Electric vehicle load identification system method, system, equipment and medium

Abstract

The invention discloses an electric vehicle load identification system and method, a system, equipment and a medium. The method comprises the following steps: constructing a load identification model; acquiring power lumped data and performing multi-stage filtering to obtain a charging load characteristic component; obtaining the starting and ending time of the charging event by adopting a change point detection algorithm and combining with the feature constraint condition; obtaining an optimal matching path by adopting a sequence matching algorithm, and reconstructing a single charging event load curve; and correspondingly combining all the reconstructed single charging event load curves according to a time axis to form an electric vehicle charging load curve. According to the method, the charging load characteristic component is extracted by constructing the load identification model and combining the multi-stage filtering, the charging event is identified by introducing the characteristic constraint condition and the change point detection algorithm, and the single charging event load curve is reconstructed through the sequence matching algorithm and integrated to form the electric vehicle charging load curve, so that the identification accuracy is improved; reliable data support is provided for power grid monitoring and charging management, and the identification precision problem of a traditional method is solved.

Description

Technical Field

[0001] This invention relates to the field of novel power system technology, and in particular to a method, system, device and medium for identifying loads in electric vehicles. Background Technology

[0002] With the accelerated development of the new energy industry, electric vehicles, with their advantages of being environmentally friendly, clean, and having low carbon emissions, have become an important force in energy transition and environmental governance, gaining policy support from governments and favor from capital markets worldwide, resulting in a continuous surge in their market share. Consequently, the impact of electric vehicle charging behavior on the power grid is becoming increasingly significant. The randomness and concentration of large-scale charging loads can easily cause grid load fluctuations and voltage instability, placing higher demands on the grid's refined planning, dynamic scheduling, and energy optimization. Therefore, developing technologies that can accurately identify and monitor electric vehicle charging loads in real time, and extract key data such as changes in charging power and duration distribution, has become a core requirement for supporting the stable operation of new power systems and improving the grid's charging load management capabilities, which is of great significance for promoting vehicle-grid collaboration.

[0003] Currently, non-intrusive load identification (NILM) technology is the mainstream method for separating and aggregating single loads in power signals. It identifies the target load by extracting high- and low-frequency characteristic signals and combining them with algorithms to decompose mixed lumped data of multiple loads. However, its application to electric vehicle charging load identification has obvious limitations: First, it requires high data accuracy and relies on high-frequency data acquisition to avoid overlapping load features, resulting in high equipment costs and heavy data processing pressure, making it difficult to promote on a large scale in low-cost scenarios. Second, model construction and optimization are complex, requiring massive amounts of data for manual annotation and training, as well as repeated adjustments to hyperparameters to adapt to different scenarios, which is time-consuming, labor-intensive, and lacks flexibility, making it difficult to meet the needs of efficient and accurate identification. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method, system, device, and medium for identifying electric vehicle charging loads to address the limitations of electric vehicle charging load identification: firstly, it relies on high-frequency data, resulting in high equipment costs and processing pressure; secondly, the model requires massive data annotation and repeated parameter tuning, which is time-consuming, labor-intensive, inflexible, and difficult to meet the requirements for efficient and accurate identification.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for an electric vehicle load identification system, comprising:

[0008] Construct a load identification model;

[0009] The load identification model is used to obtain power aggregated data, and the power aggregated data is subjected to multi-level filtering to obtain charging load characteristic components.

[0010] Based on the charging load characteristic components, a change point detection algorithm is used, combined with the charging characteristics of electric vehicles as feature constraints, to identify charging events and obtain the start and end times of the charging events.

[0011] The charging load feature component corresponding to the charging event is used as event data, and the pre-set electric vehicle charging power curve is used as a reference template. The event data is matched with the reference template using a sequence matching algorithm to obtain the optimal path. The load curve of a single charging event is reconstructed based on the optimal path.

[0012] Based on the start and end times of the charging events, all the reconstructed individual charging event load curves are combined according to the time axis to form the electric vehicle charging load curve.

[0013] As a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of constructing the load identification model includes:

[0014] Construct an initial recognition model and obtain recognition data output from the initial recognition model;

[0015] Select the mean absolute error, root mean square error, and coefficient of determination as preset evaluation indicators and thresholds;

[0016] Collect charging load data, calculate the evaluation index value of the charging load data and the identification data through evaluation index, and evaluate the recognition effect of the initial recognition model through the evaluation index value;

[0017] If the recognition effect does not reach the threshold, the initial recognition model is optimized and re-evaluated until the recognition effect reaches the threshold, and then a load recognition model is constructed.

[0018] The beneficial effects of this preferred technical solution are as follows: by constructing an initial identification model and using mean absolute error, root mean square error and coefficient of determination as evaluation indicators, the identification effect is verified by combining the collected charging load data. If the threshold is not reached, the initial identification model is optimized until the standard is met, and a load identification model is constructed to provide support for subsequent power lumped data filtering, charging event identification and reconstruction of load curves of individual charging events, thereby improving the overall accuracy and stability of electric vehicle load identification.

[0019] As a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of performing multi-level filtering on the power aggregated data to obtain charging load characteristic components includes:

[0020] A time-series decomposition algorithm is used to process the lumped power data and extract low-frequency trend components;

[0021] The low-frequency trend component is decomposed and noise information is filtered out using the discrete wavelet transform algorithm to obtain the charging load characteristic component.

