A Non-Intrusive Load Monitoring-Based Management Method for Electric Vehicle In-Home Charging

By using non-intrusive load monitoring technology and utilizing load switching events and steady-state power-harmonic characteristics, models and templates are constructed to achieve accurate identification and hierarchical management of electric vehicle charging at home. This solves the problem of misidentification in mixed environments with multiple devices and improves the accuracy and timeliness of electric vehicle charging monitoring.

CN123121998APending Publication Date: 2026-11-17DONGYING POWER SUPPLY COMPANY STATE GRID SHANDONG ELECTRIC POWER
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611334765.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In residential environments where multiple electrical devices operate together, existing technologies struggle to accurately identify electric vehicle charging behavior, which can easily lead to misidentification and mismanagement, especially when the load characteristics during the startup phase are similar to those of electric vehicle charging.

Method used

Based on non-intrusive load monitoring, starting with load switching events, and combining startup phase identification with steady-state power-harmonic verification, a load classification model and feature template are constructed. The startup phase feature vector and steady-state power-harmonic joint feature vector are used for progressive identification and hierarchical control.

Benefits of technology

It improves the accuracy and timeliness of monitoring electric vehicle charging at home, reduces the probability of false identification and accidental power outages, and realizes automated, continuous monitoring and control of electric vehicle charging behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121998A_ABST
    Figure CN123121998A_ABST
Patent Text Reader

Abstract

This invention relates to the field of non-intrusive load monitoring technology and discloses a method for managing and controlling electric vehicle (EV) in-home charging based on non-intrusive load monitoring. The method includes: constructing a load classification model for the startup phase, an EV startup phase feature template, and an EV stable charging feature template; collecting electrical data from the user's power supply line and establishing a mapping relationship between the branch to be managed and the controllable circuit breaker; detecting load switching events and constructing a startup phase event feature vector; determining candidate EV charging events based on classification results and template matching results; tracking candidate events to a stable operating state; constructing a steady-state power-harmonic joint feature vector using a continuous and effective stable operating window; confirming or rejecting candidate events; and determining the final charging identification confidence level; implementing hierarchical management and verifying the management results based on the confidence level. This invention can reduce false identification under mixed load conditions and improve the accuracy and timeliness of EV in-home charging monitoring and management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of non-intrusive load monitoring technology, and more specifically, to a method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring. Background Technology

[0002] With the increasing popularity of electric vehicles, some users are bringing their electric vehicles or removable batteries into indoor areas such as residences and hallways to charge them. During charging, safety risks may arise due to circuit overload, battery malfunction, or abnormal charging equipment. Therefore, there is a real need to promptly detect and manage the practice of charging electric vehicles indoors.

[0003] Current regulatory methods for electric vehicle charging at homes mainly include elevator image recognition, weight detection, and manual inspection. Furthermore, with the development of electricity monitoring technology, it is possible to analyze load changes by collecting electrical parameters such as current and power of user power lines to determine if specific electrical equipment is malfunctioning.

[0004] However, in residential electrical environments, refrigerators, air conditioners, electromagnetic devices, and other household appliances often operate simultaneously. The starting, stopping, and changes in the operating status of these different devices can all cause changes in line power, current, and harmonic characteristics. Some household appliances with starting impact or nonlinear electrical characteristics may even exhibit electrical characteristics similar to those of an electric vehicle charger during the startup phase.

[0005] When multiple electrical devices operate simultaneously, the characteristics of newly added loads are easily influenced by existing loads, and the accuracy of identifying electric vehicle charging behavior needs to be improved. On the other hand, some non-electric vehicle loads may have similar characteristics to electric vehicle charging loads during the startup phase. Relying solely on the identification results during the startup phase to implement alarms or power outages can easily lead to misidentification and mismanagement. Therefore, there is an urgent need to provide a non-intrusive load monitoring-based method for managing electric vehicle charging at home to solve the above problems. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a method for managing electric vehicle in-home charging based on non-intrusive load monitoring. This invention takes load switching events as the starting point, combines startup phase identification with power-harmonic verification during the stable phase, and reduces false identification under mixed loads. Through stable window reuse, longest tracking time, and confidence level control, it improves the accuracy and timeliness of electric vehicle in-home charging monitoring and power outage control.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A method for managing electric vehicle charging at home based on non-intrusive load monitoring includes:

[0009] S1. Obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and construct a load classification model for the startup phase, a feature template for the electric vehicle startup phase, and a feature template for the electric vehicle stable charging.

[0010] S2. Collect electrical data of user power supply lines and establish a mapping relationship between the branches to be managed and the corresponding controllable circuit breakers;

[0011] S3. Perform preprocessing and load switching event detection on the electrical data, determine the event occurrence time, the branch where the event is located, and the load switching direction of the load switching event, take the load switching event as the load switching event to be identified, and obtain the pre-event stability window, the start transient window, and the post-event initial operation window corresponding to the load switching event to be identified.

[0012] S4. Based on the pre-event stable window, the start-up transient window, and the post-event initial operation window, construct the start-up phase event feature vector of the current load switching event;

[0013] S5. Based on the identification results of the startup phase event feature vector by the startup phase load classification model, and the matching results of the startup phase event feature vector and the electric vehicle startup phase feature template, determine the initial confidence level of the startup phase, and identify the load switching events that meet the candidate identification conditions as candidate electric vehicle charging events.

[0014] S6. Continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous effective stable operation windows;

[0015] S7. Construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous effective stable operation windows, and determine the stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, as well as the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event; when confirming the candidate electric vehicle charging event, determine the final charging identification confidence level based on the stable phase evidence and the initial confidence level of the startup phase;

[0016] S8. Based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located, implement hierarchical control for the corresponding branch;

[0017] S9. Verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.

[0018] As a preferred embodiment of the present invention, S1 specifically includes:

[0019] Historical voltage and current data of electric vehicle charging load and non-electric vehicle load under different operating conditions are obtained, and the historical voltage and current data are labeled with equipment category and operating status to form a labeled load sample set;

[0020] Based on the load input time of each labeled load sample in the labeled load sample set, a corresponding startup phase training feature vector is formed. The startup phase training feature vector is standardized, and a binary classification load recognition model is trained using the equipment category as a supervision label to obtain the startup phase load classification model.

