Power grid assets prediction using generative adversarial networks

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using a neural network to predict locations of feeders in an electrical power grid. One of the methods includes training a generative adversarial network comprising a generator and a discriminator; and generating, by the generator, from input images, output images with feeder metadata that represents predicted locations of feeder assets, including receiving by the generator a first input image and generating by the generator a corresponding first output image with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

Claims

1. A method for using a neural network to predict locations of feeder assets in an electrical power grid, the method comprising: training a generative adversarial network comprising a generator and a discriminator, wherein training the generative adversarial network comprises: training the discriminator while holding the generator fixed, including providing to the discriminator training output images generated by the generator and ground truth images, wherein the ground truth images provided to the discriminator each include respective ground truth feeder data identifying the feeder assets and their locations on the ground truth images; and generating, by the generator, from input map data including input images, output map data with feeder metadata that represents predicted locations of feeder assets, including receiving, by the generator, a first input image and generating by the generator corresponding first output map data with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

2. The method of claim 1, wherein generating first output map data with first feeder data comprises also receiving with the first input image corresponding input feeder metadata representing one or more feeder assets in the first input image.

3. The method of claim 2, wherein generating the first output map data with first feeder data comprises also receiving with the first input image one or more asset placement rules.

4. The method of claim 3, wherein: the respective feeder data is incorporated in a respective output image.

5. The method of claim 3, wherein: the respective feeder data is generated as metadata separate from an output image.

6. The method of claim 1, wherein generating the first output map data includes generating, as first feeder data, data that identifies an underground feeder asset including a location of the underground feeder asset.

7. The method of claim 6, wherein the first feeder data identifies all feeder assets between a particular substation and a particular load.

8. The method of claim 7, wherein the particular load is a residential load.

9. The method of claim 8, wherein the one or more feeder assets comprise three or more of a line, a pole, a crossarm, a transformer, a switch, an insulator, a recloser, a sectionalizer, a capacitor bank, including switched capacitors, a load tap changer, or a tap.

10. The method of claim 9, wherein the one or more feeder assets comprise a first transformer and the first feeder data specifies a size of the first transformer.

11. The method of claim 9, wherein the one or more feeder assets comprise a first capacitor bank that includes switched capacitors.

12. A system comprising one or more computers configured to perform operations comprising: training a generative adversarial network comprising a generator and a discriminator; and generating, by the generator, from input map data including input images, output map data with feeder metadata that represents predicted locations of feeder assets, including receiving, by the generator, a first input image and generating by the generator corresponding first output map data with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

13. The method of claim 1, wherein the generator and discriminator are each a respective convolutional neural network model.

14. The system of claim 12, wherein the particular load is a residential load; wherein the one or more feeder assets comprise three or more of a line, a pole, a crossarm, a transformer, a switch, an insulator, a recloser, a sectionalizer, a capacitor bank, including switched capacitors, a load tap changer, or a tap; wherein the one or more feeder assets comprise a first transformer and the first feeder data specifies a size of the first transformer; and wherein the one or more feeder assets comprise a first capacitor bank that includes switched capacitors.

15. The system of claim 12, wherein generating first output map data with first feeder data comprises also receiving with the first input image corresponding input feeder metadata representing one or more feeder assets in the first input image.

16. The system of claim 15, wherein generating the first output map data with first feeder data comprises also receiving with the first input image one or more asset placement rules.

17. The system of claim 16, wherein: the respective feeder data is incorporated in a respective output image, and the respective feeder data is generated as metadata separate from an output image.

18. The system of claim 12, wherein generating the first output map data includes generating, as first feeder data, data that identifies an underground feeder asset including a location of the underground feeder asset.

19. The system of claim 18, wherein the first feeder data identifies all feeder assets between a particular substation and a particular load.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) FIG. 1 is a block diagram showing elements of a generative adversarial network in the context of training to predict feeders from aerial images and above-ground identified assets.

(2) FIG. 2 is a flow chart of a method for using a neural network to predict locations of feeders in an electrical power grid.

(3) Like reference numbers and designations in the various drawings indicate like elements.

DETAILED DESCRIPTION

(4) FIG. 1 is a block diagram showing elements of a generative adversarial network (“GAN”) in the context of training the network to predict feeders from aerial images and above-ground identified assets. The elements include a feeder generator network 120 and a feeder discriminator network 140.

(5) The feeder generator network and feeder discriminator network can be implemented as deep neural network models. They need not, and generally will not, have the same structure. Advantageous neural network structures for these neural networks include deep convolutional neural networks. Advantageously, the GAN is a deep convolutional generative adversarial network (“DCGAN”), and a conditional GAN in which a condition is the input imagery.

