House electric vehicle charge detection

Abstract

A system includes a processor and a non-transitory computer-readable memory including instructions executable by the processor to cause the processor to perform operations. The operations include accessing premises consumption data for premises in the power distribution network. The premises consumption data includes an indication of premises resource consumption over a period of time. The operations also include applying the machine learning model to the premises consumption data. A machine learning model is trained to generate an output corresponding to the electric vehicle classification of the premises. Further, the operation includes generating an electric vehicle classification of the premises using an output of the machine learning model, and controlling power generation of the power distribution network based on the electric vehicle classification of the premises.

Description

Technical Field

[0001] The present disclosure generally relates to consumption detection in a resource allocation system. More specifically, but without limitation, the present disclosure relates to electric vehicle charging detection at a house in a resource allocation system. Background Art

[0002] In a resource allocation system, such as an electric power grid that delivers electricity, a house may consume resources at different rates over a period of time based on devices at the house that consume the resources of the resource allocation system over the period of time. For example, when an electric vehicle is charging at a house, the house may consume a greater amount of electricity at various times of the day. To ensure sufficient power generation to meet the demand on the power grid, it may be beneficial to accurately track houses that regularly charge electric vehicles.

[0003] The demand on the power grid may change over time as additional electric vehicles need to be charged at various houses connected to the power grid. For example, the number of houses that charge electric vehicles and the various types of chargers used at the houses to charge the electric vehicles may change over time. Accurately tracking houses that charge electric vehicles with minimal input from the occupants of the houses may be useful in managing power generation in a resource allocation system. Summary of the Invention

[0004] In one embodiment, a system includes a processor and a non-transitory computer-readable memory that includes instructions executable by the processor to cause the processor to perform operations. The operations include accessing house consumption data for a house in a power distribution network. The house consumption data includes an indication of house resource consumption over a period of time. The operations also include applying a machine learning model to the house consumption data. The machine learning model is trained to generate an output corresponding to an electric vehicle classification for the house. Additionally, the operations include using the output of the machine learning model to generate an electric vehicle classification for the house and controlling power generation in the power distribution network based on the electric vehicle classification for the house.

[0005] In another embodiment, a non-transitory computer-readable medium includes instructions executable by a processor to cause the processor to perform operations. The operations include accessing house consumption data for a house in a power distribution network. The house consumption data includes an indication of house resource consumption over a period of time. The operations also include applying a machine learning model to the house consumption data. The machine learning model is trained to generate an output corresponding to an electric vehicle classification for the house. Additionally, the operations include using the output of the machine learning model to generate an electric vehicle classification for the house and controlling power generation in the power distribution network based on the electric vehicle classification for the house.

[0006] In another embodiment, a computer-implemented method includes accessing home consumption data for a home in a power distribution network. The home consumption data includes an indication of the consumption of home resources over a period of time. The method further includes applying a machine learning model to the home consumption data. The machine learning model is trained to generate an output corresponding to an electric vehicle classification for the home. Additionally, the method includes using the output of the machine learning model to generate an electric vehicle classification for the home and controlling the power generation of the power distribution network based on the electric vehicle classification of the home. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] These and other features, aspects, and advantages of the present disclosure will be better understood when the following detailed description is read with reference to the accompanying drawings.

[0008] Figure 1 An exemplary physical topology of a power distribution network in accordance with some embodiments described herein is shown, which depicts devices at various points or nodes on the network.

[0009] Figure 2 is a flowchart of a process for controlling the power supply of a power distribution network based on an electric vehicle classification for a home in the power distribution network Figure 1 in accordance with some embodiments described herein.

[0010] Figure 3 is an example of a graphical representation of a data flow for identifying a home that performs an electric vehicle charging operation in accordance with some embodiments described herein.

[0011] Figure 4 is a flowchart of a process for training a machine learning model to identify a home that performs an electric vehicle charging operation in accordance with some embodiments described herein.

[0012] Figure 5 is an example of a machine learning model environment for identifying a home that performs an electric vehicle charging operation in accordance with some embodiments described herein.

[0013] Figure 6 is an example machine learning model of an artificial neural network in accordance with some embodiments described herein.

