On-premise electric vehicle charging detection

Abstract

To provide a computing system for and a computer packaging method of detecting charging events for an electric vehicle in order to take, in advance, a measure to ensure soundness and reliability of a charging facility, power distribution facility, an electric grid facility, and other assets.SOLUTION: In a computing system, a method according to the present invention has pre-processing of encoding use values of a plurality of time intervals with one symbol among a series of symbols representing a level of electricity consumption during each of the time intervals from electricity consumption data of sets of a known electric vehicle owner, detection processing for detecting encoded consumption pattern of the symbols as to each electric vehicle owner, and verification for discriminating an EV charging motif representing an EV charging event.SELECTED DRAWING: Figure 2

Description

[Background technology]

[0001] background Recent advances in rechargeable batteries have made plug-in electric vehicles (EVs) and hybrid vehicles available to a larger portion of the population. EV adoption is increasing significantly. It is expected that over time, a large proportion of vehicles in use will be EVs. These vehicles place a significant load on city electrical grids and power sources because EVs must be periodically charged in order to operate. This creates challenges for current electric utilities, who must be able to plan for such a large penetration of electric vehicles.

[0002] For example, a typical home draws an average of about 1.0 kWh of load from their local electric utility. This may include common appliances such as washers, dryers, and ovens. However, residential load can vary depending on several variables, such as weather-related HVAC usage. Other variables may be larger, such as pools, spas, and on-site solar power generation.

[0003] In contrast, an EV can draw 3 to 4 kW or more per hour. In this case, a single EV owner can indirectly add the load of four homes to the service transformer. This is problematic because it can cause a transformer to overload, potentially creating a risk of overloading. Modern transformers can withstand some overloading, but as the popularity and adoption of EVs increases, the risk of clustering multiple EVs on a single transformer will increase the risks mentioned above. Furthermore, as the next generation of EVs prepare to hit the market, with larger batteries requiring higher kWh charging, the risk to transformers will increase even further.

[0004] When customers purchase electric vehicles and associated charging infrastructure, they are not required to notify their electric utilities about the high load that EVs pose on the electric grid, which, if sufficiently popular, could increase the load and cause electrical equipment failures and power outages on the grid.

[0005] If utilities could accurately identify EVs on the grid, they would be able to proactively take measures to ensure the health and reliability of their generation, distribution, electric grid equipment and other assets. Previous techniques for identifying EVs on the grid were inaccurate or unable to identify EVs at all.

[0006] Discovering user accounts that consume energy to charge electric vehicles is important for electric utilities. Because electric vehicle charging represents a significant load on the electric grid, there is interest in identifying the premises / locations of customers who own electric vehicles. Summary of the Invention [Means for solving the problem]

[0007] overview In one embodiment, a computer-implemented method executed by a computing device having at least one processor is disclosed, the method including retrieving a first set of electrical consumption data from a set of known electric vehicle owners, the electrical consumption data comprising a time series of usage values ​​over a period of time for each electric vehicle owner. and encoding each of the usage values ​​for the plurality of time intervals with one symbol from a sequence of symbols representing a level of electricity consumption during the corresponding time interval, wherein the encoding generates an encoded consumption pattern of the plurality of symbols for each electric vehicle owner. The method further includes identifying an EV charging motif comprising a sequence of the plurality of symbols from the encoded consumption pattern of the plurality of symbols representing an EV charging event that is distinct from a non-charging motif; repeating with additional datasets of electricity consumption data; training one or more machine learning classifiers to identify and distinguish the EV charging motif from the set of known electric vehicle owners; and configuring the one or more machine learning classifiers to identify whether an unknown dataset of electricity consumption data includes an EV charging event and to mark the unknown dataset as having electric vehicle charging or not having electric vehicle charging based on at least the EV charging motif.

[0008] In another embodiment, a computing system is disclosed that includes at least one memory, at least one processor coupled to the at least one memory, and a non-transitory computer-readable medium having stored thereon an electric vehicle detection module configured with instructions that, when executed by the at least one processor, cause the processor to:

[0009] The instructions, when executed by the at least one processor, cause the processor to retrieve a first set of electricity consumption data from a set of known electric vehicle owners, the electricity consumption data including a time series of usage values ​​over a period of time for each electric vehicle owner; convert the time series of usage values ​​for each electric vehicle owner into a data structure of multiple time intervals including corresponding usage values; and encode each of the usage values ​​for the multiple time intervals with a symbol from a set of symbols representing a level of electricity consumption during the corresponding time interval, the encoding generating a multiple-symbol encoded consumption pattern for each electric vehicle owner; and identifying an EV charging motif comprising a sequence of a plurality of symbols from the encoded consumption pattern of the plurality of symbols representing an EV charging event distinct from a non-charging motif, repeating for additional datasets of electricity consumption data; training one or more machine learning classifiers to identify the EV charging motif from the set of known electric vehicle owners and distinguish it from the non-charging motif; and configuring the one or more machine learning classifiers to identify whether an unknown dataset of electricity consumption data includes an EV charging event and mark the unknown dataset as having electric vehicle charging or not having electric vehicle charging based on at least the EV charging motif.

[0010] BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated herein and constitute a part hereof, illustrate various disclosed systems, methods, and other embodiments. It can be appreciated that the boundaries of elements shown in the figures (e.g., boxes, groups of boxes, or other shapes) represent one embodiment of those boundaries. In some embodiments, one element may be realized as multiple elements, or multiple elements may be realized as one element. In some embodiments, an element shown as an internal component of another element may be realized as an external component, and vice versa. Additionally, elements may not be drawn to scale. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 1 illustrates an embodiment of a computing system configured to detect an EV charging event. [Figure 2] FIG. 2 illustrates one embodiment of a method performed by the system of FIG. 1 to detect EV charging events and train a machine learning classifier. [Figure 3] 1 is an exemplary graph overlaying kWh charge values ​​from an account meter and a corresponding EV submeter. [Figure 4] FIG. 1 illustrates an embodiment of a method for detecting EV charging events and EV owners from an unknown dataset. [Figure 5] FIG. 1 illustrates one embodiment of a computing system configured with the disclosed example systems and / or methods. DETAILED DESCRIPTION OF THE INVENTION

[0012] Detailed Description The systems and methods described herein, in one embodiment, provide novel systems and methods that use motif discovery and detection to identify electric vehicle (EV) charging events from utility customer accounts that consume energy to charge electric vehicles. In one embodiment, the systems and methods include training a machine learning classifier to identify electric vehicle charging based on at least EV charging motifs / patterns and charging characteristics.

[0013] It is important for electric utilities to discover user accounts that consume energy to charge electric vehicles. Because electric vehicle charging places a significant load on the electric grid, identifying these customers (e.g., identifying accounts and / or charging locations) is a key concern. For example, if multiple EV owners are in the same neighborhood, these EV owners may share the same electrical transformer. In this case, the transformer could pose a safety hazard if excessive electricity is drawn simultaneously, overloading the transformer. This could result in physical damage to the transformer or even equipment failure. Threats to public safety, such as fire, could result in extreme situations.