[0021] Electric vehicle charging load samples are selected from the labeled load sample set. The electric vehicle start-up phase feature template is formed based on the standardized start-up phase training feature vector corresponding to the electric vehicle charging load sample. The electric vehicle stable charging feature template is formed based on the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of the electric vehicle charging load sample in the stable charging phase relative to the pre-event steady state.

[0022] As a preferred embodiment of the present invention, in S2, a non-intrusive load monitoring terminal is installed on the side of the user's inlet distribution box. The non-intrusive load monitoring terminal includes a voltage acquisition channel and multiple current acquisition channels. The voltage acquisition channel is used to acquire the voltage signal on the inlet side, and each of the current acquisition channels is used to acquire the current signal of the corresponding branch to be managed. A mapping relationship is established between each current acquisition channel and the corresponding controllable circuit breaker, and the voltage signal on the inlet side and the current signal of each branch to be managed are acquired synchronously based on a unified sampling clock.

[0023] As a preferred embodiment of the present invention, S3 specifically includes:

[0024] The electrical data is preprocessed and divided into continuous analysis windows. The event change amount is determined based on the changes in active power and the effective value of current between adjacent analysis windows.

[0025] When the change in the event reaches the event detection threshold, it is determined that a load switching event has occurred on the corresponding branch to be managed, and the time of occurrence of the load switching event and the branch where the event is located are determined.

[0026] Using the time of the event as the time reference, acquire stable data before the event, transient data at startup, and initial running data after the event.

[0027] As a preferred embodiment of the present invention, the event feature vector of the startup phase in S4 is composed of incremental active power, incremental reactive power, incremental current RMS value, power factor change, incremental harmonic characteristics and startup transient characteristics.

[0028] The startup transient characteristics include the peak value of the instantaneous startup current, the peak factor, the startup pulse duration, the maximum power change rate, and the startup delay time.

[0029] As a preferred embodiment of the present invention, S5 specifically includes:

[0030] Input the feature vector of the startup phase event into the startup phase load classification model to obtain the model identification probability that the current load switching event belongs to the electric vehicle charging load;

[0031] The matching degree of the start-up template is determined based on the feature distance between the feature vector of the start-up phase event and the feature template of the electric vehicle start-up phase;

[0032] The initial confidence level of the startup phase is determined based on the model recognition probability and the startup template matching degree.

[0033] When the initial confidence level of the startup phase reaches the candidate identification threshold, the current load switching event is identified as a candidate electric vehicle charging event; when the initial confidence level of the startup phase does not reach the candidate identification threshold, the electric vehicle charging identification of the current load switching event is terminated.

[0034] As a preferred embodiment of the present invention, S6 specifically includes:

[0035] The subsequent electrical data of the branch where the candidate electric vehicle charging event is located is divided into subsequent analysis windows, and the degree of change in operating status is determined based on the power and current changes between adjacent subsequent analysis windows.

[0036] When the first preset number of subsequent analysis windows meet the stable state determination conditions and no new load switching events are detected, the candidate electric vehicle charging event is determined to have entered a stable operating state, and the first preset number of subsequent analysis windows are included in the set of effective windows for the stable phase.

[0037] When the number of windows in the set of effective windows in the stable phase is less than the second preset number, the subsequent stable operation windows are acquired until the second preset number of consecutive effective stable operation windows are formed; when the consecutive effective stable operation windows are not formed within the preset longest tracking time, the candidate electric vehicle charging event is determined to be in a state of insufficient evidence.

[0038] As a preferred embodiment of the present invention, S7 specifically includes:

[0039] Based on the pre-event stability window corresponding to the candidate electric vehicle charging event, and according to the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of each continuous effective stable operation window, a steady-state power-harmonic joint feature vector corresponding to each continuous effective stable operation window is constructed respectively.

[0040] The steady-state power-harmonic joint feature vectors are matched with the electric vehicle stable charging feature template to determine the average steady-state template matching degree, and the continuous steady-state power-harmonic stability is determined based on the power change and steady-state incremental harmonic feature change between multiple continuous effective stable operation windows.

[0041] Evidence for the stable phase is determined based on the average steady-state template matching degree and the continuous steady-state power-harmonic stability. When both satisfy the corresponding stable phase confirmation conditions, the candidate electric vehicle charging event is confirmed, and the final charging identification confidence is determined based on the stable phase evidence and the initial confidence of the startup phase; otherwise, the candidate electric vehicle charging event is rejected.

[0042] As a preferred embodiment of the present invention, S8 specifically includes:

[0043] A first control threshold and a second control threshold are set, wherein the second control threshold is higher than the first control threshold;

[0044] When the final charging identification confidence level is lower than the first control threshold, the corresponding branch continues to be monitored;

[0045] When the final charging identification confidence reaches the first control threshold and is lower than the second control threshold, a charging alarm is sent, and if the stable charging characteristics of the electric vehicle are still detected after a preset response time, a disconnection command is sent to the corresponding controllable circuit breaker.

[0046] When the final charging identification confidence level reaches the second control threshold, a charging alarm is sent and a disconnection command is sent to the corresponding controllable circuit breaker.

[0047] A non-intrusive load monitoring-based electric vehicle charging management and control system includes:

[0048] The model and template building module is used to obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and to build a load classification model for the start-up phase, a feature template for the electric vehicle start-up phase, and a feature template for the electric vehicle stable charging.

[0049] The electrical data acquisition module is used to collect electrical data of the user's power supply lines and establish a mapping relationship between the branch to be managed and the corresponding controllable circuit breaker.

[0050] The load switching event detection module is used to preprocess the electrical data and detect load switching events, determine the time of occurrence of the load switching event and the branch where the event is located, and obtain the stable window before the event, the transient window after the event and the initial running window after the event.

[0051] The startup phase feature extraction module is used to construct the startup phase event feature vector of the current load switching event based on the pre-event stable window, the startup transient window, and the post-event initial running window.

[0052] The candidate event determination module is used to determine the initial confidence level of the startup phase based on the identification result of the startup phase event feature vector by the startup phase load classification model and the matching result of the startup phase event feature vector and the electric vehicle startup phase feature template, and to determine the load switching event that meets the candidate identification condition as a candidate electric vehicle charging event.