[0014] Figure 7 is an exemplary computing device for detecting an electric vehicle charging operation in accordance with some embodiments described herein. DETAILED DESCRIPTION

[0015] The present disclosure describes techniques for providing electric vehicle charging detection at a house in a resource distribution system. In an example, a resource distribution system, such as a power grid, may provide consumable resources to houses within the resource distribution system. A house may consume electricity, for example, at different rates when electric vehicle charging occurs at the house and when electric vehicle charging does not occur at the house. As houses within the power grid charge more electric vehicles, the power generation capacity of the power grid may be stressed. Additionally, because electric vehicles may charge during periods that are traditionally considered “low demand” periods, such as overnight, the increased charging during those low demand periods may unexpectedly stress the power grid when generators are offline during those low demand periods.

[0016] To maintain accurate information regarding the number of houses performing electric vehicle charging operations, a time series of power consumption data may be obtained from houses that consume power from the power grid. The data may be processed by a trained machine learning model to identify houses that are charging electric vehicles. For example, the trained machine learning model may be applied to the data to distinguish electric vehicle charging operations from other types of charging operations at the house. In some examples, the machine learning model may also be trained to identify the type of charger (e.g., level 1 or level 2) performing the electric vehicle charging operation at the house. Generally, power generation of the power grid and power grid planning may be controlled based on the information identified by the machine learning model. For example, when additional electric vehicles are detected charging at houses sharing a common transformer, a utility company may proactively replace the common transformer with a higher capacity version of the transformer. Other components of the resource distribution network may also be replaced based on a possible shift in typical demand at houses served by other components as additional electric vehicles are charged at houses served by other components. In additional examples, when additional charging operations are detected at a house, the power generated during times when the charging operations are most likely to occur may be increased. Similarly, the identified peak demand periods and low demand periods may be shifted based on additional electric vehicle charging operations, and the operation of power generation equipment may be controlled in a manner that reflects the updated peak demand periods and low demand periods.

[0017] Illustrative examples are given to introduce the reader to the general subject matter discussed herein and are not intended to limit the scope of the disclosed concepts. The following sections describe various additional features and examples with reference to the accompanying drawings, where like numerals represent like elements and directional descriptions are used to describe illustrative aspects but, like the illustrative aspects, should not be used to limit the present disclosure.

[0018] Figure 1 An exemplary physical topology of a distribution network is shown that depicts devices at various points or nodes on the network. Figure 1Depicts a house monitoring system 100 and a power distribution network 110. In an example, the house monitoring system 100 receives information from an endpoint meter in the power distribution network 110 and determines houses on the power distribution network 110 where electric vehicle charging operations are being performed. Although the house monitoring system 100 and the power distribution network 110 are described herein as part of a power distribution environment, other utility systems may include a similar house monitoring system 100. For example, the house monitoring system 100 may be employed in a gas, water, or other utility distribution environment where monitoring of specific resource consumption operations is desired.

[0019] The house monitoring system 100 includes a house monitoring application 101 and a head-end system 102. The house monitoring application 101 executes on a computing device as depicted in Figure 7 . The house monitoring application 101 may receive metering data from meters installed at customer houses, such as voltage, load, power consumption, etc. Other data from other data points may also be provided to the house monitoring application 101. The house monitoring application 101 receives metering data from the head-end system 102 or through an intermediary that reads and aggregates metering information. The meters may transmit the metering information to the head-end system via additional network devices and networks, which are not shown in the figure for simplicity. In some examples, radio frequency (RF) wireless communication, cellular communication, power line communication (PLC) communication, or any other suitable communication technology may be used to perform communication operations within the power distribution network 110. In some examples, the house monitoring application 101 may be incorporated at the edge of the power grid. In such examples, some decisions regarding the power distribution network 101 may be made in a decentralized manner (e.g., without going through the head-end system 102).

[0020] The power distribution network 110 may include a substation 112 and one or more feeders 120a-n. The substation 112 distributes the power received from a power source to the feeders 120a-n. Examples of power sources include coal-fired power plants, wind turbines, and solar panel installations. The substation 112 may include a substation transformer 113. The substation transformer 113 reduces the voltage provided to the substation 112 and outputs a lower voltage to the feeders 120a - 120n. The substation 112 may distribute polyphase (e.g., three-phase) power. Although Figure 1 a single substation 112 is depicted, the power distribution network 110 may include multiple similar substations.