[0022] The beneficial effects of this preferred technical solution are as follows: by extracting the low-frequency trend components in the lumped power data through the time-series decomposition algorithm, and then filtering out noise from the low-frequency trend components by combining the discrete wavelet transform algorithm, the features related to the electric vehicle charging load can be separated and retained, while irrelevant high-frequency noise and detail interference are filtered out. This provides a data basis for subsequent identification of charging events and change point detection based on the charging load feature components, and improves the accuracy and reliability of subsequent steps.

[0023] As a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of obtaining the start and end times of charging events using a change point detection algorithm includes:

[0024] The charging load characteristic components are arranged in chronological order to form a continuous power time series;

[0025] A change point detection algorithm is used to perform change point detection on the power time series. The change point detection algorithm calculates and identifies the time points in the power time series where the power value changes abruptly as candidate change points. The candidate change points include rising edge candidate change points that jump from the non-charging state to the charging state and falling edge candidate change points that fall back from the charging state to the non-charging state.

[0026] In the process of change point detection, feature constraints are introduced, and the detected candidate change points are filtered according to the feature constraints to obtain valid change points. The valid change points include rising edge valid change points and falling edge valid change points.

[0027] The start time of the charging event is determined by identifying the valid change point on the rising edge and the end time of the charging event by identifying the valid change point on the falling edge.

[0028] The beneficial effects of this preferred technical solution are as follows: the characteristic components of the charging load are formed into a power time series, and the time points of power change are identified by the change point detection algorithm as rising edge candidate change points and falling edge candidate change points. The effective change points are obtained by introducing feature constraints. The effective change points include rising edge effective change points and falling edge effective change points. This can distinguish the transition time between charging state and non-charging state, thereby determining the start and end time of the charging event. This provides a basis for the reconstruction of the load curve of a subsequent single charging event and the generation of the electric vehicle charging load curve, and improves the reliability of the entire load identification process.

[0029] As a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of reconstructing the load curve of a single charging event includes:

[0030] The charging load characteristic component corresponding to the charging event is used as event data, and the pre-set electric vehicle charging power curve is used as a reference template. Both the event data and the reference template are presented in time series form.

[0031] A sequence matching algorithm is used, with Euclidean distance as the basic distance. The basic distance between the event data and the corresponding data points of the reference template is calculated according to the algorithm, and a basic distance matrix is ​​constructed based on all the basic distances.

[0032] The sum of the base distance of the current data point and the minimum cumulative distance of the three adjacent data points is set as the cumulative rule. Based on the cumulative rule and the base distance matrix, a cumulative distance matrix is ​​constructed.

[0033] Set matrix constraints based on the cumulative distance matrix, and filter out the curved paths that meet the requirements in the cumulative distance matrix according to the matrix constraints;

[0034] The path with the smallest cumulative distance is selected as the optimal matching path based on the described curved path;

[0035] Based on the optimal matching path, the corresponding data points of the event data and the reference template are nonlinearly normalized and matched to obtain the power amplitude correlation relationship. Based on the power amplitude correlation relationship, the load curve of a single charging event is reconstructed.

[0036] The beneficial effects of this preferred technical solution are as follows: the optimal matching path is obtained through the sequence matching algorithm, which can match charging event data with the reference template; the power amplitude correlation is obtained by regularizing the matching based on the optimal matching path and reconstructing the load curve of a single charging event, which can correct data deviation, retain the real charging characteristics, provide basic data for subsequent combination to obtain the electric vehicle charging load curve, and improve accuracy.

[0037] In a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of obtaining the electric vehicle charging load curve includes:

[0038] Collect all the reconstructed individual charging event load curves, and retrieve the start and end times corresponding to each individual charging event;

[0039] Define a unified timeline and deploy each individual charging event load curve to the time interval covered by the start and end times of the charging event to which the individual charging event load curve belongs, thus completing the timeline deployment.

[0040] Set the load power value for the time intervals not covered on the time axis to zero, which is the zero power interval, and process it as a non-charging interval.

[0041] By deploying the time axis and processing the non-charging intervals, the load curves of all individual charging events and the zero-power intervals are integrated to obtain the electric vehicle charging load curve.

[0042] The beneficial effects of this preferred technical solution are as follows: by collecting the load curves of individual charging events and their corresponding start and end times, and by processing the corresponding deployment and non-charging intervals of the time axis, the charging characteristics of different charging events can be restored, forming a continuous and accurate electric vehicle charging load curve, presenting the overall charging load change pattern, providing intuitive and reliable data support for power grid monitoring and charging behavior management, and ensuring the integrity and practicality of the final load identification result.

[0043] As a preferred embodiment of the electric vehicle load identification system method of the present invention, the step of combining the charging characteristics of the electric vehicle as feature constraints includes:

[0044] The feature constraints include daily charging frequency constraints, power variation constraints, and charging duration constraints.

[0045] The daily charging frequency constraint is expressed as follows:

[0046]

[0047] The power variation constraint is:

[0048]

[0049]

[0050]

[0051] The charging duration constraint is:

[0052]

[0053] in, and Indicates the rising edge and falling edge times before and after the candidate change point; This represents the power difference before and after the candidate change point; Indicates the number of times.

[0054] The beneficial effects of this preferred technical solution are as follows: by using power change constraints, charging duration constraints, and daily charging frequency constraints as feature constraints of the change point detection algorithm, the charging behavior attributes can be defined, the differences between charging load and other load characteristics can be distinguished, misjudgment of non-charging loads can be avoided, and the accuracy of identifying the start and end times of charging events can be improved.