[0014] Thus, if multiple EV owners are pre-identified using the present systems and methods, measures can be taken to mitigate or eliminate potential problems with the grid and / or facilities. For example, power lines for homes with EVs may be rerouted to different transformers, transformers may be replaced or resized with higher capacity transformers, and / or EV owners may be sent warnings or notifications requesting changes to their EV charging times.

[0015] There may also be several additional benefits to utilities in being able to accurately identify significant loads on the grid caused by EVs. These benefits may enable them to take one or more resulting actions, such as (1) generating and sending messages to EV owners to change their accounts to different (better) electricity rates for EV charging to induce more favorable customer contracts; (2) identifying whether EV owners are charging during on-peak or off-peak hours and generating and sending instructions to modify charging times and / or account settings that cause one or more EV owners to implement changes to their "time of use" rates to shift their charging load to off-peak hours; (3) identifying EV "hot spots" in the grid that may represent significant load increases in specific areas where multiple EV owners are present; and (4) tracking the trend and rate of EV power uptake at a system-wide level so that corrective actions and modifications can be made to electric grid equipment to ensure safe and proper power distribution (e.g., adjusting or resizing transformers, rerouting power lines, etc.). It includes actions that identify at a more localized level.

[0016] Throughout this disclosure, the following terms are used as follows: EV: Abbreviation for "electric vehicle."

[0017] EV Owner: This represents the premises / location of the account that owns and / or charges an electric vehicle.

[0018] Non-EV Owner: This represents the premises / locations of accounts that do not own or charge electric vehicles.

[0019] Known EV Owner: A label that indicates that an account (and its associated time series data) consumes electricity to charge an electric vehicle and is known to an electric utility provider, thereby indicating that account time series data from a known EV owner has EV charging events. This information is provided to one or more machine learning classifiers of the system to identify and detect EV charging events from the account data.

[0020] Known Non-EV Owner: A label that indicates that the account does not consume electricity to charge electric vehicles and is known to the electric utility provider. Therefore, the account time series data for a known non-EV owner does not contain EV charging events. This information is provided to one or more machine learning classifiers to identify and detect non-EV charging events that can be used to distinguish them from EV charging events.

[0021] Residential Meter: This is an electronic device that records the consumption of electrical energy at a premises / location as electrical consumption data. Consumption is in kilowatts per hour (kWh), although other units may be used. The premises / location, and therefore the residential meter, is associated with a customer account. For simplicity, the term "residential" is used, but it includes any type of premises that may have a meter. The residential meter measures the electrical consumption of the entire premises, including all appliances and components that draw electricity (including an EV, if present). The measured data is typically collected at regular time intervals (e.g., 5 minutes, 15 minutes, 1 hour, etc.). A group of interval data is referred to as time-series data. In one embodiment, the residential meter is a smart meter that records electrical energy consumption and communicates the information to the electricity supplier (utility company) via network communications for monitoring and other purposes.

[0022] EV submeter: A meter installed on the premises of a known EV owner to directly monitor and record charging activity of electric vehicles only. An EV submeter is a device similar to a residential meter, but installed separately in addition to the residential meter. An EV submeter records the electricity drawn by electric vehicles only in kWh and measures charging data over time for electricity used exclusively for EV charging. No other electricity consumption components are connected to the EV submeter.

[0023] Account Time Series Data: Data indicating electricity consumption in kWh for a subject account measured and collected from residential meters overall over a period of time. Measured data is typically collected at time intervals. While 15-minute time intervals are described herein, other intervals (e.g., 5 minutes, 30 minutes, 1 hour, etc.) may also be used.

[0024] Submeter time series data (EV submeter data): Data representing electricity consumption in kWh measured and collected from a dedicated EV submeter for charging electric vehicles only. The data is collected periodically at set time intervals (for example, 15-minute intervals). , other intervals (e.g., 5 minutes, 30 minutes, 1 hour, etc.) can also be used.

[0025] EV Charging Motif / Pattern: A coded string of characters or symbols that corresponds to an identified EV charging event. The coded string may have varying lengths based on the specific EV charging event and is based on the implemented coding scheme.

[0026] Referring to FIG. 1, one embodiment of a computer-implemented EV detection system 100 is illustrated. The EV detection system 100 is configured to run on a computing system and comprises a novel technique for identifying and detecting EV charging motifs. EV charging motifs represent symbolic electricity consumption patterns associated with electric vehicle charging events that can be identified from account time-series data 110. Previous approaches look for predefined kWh-scale electricity consumption events, resulting in low accuracy and low robustness to noise and other data-related effects that can lead to inaccurate detection. The present approach is more robust, is not dependent on product brand or charger, is directly informed by charging data, and can learn and adapt over time as new EV charging motifs are discovered and added to the motif library.

[0027] The overall EV detection system 100 may be composed of three logical stages, each implemented by an executable module: (1) a pre-processing module 120, (2) a detection module 130, and (3) a verification module 140. A more detailed description of the structure and operation of each module is provided with reference to the accompanying drawings.

[0028] In one embodiment, pre-processing module 120 is configured to receive and / or retrieve time series data from one or more accounts, convert the data to interval format as needed, and encode the interval data into a string that quantizes kWh values. This module is described in more detail with reference to FIG. 2.

[0029] In one embodiment, the detection module 130 is configured to analyze time series data and encoding sequences from multiple accounts of known EV owners to detect a specified sequence (e.g., a word / character group) that best matches multiple sequences of EV charging patterns. Charging features recorded as part of EV charging events on the submeter time series are extracted from the account time series data and associated with the EV charging events as features, which are described below. The detection module 130 also performs similar processing on account time series data from a set of known non-EV owners to identify types of charging patterns and features that are not EV charging events. The detection module 130 is described in more detail with reference to FIG. 2.

[0030] In one embodiment, verification module 140 is configured to train one or more machine learning classifiers to learn what EV charging patterns look like based at least on the patterns discovered by detection module 130. The training includes inputting charging characteristics associated with each suspicious EV charging event discovered by detection module 130. The combination of character EV charging patterns and corresponding charging characteristics trains the one or more machine learning classifiers to detect and identify similar EV charging events from unknown accounts. In the following description, the one or more machine learning classifiers are referred to as classifiers.

[0031] The classifier also receives an answer key to validate the classifier's decision. The answer key includes submeter time series data for the labeled account being analyzed. As mentioned above, the submeter data is used to determine whether the corresponding account time series data represents an EV charging event. The validation module 140 identifies whether or not a vehicle has a vehicle and when an EV charging event will occur. The detection results / predictions 160 from the classifier are then compared with the submeter data to validate the accuracy of the classifier, and the classifier may be further trained if the results 160 are inaccurate. The validation module 140 is described in more detail with reference to FIG. 2.

[0032] Referring to Figure 2, one embodiment of an EV detection process 200 implemented and executed by the EV detection system 100 of Figure 1 is shown. Process 200 is a computer-implemented process executed by a computing system. The operations and functions described are performed at least by a processor that accesses data from a memory or storage device and generates data read from and / or written to memory.