[0021] Each feeder 120a-n can feed one or more houses (such as house 124 or house 125) through one or more metering devices (such as metering devices 130 and 131). House 124 or 125 can include a residence, an apartment, a commercial building, or any other end user of the power provided by the power distribution network 110. In the example, houses 124 and 125 are coupled to metering devices 130 and 131. Metering devices 130 and 131 can respectively measure the power consumption of houses 124 and 125 over time. Although only houses 124 and 125 are shown in the power distribution network 110, it can be understood that additional houses of an entire network such as houses can also be located in the power distribution network 110.

[0022] The house monitoring application 101 can derive a set of details about houses 124 and 125 based on the power consumption data observed by metering devices 130 and 131. For example, metering devices 130 and 131 can send data related to the power consumption at houses 124 and 125 to the front-end system 102. In the example, the house monitoring application 101 can determine the condition of a specific target house (e.g., no electric vehicle charging, level 1 electric vehicle charging, level 2 electric vehicle charging, etc.). For example, the house monitoring application 101 can utilize data obtained from metering devices 130 and 131 and other data sources in the power distribution network 110 to predict the condition of houses in the power distribution network 110. In the example, the house monitoring application 101 can apply one or more machine learning models to data obtained from metering devices 130 and 131, the power distribution network 110, other relevant data sources, or a combination thereof to generate a predictive indication of the house condition. In some examples, new machine learning models can be generated and trained for the purposes described herein. In other examples, the data obtained from metering devices 130 and 131, the power distribution network 110, and other relevant data sources can have existing machine learning models applied to it, which are trained using data similar to the data collected by metering devices 130 and 131. For example, supervised regression machine learning models or reinforcement learning models can be used for the purposes described herein.

[0023] Using the predicted condition of a house, the house monitoring application 101 can be used for power generation scheduling of the power distribution network 110 or other resource supply management operations. For example, the predicted condition of a house can provide information for predicting the power demand throughout the day. In such an example, a house with electric vehicle charging may consume more power at night than a house without electric vehicle charging. By tracking houses that perform electric vehicle charging operations and the type of charging operations (e.g., level 1 or level 2), the power distribution network 110 may be able to control power generation to meet the demand for electric vehicle charging operations on the power distribution network 110 at times that are typically considered off-peak hours. By controlling power generation based on the condition of the house, the power distribution network 110 can avoid over-consuming the pressure on the network and avoid over-generation caused by over-estimating power consumption. In another example, the power distribution network 110 can intelligently schedule the charging of detected electric vehicles using the common power grid components to ensure that the common power grid components do not become overloaded during the charging operation.

[0024] When it is detected that an additional electric vehicle is charging at a house sharing a common transformer (or a future charging operation is predicted for the house), the utility company can also proactively replace the common transformer with a higher-capacity version of the transformer. When an additional electric vehicle is charged at a house served by other components, other components of the resource allocation network can also be replaced or otherwise updated based on the possible deviation from the typical demand at the house served by other components.

[0025] In an example, the machine learning model of the house monitoring application 101 can be trained based on a historical data corpus obtained from one or more power distribution networks. For example, training and validation data with real states can be used to train the machine learning model and verify the accuracy of the trained machine learning model. The machine learning model can be trained in such a way that the historical data provided to the machine learning model results in an output that matches the representation of the real state. After initial training, and as the state of the power distribution network 110 evolves over time, additional data points and additional real states obtained from houses 125 and 125 of the power distribution network 110 can be used to further adjust the machine learning model.

[0026] In some examples, the trained machine learning model of the house monitoring application 101 can be trained to predict the future condition of the houses of the power distribution network 110. For example, the machine learning model can utilize the trend of data provided by the condition of the houses of the power distribution network 110 over time to predict electric vehicle charging operations at the houses within the power distribution network 110 at future time points. These predictions can enable the utility to maintain a plan for the capacity change within the power distribution network 110 over time.

[0027] In an example, an additional machine learning model can be trained to generate predictions of electric vehicle charging operations within the distribution network 110. In such an example, the output over time of a machine learning model that detects the presence of an electric vehicle charging operation can be used as input to the additional machine learning model. Thus, the additional machine learning model can be trained to identify trends that can predict future electric vehicle charging operations within the distribution network 110.