[0055] In a second aspect, the present invention provides an electric vehicle load identification system, comprising:

[0056] The data acquisition module is used to collect aggregated power data and actual charging load data;

[0057] The model building module is used to build a load identification model. It evaluates the identification effect through preset evaluation indicators and thresholds. If the threshold is not reached, it is optimized until the load identification model is built.

[0058] The data filtering module is used to perform multi-level filtering on the lumped power data to obtain the characteristic components of the charging load.

[0059] The event recognition module is used to identify the start and end times of charging events by combining a change point detection algorithm with feature constraints.

[0060] The curve reconstruction module is used to reconstruct the load curve of a single charging event by matching event data with a reference template using a sequence matching algorithm.

[0061] The integrated generation module is used to combine all individual charging event load curves along the time axis to obtain the electric vehicle charging load curve.

[0062] Thirdly, the present invention provides an electronic device, comprising:

[0063] Memory and processor;

[0064] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the electric vehicle load identification system method.

[0065] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the electric vehicle load identification system method.

[0066] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a load identification model, combining multi-level filtering to extract charging load feature components, introducing feature constraints and change point detection algorithms to identify charging events, and reconstructing the load curve of a single charging event using a sequence matching algorithm, the two methods are integrated to form an electric vehicle charging load curve. This solution effectively improves identification accuracy, reduces noise interference and event misjudgment, and provides reliable data support for power grid monitoring and charging management. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a schematic diagram of the overall process of an electric vehicle load identification system method according to an embodiment of the present invention. Detailed Implementation

[0069] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0070] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for identifying the load of an electric vehicle is provided, comprising the following steps S100~S500:

[0071] S100, Construct a load identification model.

[0072] S200. Obtain power aggregated data through the load identification model, and perform multi-level filtering on the power aggregated data to obtain charging load characteristic components.

[0073] S300. Based on the charging load characteristic components, a change point detection algorithm is used, combined with the charging characteristics of electric vehicles as characteristic constraints, to identify charging events and obtain the start and end times of the charging events.

[0074] S400. The charging load feature component corresponding to the charging event is used as event data, and the pre-set electric vehicle charging power curve is used as a reference template. The event data is matched with the reference template using a sequence matching algorithm to obtain the optimal path. The load curve of a single charging event is reconstructed based on the optimal path.

[0075] S500. Based on the start and end times of the charging events, combine all the reconstructed individual charging event load curves according to the time axis to form an electric vehicle charging load curve.

[0076] It should be noted that the current electric vehicle charging load is often mixed with other household electricity loads in the aggregated power data, making it difficult to separate them directly. Moreover, the charging load is easily affected by users' charging habits and grid noise interference, resulting in great difficulty in feature extraction. Traditional identification methods mostly rely on single filtering or matching algorithms, which are prone to problems such as misjudgment of charging events and deviations in load curve reconstruction. They cannot accurately output independent charging load information for electric vehicles, thereby affecting the grid's monitoring, scheduling, and resource allocation efficiency for charging loads.

[0077] Therefore, to address the aforementioned problems of difficulty in separating aggregated power data, low accuracy in extracting charging features, and inaccurate load curve reconstruction, the following steps (S100-S500) are employed: first, a reliable load identification model is constructed as a foundation; then, noise information is filtered out from the aggregated power data through multi-level filtering to obtain the charging load feature components; next, charging events are identified by combining charging feature constraints and change point detection to obtain start and end times; subsequently, the load curve of a single charging event is reconstructed through a sequence matching algorithm; finally, the data are integrated to form the electric vehicle charging load curve, providing data support and technical basis for power grid monitoring of charging load, optimization of charging scheduling strategies, and ensuring the stable operation of the distribution network.

[0078] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, an electric vehicle load identification system method is provided.

[0079] In this embodiment of the application, taking a typical residential community in a third-tier city (including 4 decentralized charging piles and 1 centralized charging shed) as the application scenario, the specific implementation of the steps for constructing the load identification model for A1~A4 in S100 is as follows:

[0080] A1. Construct an initial recognition model and obtain recognition data output by the initial recognition model.

[0081] Specifically, the initial identification model uses 2,160 mixed load data points collected by community smart meters at 30-minute intervals over 45 days as input for data preprocessing.

[0082] The system presets typical characteristics of electric vehicle charging loads, such as power of 1.5-15kW, single charging duration of 0.5-12h, and power fluctuation of ≤8% / 30min. It then matches the pre-processed mixed load data point by point in time and outputs 45 days of identification data, such as identifying power of 3.8kW at 18:00 and identifying power of 5.6kW at 18:30.

[0083] A2. Select the mean absolute error, root mean square error, coefficient of determination, and preset evaluation indicators and thresholds.

[0084] Specifically, mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are selected as evaluation indicators, with preset thresholds: MAE≤0.7kW, RMSE≤1.0kW, and R²≥0.93.

[0085] A3. Collect charging load data, calculate the evaluation index value of the charging load data and the identification data through the evaluation index, and evaluate the recognition effect of the initial recognition model through the evaluation index value.

[0086] Specifically, through the metering modules of four distributed charging piles and one centralized charging shed in the community, charging load data is collected at 30-minute intervals over a period of 45 days. The actual power at each time point is recorded, such as 3.0kW at 6 PM and 5.2kW at 6:30 PM. Evaluation index values ​​are then calculated based on the identified data.