[0033] Figure 2 shows three columns of processing flow, each representing a different type of data being processed depending on data availability. In block 210 (left-most column), account time series data from known EV owners is processed. In block 210a (center column), account time series data from known non-EV owners is processed. In block 215 (right-most column), EV submeter data is processed. The three processing columns may be executed at different times from each other and are not necessarily executed simultaneously.

[0034] In the processing of Figure 2, blocks 210, 220, 230, and 240 correspond to pre-processing module 120 of Figure 1. Blocks 250 and 260 correspond to detection module 130 of Figure 1, and blocks 270 and 280 correspond to verification module 140. Note that pre-processing and detection functions are performed on both types of data sets: account time-series data from known EV owners 210 and account time-series data from known non-EV owners 210a. Corresponding functions performed on data from known non-EV owners are designated with an "a" in 220a, 230a, 240a, etc. Thus, the functionality of blocks 220 and 220a is the same except for the data set being processed.

[0035] Pre-processing Stage (Blocks 220, 230, 240) 2, pre-processing begins in block 210, where raw account time series data 110 from known EV owners is retrieved and input into pre-processing module 120. This process is repeated for multiple known EV owners, with each time series from each EV owner account being processed individually if a set of time series data is available. For each known EV owner, their corresponding submeter time series data is also input at the same time or at a later time (block 215). The account time series data 110 is read and retrieved by a processor in the computing system from a storage device or memory that stores the data in a data structure. In the following example, the account time series data 110 is recorded in kWh charging values ​​at 15-minute intervals.

[0036] In block 220, the raw 15-minute interval kWh data from the homes of known EV owners is converted to a long-format time series (i.e., for 15-minute interval data, there are 96 intervals in a day). This results in a 1 row by 96 column data vector. If 30 days of data are available, the vector for each home meter will be 1 row by (96 * 30) column. In one embodiment, only time series data from days with less than 4 unknown or unavailable kWh values ​​are considered. This function is also performed on 15-minute interval kWh data from known non-EV owners in block 220a.

[0037] If the corresponding EV submeter-level data is available to the EV owner, a response key is generated that labels and identifies the time intervals when the EV was charging in block 225. The identified time intervals can then be used to verify whether and when EV charging occurred by identifying the corresponding time intervals in the account time series data, which is done later in the process.

[0038] First, the account time series data is analyzed without looking at the EV submeter data. The pre-processing module 120 analyzes the account time series data 110 to detect whether there is EV charging activity during the day from a known EV owner. As described above, the account time series data 110 represents residential electricity usage (electricity consumption) in kWh from a known EV owner over a period of time. A residential meter records electricity drawn and used by the entire residence, and therefore includes electricity used for EV charging combined with all other charging, if any. This is referred to herein as the account time series data.

[0039] In one embodiment, EV submeter data 215 is a second set of retrieved time-series charging data collected from an EV submeter. For example, each known EV owner also includes a respective EV submeter installed on their premises to directly monitor electric vehicle charging activity. The EV submeter is in addition to the residential meter. As described above, the EV submeter records electricity drawn only by electric vehicles and therefore includes charging data of electricity used for EV charging in a time series. This is referred to as submeter time-series data or EV submeter data. Thus, by analyzing the submeter data, the system can easily identify the time intervals between the start and end of EV charging. These time intervals are marked and become part of the answer key generated in block 225.

[0040] At 230, the account time series data may be normalized. Residential electricity usage throughout the day can be quite noisy, with upward and / or downward spikes in kWh values. In one embodiment, the preprocessing module 120 is configured to convert the time series into a sequence that can represent the specific magnitude and variance of each meter throughout the day. To accomplish this, for example, the kWh value in the account time series for each meter of one or more known EV owners is normalized to itself, and each raw kWh value is transformed to be a measure of several standard deviations away from the mean kWh value for the given account time series. Other techniques for normalizing these values ​​may also be implemented.

[0041] 2, the process shows arrowed lines exiting blocks 220 and 230 and entering decision block 260, which determines charging characteristics. These lines represent that time series data from blocks 220 and / or 230 is provided to decision block 260 so that charging data from the time series can be extracted and used to generate charging characteristics, as described below.

[0042] At 240, the kWh values ​​of the time series data are encoded in an encoding scheme. In one embodiment, the encoding involves encoding each 15-minute kWh interval of electricity usage collected at the meter as one character from a series of letters / symbols. Thus, a series of consecutive time intervals becomes a series of symbols. For example, one type of encoding may be performed on groups of consecutive letters.

[0043] For example, in one embodiment, six consecutive alphabetic characters may be selected to represent low to high usage level characters. One group of six characters may be "a, b, c, d, e, f," with "a" being the lowest usage level. , "f" is the highest usage level. Thus, time intervals during the day with high usage levels may be coded as "f," while lower mid-range levels may be coded as "c." Naturally, more or fewer letters can be used for coarser or finer approximations and representations of usage levels. The sequence of letters also need not be alphabetically consecutive. For example, a sequence could skip letters and include a, c, e, g, i, k.

[0044] Further, an example of an "a,b,c,d,e,f" based encoding scheme is shown in FIG. 3, where kWh values ​​from account time series data 300 are displayed in a graph of time intervals. Each kWh value in each time interval is encoded with one of the encoding characters based on usage level, as described above. The resulting encoding sequence 325 for a series of time intervals for account data 300 is shown below the graph in FIG. 3. Each encoding character in encoding sequence 325 is approximately aligned below its corresponding kWh value and time interval. FIG. 3 is described in more detail below.

[0045] In one embodiment, parameters of the encoding scheme may be adjusted or modified, including (a) the interval length, (b) the probability distribution function to be followed during encoding, and / or (c) the number of characters to be utilized during encoding.

[0046] The low-high usage levels are relative to the overall usage level (e.g., a range) for a particular time series and, therefore, are not dependent on a particular kWh value. For example, if an account time series for a particular day had a kWh value range of 0 kWh to 3 kWh, any time interval with a value of 3 kWh would be assigned the letter "f" as the maximum usage level. Similarly, if another time series from a larger home had a range of 0 kWh to 10 kWh, any time interval with a value of 10 kWh would be assigned the letter "f" as the maximum usage level. This encoding is performed on the time series for all 96 time intervals to generate a string of coded characters (symbols) representing the electricity consumption pattern. The string of coded characters is stored in memory for use in the detection phase. Coding is also performed on account time series data from a set of known non-EV owners in block 240a.

[0047] Thus, the coded consumption pattern of the coding characters for any given time series contains a charging pattern that can be identified and matched with other charging patterns. The coding characters allow the system to detect and match similar charging patterns across different time series data (different consumer accounts) rather than attempting to match kWh values. In the example above, the three time interval values ​​of 3 kWh-3 kWh-3 kWh from the first account would not match the other three time interval values ​​of 10 kWh-10 kWh-10 kWh from the second account. However, after encoding, the three time intervals in both data sets would match by having the same coding sequence pattern fff because each kWh value is the highest value in their corresponding time series data.