[0053] The stable operation tracking module is used to continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous and effective stable operation windows.

[0054] The stable phase confirmation module is used to construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous and effective stable operation windows, determine stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, and the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event based on the stable phase evidence; when confirming the candidate electric vehicle charging event, the final charging identification confidence level is determined based on the stable phase evidence and the initial confidence level of the startup phase.

[0055] The hierarchical control module is used to implement hierarchical control on the corresponding branch based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located;

[0056] The control result verification module is used to verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.

[0057] The beneficial effects of this invention are:

[0058] This invention takes load switching events as the starting point for identification. Based on the stable data before the event, the start-up transient data, and the initial operating data after the event, it extracts incremental power, incremental current, incremental harmonics, and start-up transient features. This allows the identification during the start-up phase to mainly reflect the changes in electrical state caused by the newly added load, thereby reducing the interference of the original household loads on the characteristics of the load to be identified. This is beneficial to improving the load identification capability under mixed operation conditions of multiple devices.

[0059] This invention divides electric vehicle charging identification into a startup phase and a stable operation phase. During the startup phase, candidate electric vehicle charging events are screened using a load classification model and electric vehicle startup phase feature templates, without directly cutting off power based on the startup phase identification results. For candidate events, the same branch is tracked until a stable operation state is reached, and then the power-harmonic joint characteristics of the stable operation phase are used for confirmation or rejection. This allows for progressive identification using the different electrical characteristics exhibited by the electric vehicle charging load at different operation phases, reducing misidentification caused by other devices exhibiting similar characteristics only during the startup phase.

[0060] In the stabilization phase, this invention not only compares the matching degree between candidate loads and the characteristic templates for stable charging of electric vehicles, but also determines steady-state power-harmonic stability based on power fluctuations and harmonic variations between multiple consecutive effective stable operating windows. Therefore, the stabilization phase simultaneously considers the category similarity between candidate loads and target loads, as well as the characteristic consistency of candidate loads themselves over continuous time. This allows the stabilization phase to independently verify the candidate identification results formed in the startup phase, which helps to further reduce the probability of false alarms and false power outages.

[0061] This invention, when determining whether a candidate load has entered a stable operating state, directly incorporates the continuous analysis window used for stable state determination into the effective window of the stable phase. This eliminates the need to wait for another complete stable window after confirming entry into a stable state, and sets a maximum tracking time. While ensuring sufficient identification data during the stable phase, it reduces redundant waiting, enabling candidate events to be confirmed within a limited time, thus meeting the requirements for timely monitoring and control of electric vehicle charging at homes.

[0062] This invention establishes a mapping relationship between the controlled branch and the corresponding controllable circuit breaker, and implements hierarchical control based on the final charging identification confidence level. For identification results with different confidence levels, different processing methods are adopted: continued monitoring, delayed control after alarm, or alarm and power outage. This ensures that the identification results correspond to the actual control intensity, promptly addressing electric vehicle charging behavior at homes while reducing unnecessary impacts caused by power outages directly triggered by low-confidence identification results.

[0063] This invention acquires electrical data of user power supply lines in a non-intrusive manner, without the need to install identification devices on specific household appliances. It can continuously detect load switching events and identify electric vehicle charging behavior, reducing reliance on manual inspections and improving the continuous monitoring and automated management capabilities of electric vehicle charging behavior at home. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0065] Figure 2 This is a schematic diagram of a typical non-electric vehicle load current waveform.

[0066] Figure 3 This is a schematic diagram of the load current waveform for electric vehicle charging.

[0067] Figure 4 This is a schematic diagram of the non-intrusive load identification principle. Detailed Implementation

[0068] In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0069] Combination Figures 1-4 The present invention provides the following embodiments:

[0070] Example 1:

[0071] A method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring includes:

[0072] S1. Obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and construct a load classification model for the startup phase, a feature template for the electric vehicle startup phase, and a feature template for the electric vehicle stable charging.

[0073] S2. Collect electrical data of user power supply lines and establish a mapping relationship between the branches to be managed and the corresponding controllable circuit breakers;

[0074] S3. Perform preprocessing and load switching event detection on the electrical data, determine the event occurrence time, the branch where the event is located, and the load switching direction of the load switching event, take the load switching event as the load switching event to be identified, and obtain the pre-event stability window, the start transient window, and the post-event initial operation window corresponding to the load switching event to be identified.

[0075] S4. Based on the pre-event stable window, the start-up transient window, and the post-event initial operation window, construct the start-up phase event feature vector of the current load switching event;

[0076] S5. Based on the identification results of the startup phase event feature vector by the startup phase load classification model, and the matching results of the startup phase event feature vector and the electric vehicle startup phase feature template, determine the initial confidence level of the startup phase, and identify the load switching events that meet the candidate identification conditions as candidate electric vehicle charging events.

[0077] S6. Continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous effective stable operation windows;

[0078] S7. Construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous effective stable operation windows, and determine the stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, as well as the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event; when confirming the candidate electric vehicle charging event, determine the final charging identification confidence level based on the stable phase evidence and the initial confidence level of the startup phase;

[0079] S8. Based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located, implement hierarchical control for the corresponding branch;

[0080] S9. Verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.

[0081] Furthermore, S1 specifically includes:

[0082] Historical voltage and current data of electric vehicle charging load and non-electric vehicle load under different operating conditions are obtained, and the historical voltage and current data are labeled with equipment category and operating status to form a labeled load sample set;

[0083] Based on the load input time of each labeled load sample in the labeled load sample set, a corresponding startup phase training feature vector is formed. The startup phase training feature vector is standardized, and a binary classification load recognition model is trained using the equipment category as a supervision label to obtain the startup phase load classification model.

[0084] Electric vehicle charging load samples are selected from the labeled load sample set. The electric vehicle start-up phase feature template is formed based on the standardized start-up phase training feature vector corresponding to the electric vehicle charging load sample. The electric vehicle stable charging feature template is formed based on the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of the electric vehicle charging load sample in the stable charging phase relative to the pre-event steady state.