[0087]

[0088]

[0089]

[0090] in, This is the actual power. To identify power, N is the number of samples. This represents the average actual power.

[0091] The calculated values ​​are: MAE = 0.95kW, RMSE = 1.3kW, and R² = 0.89. Because the electric water heater was not distinguished from short-term high-power power interference, the initial identification model did not meet the standard.

[0092] A4. If the recognition effect does not reach the threshold, optimize the initial recognition model and re-evaluate until the recognition effect reaches the threshold, and then construct the load recognition model.

[0093] Specifically, the optimization scheme involves supplementing weekday and weekend scenario-based feature weights. After optimization, the MAE=0.62kW, RMSE=0.85kW, and R²=0.94 are obtained. The initial identification model meets the standards, completing the construction of the load identification model and providing a reliable foundation for subsequent multi-level filtering.

[0094] In an optional implementation, step S100 may further include sample expansion and cross-validation. The steps are as follows: In step A3, in addition to the charging pile metering data, charging load data from commercial complexes (including two fast-charging stations) in the same area are added as extended samples. A 5-fold cross-validation method is used to evaluate the initial identification model: the charging load data from the community and commercial complex for a total of 75 days are randomly divided into 5 groups, with 4 groups used as the training set to optimize parameters and 1 group used as the test set to verify accuracy. After cross-validation, =0.68kW =0.93kW =0.935, where The average value of the mean absolute error The average value of the root mean square error, The average value of the determination coefficients is used to avoid overfitting of the constructed load identification model, while also enabling the load identification model to adapt to scenarios with a higher proportion of fast charging, thus expanding its applicability.

[0095] In another optional implementation, the threshold mechanism can be adjusted in step S100. The steps are as follows: In step A4, the seasonal electricity consumption characteristics of the community can be taken into account. For example, the heating load of residents in winter is 2-3kW and lasts for 4-6 hours. It is easy to be confused with the slow charging load. The threshold can be adjusted: the winter MAE threshold is relaxed to 0.8kW and the RMSE threshold is relaxed to 1.1kW. The charging load data of 30 days in winter is collected for verification. After optimization, MAE=0.75kW, RMSE=1.02kW and R²=0.92 are obtained, which meet the threshold requirements and improve the adaptability to seasonal electricity consumption fluctuations.

[0096] In this embodiment of the application, step S200, which involves performing multi-level filtering on the power aggregated data to obtain the charging load characteristic components, includes steps B1 to B2:

[0097] B1. The time-series decomposition algorithm is used to process the lumped power data and extract the low-frequency trend component.

[0098] Specifically, the power aggregated data initially screened by the load identification model is used as input, and the STL algorithm is used for processing according to the formula. ;in The power lumped data at time t, As a low-frequency trend component, This represents the daily electricity consumption cycle component. This represents high-frequency noise components.

[0099] Based on the community's electricity consumption patterns, the cycle The number of cycles is set to 24, with 15 cycles for the inner cycle and 5 cycles for the outer cycle. In the outer cycle, robustness weights are calculated using a formula to exclude those caused by sudden high-power consumption by residents. Outliers are ultimately extracted, and low-frequency trend components are identified. The formula is as follows:

[0100]

[0101]

[0102]

[0103] in denoted by t, representing the robust weights calculated at time t; B represents the defined quadratic weight function; h represents the defined median function. express The smoothed remainder; 6 represents the coefficient related to the robustness weights; This represents the median function.

[0104] B2. The low-frequency trend component is decomposed using the discrete wavelet transform algorithm, and noise information is filtered out to obtain the charging load characteristic component.

[0105] Specifically, through the extraction of low-frequency trend components The input is the DWT algorithm for secondary filtering.

[0106] Select the db4 wavelet basis function, set the decomposition level to 3, and follow the formula. and formula ,Will It is decomposed into low-frequency approximate components and high-frequency detail components; among which... The low-frequency component signal represents the overall trend; This represents the high-frequency component detail signal, indicating a disturbance; It is a low-pass filter; It is a high-pass filter; Given a set of integers, the high-frequency components are ultimately obtained. Approximate components of the fundamental low frequency .

[0107] High frequency components After setting it to zero, substitute the fundamental low-frequency approximation component. ,in accordance with The reconstructed signal yields a low-frequency approximate component. This is the characteristic component of the charging load.

[0108] In an optional implementation, step S200 may further include a filter accuracy feedback adjustment step, which involves: retrieving the collected charging load data and calculating the average absolute error between the collected data and the charging load characteristic components output by B2. , This is the actual power. Let N be the power characteristic component of the charging load, and N be the number of samples; if Then adjust the outer loop count of the STL algorithm to 8 times or the decomposition level of the DWT algorithm to 4 levels, and re-execute B1-B2 to calculate the result. No adjustments are needed, ensuring consistency between the charging load characteristic components and the charging load data.

[0109] In another optional implementation, step S200 can also include setting time-period differences. The steps are as follows: to address the differences in charging time periods between 6 PM to 11 PM on weekdays and between 7 PM to 10 PM on weekends, the STL algorithm is set to loop 15 times and the DWT algorithm is set to decompose 3 layers on weekdays to enhance the extraction of nighttime charging load feature components; the STL algorithm is set to loop 12 times and the DWT algorithm is set to decompose 2 layers on weekends to reduce over-smoothing of the dispersed charging load, thereby improving the accuracy of identifying charging load feature components and adapting to the distribution characteristics of charging load at different times.