[0048] Therefore, quantizing time series data into strings allows for at least two benefits: (1) collapsing the dimensions of the time series into a single string, and (2) more easily detecting anomalies based on a particular string.

[0049] Regarding the first advantage mentioned above, rather than relying on the actual magnitude or value from the kWh usage time series (which can easily vary depending on the size of the home or the number of occupants in the dwelling), the letter coding already normalizes each meter's time series to itself. Thus, high usage for a particular meter is easily recognized by the "higher" letter (i.e., low usage might correspond to "a" or "b," and high usage might correspond to "e" or "f"). Thus, the encoding not only preserves the unique characteristics of the meter over time, but also allows comparisons between different meters.

[0050] When combined with a sequence search algorithm, sequence encoding naturally provides the second advantage mentioned above: it is easier to search a string of characters for a particular pattern (e.g., search for a "word") than to identify anomalies by kWh count. Once suspicious event detection is complete, a classifier can be easily trained on charging features associated with suspicious EV events to distinguish true positives from false positives, as described below.

[0051] Detection Module (Blocks 250 and 260) After the kWh data is encoded into characters in block 240, the encoded characters in the time series are read from memory and analyzed to detect EV charging patterns in block 250. If EV submeter data is not available for a given account time series, the detection module 130 is configured to analyze the encoded characters from the account time series data. Given that electric vehicle charging is typically characterized by some sustained period of higher levels of kWh usage, it is highly likely that some of the time series data indicative of EV charging will be a string of consecutive high usage characters.

[0052] Based on the encoding scheme described above (using low-high usage level characters "a," "b," "c," "d," "e," and "f"), known EV owner accounts will contain high-usage strings that form a character pattern representative of EV charging (e.g., mostly f's with some e's). Thus, the system searches the encoded characters to identify suspicious charging of electric vehicles by identifying a sequence of characters / symbols in the encoded consumption pattern of symbols that represent a sequence of sustained high-usage level symbols (e.g., multiple consecutive f's and possibly some e's).

[0053] This character pattern of high-use characters over a sustained period is stored as a motif and stored in a motif library of EV usage patterns. The high-use character patterns are found across multiple known EV owner accounts to generate several different EV charging patterns (motifs). The EV charging patterns can then be used to identify and detect similar EV charging patterns from other account time series datasets where EV charging is unknown.

[0054] Thus, for a given time series, the system is configured to search for high-use characters of various lengths that occur nearly consecutively throughout the entire sequence of characters over a period of time. In one embodiment, the system searches for sequence similarity with motifs in a motif library to return potential matches for high-use characters throughout an account's encoded time series. If the search returns potential matches for one or more known EV charging motifs during a day, the system marks that day as having EV charging activity. In one embodiment, the search function is a component of the system's detection module 130.

[0055] Accounting time-series data from a location / premises is compared with EV submeter data collected at the same location / premises. If a suspect charge in the accounting data overlaps in time with any of the charges from the submeter data, the suspect charge includes an EV charge. As mentioned above, any charge occurring in the EV submeter data should be an EV charge. The system then marks the suspect charge as an EV charge. An example is shown in Figure 3.

[0056] In one embodiment, in block 250 of FIG. 2, the system or a string of highly used coded characters. Suspicious EV charging can then be correctly identified and verified by using the corresponding EV submeter data (e.g., the generated answer key) for that account. For a known EV owner, suspicious EV charging can be identified and verified in the account time series data by matching a time interval from the account time series data with the corresponding EV submeter time series data. The submeter's 15-minute kWh interval will indicate when the EV owner approximately started charging their electric vehicle and when EV charging stopped.

[0057] The start and stop time intervals from the EV submeter data can be compared to the suspected EV charging and found marked in the corresponding account time series data. If the time intervals match (overlap with each other), the suspected EV charging in the account data is truly EV charging. If the suspected EV charging does not overlap with the charging time interval from the EV submeter, the suspected EV charging is not EV charging. An example of account time series data 300 matched (e.g., overlaid) with its corresponding EV submeter time series data 305 is shown in FIG. 3.

[0058] In FIG. 3, the x-axis is a 15-minute time interval, and the y-axis shows the kWh values ​​recorded at the corresponding time interval. Over a 24-hour day, there are 96 kWh values ​​in the time series (e.g., a 1×96 vector of kWh values). FIG. 3 shows only a portion of a one-day time interval. Generally, when EV charging occurs, the kWh values ​​in EV submeter data 305 increase sharply from 0 kWh to a much higher value. This is indicated by the “rising spike” point. Furthermore, the kWh level remains at that high level for a sustained period (represented by the “sustained level” area). When EV charging stops, the magnitude of the kWh values ​​in EV submeter data 305 drops sharply (indicated by the “falling spike” point).

[0059] During the time intervals of the rising spike and the falling spike, the system recognizes that these are start and stop time intervals that designate an EV charging event 310. The time intervals where EV charging from EV submeter data 305 overlaps with charging in accounting data 300 identify EV charging in accounting data 300. Note that the kWh values ​​in accounting data 300 include all electric charging from the entire premises, with EV charging being one component of charging. Using the start and stop time intervals from the EV submeter data, the system can identify where in accounting time series data 300 the actual EV charging event is occurring, which is identified as EV charging event 310.

[0060] Once the start and stop time intervals of the EV charging event 310 are discovered, the system identifies the corresponding coded characters assigned to the same time periods of the EV charging event 310. As shown in Figure 3, the coded characters 325 for the account data 300 include the string "ffffffefffffe" during the EV charging event 310. Therefore, this coded string represents a known EV charging event and is stored as one recognized EV charging pattern / motif among the coded characters. The spaces within the string are unnecessary and are simply shown to make the coded characters easier to read.

[0061] Referring again to FIG. 2 , in block 260, once an EV charging event and its start and stop time intervals are identified, the detection module 130 identifies charging features associated with the EV charging event. For example, data collected in the account time series data 300 occurring before, during, and at the end of the suspected or identified EV charging event is read. The charging features are extracted and / or calculated from the EV charging event to help characterize EV charging from non-EV charging. The charging features identify what EV charging resembles and what it does not resemble, so that EV charging can be distinguished. This will help the system learn.

[0062] For example, the system may extract kWh values ​​and calculate various charging features, such as the average consumption of electricity during the EV charging event, the magnitude of the initial kWh spike at the beginning of EV charging, the magnitude of the kWh drop at the end of EV charging, the duration of the EV charging event, the standard deviation of kWh values ​​during EV charging, the total kWh usage during the EV charging event, the maximum and / or minimum kWh values ​​during the EV charging event, etc. Any combination of these features may be used, and other features may be calculated based on the charging data available for a given dataset. The one or more charging features used and stored with the encoded EV charging pattern serve as a charging characteristic or signature of the associated EV charging event.

[0063] If the suspected EV charge does not overlap with an EV charge from the EV submeter data, the suspected EV charge is not an actual EV charge. The system may extract the kWh value and calculate various charging characteristics for this non-EV charging pattern, as described above. However, these charging characteristics and the associated coded character patterns are stored as non-EV charging characteristics that can be used to distinguish charging events during subsequent determinations.