[0085] Furthermore, in S2, a non-intrusive load monitoring terminal is installed on the user's inlet distribution box side. The non-intrusive load monitoring terminal includes a voltage acquisition channel and multiple current acquisition channels. The voltage acquisition channel is used to acquire the voltage signal on the inlet side, and each of the current acquisition channels is used to acquire the current signal of the corresponding branch to be managed. A mapping relationship is established between each current acquisition channel and the corresponding controllable circuit breaker, and the voltage signal on the inlet side and the current signal of each branch to be managed are acquired synchronously based on a unified sampling clock.

[0086] The voltage signal at the inlet side serves as a common voltage reference for all branches to be managed. Each current acquisition channel reflects the current changes of its corresponding branch. After synchronous acquisition based on a unified sampling clock, the power and its changes in each branch can be determined under the same time reference. The mapping relationship between the current acquisition channels and the controllable circuit breaker allows the branch where the identified event occurred to be further associated with the corresponding physical control object.

[0087] Furthermore, S3 specifically includes:

[0088] The electrical data is preprocessed and divided into continuous analysis windows. The event change amount is determined based on the changes in active power and the effective value of current between adjacent analysis windows.

[0089] When the change in the event reaches the event detection threshold, it is determined that a load switching event has occurred on the corresponding branch to be managed, and the time of occurrence of the load switching event and the branch where the event is located are determined.

[0090] Using the time of the event as the time reference, acquire stable data before the event, transient data at startup, and initial running data after the event.

[0091] The introduction of new loads will cause sudden changes in the active power and current of the corresponding branches. By comparing adjacent analysis windows, the load switching event and its occurrence time can be determined. By using the occurrence time of the event as a unified time reference to obtain data before the event, the start-up transient, and after the event, the features of the start-up phase extracted subsequently can all be formed around the same load introduction, thereby reducing the interference of the original continuously operating load on the features of the newly introduced load.

[0092] Furthermore, the event feature vector of the startup phase described in S4 consists of incremental active power, incremental reactive power, incremental current RMS value, power factor change, incremental harmonic characteristics, and startup transient characteristics.

[0093] The startup transient characteristics include the peak value of the instantaneous startup current, the peak factor, the startup pulse duration, the maximum power change rate, and the startup delay time.

[0094] Incremental active power, incremental reactive power, incremental current, and incremental harmonic characteristics are used to characterize the electrical changes produced by the newly added load relative to the original background load. The peak value of the instantaneous starting current, the duration of the starting pulse, and the starting delay time reflect the dynamic response at the moment the load is added. Combining these two types of characteristics to form the event feature vector of the starting phase can describe the newly added load from both incremental changes and transient responses, thereby improving the distinguishability of the starting phase.

[0095] Furthermore, S5 specifically includes:

[0096] Input the feature vector of the startup phase event into the startup phase load classification model to obtain the model identification probability that the current load switching event belongs to the electric vehicle charging load;

[0097] The matching degree of the start-up template is determined based on the feature distance between the feature vector of the start-up phase event and the feature template of the electric vehicle start-up phase;

[0098] The initial confidence level of the startup phase is determined based on the model recognition probability and the startup template matching degree.

[0099] When the initial confidence level of the startup phase reaches the candidate identification threshold, the current load switching event is identified as a candidate electric vehicle charging event; when the initial confidence level of the startup phase does not reach the candidate identification threshold, the electric vehicle charging identification of the current load switching event is terminated.

[0100] The model recognition probability reflects the statistical discrimination results obtained from training on labeled load samples, while the startup template matching degree reflects the degree of similarity between the current load switching event and the typical startup characteristics of electric vehicles. Using both together to determine the initial confidence level in the startup phase can reduce the judgment bias caused by relying solely on the classification model or template matching; this stage only filters candidate events, reserving conditions for subsequent verification using different physical evidence from the stable operation phase.

[0101] Furthermore, S6 specifically includes:

[0102] The subsequent electrical data of the branch where the candidate electric vehicle charging event is located is divided into subsequent analysis windows, and the degree of change in operating status is determined based on the power and current changes between adjacent subsequent analysis windows.

[0103] When the first preset number of subsequent analysis windows meet the stable state determination conditions and no new load switching events are detected, the candidate electric vehicle charging event is determined to have entered a stable operating state, and the first preset number of subsequent analysis windows are included in the set of effective windows for the stable phase.

[0104] When the number of windows in the set of effective windows in the stable phase is less than the second preset number, the subsequent stable operation windows are acquired until the second preset number of consecutive effective stable operation windows are formed; when the consecutive effective stable operation windows are not formed within the preset longest tracking time, the candidate electric vehicle charging event is determined to be in a state of insufficient evidence.

[0105] The power and current changes between adjacent subsequent analysis windows can reflect whether the candidate load has transitioned from the startup state to a relatively stable operating state. Extracting evidence of the stable phase after the stability conditions are met through consecutive windows can avoid mistaking the startup transition process as a steady-state feature. Continuing to use windows already used for steady-state determination as valid windows and setting a maximum tracking time can reduce redundant waiting while ensuring the reliability of steady-state evidence and improving the timeliness of identification.

[0106] Furthermore, the S7 specifically includes:

[0107] Based on the pre-event stability window corresponding to the candidate electric vehicle charging event, and according to the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of each continuous effective stable operation window, a steady-state power-harmonic joint feature vector corresponding to each continuous effective stable operation window is constructed respectively.

[0108] The steady-state power-harmonic joint feature vectors are matched with the electric vehicle stable charging feature template to determine the average steady-state template matching degree, and the continuous steady-state power-harmonic stability is determined based on the power change and steady-state incremental harmonic feature change between multiple continuous effective stable operation windows.

[0109] Evidence for the stable phase is determined based on the average steady-state template matching degree and the continuous steady-state power-harmonic stability. When both satisfy the corresponding stable phase confirmation conditions, the candidate electric vehicle charging event is confirmed, and the final charging identification confidence is determined based on the stable phase evidence and the initial confidence of the startup phase; otherwise, the candidate electric vehicle charging event is rejected.