[0110] In this embodiment of the application, step S300, which involves using a change-point detection algorithm to obtain the start and end times of the charging event, includes steps C1 to C4:

[0111] C1. Arrange the characteristic components of the charging load in chronological order to form a continuous power time series.

[0112] Specifically, the charging load characteristic components obtained from S200 are arranged in chronological order to form a continuous power time series. t represents a time point, corresponding to a 30-minute interval, with a total of 2160 data points. This power time series reflects the changes in charging power at various times within the community, such as 6 PM on a weekday. 18:30 When not charging This provides a well-organized data foundation for subsequent change point detection.

[0113] C2. A change point detection algorithm is used to perform change point detection on the power time series. The change point detection algorithm is used to calculate and identify the time points in the power time series where the power value changes abruptly as candidate change points. The candidate change points include rising edge candidate change points that jump from the non-charging state to the charging state and falling edge candidate change points that fall back from the charging state to the non-charging state.

[0114] Specifically, the PELT algorithm is used to calculate the time point at which the power value in the sequence undergoes a sudden change—when From non-charging state ( ) Jumps to charging state ( When this occurs, it is marked as a candidate change point for the rising edge, such as 6 PM on a weekday. The power output suddenly increased from 0.3kW to 3.2kW; when When the state transitions from charging to non-charging, it is marked as a candidate turning point of the falling edge, such as 22:30 on a weekday evening. The voltage abruptly changed from 4.8kW to 0.2kW. A total of 420 candidate change points were detected, including 212 rising edge candidate change points and 208 falling edge candidate change points.

[0115] C3. In the process of change point detection, feature constraints are introduced, and the detected candidate change points are screened according to the feature constraints to obtain valid change points. The valid change points include rising edge valid change points and falling edge valid change points.

[0116] Specifically, the feature constraints include daily charging frequency constraints, power variation constraints, and charging duration constraints; the candidate change point screening for C2 detection includes steps C3.1 to C3.3:

[0117] C3.1 The daily charging frequency constraint is expressed as follows:

[0118]

[0119] Specifically, The maximum threshold for the number of charges per day; based on the statistical settings of S100, weekdays... ,weekend Excluding single-day excess Invalid candidate changes were identified. For example, 35 candidate changes were falsely detected on a weekend. Five isolated candidate changes that exceeded 30 were removed. Most of the excess candidate changes were false changes caused by minor fluctuations in the electricity meter.

[0120] C3.2, The power variation constraint is:

[0121]

[0122]

[0123]

[0124] Specifically, Rising edge candidate variable point The power difference before and after, ; Falling edge candidate variable point The power difference before and after, ; To avoid excessively large power fluctuations during a single charge and to eliminate non-charging interference.

[0125] Candidate change points whose power changes do not meet the threshold are eliminated, for example, a certain candidate change point. They were removed.

[0126] C3.3, The charging duration constraint is as follows:

[0127]

[0128] in, and Indicates the rising edge and falling edge times before and after the candidate change point; This represents the power difference before and after the candidate change point; Indicates the number of times.

[0129] Specifically, This corresponds to 0.5h, excluding short-term false starts. This corresponds to 6 hours, which aligns with the community's longest slow charging duration. Candidate variables with time intervals exceeding this range were removed, such as a certain group of candidate variables. If the time interval is 0.25h < 0.5h, it is considered a false trigger and is removed.

[0130] After screening based on the above three feature constraints, 386 valid change points were retained out of 420 candidate change points, including 193 valid change points on the rising edge and 193 valid change points on the falling edge.

[0131] C4. Determine the start time of the charging event by identifying the valid change point of the rising edge and the end time of the charging event by identifying the valid change point of the falling edge, thereby obtaining the start and end times of the charging event.

[0132] Specifically, the time corresponding to the valid rising edge change point selected by C3 is determined as the start time of the charging event. For example, 18:00 on a weekday corresponds to the start of one charging event. The time corresponding to the valid falling edge change point is determined as the end time of the charging event. For example, 22:30 on a weekday corresponds to the end of one charging event. A total of 193 charging events were identified. For example: Weekday charging event 1: Start time 18:00, End time 22:30, lasts 4.5 hours, power 2.8-5.2kW (slow charging); Weekend charging event 2: Start time 12:30, End time 13:30, lasts 1 hour, power 8.5-11.8kW (fast charging).

[0133] In an optional implementation, step S300 may also include a candidate variable point clustering deduplication step, which is as follows: the candidate variable points detected by C2 are clustered by time, and the clustering radius is set to one time interval, i.e., 30 minutes. If there are two or more candidate variable points in the same cluster, the power change amplitude of each point in the cluster is calculated, and the one with the largest change amplitude is retained as the final candidate variable point, and the rest are removed, so as to improve the screening efficiency of C3 and reduce the misjudgment of charging events caused by duplicate variable points.

[0134] In another optional implementation, step S300 may further include time-based differential constraint adjustment, which involves adjusting the constraint threshold for community charging time-based differences, such as adjusting the power variation constraint for weekday evenings from 6 PM to 11 PM. Reduced to 0.8kW, charging duration constraint Increased to 8 hours (to accommodate some residents' overnight slow charging); weekend charging hours from 12:00 to 16:00 will be... Increased to 1.2kW Increasing the interval to 1.5 (i.e. 0.75h, excluding short-term charging interference) can improve the accuracy of identifying charging events and adapt to the charging behavior characteristics at different times.