[0028] Figure 2 is a flowchart of a process 200 for implementing a home monitoring application 101 to control the power supply of a distribution network 110. At block 202, the process 200 involves accessing consumption data for a set of homes 124 and 125 within the distribution network 110. The data can be received at the head-end system 102 from metering devices 130 and 131 or from any other data collection point within the distribution network 110. Some data can be automatically collected by the metering meters 130 and 131, and other data can be manually collected in response to maintenance operations. For example, maintenance operations such as installing distributed energy metering meters at homes, installing electric vehicle chargers at homes, or other related maintenance operations can be manually collected and reported to the home monitoring application 101.

[0029] At block 204, the process 200 involves applying a trained machine learning model to the collected data. The trained machine learning model can be trained to classify whether an electric vehicle charging operation is occurring at an individual home within the distribution network 110. For example, the trained machine learning model can be applied to the consumption data obtained at block 202, which can be a time series of power consumption at each home. In some examples, the time series of power consumption can be an indication of the instantaneous consumption at intervals. For example, the instantaneous consumption can occur at intervals of 10 or 15 minutes over a 7-day period. The trained machine learning model can be trained on similar data in the time domain to identify a set of homes that perform electric vehicle charging operations.

[0030] At block 206, process 200 involves generating classification information for the set of houses based on the output of a trained machine learning model. In an example, the trained machine learning model can be trained to output an indication that a house performs an electric vehicle charging operation or does not perform an electric vehicle charging operation. The electric vehicle classification of a house can be a one-hot encoding of a SoftMax function of a set of possible electric vehicle classifications. In other words, the classification can be an indication of a class among a set of possible classifications for each house. In some examples, the classification of a house that performs an electric vehicle charging operation can also involve identifying the type of charger that performs the electric vehicle charging operation. For example, the classification of a house can include an indication that a level 1 electric vehicle charger or a level 2 electric vehicle charger operates at the house. In an example, a level 1 charger can deliver approximately 1.2 kW and operate directly from a standard 120 VAC outlet. A level 2 charger can be a charger ranging from 6.2 kW to 19.2 kW and can rely on a 208 - 240 V, 12 - 80 Amp circuit. In some examples, each house within the distribution network 110 can be classified based on whether an electric vehicle charging operation occurs at the house.

[0031] At block 208, process 200 involves controlling the power supply of the distribution network 110 based on the classification of the houses generated by the machine learning model. In an example, the classification generated by the machine learning model identifies the number of houses within the distribution network 110 that perform an electric vehicle charging operation, and in some examples, identifies the type of charger that performs the electric vehicle charging operation. Information associated with the amount of houses that perform an electric vehicle charging operation can be used to control a generator that supplies power to the distribution network 110 to meet the expected demand of the distribution network 110. For example, as the number of houses performing a charging operation grows, the generator can increase the power supply provided to the distribution network 110 during a time period when a charging operation is likely to occur.

[0032] In an example, the power supply of the power distribution network 110 can be adjusted while detecting new homes performing electric vehicle charging operations. In some examples, the power supply of the power distribution network 110 can be adjusted only when a threshold number of new homes performing electric vehicle charging operations are detected. For example, the control of the power supply can occur when the total number of new homes performing electric vehicle charging operations (since the previous adjustment of the power supply) is expected to have an impact on the ability of the power distribution network 110 to meet the demand at the homes. In such an example, if the total number of new homes performing electric vehicle charging operations is expected to cause the power distribution network 110 to consume a threshold percentage more power (e.g., 1% more, 5% more, etc.) than after the previous adjustment, then the power supply can be adjusted. Other threshold percentages can also be used, and other trigger events associated with adding homes performing electric vehicle charging operations can also trigger an adjustment of the power supply at block 208. In some examples, classification information can also be used by the power distribution network 110 to determine whether components of the power distribution network 110 should be updated to obtain a higher capacity. For example, a transformer operating with several homes classified as performing electric vehicle charging operations can be updated to a model with a higher capacity than a typical standard capacity transformer. Other types of components of the power distribution network 110 can also be updated.

[0033] In some examples, process 200 can be implemented at any utility having a sensor network providing sensing data. A machine learning model can utilize the sensing data from the sensor network to predict the power consumption demand of the utility.