[0110] Steady-state template matching is used to determine whether candidate loads have category characteristics consistent with stable electric vehicle charging loads during the stable operation phase. Steady-state power-harmonic stability is used to evaluate the consistency of the above characteristics over a continuous time window. Candidate events are confirmed only when both category similarity and temporal stability conditions are met. This allows the stable phase to independently verify candidate results using physical evidence distinct from that used in the startup phase, and to exclude non-target loads that are only similar to electric vehicle charging loads during the startup phase.

[0111] Furthermore, S8 specifically includes:

[0112] A first control threshold and a second control threshold are set, wherein the second control threshold is higher than the first control threshold;

[0113] When the final charging identification confidence level is lower than the first control threshold, the corresponding branch continues to be monitored;

[0114] When the final charging identification confidence reaches the first control threshold and is lower than the second control threshold, a charging alarm is sent, and if the stable charging characteristics of the electric vehicle are still detected after a preset response time, a disconnection command is sent to the corresponding controllable circuit breaker.

[0115] When the final charging identification confidence level reaches the second control threshold, a charging alarm is sent and a disconnection command is sent to the corresponding controllable circuit breaker.

[0116] The final charging identification confidence level comprehensively reflects the reliability of the identification results during the startup phase and the confirmation results during the stable phase. Depending on the confidence level range, different measures are adopted: continued monitoring, delayed power-off after alarm, and direct power-off after alarm, to ensure that the control intensity matches the identification reliability. For medium-confidence events, a response time is retained to reduce false power-offs caused by boundary identification results; for high-confidence events, the corresponding branch can be cut off in a timely manner.

[0117] Example 2:

[0118] A non-intrusive load monitoring-based in-home charging management system for electric vehicles includes:

[0119] The model and template building module is used to obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and to build a load classification model for the start-up phase, a feature template for the electric vehicle start-up phase, and a feature template for the electric vehicle stable charging.

[0120] The electrical data acquisition module is used to collect electrical data of the user's power supply lines and establish a mapping relationship between the branch to be managed and the corresponding controllable circuit breaker.

[0121] The load switching event detection module is used to preprocess the electrical data and detect load switching events, determine the time of occurrence of the load switching event and the branch where the event is located, and obtain the stable window before the event, the transient window after the event and the initial running window after the event.

[0122] The startup phase feature extraction module is used to construct the startup phase event feature vector of the current load switching event based on the pre-event stable window, the startup transient window, and the post-event initial running window.

[0123] The candidate event determination module is used to determine the initial confidence level of the startup phase based on the identification result of the startup phase event feature vector by the startup phase load classification model and the matching result of the startup phase event feature vector and the electric vehicle startup phase feature template, and to determine the load switching event that meets the candidate identification condition as a candidate electric vehicle charging event.

[0124] The stable operation tracking module is used to continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous and effective stable operation windows.

[0125] The stable phase confirmation module is used to construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous and effective stable operation windows, determine stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, and the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event based on the stable phase evidence; when confirming the candidate electric vehicle charging event, the final charging identification confidence level is determined based on the stable phase evidence and the initial confidence level of the startup phase.

[0126] The hierarchical control module is used to implement hierarchical control on the corresponding branch based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located;

[0127] The control result verification module is used to verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.

[0128] Application example:

[0129] A method for managing electric vehicle charging at home based on non-intrusive load monitoring is proposed. This method acquires electrical data of the user's power supply line through non-intrusive load monitoring. Taking load switching events as the identification starting point, the identification process of the same candidate load is divided into a startup phase and a subsequent stable operation phase. The startup phase is used to screen candidate electric vehicle charging events from newly connected loads; the subsequent stable operation phase is used to track the same candidate event and use different steady-state power-harmonic characteristics than the startup phase to confirm or reject the candidate judgment in the startup phase. After completing the two-stage progressive confirmation, hierarchical control is implemented according to the final identification result and the branch where the event is located, and the control results are verified in a closed loop. The specific process is as follows.

[0130] S1. Construct a startup phase identification model and a stable charging phase characteristic benchmark.

[0131] Historical voltage and current data of electric vehicle charging loads and non-electric vehicle loads under different operating conditions are acquired and labeled with equipment category and operating status to form a labeled load sample set. The non-electric vehicle loads include compressor loads, electromagnetic loads, and other household appliances with starting inrush current or harmonic distortion characteristics that may be confused with electric vehicle charging loads. The equipment category label is used to distinguish between electric vehicle charging loads and non-electric vehicle loads; the operating status label includes at least the pre-operation state, the startup state, and the stable operating state.

[0132] For each labeled load sample in the labeled load sample set, the corresponding load's activation time is determined based on the operating status label. Using this activation time as the time reference, the sample data is divided into pre-event stable data, startup transient data, and post-event initial operating data. Following the same feature extraction rules as the online identification stage, incremental active power, incremental reactive power, incremental current RMS value, power factor change, incremental harmonic characteristics, startup instantaneous current peak value, peak factor, startup pulse duration, maximum power change rate, and startup delay time are extracted from the above data to form the startup stage training feature vector corresponding to each labeled load sample.

[0133]

[0134] in, This indicates the serial number of the labeled load sample; Indicates the first The instantaneous peak current of a single labeled load sample during the startup phase; This indicates the duration of the start pulse; the definitions of the other features are consistent with those of the features corresponding to the online identification stage.

[0135] The training sample set is constructed using the samples used for model training from the labeled load sample set. The mean of each dimension of the training feature vector during the startup phase is calculated based on the training sample set. and standard deviation The training feature vectors for each startup phase are standardized. A binary load identification model is trained using equipment category labels as supervisory labels. The load classification model for the startup phase is obtained. This model receives the standardized feature vectors of online load switching events and outputs the model identification probability that the load switching event belongs to the electric vehicle charging load. The model training process can use the cross-entropy loss function:

[0136]

[0137] in, Indicates the number of training samples. Indicates the first Device category labels for each training sample This indicates the charging load of the electric vehicle. Indicates non-electric vehicle load. The model represents the first The probability of electric vehicle charging output from each training sample. This represents the model parameters after training.