[0135] In this embodiment of the application, step S400, the step of reconstructing the load curve of a single charging event, includes D1 to D6:

[0136] D1. The charging load characteristic component corresponding to the charging event is taken as event data, and the pre-set electric vehicle charging power curve is taken as a reference template. Both the event data and the reference template are presented in time series form.

[0137] Specifically, based on the start and end times of the charging events acquired by S300, corresponding time segments are extracted from the charging load characteristic components of S200 as event data. Taking a slow charging event from 6:00 PM to 10:30 PM on a weekday as an example, power data at 30-minute intervals during this period are extracted to form an event data sequence. A total of 10 data points, corresponding to a 5-hour charging time;

[0138] Based on the community's charging load type (80% slow charging and 20% fast charging), two reference templates are preset: Slow charging reference template. Seven data points, corresponding to a standard 3.5-hour slow charging power curve; fast charging reference template. Five data points correspond to a standard fast charging power curve over 2.5 hours. Event data and reference templates are presented in time series format, ensuring that the time intervals are all 30 minutes.

[0139] D2. Using a sequence matching algorithm, Euclidean distance is used as the basic distance. The basic distance between the event data and the corresponding data points of the reference template is calculated according to the algorithm, and a basic distance matrix is ​​constructed based on all the basic distances.

[0140] Specifically, regarding slow charging event data S and the slow charging reference template According to the Euclidean distance formula , Let S be the i-th data point. for For the j-th data point, calculate the distance to the corresponding point. For example... , ;

[0141] Arrange all basic distances according to the dimension of "event data row × reference template column" to form The fundamental distance matrix D, matrix elements .

[0142] D3. Set the sum of the basic distance of the current data point and the minimum cumulative distance of the three adjacent data points as the cumulative rule, and construct the cumulative distance matrix according to the cumulative rule and the basic distance matrix.

[0143] Specifically, for the cumulative distance matrix element in row i and column j Its calculation logic is as follows: The path is restricted to the three adjacent points in the "top left, top, left" directions to avoid skipping or jumping.

[0144] Set boundary conditions , Ensure the path starts from the matrix starting point (1,1), and then calculate row by row and column by column according to the cumulative rule, for example... Ultimately formed Cumulative distance matrix .

[0145] D4. Set matrix constraints based on the cumulative distance matrix, and filter out the curved paths that meet the requirements in the cumulative distance matrix according to the matrix constraints.

[0146] Specifically, the matrix constraints are defined as follows: Boundary condition: the path must start from the matrix starting point (1,1) and end point (10,7) to ensure a complete sequence match; Continuity: coordinates of adjacent path points to avoid cross-point matching, for example, from... Only redirect to , or Monotonicity: Path coordinates increase as the matching process progresses, avoiding time backtracking, such as from... Jump back .

[0147] In the cumulative distance matrix The algorithm iterates through all paths that satisfy the constraints and selects 12 valid curved paths, such as path 1: (1,1)→(2,2)→(3,3)→…→(10,7) and path 2: (1,1)→(2,1)→(3,2)→…→(10,7), which meet the matrix constraint requirements.

[0148] D5. Select the path with the smallest cumulative distance as the optimal matching path based on the curved path.

[0149] Specifically, calculate the cumulative distance of the 12 curved paths selected by D4, that is, the total distance along the path. The sum of the values ​​is used to select the path with the smallest cumulative distance as the optimal matching path.

[0150] Total distance of path 1: ;

[0151] Total distance along path 2: ;...

[0152] Total distance along path 7: (Minimum cumulative distance is the optimal matching path) can achieve the minimum difference matching between event data and slow charging reference template. For example, the 4th point of the event data, 4.9kW, corresponds to the 4th point of the template, 5.2kW, and the 8th point, 4.7kW, corresponds to the 6th point of the template, 3.0kW, which is in line with the slow charging power change pattern.

[0153] D6. Based on the optimal matching path, perform nonlinear regularization matching on the corresponding data points of the event data and the reference template to obtain the power amplitude correlation relationship, and reconstruct the load curve of a single charging event based on the power amplitude correlation relationship.

[0154] Specifically, the one-to-one correspondence between event data and templates is determined through the optimal path. For example, the 10 points of event data S and the template... The seven points are associated in the order of “1→1, 2→2, 3→3, 4→4, 5→4, 6→5, 7→5, 8→6, 9→7, 10→7” to correct the power deviation caused by noise in the event data and obtain the power amplitude correlation relationship.

[0155] Based on the correlation of power amplitude, a continuous curve is plotted along the time axis to form the load curve of a single charging event for this slow charging event.

[0156] In an optional implementation, step S400 may also include a template dynamic update step, which involves: after every 10 sets of curve reconstructions for the same type of charging events, calculating the average deviation between the reconstructed curves and the initial reference template. , where n is the number of event data points; if The average of the 10 reconstructed curves will then be used as a new reference template, such as the new slow-charging template. Replace the initial template to improve template compatibility.

[0157] In another optional implementation, step S400 can also include abnormal event data correction. The steps are as follows: after extracting event data in D1, first detect abnormal data values, replace them with the average of the preceding and following data, and then execute D2-D6 to avoid matching deviations caused by abnormal data.

[0158] In this embodiment of the application, step S500, the step of obtaining the electric vehicle charging load curve, includes E1 to E4:

[0159] E1. Collect all reconstructed individual charging event load curves and retrieve the start and end times corresponding to each individual charging event.