[0034] Figure 3 is an example of a graphical representation of a data flow 300 for identifying homes performing electric vehicle charging operations according to some embodiments described herein. Power consumption data 302 can be received at the home monitoring system 100. In an example, the power consumption data 302 is a collection of consumption data in the time domain of homes within the power distribution network 110. For each home within the power distribution network 110, the power consumption data 302 can be obtained at intervals of 10 or 15 minutes over a period of 7 days, where a determination regarding whether the home is performing an electric vehicle charging operation is desired. Other time periods and interval lengths can also be used.

[0035] A trained machine learning model, such as a convolutional neural network, can be applied to the power consumption data 302 to detect low-level features 304 of the power consumption data 302. The low-level features 304 can include relatively coarse features represented by the power consumption data 302. For example, the low-level features 304 can include consumption peaks of the home over time, consumption floors of the home over time, or any other data anomalies associated with the power consumption data 302.

[0036] When a low-level feature 304 is detected, a machine learning model can detect a high-level feature 306 of the power consumption data 302. For example, the machine learning model can be used to interpret or otherwise classify the low-level features 304 of the power consumption data 302. In some examples, the high-level feature 306 can be identified by a layer of the machine learning model that is closer to the output layer than the layer used to generate the low-level feature 304.

[0037] The time ordering 308 of the power consumption data 302 can also be used by the machine learning model. Since the power consumption data 302 is provided in the time domain, various features of the power consumption data 302 may be related to the time series of the power consumption data 302. For example, the machine learning model can use the time ordering 308 to identify features of the power consumption data 302 that may occur consecutively with a particular event or at certain times during a particular day. Electric vehicle charging that occurs at a particular time (e.g., overnight) can be an important indicator for the machine learning model to identify an electric vehicle charging operation at a house. Additionally, the length of time of increased power consumption identified by the power consumption data 302 can also be an indicator for the machine learning model to identify the type of charger used for an electric vehicle charging operation at the house. For example, a charging operation performed by a Level 1 charger may have increased power consumption over a longer amount of time than a charging operation performed by a Level 2 charger.

[0038] A classification 310 can be generated for each house represented in the power consumption data 302 based on the low-level feature detection 304, the high-level feature detection 306, and the time ordering 308. In an example, the classification can include an indication that the house does not perform an electric vehicle charging operation, an indication that the house performs a Level 1 electric vehicle charging operation, or an indication that the house performs a Level 2 electric vehicle charging operation. Other classifications 310 are possible, such as identifying other types of chargers, identifying the charging capacity of the electric vehicle being charged, detecting other types of charging operations (e.g., stationary batteries at the house), etc. Once the classification 310 is determined by the machine learning model, an output 312 can be generated to control the power generation operation to meet the demand requirements of the houses within the distribution network 110.

[0039] Figure 4FIG. 400 is a flow chart of a process 400 for training a machine learning model to identify a house performing an electric vehicle charging operation according to some embodiments described herein. At block 402, process 400 involves accessing a corpus of training and validation power consumption data. In an example, the corpus of data can be partitioned into a training data set and a validation data set, and the validation data set is used to validate the machine learning model trained using the training data set. The corpus of training data can be a National Renewable Energy Laboratory (NREL) data set, which is labeled with electric vehicle charging operation information of 600 houses obtained within 52 weeks, where readings are obtained at 10-minute intervals. In some examples, the NREL data set can be further broken down into data subsets with 7-day consumption measured at 10-minute intervals, and these subsets of the NREL data set can be used to train the machine learning model at block 406 below. The training labels can be one-hot training labels, which identify each individual data set as not performing an electric vehicle charging operation, performing a level 1 charging operation, or performing a level 2 charging operation. Other data sets can also be used as training and validation data.

[0040] At block 404, process 400 involves normalizing the training consumption data and the validation consumption data. In an example, normalization can involve assigning values between 0 and 1 corresponding to the kW values in the consumption data. For example, 0 can represent 0 kW, 1 can represent 35 kW, and all values between 0 kW and 35 kW can be assigned corresponding values between 0 and 1. Min-max scaling can be used to perform the normalization of the training and validation data. In an example, min-max scaling operations can also be used to normalize the house consumption data to which the trained machine learning model is applied. Other normalization scaling operations can also be used.