[0138] Electric vehicle charging load samples are selected from the labeled load sample set. For the standardized start-up phase training feature vector corresponding to the electric vehicle charging load sample, the average value of each feature dimension is calculated to form the electric vehicle start-up phase feature template:

[0139]

[0140] in, This indicates the number of electric vehicle charging load samples used to form the characteristic template for the electric vehicle startup phase. Indicates the first The standardized start-up phase training feature vector corresponding to each electric vehicle charging load sample.

[0141] For each electric vehicle charging load sample, after entering a stable charging state, the incremental active power, incremental current RMS value, power factor change, and steady-state incremental harmonic characteristics relative to the stable state before the corresponding event are used to construct a stable charging feature vector for each electric vehicle charging load sample. And calculate the average value of each dimension of features to form a feature template for stable charging of electric vehicles:

[0142]

[0143] in, Indicates the first Stable charging feature vector corresponding to each electric vehicle charging load sample.

[0144] The electric vehicle stable charging feature template can be represented as follows:

[0145]

[0146] in, , and These represent the average active power increment, average current RMS value increment, and average power factor change of the electric vehicle charging load sample relative to the steady state before the corresponding event, respectively. Indicates the first The average steady-state incremental harmonic characteristics corresponding to the subharmonics.

[0147] The candidate identification threshold for the startup phase, the template matching threshold for the stable phase, the steady-state stability threshold, and the subsequent control threshold are determined using the training set and the validation set.

[0148] S2. Establish the correspondence between monitoring branches and controlled objects and perform synchronous sampling.

[0149] A non-intrusive load monitoring terminal is installed at the user's inlet distribution box. This non-intrusive load monitoring terminal includes a voltage acquisition channel and multiple current acquisition channels. The voltage acquisition channel is used to acquire the voltage signal at the inlet side. Each of the aforementioned current acquisition channels is used to acquire the current signal of the corresponding branch to be controlled. .

[0150] Establish the first Each current acquisition channel and its corresponding controllable circuit breaker Mapping relationship between them:

[0151]

[0152] in, Indicates the first The current acquisition channel number corresponding to each branch to be managed. This indicates the controllable circuit breaker corresponding to the branch to be controlled.

[0153] The voltage signal at the inlet side is collected synchronously based on a unified sampling clock. and the current signals of each branch to be controlled It also records the corresponding sampling time information, so that the calculation of electrical parameters of different branches, the detection of load switching events and the location of branches are all based on a unified time reference.

[0154] S3. Preprocess electrical data and detect load switching events.

[0155] The voltage and current sequences obtained in step S2 are calibrated, outlier-handled, and digitally filtered to obtain the preprocessed voltage sequence. and current sequence The continuous data is divided into multiple analysis windows according to a preset window length, and the analysis window preferably contains an integer number of power frequency cycles.

[0156] For the The first branch road An analysis window calculates active power. and the effective value of current :

[0157]

[0158] in, This indicates the number of sampling points in the current analysis window.

[0159] Construct event change quantities based on power and current changes between adjacent analysis windows:

[0160]

[0161] in, and These represent the weights of the power change and the current change, respectively. ; and These represent the power reference value and current reference value corresponding to the previous stable operating period of the current branch, respectively; A positive number set to prevent the denominator from being zero.

[0162] when At that time, the judgment of the first Load switching events occurred on several branches, among which... This represents the event detection threshold. An event record is created for each branch that meets the event detection criteria to avoid retaining only a single event when multiple branches switch over at the same time.

[0163] For load switching events that reach the event detection threshold, the load switching direction is determined based on the changes in active power and effective current before and after the event; the load switching event is taken as the current load switching event to be identified and enters the subsequent start-up phase for identification, while the load switching event is not entered into the start-up phase for identification after the event record is formed.

[0164] For the event to be identified, its branch is denoted as The time of the event is recorded as .by Extracting a stable window before an event as a time reference The startup transient window includes the load input process. and the initial running window after the event .

[0165] S4. Constructing event characteristics for the startup phase based on changes before and after the event.

[0166] For the current load switching event determined in step S3, the pre-event stability window is utilized. and the initial running window after the event Calculate the incremental characteristics of load input, including incremental active power. Incremental reactive power Incremental current RMS value and power factor change .in , , , These represent the active power, reactive power, RMS current, and power factor corresponding to the stability window before the event, respectively. , , , These represent the active power, reactive power, RMS current, and power factor corresponding to the initial operating window after the event.

[0167] Frequency domain transformation is performed on the current data before and after the event, aligned with the power frequency phase, to obtain the amplitude of each harmonic, and the 1st harmonic is calculated. The proportion of sub-incremental harmonics:

[0168]

[0169] in, and These represent the events before and after the event, respectively. Second harmonic amplitude and These represent the fundamental current amplitudes before and after the event, respectively. Indicates the harmonic order. This indicates the highest harmonic order involved in feature extraction during the startup phase.

[0170] At the same time, start the transient window Extract the peak value of the instantaneous starting current. Peak factor Start-up pulse duration Startup delay time from load input to reaching the initial steady state and maximum power change rate ,in , This indicates the time interval between adjacent power analysis points. and Δt represents the active power corresponding to two adjacent power analysis points, and Δt represents the time interval between adjacent power analysis points.

[0171] The above features are combined into a startup phase event feature vector:

[0172]

[0173] The features of each dimension are standardized using the feature standardization parameters of the startup phase saved in step S1, i.e. The standardized initiation phase event feature vector is obtained. .

[0174] S5. Form candidate electric vehicle charging events based on evidence from the initiation phase.

[0175] The standardized initiation phase event feature vector obtained in step S4 Input the load classification model for the startup phase constructed in step S1 to obtain the model recognition probability corresponding to the electric vehicle charging load. .

[0176] At the same time, Electric vehicle start-up phase feature template established in step S1 Perform matching, calculate weighted feature distance and starting template matching degree:

[0177]

[0178] in, This represents the feature dimension of the startup phase. Indicates the first The matching weights corresponding to the features in the initiation phase of the item satisfy the following conditions: and .

[0179] The model recognition probability and the startup template matching degree are fused to obtain the initial confidence level in the startup phase:

[0180]

[0181] in, This represents the fusion weight between the model recognition result and the starting template matching result.