[0160] Specifically, collect the load curves of all individual charging events reconstructed by S400, and synchronously retrieve the corresponding start and end times determined by S300:

[0161] Examples of slow charging events: Curve 1, power 2.8-5.2kW, start and end time is weekdays 18:00-22:30; Curve 2, power 3.0-5.0kW, start and end time is weekends 10:00-15:30;

[0162] Examples of fast charging events: Curve 3, power 8.5-11.8kW, start and end time is weekend 12:30-13:30; Curve 4, power 9.0-11.5kW, start and end time is weekday 19:00-20:00.

[0163] E2. Determine a unified time axis and deploy each individual charging event load curve to the time interval covered by the start and end times of the charging event to which the individual charging event load curve belongs, thus completing the time axis deployment.

[0164] Specifically, a time axis T is established with a 45-day monitoring cycle in the community, with a time interval of 30 minutes. The total length is 45 days × 48 intervals = 2160 time points, consistent with S200, covering the entire period from 0:00 on day 1 to 23:30 on day 45.

[0165] Map each individual charging event load curve to the time axis T according to its start and end times, for example, the power sequence of slow charging curve 1. This corresponds to deployment to T from "18:00" (point i) to "22:30" (point i) on weekdays. (Point) range; power sequence of fast charging curve 3 Deployed to T from "12:30 on weekend" (point j) to "13:30" (point j). (Point) interval. Each charging period on the timeline is filled with the corresponding power value, completing the deployment corresponding to the timeline.

[0166] E3. Set the load power value for the time intervals not covered on the time axis to zero, i.e., the zero power interval, and perform non-charging interval processing.

[0167] Specifically, zero-power processing is performed on the intervals on the time axis T that are not covered by any charging events (i.e., non-charging periods). For example, during the intervals of 0:00-6:00 on weekdays and 8:00-10:00 on weekends, there are no charging events, so the power value at the corresponding time point is uniformly set to 0kW.

[0168] E4. By deploying the time axis and processing the non-charging interval, the load curves of all individual charging events and the zero-power interval are integrated to obtain the electric vehicle charging load curve.

[0169] Specifically, the load curves of all individual charging events deployed by E2 are integrated with the zero-power intervals processed by E3 to form an electric vehicle charging load curve covering the entire 45-day period.

[0170] In an optional implementation, step S500 may further include curve smoothing and noise reduction processing, the steps of which are: applying a moving average filter to the electric vehicle charging load curve, with the window size set to 3 time intervals, i.e., 1.5 hours, and the calculation formula is as follows: , The original curve represents the power at time t; for example, power fluctuations over a certain period. After smoothing, it becomes By eliminating short-term power fluctuations, the smoothness of the curve is improved after processing, which is beneficial for subsequent load trend analysis.

[0171] In another optional implementation, step S500 can also output load curves by type. The steps are as follows: based on the electric vehicle charging load curve, sub-curves are constructed according to charging type (slow charging / fast charging). The slow charging load curve contains only 155 sets of slow charging event data, and the fast charging load curve contains only 38 sets of fast charging event data. At the same time, the proportion and time period distribution of the two types of curves are calculated. These sub-curves can provide data support for the expansion of community slow charging piles and the setting of time-of-use electricity prices for fast charging piles, thereby enhancing the application value of the curves.

[0172] In summary, this invention first constructs a load identification model to extract charging load feature components from power aggregated data; then, by combining charging feature constraints and change point detection, it identifies charging events and obtains their start and end times; it reconstructs the load curve of a single charging event through a sequence matching algorithm; and finally, it forms an electric vehicle charging load curve, providing data and technical support for power grid monitoring and scheduling and optimized configuration of charging piles, thereby improving the accuracy of charging load identification and the efficiency of power grid operation.

[0173] Example 3 illustrates an electric vehicle load identification system method. It should be noted that the technical solution of this electric vehicle load identification system belongs to the same concept as the technical solution of the electric vehicle load identification system method described above. Details not described in detail in this example can be found in the description of the technical solution of the electric vehicle load identification system method described above.

[0174] This embodiment also provides an electric vehicle load identification system, including:

[0175] The data acquisition module is used to collect aggregated power data and actual charging load data;

[0176] The model building module is used to build a load identification model. It evaluates the identification effect through preset evaluation indicators and thresholds. If the threshold is not reached, it is optimized until the load identification model is built.

[0177] The data filtering module is used to perform multi-level filtering on the lumped power data to obtain the characteristic components of the charging load.

[0178] The event recognition module is used to identify the start and end times of charging events by combining a change point detection algorithm with feature constraints.

[0179] The curve reconstruction module is used to reconstruct the load curve of a single charging event by matching event data with a reference template using a sequence matching algorithm.

[0180] The integrated generation module is used to combine all individual charging event load curves along the time axis to obtain the electric vehicle charging load curve.

[0181] This embodiment also provides an electronic device suitable for an electric vehicle load identification system, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for implementing an electric vehicle load identification system as proposed in the above embodiment.

[0182] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for implementing an electric vehicle load identification system as proposed in the above embodiments.