(6) The feeder discriminator network is used in training the feeder generator network. During training and in production, the feeder generator network receives as inputs map data 110, including, e.g., aerial images 102, data identifying locations of above-ground assets 104, which will be generally be incomplete and often entirely absent during production but present during training, and, during training, in addition to map data, noise 106. In some implementations, the feeder generator network receives asset placement rules 108, which will be described later in this specification. Map data can include other kinds of useful data as well, in particular, data that cannot be seen in aerial imagery. For example, when it is being trained, the GAN will learn about roads and houses that appear in the imagery. Additional information that can be included in the input map includes the age of the buildings, because buildings built around the same time are more likely to be on the same feeder, which buildings have air conditioning, because it requires a large amount of power, and the type of industry an industrial building belongs to, because that indicates the building's power requirements. The identified assets data and the images include geolocation by which the feeder generator network or a front-end to the feeder generator network can locate the assets on the images. The feeder generator network generates for each input image output map data, and in particular generated feeder metadata 132, which represents predicted locations of feeders, including data identifying assets and their locations. The generated feeder metadata can be overlaid on an input image to show where the predicted feeder and its assets are. The generated map data will in general be the same as the input map data, insofar as it relates to information that can be seen in imagery. The location coordinates could contain an optional elevation term. Wires could be represented by the assets that they connect. A 3-phase power wire would be represented as a single entity; but as phases split off farther along the feeder, each phase is represented as its own entity. The input and output metadata can be embedded in image data or it can be in a separate data entity, e.g., a comma-separated value file, a JSON file, or a protocol buffer. The combination of an image and corresponding feeder metadata, whether or not in a single file, may be called map data or generated map data.

(7) During training, generated map data is input into the feeder discriminator network.

(8) Images including corresponding ground truth feeder metadata 130, which may be called “ground truth feeder images,” are another input into the feeder discriminator network during training. Ground truth feeder metadata is ground truth data representing actual locations of feeders, including locations of their assets. Generated and ground truth feeder images are input during training to the feeder discriminator network in batches, which may include both types of images, advantageously half of each type, or only one of the two types.

(9) Aerial images can include, or be associated with, further data that identifies structures that have an effect on where a utility would located feeder assets. Such other structures include, for example, houses, apartments, business offices, and other loads, as well as structures that constrain the locations where a utility can place assets and route feeders, e.g., roads, highways, or rivers.

(10) The ground truth map data can be any kind of map data that the feeder generator network is trained to generate.

(11) In some implementations, the data format of the map data represents nodes and connections in the feeder network, where each node contains its type and location, and the connections specify which nodes are connected to each other.

(12) The goal of training is to train the feeder generator network to generate generated map data, and in particular to do so from aerial images 102, that is indistinguishable from ground truth map data insofar as the locations, connections, and identities of assets are concerned. In production, the input to the feeder generator network includes aerial images, which aerial images are used as a conditioning image for an implementation based on a conditional GAN architecture, for which the feeder generator network will generate feeder metadata data identifying known assets in the areas covered by the images.

(13) The process of training trains the feeder discriminator network to determine whether an feeder metadata is ground truth or not. Correspondingly, the process of training the feeder generator network trains it to create an image with feeder metadata that will fool the feeder discriminator network, i.e., that the feeder discriminator network will categorize as ground truth. In effect, the feeder discriminator network implements a loss function for the feeder generator network. As a result, this generative adversarial network structure and training process has the great advantage that no hand-crafted loss function needs to be created for the feeder generator network. The training data includes ground truth data for all the different kinds of assets that appear on feeders.

(14) The training process alternates between training the generator and training the discriminator. When training the generator, the discriminator is fixed and the generator trains to trick the discriminator. When training the discriminator, the generator is fixed and the discriminator trains to identify the fakes produces by the generator. The process repeats until an end condition occurs, e.g., the engineer managing the training sees overfitting, or an equilibrium is reached between the generator and the discriminator. In this training process, there are two losses—the generator's, when a generated feeder image is detected as a fake, and the discriminator's, when a generated feeder image is accepted or a ground truth image is rejected.

(15) The discriminator will have a set of convolutions and pooling layers applied to the conditioning image to extract features that will then be concatenated with the feeder metadata and passed into a deep classifier to determine whether the feeder metadata is real or generated.

(16) The generator will have a similar set, generally but not necessarily the same, of convolutions and pooling layers applied to the conditioning image and then some combination of up convolutions and deep layers to predict the feeder metadata. One can think of the feeder metadata as being like the output of an object detection network like FasterRCNN. The output of the generator predicts multiple nodes, each with a location, type, and confidence, and multiple connections, each with a confidence, and each between a pair of nodes.