[0064] 2 , if there is no non-EV charging event, the detection module 130 performs similar processing at blocks 250a and 260a for known non-EV data. The system identifies coded character patterns known to be non-EV charging and determines their associated charging characteristics, as described above. The non-EV charging coding patterns and their charging characteristics characterize what non-EV charging resembles. Thus, the non-EV charging data can be used to assist in determining whether a suspected EV charge is truly an EV charge or a non-EV charge.

[0065] Thus, the system uses account time series data from known non-EV owners as a control group to compare with known EV owners to create "true negative" samples. The kWh data and time intervals from members of the control group are processed in the same or similar manner as the charging data and charging characteristics from known EV owners described above.

[0066] Verification blocks 270 and 280 At block 270, EV charging patterns and their associated charging features are input to one or more machine learning classifiers (simply, classifiers). This data trains the classifiers to learn from the charging features to identify and detect EV charging events from an unknown dataset. Similarly, non-EV charging patterns and their associated charging features are input to train the classifiers to identify non-EV charging patterns and distinguish them from EV charging events.

[0067] The classifier can be trained and run any number of times on the account time-series data. The classifier is configured to generate a list of detected EV events that the classifier has identified as EV charging based on known EV charging patterns / motifs and charging characteristics. In block 280, the generated list can be compared and validated against the answer key generated in block 225 from actual EV submeter data. Results of the validation can be fed back to the classifier to correct incorrect or missed EV charging events, thereby further training the classifier.

[0068] In one embodiment, the classifier includes a machine learning algorithm configured to determine how likely it is that the electric charge being analyzed is a true EV charge. The classifier uses the detected EV charging patterns and EV charging features from the detection module 130 to The classifier is trained to learn the difference between EV and non-EV charging using the data. The classifier also uses charging characteristics from known non-EV owners to distinguish between true EV charging and non-EV charging. The answer key generated in block 225 from the EV submeter data allows the system to test the results of the classifier. The answer key may also be used to train the classifier to identify which charges are from EV owners and which are from non-EV owners.

[0069] Using detected EV charging patterns and charging characteristics from known EV owners and charging characteristics from known non-EV owners, the machine learning classifier learns based on observed experience. In this way, the machine learning classifier can be programmed by providing the detected information used to train it, and the classifier gains the ability to identify EV charging elements or characteristics with a high degree of probability.

[0070] Training a classifier to look for EV charging patterns is part of a supervised learning process. Supervised learning provides a computer with the ability to recognize factors based on provided samples. In this case, these samples include EV charging patterns and charging features that characterize associated EV charging events. The computer studies the samples and develops the ability to recognize new data based on the provided data. In one or more embodiments, supervised learning algorithms may include decision trees, support vector machines, naive Bayes classifiers, k-nearest neighbors, and / or linear regression.

[0071] In one embodiment, the machine learning classifier is implemented with XGBoost, which implements machine learning algorithms under the gradient boosting framework. XGBoost provides parallel tree boosting that solves many data science problems quickly and accurately. The classifier code can be implemented and run on a variety of distributed environments, including Hadoop, SGE, and MPI.

[0072] In one embodiment, gradient boosting is a machine learning technique for regression and classification problems that generates weak predictive models, typically in the form of a collection of decision trees. Predictive modeling is the process in which a model is generated to predict an outcome. If the outcome is categorical, it is called classification; if the outcome is numerical, it is called regression.

[0073] Once the machine learning classifier is trained, it can be used to identify whether an unknown account has suspicious EV charging. The detection process can be systematically performed on a large number (thousands, tens of thousands, etc.) of unknown accounts. Accordingly, account time series data collected from meters of unknown accounts may be input to the machine learning classifier for identification of suspicious EV charging. If suspicious EV charging is found in a given time series dataset, the associated account is marked and labeled as a suspicious EV owner. One or more resulting actions can then be performed based on the identified EV owner. An example of this process is described with reference to FIG. 4. Other classifiers using additional inputs may be cascaded to the classifier to improve performance.

[0074] The novel technology implemented by the present system and method responds to EV detection questions with a "yes / no" answer. Using one or more steps from the process of Figure 2, the system can solve for:

[0075] (1) For known electric vehicle owners, detect whether there is any charging activity during the day.

[0076] (2) For known electric vehicle owners, detect when their electric vehicles are plugged in for charging based on residential electricity usage (measured by account time-series data).

[0077] (3) Regarding the time series of the account, whether the EV is being charged or not. In one embodiment, the system implements and uses a logistic classifier and a decision tree-based algorithm (e.g., Support Vector Machine (SVM)) to identify EV charging. For the logistic classifier, suspicious EV events that are assigned a probability greater than 0.5 are classified as EV charging. Of course, other values ​​may be used.

[0078] Exemplary results observed for the present detection system to address (1) above were as follows:

[0079] precision=0.852564102564

[0080] [Table 1]

[0081] For question (3) above, based on a given set of account time series data, a machine learning classifier implemented in an SVM model produced the following results:

[0082] [Table 2]

[0083] Referring to Figure 4, one embodiment of an EV detection process 400 for identifying and detecting suspicious EV charging from account meter data for unknown accounts is shown. Here, an unknown account refers to an account for which the EV owner and the existence or nonexistence of EV charging associated with the account are unknown. The detection process of method 400 uses the trained machine learning classifier described above to predict whether an unknown account is associated with an electric vehicle.

[0084] The detection process can be performed systematically on selected accounts or on a large number (thousands, tens of thousands, etc.) of unknown accounts. To this end, account time-series data collected from meters of unknown accounts may be input into a machine learning classifier for identification of suspicious EV charging. If suspicious EV charging is found in a given time-series dataset, the associated account is marked and labeled as a suspicious EV owner. One or more resulting actions can then be performed based on the identified EV owner.

[0085] At 410, an account time series data set from an unknown account is read from memory or data storage and provided as input. At 420 and 430, the account time series data is converted to a long-format time series (e.g., a data vector) and the kWh values ​​are normalized as performed in blocks 220 and 230 of Figure 2. At 440, the normalized kWh values ​​are encoded with an encoding character scheme as performed in block 240 of Figure 2. As mentioned above, the exemplary scheme uses six characters for low-high usage levels: a, b, c, d, e, and f.

[0086] At 450, the coded characters from the data vector are input to a machine learning classifier. The coded characters are searched for character patterns of high-use characters, such as a string of "f" characters in a row, over a duration. As is known, a string of high-use characters can represent EV charging. If such a string is not found, the account is marked as a non-EV owner, and the next account time series can be input for analysis. If a string of high-use characters is found, the machine learning classifier identifies the corresponding time interval and kWh charging as suspicious EV charging.