[0183] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for implementing an electric vehicle load identification system proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0184] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0185] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying the load of an electric vehicle, characterized in that, include: Construct a load identification model; The load identification model is used to obtain power aggregated data, and the power aggregated data is filtered in multiple stages to obtain charging load characteristic components. Based on the charging load characteristic components, a change point detection algorithm is used, combined with the charging characteristics of electric vehicles as characteristic constraints, to identify charging events and obtain the start and end times of the charging events. The charging load feature component corresponding to the charging event is used as event data, and the pre-set electric vehicle charging power curve is used as a reference template. The event data is matched with the reference template using a sequence matching algorithm to obtain the optimal path. The load curve of a single charging event is reconstructed based on the optimal path. Based on the start and end times of the charging events, all the reconstructed individual charging event load curves are combined according to the time axis to form the electric vehicle charging load curve.

2. The electric vehicle load identification system method as described in claim 1, characterized in that, The steps to build a load identification model include: Construct an initial recognition model and obtain recognition data output from the initial recognition model; Select the mean absolute error, root mean square error, and coefficient of determination as preset evaluation indicators and thresholds; Collect charging load data, calculate the evaluation index value of the charging load data and the identification data through evaluation index, and evaluate the recognition effect of the initial recognition model through the evaluation index value; If the recognition effect does not reach the threshold, the initial recognition model is optimized and re-evaluated until the recognition effect reaches the threshold, and then a load recognition model is constructed.

3. The electric vehicle load identification system method as described in claim 2, characterized in that, The steps of performing multi-level filtering on the lumped power data to obtain the charging load characteristic components include: A time-series decomposition algorithm is used to process the lumped power data and extract low-frequency trend components; The low-frequency trend component is decomposed using a discrete wavelet transform algorithm, and noise information is filtered out to obtain the charging load characteristic component.

4. The electric vehicle load identification system method as described in claim 3, characterized in that, The steps for obtaining the start and end times of a charging event using a change-point detection algorithm include: The charging load characteristic components are arranged in chronological order to form a continuous power time series; A change point detection algorithm is used to perform change point detection on the power time series. The change point detection algorithm calculates and identifies the time points in the power time series where the power value changes abruptly as candidate change points. The candidate change points include rising edge candidate change points that jump from the non-charging state to the charging state and falling edge candidate change points that fall back from the charging state to the non-charging state. In the process of change point detection, feature constraints are introduced, and the detected candidate change points are filtered according to the feature constraints to obtain valid change points. The valid change points include rising edge valid change points and falling edge valid change points. The start time of the charging event is determined by identifying the valid change point on the rising edge and the end time of the charging event by identifying the valid change point on the falling edge.

5. The electric vehicle load identification system method as described in claim 4, characterized in that, The steps to reconstruct the load curve for a single charging event include: The charging load characteristic component corresponding to the charging event is used as event data, and the pre-set electric vehicle charging power curve is used as a reference template. Both the event data and the reference template are presented in time series form. A sequence matching algorithm is used, with Euclidean distance as the basic distance. The basic distance between the event data and the corresponding data points of the reference template is calculated according to the algorithm, and a basic distance matrix is ​​constructed based on all the basic distances. The sum of the base distance of the current data point and the minimum cumulative distance of the three adjacent data points is set as the cumulative rule. Based on the cumulative rule and the base distance matrix, a cumulative distance matrix is ​​constructed. Set matrix constraints based on the cumulative distance matrix, and filter out the curved paths that meet the requirements in the cumulative distance matrix according to the matrix constraints; The path with the smallest cumulative distance is selected as the optimal matching path based on the described curved path; Based on the optimal matching path, the corresponding data points of the event data and the reference template are nonlinearly normalized and matched to obtain the power amplitude correlation relationship. Based on the power amplitude correlation relationship, the load curve of a single charging event is reconstructed.

6. The electric vehicle load identification system method as described in claim 5, characterized in that, The steps for obtaining the charging load curve of an electric vehicle include: Collect all the reconstructed individual charging event load curves, and retrieve the start and end times corresponding to each individual charging event; Define a unified timeline and deploy each individual charging event load curve to the time interval covered by the start and end times of the charging event to which the individual charging event load curve belongs, thus completing the timeline deployment. Set the load power value for the time intervals not covered on the time axis to zero, which is the zero power interval, and process it as a non-charging interval. By deploying the time axis and processing the non-charging intervals, the load curves of all individual charging events and the zero-power intervals are integrated to obtain the electric vehicle charging load curve.

7. The electric vehicle load identification system method as described in claim 6, characterized in that, The steps of incorporating the charging characteristics of electric vehicles as feature constraints include: The feature constraints include daily charging frequency constraints, power variation constraints, and charging duration constraints. The daily charging frequency constraint is expressed as follows: The power variation constraint is: The charging duration constraint is: in, and Indicates the rising edge and falling edge times before and after the candidate change point; This represents the power difference before and after the candidate change point; Indicates the number of times.

8. An electric vehicle load identification system, using the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect aggregated power data and actual charging load data; The model building module is used to build a load identification model. It evaluates the identification effect through preset evaluation indicators and thresholds. If the threshold is not reached, it is optimized until the load identification model is built. The data filtering module is used to perform multi-level filtering on the lumped power data to obtain the characteristic components of the charging load. The event recognition module is used to identify the start and end times of charging events by combining a change point detection algorithm with feature constraints. The curve reconstruction module is used to reconstruct the load curve of a single charging event by matching event data with a reference template using a sequence matching algorithm. The synthesis generation module is used to combine all individual charging event load curves according to the time axis to obtain the electric vehicle charging load curve.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the electric vehicle load identification system method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the electric vehicle load identification system method according to any one of claims 1 to 7.

Images