[0041] At block 406, process 400 involves using the normalized training data to train a machine learning model to classify the presence of an electric vehicle charging operation. Using the labeled and normalized training data, a machine learning model can be trained to identify houses that are performing an electric vehicle charging operation. In an example, the training data is provided to the machine learning model in the time domain. Thus, the trained machine learning model can also be applied to the consumption data in the time domain to classify whether an electric vehicle charging operation is occurring at a house and, if so, the type of the charging operation.

[0042] At block 408, process 400 involves validating a trained machine learning model using normalized validation data. The normalized validation data can be a subset of a corpus of the NREL dataset that was not used to train the machine learning model. By validating the trained machine learning model, the accuracy of the machine learning model can be evaluated. In some examples, the validation process can involve determining that the classification accuracy of the validation data exceeds an accuracy threshold. For example, when the validation accuracy is greater than 95%, the trained machine learning model can be implemented. Other threshold accuracy percentages can also be used.

[0043] At block 410, process 400 involves updating the trained machine learning model using additional real-world consumption data. In some examples, the trained machine learning model can continuously learn based on additional data received by the machine learning model. In such examples, the machine learning model can be updated based on the additional real-world consumption data that is provided to the machine learning model for further updating. Such continuous learning can be beneficial as electric vehicles and electric vehicle charging operations evolve over time.

[0044] Exemplary machine learning environment

[0045] Figure 5 is an example of a machine learning environment for identifying a house that performs an electric vehicle charging operation according to some embodiments described herein. Training vector 502 is shown as having a query 504 and a known response 506. As an example, the query can be a request to classify the electric vehicle charging status of a house in a distribution network 110. For example, query 504 can be a classification problem such as an indication of whether the house is performing a charging operation and, if so, the type of charging operation being performed by the house. For ease of illustration, only two training vectors 502 are shown, but the number of training vectors can be much larger, such as 10, 50, 100, 1000, 10000, 100000 or more.

[0046] Learning module 508 can use training vector 502 to perform training 510 of model 512. Learning module 508 can optimize the parameters of model 512 (such as a machine learning model) such that a quality metric (e.g., the accuracy of model 512) is achieved using one or more specified criteria. The accuracy can be measured by comparing the known response 506 with the predicted output of model 512. The parameters of model 512 can be iteratively changed to improve the accuracy. The determination of the quality metric can be implemented for any arbitrary function that includes a set of all risk, loss, utility, and decision functions.

[0047] In some embodiments of training, it can be determined how changing a parameter affects the gradient of a cost function, which can provide a measure of the accuracy of the current state of model 512. The gradient can be used in conjunction with a learning step (e.g., a measure of how much the parameters of model 512 should be updated for a given time step of an optimization process). Thus, the parameters (which can include weights, matrix transformations, and probability distributions) can be optimized to provide an optimal value of the cost function, which, as an example, can be measured as being above or below a threshold (i.e., exceeding the threshold) or the cost function not changing significantly over several time steps. In other embodiments, training can be implemented using methods that do not require Hessian or gradient calculations, such as dynamic programming or evolutionary algorithms.

[0048] The prediction phase 514 can provide a prediction response 516 for query vector 518 based on a new query 520. The new query 520 can be of a type similar to query 504 of training vector 502. If the new query record has a different type, then a transformation can be performed on the data to obtain data in a format similar to the format of training vector 502. The prediction response 516 can correspond to the question encoded in query vector 518. In some examples, the predicted response 516 can be an indication of whether a house 124 in the power distribution network 110 is performing an electric vehicle charging operation. In additional examples, the prediction response 516 can also indicate the type of charging operation being performed at the house (e.g., level 1 or level 2 electric vehicle charging). The model 512 can also be trained to generate other predictions related to the identification of electric vehicle charging on the power distribution network 110.

[0049] The model 512 can include a machine learning model, such as a deep learning model, a neural network (e.g., a deep learning neural network), kernel-based regression, adaptive basis regression or classification, Bayesian methods, ensemble methods, logistic regression and extensions, Gaussian processes, support vector machines (SVMs), probabilistic models, and probabilistic graphical models. Embodiments using neural networks can employ deep architectures that are wide and tensorized, convolutional layers, dropout, various neural activations, and regularization steps.