[0087] At 460, the machine learning classifier determines charging features from the suspect EV charge in a manner similar to the charging features described above for block 260 of FIG. 2 . For example, kWh data collected in the account time-series data occurring before, during, and / or at the end of the suspect EV charge is read from memory or storage. One or more charging features are generated. The one or more charging features may be, for example, the average consumption of electricity during the suspect EV charge, the magnitude of the initial spike in kWh at the start of the suspect EV charge, the magnitude of the drop in kWh at the end of the suspect EV charge, the duration of the suspect EV charging event, the standard deviation of the kWh values ​​during the suspect EV charge, the total kWh usage during the suspect EV charge, the maximum kWh value and / or the minimum kWh value during the suspect EV charge, etc. Any combination of these charging features may be generated, which represent charging characteristics (e.g., a signature) of the associated suspect EV charge.

[0088] At 470, the machine learning classifier determines whether the charging features and coding characters of the suspect EV charging match any of the known EV charging patterns / motifs and their corresponding known charging features. This match may be based on comparing the corresponding types of charging features and determining whether these values ​​match to some extent (e.g., a set threshold amount). For example, the classifier determines whether the average consumption of electricity used during the suspect EV charging matches the average consumption of electricity used during known EV charging events from the matching EV charging pattern / motif.

[0089] This comparison may be performed for other different types of charging features, and a total score of the match results may be generated. A threshold amount of match similarity for each type of charging feature may be defined and set in the classification algorithm. In one embodiment, this matching involves pattern matching using learned observations from known EV charging patterns / motifs and their corresponding charging features, as learned from the process of FIG. 2.

[0090] At 480, if the suspected EV charging matches one or more known EV charging patterns and matches associated EV charging characteristics to a specified degree, the suspected EV charging is marked as an EV charging and the account is marked as a potential EV owner account. The process may be repeated for additional accounts. A list or other data output is generated identifying accounts likely to be associated with electric vehicle charging. These accounts are marked as EV owners.

[0091] If the charging characteristics of the suspect EV charge do not match within a threshold of known EV charging characteristics, the suspect charge is marked as a non-EV charge and the analysis is repeated for any other suspect EV charges in the account time series dataset. If no EV charges are identified, the account is marked as a non-EV owner. Processing of method 400 may continue and be repeated for further time series datasets.

[0092] One or more resulting actions may then be performed at 490 based on the identified EV owner. In one embodiment, the one or more resulting actions include, for example, generating and sending a message to the EV owner's remote device to change the EV owner's account to a different (better) electricity rate for EV charging (providing a more desirable customer agreement). The system may identify whether the EV owner is charging during on-peak or off-peak hours. Based on the charging time being an on-peak hour, the system may generate and send an electronic message with instructions to the EV owner's associated remote device. The electronic message may include instructions to modify the charging time to an off-peak hour and / or instructions to modify account settings that cause one or more EV owners to implement changes to their "time of use" rates to shift their EV charging load to off-peak hours.

[0093] Another resulting action may include identifying the location of the EV owner within the electrical grid and determining the location of the EV owners relative to each other. The system can then identify EV "hot spots" within the grid that may represent significant load increases in particular areas with multiple EV owners.

[0094] Based on the periodic monitoring and identification of EV owners using the system, the system may identify trends and rates of EV power uptake at a system-wide level and / or at a more localized level. If a certain increase in the number of EV owners detected in a particular area of ​​the grid is detected, corrective actions and modifications can be made to electrical grid equipment to ensure safe and proper power distribution. These actions may include adjusting or resizing transformers, rerouting power lines, etc.

[0095] In one embodiment, generating the instructions includes generating an electronic message including information regarding the detected EV charging and information regarding changing EV charging behavior (e.g., changing charging times to off-peak hours). Changing the charging behavior can reduce load on the electric grid at a particular time. The generated message, such as an email, text message, electronic phone message, or other electronic format, can then be sent via network communication to a remote device associated with the detected EV owner.

[0096] In another embodiment, the system and method may instruct and control a printer to print a hard copy of the generated message to be delivered in physical form to a location associated with the detected EV owner.

[0097] Another embodiment of an electric vehicle detection system based on the above description may perform the following processes: The detection system may receive a first time series of electrical consumption data from an electric vehicle (EV) owner and encode (by a processor accessing a memory) the time series into a symbolic EV usage pattern including an EV charging pattern, as described above. The detection system may then encode a second time series of electrical consumption data from a non-electric vehicle (non-EV) owner into a multi-symbol non-EV usage pattern by the processor accessing the memory. The detection system may then encode the multi-symbol EV usage pattern (including the EV charging pattern) to detect and identify similar electric vehicle charging patterns from unknown time series of electrical consumption data. The machine learning classifiers can be constructed and trained using multi-symbol non-EV usage patterns (including EVs).

[0098] After one or more rounds of training, a machine learning classifier may be used to analyze an unknown data set. For example, in response to receiving a first unknown time series of electricity consumption data, the machine learning classifier detects whether the first unknown time series of electricity consumption data contains similar electric vehicle charging patterns based on the information used to train the classifier. Based on the detection, the machine learning classifier generates an output that identifies the first unknown time series as an EV owner or a non-EV owner. This may be repeated for any number of unknown time series of data.

[0099] Electric utilities can use the improved accuracy of the electric vehicle detection system and method of the present invention to identify accounts with electric vehicles. As described above, various actions can be taken after one or more EV owners are identified. Based on the location of the identified EV owners, network elements on the electric grid that may be subject to overload can be identified. Overloads can occur when multiple EV owners are detected and discovered on the electric grid in the same neighborhood. This directly facilitates the utility's ability to better control loads on the electric grid, target energy efficiency programs to identified EV owners, create and target demand response programs to identified EV owners, and / or perform delayed or just-in-time upgrades to network elements in the grid.

[0100] Cloud or Enterprise Implementation In one embodiment, the electric vehicle charging detection system is a computing / data processing system that includes an application or collection of distributed applications for an enterprise organization, the applications and computing system being a cloud-based networking system, a software-as-a-service (SaaS) architecture, or other type of networking. The detection system may be configured to operate using a computing solution, or may be implemented as a cloud-based networking system, a SaaS architecture, or other type of networked computing solution. In one embodiment, the detection system is implemented as part of a centralized server-side application that provides at least the functionality disclosed herein and is accessed by many users via computing devices / terminals that communicate with computing system 100 (acting as a server) over a computer network.

[0101] In one embodiment, one or more of the components described herein, including EV detection system 100, are configured as program modules stored on a non-transitory computer-readable medium. The program modules are comprised of stored instructions that, when executed by at least one processor, cause a computing device to perform the corresponding functions described herein.

[0102] Computing Device Embodiments In one embodiment, FIG. 5 illustrates a computing device configured and / or programmed as a dedicated computing device using one or more components of the electric vehicle detection system 100 and the methods and / or equivalents described herein.

[0103] The exemplary computing device includes a hardware processor 510 and memory 512. 15 and an input / output port 520 operably connected by a bus 525. In one example, the computer 505 comprises the EV detection system 100 shown and described with reference to Figures 1-4. In various examples, the EV detection system 100 may be implemented in hardware, a non-transitory computer-readable medium having instructions stored thereon, firmware, and / or combinations thereof.