(17) In some implementations, the feeder generator network includes asset placement rules 108. These rules are associates with particular constituents of a feeder. The rules can represented in code that alters a predicted feeder metadata to be compliant with the rules. After the generator generates the feeder metadata, the rule-implementing code is run to make the metadata compliant, and then the compliant metadata is run through the discriminator and back propagation. Example of a rule include: all lines must be connected to a substation, i.e., no islands; a unique path must exist from every point on the feeder back to the substation, i.e., no cycles; a riser must be connected to something else; a residential load must be no more than a maximum distance from a transformer, which distance depends on whether the region is rural, suburban, or urban; a pole must be no more than a maximum distance from another pole or other line-carrying structure.

(18) FIG. 2 is a flow chart of a method 200 for using a neural network to predict locations of feeders in an electrical power grid.

(19) The method includes training (210) a generative adversarial network that has a generator and a discriminator. The method includes generating (220), by the generator, from input map data including input images, output map data with feeder metadata that represents predicted locations of feeder assets, including receiving (222) by the generator a first input image and generating by the generator a corresponding first output map data with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

(20) In some implementations, generating first output map data with first feeder data also includes receiving (224) with the first input image corresponding input feeder metadata representing one or more feeder assets in the first input image.

(21) In some implementations, generating the first output map data with first feeder data also includes receiving (226) with the first input image one or more asset placement rules.

(22) In some implementations, generating the first output map data includes generating (228), as first feeder data, data that identifies an underground feeder asset including a location of the underground feeder asset.

(23) In some implementations, training the generative adversarial network includes training (230) the generator while holding the discriminator fixed, including (i) providing training input to the generator, the training input comprising training input map data, (ii) providing corresponding training outputs generated by the generator to the discriminator, the training outputs comprising map data including feeder metadata, and (iii) training the generator based on a respective discriminator output from the discriminator for each training output received by the discriminator.

(24) In some implementations, training the generative adversarial network includes training (240) the discriminator while holding the generator fixed, including providing to the discriminator training output images generated by the generator and ground truth images, wherein the ground truth images provided to the discriminator each include respective ground truth feeder data identifying feeder assets and their locations on the ground truth images.

(25) Embodiments of the subject matter and the actions and operations described in this specification can be implemented in digital electronic circuitry, in tangibly-embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, e.g., one or more modules of computer program instructions, encoded on a computer program carrier, for execution by, or to control the operation of, data processing apparatus. The carrier may be a tangible non-transitory computer storage medium. Alternatively or in addition, the carrier may be an artificially-generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be or be part of a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them. A computer storage medium is not a propagated signal.

(26) The term “data processing apparatus” encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. Data processing apparatus can include special-purpose logic circuitry, e.g., an FPGA (field programmable gate array), an ASIC (application-specific integrated circuit), or a GPU (graphics processing unit). The apparatus can also include, in addition to hardware, code that creates an execution environment for computer programs, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

(27) A computer program, which may also be referred to or described as a program, software, a software application, an app, a module, a software module, an engine, a script, or code, can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages; and it can be deployed in any form, including as a stand-alone program or as a module, component, engine, subroutine, or other unit suitable for executing in a computing environment, which environment may include one or more computers interconnected by a data communication network in one or more locations.

(28) A computer program may, but need not, correspond to a file in a file system. A computer program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub-programs, or portions of code.

(29) The processes and logic flows described in this specification can be performed by one or more computers executing one or more computer programs to perform operations by operating on input data and generating output. The processes and logic flows can also be performed by special-purpose logic circuitry, e.g., an FPGA, an ASIC, or a GPU, or by a combination of special-purpose logic circuitry and one or more programmed computers.

(30) Computers suitable for the execution of a computer program can be based on general or special-purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a central processing unit for executing instructions and one or more memory devices for storing instructions and data. The central processing unit and the memory can be supplemented by, or incorporated in, special-purpose logic circuitry.

(31) Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to one or more mass storage devices. The mass storage devices can be, for example, magnetic, magneto-optical, or optical disks, or solid state drives. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.

(32) To provide for interaction with a user, embodiments of the subject matter described in this specification can be implemented on a computer having, or configured to communicate with, a display device, e.g., a LCD (liquid crystal display) or organic light-emitting diode (OLED) monitor, a virtual-reality (VR) or augmented-reality (AR) display, for displaying information to the user, and an input device by which the user can provide input to the computer, e.g., a keyboard and a pointing device, e.g., a mouse, a trackball or touchpad. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user's device in response to requests received from the web browser, or by interacting with an app running on a user device, e.g., a smartphone or electronic tablet. Also, a computer can interact with a user by sending text messages or other forms of message to a personal device, e.g., a smartphone that is running a messaging application, and receiving responsive messages from the user in return.