[0050] Figure 6 is an example machine learning model of an artificial neural network 600 according to some embodiments described herein. As an example, the model 512 can be an artificial neural network 600 that includes a plurality of neurons 602 (e.g., adaptive basis functions) organized in layers. These layers can include an input layer 608, a first hidden layer 604, a second hidden layer 610, and an output layer 612. Other layer arrangements are also contemplated. For example, the artificial neural network 600 can have more than Figure 6Two more or less hidden layers depicted therein. The neurons 602 or nodes can be connected by edges 606. Training of the artificial neural network 600 can iteratively search for an optimal configuration of the parameters of the neural network for feature recognition, classification, and / or prediction performance. A variety of numbers of layers and nodes can be used. One of ordinary skill in the art can readily recognize variations in the design of neural networks and the design of other machine learning models.

[0051] Exemplary computing device for house monitoring

[0052] Figure 7 An exemplary computing device for detecting an electric vehicle charging operation in accordance with some embodiments described herein is shown. Any suitable computing system can be used to perform the operations described herein. Examples of the depicted computing device 700 include a processor 702 communicatively coupled to one or more memory devices 704. The processor 702 executes computer-executable program code 730 stored in the memory device 704, accesses data 720 stored in the memory device 704, or both. Examples of the processor 702 include a microprocessor, an application specific integrated circuit (“ASIC”), a field programmable gate array (“FPGA”), or any other suitable processing device. The processor 702 can include any number of processing devices or cores, including a single processing device. The functionality of the computing device can be implemented in hardware, software, firmware, or a combination thereof.

[0053] The memory device 704 includes any suitable non-transitory computer-readable medium for storing data, program code, or both. The computer-readable medium can include any electronic, optical, magnetic, or other storage device capable of providing computer-readable instructions or other program code to the processor. Non-limiting examples of the computer-readable medium include flash memory, ROM, RAM, an ASIC, or any other medium from which the processing device can read instructions. The instructions can include processor-specific instructions generated by a compiler or interpreter from code written in any suitable computer programming language, including, for example, C, C++, C#, Visual Basic, Java, or a scripting language.

[0054] The computing device 700 can also include a number of external or internal devices, such as input or output devices. For example, the computing device 700 is shown as having one or more input / output (“I / O”) interfaces 708. The I / O interface 708 can receive input from an input device or provide output to an output device. One or more buses 706 are also included in the computing device 700. The bus 706 communicatively couples the respective one of the one or more components in the computing device 700.

[0055] The computing device 700 executes program code 730 that configures the processor 702 to perform one or more of the operations described herein. For example, the program code 730 causes the processor to execute Figures 1-6 the operations described therein.

[0056] The computing device 700 also includes a network interface device 710. The network interface device 710 includes any device or group of devices suitable for establishing a wired or wireless data connection to one or more data networks. The network interface device 710 may be a wireless device and have an antenna 714. The computing device 700 may use the network interface device 710 to communicate with one or more other computing devices implementing computing devices or other functions via a data network.

[0057] The computing device 700 may also include a display device 712. The display device 712 may be an LCD, LED, touch screen, or other device operable to display information about the computing device 700. For example, the information may include the operating state of the computing device, the network state, and the like.

[0058] Although the present subject matter has been described in detail with respect to specific aspects thereof, it should be understood that those skilled in the art can readily generate alternatives, variations, and equivalents to these aspects upon understanding the foregoing and the following. Accordingly, it should be understood that the present disclosure has been presented for purposes of illustration and not limitation, and does not exclude inclusion of such modifications, variations, and / or additions to the present subject matter that would be apparent to a person of ordinary skill in the art.

Claims

1. A system, comprising: a processor; and a non - transitory computer - readable memory including instructions executable by the processor to cause the processor to perform operations, the operations including: accessing house consumption data of a house in a power distribution network, wherein the house consumption data includes an indication of house resource consumption over a period of time; applying a machine - learning model to the house consumption data, wherein the machine - learning model is trained to generate an output corresponding to an electric - vehicle classification of the house; using the output of the machine - learning model to generate the electric - vehicle classification of the house; and controlling power generation of the power distribution network based on the electric - vehicle classification of the house.