[0104] In one embodiment, the EV detection system 100 and / or the computer 505 are means (e.g., structures: hardware, non-transitory computer-readable media, firmware) for performing the actions described above. In some embodiments, the computing device may be a server operating within a cloud computing system, a server configured in a Software as a Service (SaaS) architecture, a smartphone, a laptop, a tablet computing device, etc.

[0105] The EV detection system 100 may also be implemented as stored computer-executable instructions presented to the computer 505 as data 540 that are temporarily stored in the memory 515 and then executed by the processor 510 .

[0106] Briefly describing an exemplary configuration of computer 505, processor 510 may be a wide variety of processors, including dual microprocessors and other multi-processor architectures. Memory 515 may include volatile memory and / or non-volatile memory. Non-volatile memory may include, for example, ROM, PROM, EPROM, EEPROM, etc. Volatile memory may include, for example, RAM, SRAM, DRAM, etc.

[0107] The storage disk 535 may be operatively connected to the computer 505, for example, via an input / output (I / O) interface (e.g., card, device) 545 and an input / output port 1020. The disk 535 may be, for example, a magnetic disk drive, a solid-state disk drive, a floppy disk drive, a tape drive, a Zip drive, a flash memory card, a memory stick, etc. Additionally, the disk 535 may be a CD-ROM drive, a CD-R drive, a CD-RW drive, a DVD ROM, etc. The memory 515 may store, for example, process 550 and / or data 540. The disk 535 and / or memory 515 may store an operating system that controls and allocates resources of the computer 505.

[0108] Computer 505 may interact with input / output (I / O) devices through I / O interface 545 and input / output ports 520. Communication between processor 510 and I / O interface 545 and ports 520 is managed by input / output controller 547. I / O ports 520 may include, for example, serial ports, parallel ports, and USB ports.

[0109] The computer 505 can operate in a networked environment and, as such, can be connected to a network device 555 via the I / O interface 545 and / or the I / O port 520. The computer 505 can interact with a network 560 via the network device 555. The computer 505 can also be logically connected to a remote computer 565 via the network 560. Networks with which the computer 505 can interact include, but are not limited to, a LAN, a WAN, and other networks.

[0110] The computer 505 can send and receive information and signals from one or more output or input devices via I / O ports 520. Output devices include one or more displays 570, printers 572 (such as inkjet, laser, or 3D printers), and audio output devices 574 (such as speakers or headphones). Input devices include one or more text input devices 580 (such as a keyboard), cursor controllers 582 (such as a mouse, touchpad, or touchscreen), audio input devices 584 (such as a microphone), video input devices 586 (such as video and still cameras), or other input devices such as a scanner 588. Input / output devices may also include disks 535, network devices 555, and the like. In some cases, the computer 505 may be controlled by information or signals generated or provided by input or output devices, such as by the text input device 580, cursor controllers 582, voice input device 584, disks 535, and network devices 555.

[0111] Definitions and Other Embodiments None of the functions described or claimed herein are performed manually by a human being or performed by mental consciousness, and any such interpretation is inconsistent with this disclosure.

[0112] In one embodiment, the described methods and / or their equivalents are implemented with stored computer-executable instructions. Thus, in one embodiment, a non-transitory computer-readable / storage medium comprises stored computer-executable instructions for an algorithm / executable application that, when executed by a machine, causes the machine (and / or associated components) to perform the method. Exemplary machines include, but are not limited to, processors, computers, servers operating in a cloud computing system, servers configured in a Software-as-a-Service (SaaS) architecture, smartphones, etc. In one embodiment, a computing device is implemented with one or more executable algorithms configured to perform any of the disclosed methods.

[0113] In one or more embodiments, the disclosed methods or equivalents thereof are performed by either computer hardware configured to perform the methods or by computer instructions embodied in modules stored on a non-transitory computer-readable medium or in a memory, the instructions being configured as an executable algorithm configured to perform the methods when executed by at least a processor of a computing device.

[0114] For simplicity of explanation, the example method shown in the figures is illustrated and described as a series of algorithmic blocks, but it should be understood that the method is not limited by the order of the blocks. Some blocks may occur in a different order than illustrated and described and / or concurrently with other blocks. Furthermore, fewer than all of the illustrated blocks may be used to implement the example method. Blocks may be combined or separated into multiple actions / components. Furthermore, additional and / or alternative methods may use additional actions not shown in the blocks.

[0115] The following contains definitions of selected terms used herein. The definitions include various examples and / or forms of components that fall within the scope of the term and that may be used for implementation. These examples are not intended to be limiting. Both singular and plural forms of a term may be within the scope of the definition.

[0116] References to "one embodiment," "an embodiment," "an example," "an example," etc., indicate that the embodiment or example so described may include a particular feature, structure, characteristic, property, element, or limitation, but not all embodiments or examples necessarily include that particular feature, structure, characteristic, property, element, or limitation. Furthermore, repeated use of the phrase "in one embodiment" does not necessarily refer to the same embodiment, although it may.

[0117] As used herein, a "data structure" is an organization of data within a computing system that is stored in memory, storage, or other computerized system. A data structure may be, for example, any one of a data field, a data file, a data array, a data record, a database, a data table, a graph, a tree, a linked list, etc. A data structure may be formed from and may contain many other data structures (e.g., a database contains many data records). Other examples of data structures are possible according to other embodiments.

[0118] As used herein, "computer-readable medium" or "computer storage medium" refers to a non-transitory medium that stores instructions and / or data that, when executed, are configured to perform one or more of the disclosed functions. Data may function as instructions in some embodiments. Computer-readable media may take forms including, but not limited to, non-volatile media and volatile media. Non-volatile media may include, for example, optical disks, magnetic disks, and the like. Volatile media may include, for example, semiconductor memory, dynamic memory, and the like. Common forms of computer-readable media are floppy disks, flexible disks, hard disks, magnetic tape, other magnetic media, application specific integrated circuits (ASICs), Programmable logic devices, compact disks (CDs), other optical media, random access memory (RAM), read-only memory (READ-ONLY MEMORY), The media may include, but are not limited to, read only memory (ROM), memory chips or cards, memory sticks, solid state storage devices (SSD), flash drives, and other media on which a computer, processor, or other electronic device can function. When selected for implementation in an embodiment, each type of media may store instructions for an algorithm configured to perform one or more of the disclosed and / or claimed functions.

[0119] As used herein, "logic" refers to components embodied in computer or electrical hardware, non-transitory media having executable application or program module instructions stored thereon, and / or combinations thereof, to perform any of the functions or actions disclosed herein and / or to cause other logic, methods, and / or systems disclosed herein to perform functions or actions. Equivalent logic may include firmware, a microprocessor programmed with an algorithm, discrete logic (e.g., ASIC), at least one circuit, analog circuit, digital circuit, programmed logic device, memory device containing algorithmic instructions, etc., any of which may be configured to perform one or more of the disclosed functions. In one embodiment, logic may include one or more gates, combinations of gates, or other circuit components configured to perform one or more of the disclosed functions. Where multiple logics are described, it may be possible to incorporate the multiple logics into one logic. Similarly, where a single logic is described, it may be possible to distribute the single logic among multiple logics. In one embodiment, one or more of these logics are corresponding structures associated with performing the disclosed and / or claimed functions. The choice of which type of logic to implement depends on the desired system requirements or specifications. For example, if higher speed is a consideration, hardware would be selected to implement the function. If lower cost is a consideration, stored instructions / executable applications would be selected to implement the function.