(33) This specification uses the term “configured to” in connection with systems, apparatus, and computer program components. That a system of one or more computers is configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions. That one or more computer programs is configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions. That special-purpose logic circuitry is configured to perform particular operations or actions means that the circuitry has electronic logic that performs the operations or actions.

(34) In addition to the embodiments of the attached claims and the embodiments described above, the following numbered embodiments are also innovative:

(35) Embodiment 1 is a method for using a neural network to predict locations of feeders in an electrical power grid, the method comprising:

(36) training a generative adversarial network comprising a generator and a discriminator; and

(37) generating, by the generator, from input map data including input images, output map data with feeder metadata that represents predicted locations of feeder assets, including receiving, by the generator, a first input image and generating by the generator a corresponding first output map data with first feeder data that identifies one or more feeder assets and their respective locations, wherein the one or more feeder assets had not been identified in any input to the generator.

(38) Embodiment 2 is a method of embodiment 1, wherein generating first output map data with first feeder data comprises also receiving with the first input image corresponding input feeder metadata representing one or more feeder assets in the first input image.

(39) Embodiment 3 is a method of any one of embodiments 1-2, wherein generating the first output map data with first feeder data comprises also receiving with the first input image one or more asset placement rules.

(40) Embodiment 4 is a method of any one of embodiments 1-3, wherein: the respective feeder data is incorporated in a respective output image.

(41) Embodiment 5 is a method of any one of embodiments 1-3, wherein: the respective feeder data is generated as metadata separate from an output image.

(42) Embodiment 6 is a method of any one of embodiments 1-5, wherein generating the first output map data includes generating, as first feeder data, data that identifies an underground feeder asset including a location of the underground feeder asset.

(43) Embodiment 7 is a method of any one of embodiments 1-6, wherein the first feeder data identifies all feeder assets between a particular substation and a particular load.

(44) Embodiment 8 is a method of embodiment 7, wherein the particular load is a residential load.

(45) Embodiment 9 is a method of any one of embodiments 1-8, wherein the one or more feeder assets comprise three or more of a line, a pole, a crossarm, a transformer, a switch, an insulator, a recloser, a sectionalizer, a capacitor bank, including switched capacitors, a load tap changer, or a tap.

(46) Embodiment 10 is a method of embodiment 9, wherein the one or more feeder assets comprise a first transformer and the first feeder data specifies a size of the first transformer.

(47) Embodiment 11 is a method of embodiment 9, wherein the one or more feeder assets comprise a first capacitor bank that includes switched capacitors.

(48) Embodiment 12 is a method of any one of embodiments 1-11, wherein the generator and discriminator are each a respective convolutional neural network model.

(49) Embodiment 13 is a method of any one of embodiments 1-12, wherein training the generative adversarial network comprises:

(50) training the generator while holding the discriminator fixed, including providing training input to the generator, the training input comprising training input map data, and providing corresponding training outputs generated by the generator to the discriminator, the training outputs comprising map data including feeder metadata, and training the generator based on a respective discriminator output from the discriminator for each training output received by the discriminator.

(51) Embodiment 14 is a method of embodiment 13, wherein:

(52) the training input includes, for a first plurality of training inputs, data representing respective identified above-ground feeder assets corresponding to training input images.

(53) Embodiment 15 is a method of any one of embodiments 13-14, wherein:

(54) the training input includes one or more asset placement rules.

(55) Embodiment 16 is a method of any one of embodiments 1-15, wherein training the generative adversarial network comprises:

(56) training the discriminator while holding the generator fixed, including providing to the discriminator training output images generated by the generator and ground truth images, wherein the ground truth images provided to the discriminator each include respective ground truth feeder data identifying feeder assets and their locations on the ground truth images.

(57) Embodiment 17 is a system comprising one or more computers implementing the method of any one of embodiments 1-16.

(58) Embodiment 18 is a computer storage medium encoded with a computer program, the program comprising instructions that are operable, when executed by data processing apparatus, to cause the data processing apparatus to perform the method of any one of embodiments 1-16.

(59) While this specification contains many specific implementation details, these should not be construed as limitations on the scope of what is being claimed, which is defined by the claims themselves, but rather as descriptions of features that may be specific to particular embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially be claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claim may be directed to a subcombination or variation of a subcombination.

(60) Similarly, while operations are depicted in the drawings and recited in the claims in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

(61) Particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results. As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some cases, multitasking and parallel processing may be advantageous. Also, rather than using a GAN to predict feeders, a recurrent neural network with a beam search could be used. The beam search would allow for possible next moves and the recurrent neural network would give the probabilities and locations of each feeder asset.