. The concepts illustrated in the examples may be applied to other examples and embodiments.

[0178] Although preferred embodiments of the present invention have been shown and described herein, it is readily understood by those skilled in the art that such embodiments are provided by way of example only. It is not intended to limit the present invention by the specific examples provided in the specification. Although the present invention has been described with reference to the foregoing description, the description and illustration of the embodiments herein are not meant to be interpreted in a restrictive sense. Without departing from the present invention, those skilled in the art will now think of many changes, modifications and substitutions. In addition, it should be understood that all aspects of the present invention are not limited to the specific description, configuration or relative proportions set forth herein, which depend on various conditions and variables. It should be understood that various alternatives to the embodiments of the present invention described herein may be used to implement the present invention. Therefore, it is contemplated that the present invention will also cover any such substitutions, modifications, changes or equivalents. It is intended that the scope of the present invention is defined by the claims, and methods and structures and their equivalents within the scope of these claims are thus covered.

Claims

1. A computer-implemented method for training a machine learning model, comprising: Island-level event data is received, the island-level event data describing island-level events occurring at an island, the island being capable of dynamically changing by splitting or merging with another island, wherein: The island is part of a network and includes one or more individual components, the network including a plurality of islands, the plurality of islands including the another island; and The island-level event data cannot immediately identify a source single component of the island-level event from among the one or more single components; estimating a priori probabilities of each of the one or more individual components causing the island-level event; performing a posteriori inference using the machine learning model and based on the island-level event data and the prior probabilities of each of the one or more individual components to determine a probability estimate of each of the one or more individual components causing the island-level event; updating the prior probability of each of the one or more individual components based on the result of the execution of the a posteriori inference; and repeatedly performing the receiving of the island-level event data, the performing of the a priori inference, and the updating of the prior probability until a difference between the updated prior probability and the estimated prior probability of each of the one or more individual components is less than a threshold difference, Wherein, after the execution, the machine learning model is trained to receive the island-level event data describing the island-level event occurring at the island, and based on the island-level event data received when the island dynamically changes, the source single component is identified as the cause of the island-level event. The method according to claim 1 , wherein the a posteriori inference is an exact a posteriori inference. 3 . The method of claim 1 , wherein the a priori probability comprises an initially known, actual, or estimated probability of each of the one or more individual components causing the island-level event based on a priori event data. The method of claim 1 , wherein the prior probability of each of the one or more individual components is iteratively updated using an Expectation-Maximization (EM) algorithm.

5. The method of claim 1, wherein the network comprises a power distribution network including a plurality of nodes and one or more branches, and the one or more individual components of the island comprise at least a portion of the plurality of nodes and the one or more branches. 6 . The method of claim 1 , wherein the island-level event data describing the island-level events occurring at the island is collected from a subset of the one or more individual components of the island.

7. The method of claim 1, wherein the machine learning model is a machine learning latent variable model, including a Bayesian model or a hybrid model.

8. A system for training a machine learning model, comprising: processor; as well as A storage device coupled to the processor and comprising instructions that, when executed by the processor, configure the processor to perform operations comprising: Island-level event data is received, the island-level event data describing island-level events occurring at an island, the island being capable of dynamically changing by splitting or merging with another island, wherein: The island is part of a network and includes one or more individual components, the network including a plurality of islands, the plurality of islands including the another island; and The island-level event data cannot immediately identify a source single component of the island-level event from among the one or more single components; estimating a priori probabilities of each of the one or more individual components causing the island-level event; performing a posteriori inference using the machine learning model and based on the island-level event data and the prior probabilities of each of the one or more individual components to determine a probability estimate of each of the one or more individual components causing the island-level event; updating the prior probability of each of the one or more individual components based on the result of the execution of the a posteriori inference; and The receiving of the island-level event data, the performing of the a priori inference, and the updating of the prior probability are repeatedly performed until a difference between the updated prior probability and the estimated prior probability of each of the one or more individual components is less than a threshold difference, wherein: After the execution, the machine learning model is trained to receive the island-level event data describing the island-level event occurring at the island, and based on the island-level event data received when the island dynamically changes, the source single component is identified as the cause of the island-level event.

9. The system of claim 8, wherein the a posteriori inference is an exact a posteriori inference.

10. The system of claim 8, wherein the a priori probability comprises an initially known, actual, or estimated probability of each of the one or more individual components causing the island-level event based on a priori event data. 11 . The system of claim 8 , wherein the prior probability of each of the one or more individual components is iteratively updated using an expectation-maximization (EM) algorithm.

12. The system of claim 8, wherein the network comprises a power distribution network including a plurality of nodes and one or more branches, and the one or more individual components of the island comprise at least a portion of the plurality of nodes and the one or more branches.

13. The system of claim 8, wherein the island-level event data describing the island-level events occurring at the island is collected from a subset of the one or more individual components of the island.

14. The system of claim 8, wherein the machine learning model is a machine learning latent variable model, including a Bayesian model or a hybrid model.

15. A non-transitory event distribution computer readable medium (CRM) storing a program for causing a computer to perform operations including: Island-level event data is received, the island-level event data describing island-level events occurring at an island, the island being capable of dynamically changing by splitting or merging with another island, wherein: The island is part of a network and includes one or more individual components, the network including a plurality of islands, the plurality of islands including the another island; and The island-level event data cannot immediately identify a source single component of the island-level event from among the one or more single components; estimating a priori probabilities of each of the one or more individual components causing the island-level event; performing a posteriori inference using a machine learning model and based on the island-level event data and the prior probabilities of each of the one or more individual components to determine a probability estimate of each of the one or more individual components causing the island-level event; updating a priori probabilities of each of the one or more individual components based on results of the execution of the a posteriori inference; as well as The receiving of the island-level event data, the performing of the a priori inference, and the updating of the prior probability are repeatedly performed until a difference between the updated prior probability and the estimated prior probability of each of the one or more individual components is less than a threshold difference, wherein: After the execution, the machine learning model is trained to receive the island-level event data describing the island-level event occurring at the island, and based on the island-level event data received when the island dynamically changes, the source single component is identified as the cause of the island-level event.

16. The non-transitory event allocation computer-readable medium of claim 15, wherein the a posteriori inference is an exact a posteriori inference.

17. The non-transitory event allocation computer readable medium of claim 15, wherein the a priori probability comprises an initially known, actual, or estimated probability of each of the one or more individual components causing the island-level event based on a priori event data. 18 . The non-transitory event allocation computer readable medium of claim 15 , wherein the a priori probability of each of the one or more individual components is iteratively updated using an expectation-maximization (EM) algorithm.

19. The non-transitory event allocation computer-readable medium of claim 15, wherein the network comprises a power distribution network comprising a plurality of nodes and one or more branches, and the one or more individual components of the island comprise at least a portion of the plurality of nodes and the one or more branches.

20. The non-transitory event allocation computer-readable medium of claim 15, wherein the island-level event data describing the island-level events occurring at the island is collected from a subset of the one or more individual components of the island.

21. The non-transitory event allocation computer-readable medium of claim 15, wherein the machine learning model is a machine learning latent variable model, including a Bayesian model or a hybrid model.

Images