[0120] An "operable connection," or a connection through which entities are "operably connected," is a connection through which signals, physical communications, and / or logical communications may be sent and / or received. An operable connection may include physical interfaces, electrical interfaces, and / or data interfaces. An operable connection may include various combinations of interfaces and / or connections sufficient to enable operable control. For example, two entities may be operably connected to communicate signals to each other directly or through one or more intermediate entities (e.g., processors, operating systems, logic, non-transitory computer-readable media). Logical and / or physical communication channels may be used to create an operable connection.

[0121] As used herein, a "user" includes, but is not limited to, one or more people, computers or other devices, or combinations thereof.

[0122] Although the disclosed embodiments have been illustrated and described in considerable detail, it is not intended to restrict or in any way limit the scope of the appended claims to such detail. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing various aspects of the subject matter. Accordingly, the disclosure is not limited to the specific details or specific examples shown and described. Accordingly, the disclosure is intended to embrace those changes, modifications, and variations that satisfy statutory subject matter requirements and fall within the scope of the appended claims.

[0123] If the term "includes" or "including" is used in the detailed description or To the extent that this term is used in the claims, it is intended to be as inclusive as the term "comprising," as interpreted when used as a transitional term in the claims.

[0124] To the extent the term "or" is used in the detailed description or claims (e.g., A or B), it is intended to mean "A or B or both." If applicant intends to indicate "only A or B, but not both," the phrase "only A or B, but not both" would be used. Thus, use of the term "or" herein is inclusive, not exclusive.

Claims

1. 1. A computer-implemented method executed by a computing device having at least one processor, comprising: receiving unknown time series data of electricity consumption usage values ​​from a plurality of accounts, wherein it is unknown whether the unknown time series data corresponding to the accounts include an electric vehicle (EV) charging event; and the computer-implemented method further comprising: For a given account of the plurality of accounts: converting the unknown time series data into a data structure of multiple time intervals including corresponding used values; encoding each of the plurality of time intervals with one symbol from a series of symbols representing a level of electrical consumption during the corresponding time interval; The encoding step generates a coding consumption pattern of a plurality of symbols, and the computer-implemented method further comprises: running a machine learning classifier using the encoded consumption pattern of the plurality of symbols, the machine learning classifier being trained to identify a set of known electric vehicle (EV) charging motifs comprising sequences of a plurality of symbols representing EV charging events that are distinct from non-charging motifs; and the machine learning classifier detecting whether the encoded consumption pattern of the plurality of symbols from the given account includes a sequence of high-use symbols similar to one of the set of known EV charging motifs; the machine learning classifier marking the given account as having an electric vehicle charging event or as not having an electric vehicle charging event based at least on the detecting step; Detecting whether the coding consumption pattern of the plurality of symbols includes a sequence of high-use symbols comprises: searching for a symbol pattern of high usage symbols over a duration of a plurality of time intervals from among the coding consumption patterns of the plurality of symbols generated from the unknown time series data; identifying the signature pattern as a suspect EV charging event; determining a charging characteristic from the usage values ​​for the duration of the plurality of time intervals; and determining, with the machine learning classifier, whether the signature pattern and the charging characteristics of the suspicious EV charging event match any of the known EV charging motifs and corresponding known charging characteristics of known EV charging events.

2. The encoding step further comprises: The computer-implemented method of claim 1 , comprising applying a character encoding scheme representing the level of electricity consumption ranging from a low usage level to a high usage level for the range of usage values ​​in the unknown time series data.

3. The encoding step further comprises:

2. The computer-implemented method of claim 1, further comprising encoding each of the plurality of time intervals of electricity usage into one symbol of the series of symbols representing the level of electricity consumption at the given time interval, wherein a series of consecutive time intervals becomes an encoded consumption pattern of the plurality of symbols.

4. The marking step further comprises: identifying the given account as having an electric vehicle in response to detecting that the encoded consumption pattern of the plurality of symbols matches one motif of the set of known EV charging motifs; and identifying the given account as a non-electric vehicle owner in response to detecting that the encoded consumption pattern of the plurality of symbols does not match one motif of the set of known EV charging motifs.

5. The computer-implemented method further comprises:

5. The computer-implemented method of claim 1, further comprising generating and sending an electronic message to a remote device associated with the account, the electronic message including instructions to change EV charging times to off-peak hours.

6. 1. A computing system comprising: at least one memory; at least one processor coupled to said at least one memory; an electric vehicle detection module stored on a computer-readable medium and including instructions that, when executed by the at least one processor, cause the at least one processor to: receiving unknown time series data of electricity consumption usage values ​​from a given account of a plurality of accounts, wherein it is unknown whether the unknown time series data includes an electric vehicle (EV) charging event; and the instructions cause the at least one processor to: converting the unknown time series data into a data structure of a plurality of time intervals including corresponding used values; encoding each of the plurality of time intervals with a symbol from a series of symbols representing a level of electrical consumption during the corresponding time interval; The encoding generates a coded consumption pattern of a plurality of symbols, and the instructions cause the at least one processor to: and further causing the at least one processor to execute a machine learning classifier using the encoded consumption pattern of the plurality of symbols, the machine learning classifier being trained to identify a set of known electric vehicle (EV) charging motifs comprising sequences of a plurality of symbols representing EV charging events that are distinguished from non-charging motifs. the machine learning classifier detecting whether the coded consumption pattern of the plurality of symbols from the given account includes a sequence of high-use symbols similar to one of the set of known EV charging motifs; further causing the machine learning classifier to mark the given account as having an electric vehicle charging event or as not having an electric vehicle charging event based at least on the detecting; Detecting whether the coding consumption pattern of the plurality of symbols includes a sequence of high-use symbols comprises: searching for a symbol pattern of high usage symbols over a duration of a plurality of time intervals from among the coding consumption patterns of the plurality of symbols generated from the unknown time series data; identifying the signature pattern as a suspect EV charging event; determining a charging characteristic from the usage values ​​for the duration of the plurality of time intervals; and determining, with the machine learning classifier, whether the signature pattern and the charging characteristics of the suspicious EV charging event match any of the known EV charging motifs and corresponding known charging characteristics of known EV charging events.

7. The instructions further include causing the at least one processor to: identifying a location of the EV owner within an electric grid based on the plurality of accounts marked as having the electric vehicle charging event; and identifying EV hotspots in the electrical grid that represent increased load in areas with multiple EV owners.

8. A computer readable program for causing at least one processor to carry out the method of any one of claims 1 to 5.

Images