[0182] when When a load switching event occurs, it is marked as a candidate electric vehicle charging event. This only indicates that the newly added load has similar event characteristics to an electric vehicle charger during the startup phase; it is not directly confirmed as an electric vehicle charging load, nor is a power outage control directly triggered. Events that do not reach the candidate identification threshold end the current electric vehicle identification process, and new load switching events continue to be monitored.

[0183] S6. Track candidate events and determine stable operation phases.

[0184] For the candidate electric vehicle charging events obtained in step S5, continue to collect data on the corresponding branches. The subsequent electrical data is collected, and the changes of the same candidate event from the startup state to the stable operating state are tracked.

[0185] For continuous analysis windows where no new significant load switching events occur, the degree of change in operating status is determined based on the power and current changes between adjacent windows:

[0186]

[0187] in, and These represent the state determination weights for power change and current change, respectively. .

[0188] When continuous Each analysis window satisfies And the continuous If no new load switching events are detected within a single analysis window, the candidate event is determined to have entered a stable operation phase. This represents the threshold for determining a stable state.

[0189] The continuous [measures] used to determine whether a system has entered a stable operating phase The start time corresponding to the first analysis window among the analysis windows is determined as the stable operation start time. and the continuous Each analysis window is directly included in the set of valid windows in the stable phase, and another set of stable state determination windows is no longer collected.

[0190] The total number of valid stable operating windows required for stable phase identification is denoted as . When the number of windows included in the effective window set of the stable phase is less than Then, continue to acquire subsequent stable operation windows that are not affected by new load switching events, until the set of effective stable operation windows contains A number of consecutive windows used to determine a stable operating state; when the number of consecutive windows has reached or exceeded [a certain threshold]. When, select from them A continuous, effective, and stable operating window is used for identifying the stable phase.

[0191] If a new significant load switching event is detected during the formation of the set of effective stable operating windows, the window affected by the new load switching event will be excluded from the stable phase identification data of the current candidate events, and the new load switching event will be treated as an independent event.

[0192] The timing begins from step S5, when the current event is identified as a candidate electric vehicle charging event. If the preset maximum tracking time is exceeded... Still unable to obtain If a candidate event has a continuous, valid, and stable operating window, it will be marked as insufficient evidence, power outage control will not be implemented, and the stable phase confirmation process for this candidate event will end.

[0193] In one implementation, the analysis window duration can be set to... The number of consecutive windows required for determining a stable state can be set to The total number of valid stable operating windows required for stable phase identification can be set to [value]. The preset maximum tracking time can be set to Under this parameter setting, the parameters used to determine the steady state are... Each window can be directly used as the aforementioned This identifies a portion of the stable phase within a window, thus eliminating the need to wait for another complete stable window once a candidate load reaches a stable operating state. The parameters above are example parameters and can be adjusted based on the sampling frequency, analysis window length, and the startup characteristics of the electric vehicle charger.

[0194] S7. Confirming or denying candidate events based on continuous steady-state power-harmonic characteristics.

[0195] When step S6 is within the preset maximum tracking time Internal acquisition When there are consecutive valid and stable operating windows, the stable window before the event corresponding to the current candidate event in step S3 is used. As a baseline, the incremental active power of each stable operating window relative to the pre-event stable state is extracted. Incremental current RMS value and power factor change .

[0196] For the The stable operating window is used to calculate the first [number]th ... Substeady-state incremental harmonic ratio:

[0197]

[0198] in, Indicates the first The first stable operating window Second harmonic amplitude The first stable state before the event Second harmonic amplitude. Indicates the first The fundamental current amplitude within a stable operating window, The amplitude of the fundamental current in the steady state before the event; Indicates the harmonic order. This indicates the highest harmonic order involved in the feature extraction of the stable phase.

[0199] Based on this, the first... Steady-state power-harmonic joint eigenvectors for a stable operating window:

[0200]

[0201] First, a weighted feature distance calculation method corresponding to the template matching in the startup phase is adopted to calculate the first... Steady-state power-harmonic joint eigenvector of a stable operating window The electric vehicle stable charging feature template established in step S1 Perform matching and calculate steady-state feature distance. .

[0202] And determine the steady-state template matching degree based on the steady-state feature distance:

[0203]

[0204] in, The characteristic dimension of the steady-state power-harmonic joint eigenvector; Indicates the first The matching weights corresponding to the steady-state features of the terms satisfy the following conditions: and .

[0205] Furthermore, the average steady-state template matching degree corresponding to multiple consecutive effective stable operating windows is calculated:

[0206]

[0207] in, This indicates the number of consecutive, valid, stable operating windows used for identifying the stable phase; the This is used to characterize the degree of class similarity between candidate loads during stable operation and electric vehicle stable charging states.

[0208] Secondly, construct the harmonic characteristic vectors for each stable operating window. And calculate the average harmonic eigenvector. Harmonic stability is determined based on the degree of deviation of the harmonic characteristics of each stable operating window from the average harmonic characteristics.

[0209]

[0210] in, Represents the L2 norm, The larger the value, the more stable the harmonic spectrum of the candidate load is within the continuous stable operating window.

[0211] Further calculate the average incremental active power during the continuous stable operation window. and determine power stability .

[0212] By fusing power stability and harmonic stability, we obtain continuous steady-state power-harmonic stability. ,in, .

[0213] Based on the average steady-state template matching degree and evidence for the formation of a steady-state phase in continuous steady-state power-harmonic stability:

[0214]

[0215] in, .

[0216] when and When a candidate load is determined to have sufficient class similarity to the stable charging characteristics of electric vehicles during the stable operation phase, and also maintains corresponding power-harmonic stability over continuous time, the stable phase confirmation result is recorded as follows: Otherwise, record as .

[0217] when When this occurs, the candidate judgment formed in step S5 is rejected, and the current event is identified as a non-target load with similar startup phase but whose stable operation characteristics do not conform to the stable charging characteristics of electric vehicles, and electric vehicle charging control is not executed. At that time, the initial confidence level during the startup phase is fused with the evidence from the stable phase to obtain the final charging recognition confidence level:

[0218]

[0219] in, .

[0220] Therefore, step S5 only uses event evidence from the load input and startup phases to form candidate events; this step uses another set of physical evidence after the same candidate event enters stable operation to independently verify the candidate judgment, and has the function of negating the candidate results from the startup phase.