2. The system according to claim 1, wherein the operations further comprise: training the machine - learning model to generate an output corresponding to the electric - vehicle classification of the house using training vectors of ground - truth data of multiple houses of an additional power distribution network.

3. The system according to claim 2, wherein the operations further comprise: updating the machine - learning model using additional consumption data of multiple houses from the power distribution network.

4. The system according to claim 1, wherein the electric - vehicle classification of the house includes an indication that the house does not charge an electric vehicle, an indication that the house charges the electric vehicle using a first type of electric - vehicle charger, or an indication that the house charges the electric vehicle using a second type of electric - vehicle charger.

5. The system according to claim 1, wherein the electric - vehicle classification of the house includes a one - hot encoding of a SoftMax function of a set of possible electric - vehicle classifications.

6. The system according to claim 1, wherein the indication of house resource consumption over the period of time includes a time series of power consumption performed by the house at regular time intervals over multiple days.

7. The system according to claim 1, wherein the operations further comprise: normalizing the house consumption data using min - max scaling, and the min - max scaling is also used to normalize training consumption data for training the machine - learning model.

8. The system according to claim 1, wherein the house consumption data includes time - domain data.

9. A non - transitory computer - readable medium including instructions executable by a processor to cause the processor to perform operations, the operations comprising: accessing house consumption data of a house in a power distribution network, wherein the house consumption data includes an indication of house resource consumption over a period of time; applying a machine - learning model to the house consumption data, wherein the machine - learning model is trained to generate an output corresponding to an electric - vehicle classification of the house; using the output of the machine - learning model to generate the electric - vehicle classification of the house; and controlling power generation of the power distribution network based on the electric - vehicle classification of the house.

10. The non - transitory computer - readable medium according to claim 9, wherein the operations further comprise: Train the machine learning model using training vectors of ground truth data of multiple houses using an additional power distribution network to generate an output corresponding to the electric vehicle classification of the houses.

11. The non-transitory computer-readable medium according to claim 10, wherein the operation further comprises: Updating the machine learning model using additional consumption data of multiple houses from the power distribution network.

12. The non-transitory computer-readable medium according to claim 9, wherein the electric vehicle classification of the house includes an indication that the house does not charge an electric vehicle, an indication that the house charges the electric vehicle using a level 1 electric vehicle charger, or an indication that the house charges the electric vehicle using a level 2 electric vehicle charger.

13. The non-transitory computer-readable medium according to claim 9, wherein the electric vehicle classification of the house includes a one-hot encoding of a SoftMax function of a set of possible electric vehicle classifications.

14. The non-transitory computer-readable medium according to claim 9, wherein the operation further comprises: Applying an additional machine learning model to the electric vehicle classification of the house; and Using the output of the additional machine learning model to generate a prediction of future electric vehicle charging operations.

15. A computer-implemented method, comprising: Accessing house consumption data of a house in a power distribution network, wherein the house consumption data includes an indication of the consumption of house resources over a period of time; Applying a machine learning model to the house consumption data, wherein the machine learning model is trained to generate an output corresponding to the electric vehicle classification of the house; Generating the electric vehicle classification of the house using the output of the machine learning model; and Controlling the power generation of the power distribution network based on the electric vehicle classification of the house.

16. The computer-implemented method according to claim 15, further comprising: Applying an additional machine learning model to the electric vehicle classification of the house and multiple additional electric vehicle classifications of additional houses in the power distribution network; and Using the output of the additional machine learning model to generate a prediction of future electric vehicle charging operations in the power distribution network.

17. The computer-implemented method according to claim 15, wherein the electric vehicle classification of the house includes an indication that the house does not charge an electric vehicle, an indication that the house charges the electric vehicle using a level 1 electric vehicle charger, or an indication that the house charges the electric vehicle using a level 2 electric vehicle charger.

18. The computer-implemented method according to claim 15, wherein the electric vehicle classification of the house includes a one-hot encoding of a SoftMax function of a set of possible electric vehicle classifications.

19. The computer-implemented method according to claim 15, wherein the indication of the consumption of house resources over the period of time includes a time series of power consumption performed by the house at regular time intervals over multiple days.

20. The computer-implemented method according to claim 15, wherein the house consumption data includes time-domain data.

Images