[0221] S8. Implement hierarchical branch control based on the final identification results.

[0222] Only when the result is confirmed during the stable phase Enter the control process. Set the first control threshold. Second control threshold ,and .

[0223] when When, do not perform a power outage operation, continue monitoring the current branch; when When the event occurs, a charging alarm is triggered, and the event time, corresponding branch, final charging identification confidence level, and current charging status are sent to the user terminal and / or management platform, and a response timer is started.

[0224] Continue monitoring the corresponding branch after an alarm is triggered. If within the preset response time... If the stable charging characteristic described in step S7 is detected to have disappeared, it is determined that the user has stopped charging, and no power-off is performed; if the condition is met... If the stable charging characteristic is still detected, then according to the branch-circuit breaker mapping relationship... To the controllable circuit breaker corresponding to the branch where the event occurred. Send a disconnect command.

[0225] when If the current charging behavior is determined to have a high degree of recognition confidence, an alarm message is sent to the corresponding controllable circuit breaker. Send a disconnect command.

[0226] S9. Perform closed-loop verification of the power-off action and the charging termination result.

[0227] To the controllable circuit breaker After sending the disconnect command, obtain the status feedback of the circuit breaker and continue to collect data on the branch where the event occurred. Electrical data.

[0228] When the circuit breaker feedback is in the open state, and the electric vehicle stable charging characteristics determined in step S7 are no longer detected in a consecutive preset number of posterior analysis windows, the electric vehicle in-home charging management is determined to be successful.

[0229] When the circuit breaker does not report an open state, the control execution is deemed abnormal; when the circuit breaker reports that it has been opened but the corresponding branch still exhibits the stable charging characteristics of the electric vehicle, the actual electrical result is deemed inconsistent with the control state, and abnormal handling information is generated.

[0230] The time of occurrence of this event The side road where the incident occurred Initial confidence level during the startup phase Evidence of the stable phase Final charging identification confidence level The alarm status, circuit breaker operation status, and final electrical verification results form a complete event record.

[0231] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring, characterized in that, include: S1. Obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and construct a load classification model for the startup phase, a feature template for the electric vehicle startup phase, and a feature template for the electric vehicle stable charging. S2. Collect electrical data of user power supply lines and establish a mapping relationship between the branches to be managed and the corresponding controllable circuit breakers; S3. Perform preprocessing and load switching event detection on the electrical data, determine the event occurrence time, the branch where the event is located, and the load switching direction of the load switching event, take the load switching event as the load switching event to be identified, and obtain the pre-event stability window, the start transient window, and the post-event initial operation window corresponding to the load switching event to be identified. S4. Based on the pre-event stable window, the start-up transient window, and the post-event initial operation window, construct the start-up phase event feature vector of the current load switching event; S5. Based on the identification results of the startup phase event feature vector by the startup phase load classification model, and the matching results of the startup phase event feature vector and the electric vehicle startup phase feature template, determine the initial confidence level of the startup phase, and identify the load switching events that meet the candidate identification conditions as candidate electric vehicle charging events. S6. Continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous effective stable operation windows; S7. Construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous effective stable operation windows, and determine the stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, as well as the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event; when confirming the candidate electric vehicle charging event, determine the final charging identification confidence level based on the stable phase evidence and the initial confidence level of the startup phase; S8. Based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located, implement hierarchical control for the corresponding branch; S9. Verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.

2. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, S1 specifically includes: Historical voltage and current data of electric vehicle charging load and non-electric vehicle load under different operating conditions are obtained, and the historical voltage and current data are labeled with equipment category and operating status to form a labeled load sample set; Based on the load input time of each labeled load sample in the labeled load sample set, a corresponding startup phase training feature vector is formed. The startup phase training feature vector is standardized, and a binary classification load recognition model is trained using the equipment category as a supervision label to obtain the startup phase load classification model. Electric vehicle charging load samples are selected from the labeled load sample set. The electric vehicle start-up phase feature template is formed based on the standardized start-up phase training feature vector corresponding to the electric vehicle charging load sample. The electric vehicle stable charging feature template is formed based on the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of the electric vehicle charging load sample in the stable charging phase relative to the pre-event steady state.

3. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, In S2, a non-intrusive load monitoring terminal is installed on the user's inlet distribution box side. The non-intrusive load monitoring terminal includes a voltage acquisition channel and multiple current acquisition channels. The voltage acquisition channel is used to acquire the voltage signal on the inlet side, and each of the current acquisition channels is used to acquire the current signal of the corresponding branch to be managed. A mapping relationship is established between each current acquisition channel and the corresponding controllable circuit breaker, and the voltage signal on the inlet side and the current signal of each branch to be managed are acquired synchronously based on a unified sampling clock.

4. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 3, characterized in that, S3 specifically includes: The electrical data is preprocessed and divided into continuous analysis windows. The event change amount is determined based on the changes in active power and the effective value of current between adjacent analysis windows. When the change in the event reaches the event detection threshold, it is determined that a load switching event has occurred on the corresponding branch to be managed, and the time of occurrence of the load switching event and the branch where the event is located are determined. Using the time of the event as the time reference, acquire stable data before the event, transient data at startup, and initial running data after the event.

5. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, The startup phase event feature vector described in S4 consists of incremental active power, incremental reactive power, incremental current RMS value, power factor change, incremental harmonic characteristics, and startup transient characteristics. The startup transient characteristics include the peak value of the instantaneous startup current, the peak factor, the startup pulse duration, the maximum power change rate, and the startup delay time.

6. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, S5 specifically includes: Input the feature vector of the startup phase event into the startup phase load classification model to obtain the model identification probability that the current load switching event belongs to the electric vehicle charging load; The matching degree of the start-up template is determined based on the feature distance between the feature vector of the start-up phase event and the feature template of the electric vehicle start-up phase; The initial confidence level of the startup phase is determined based on the model recognition probability and the startup template matching degree. When the initial confidence level of the startup phase reaches the candidate identification threshold, the current load switching event is identified as a candidate electric vehicle charging event; when the initial confidence level of the startup phase does not reach the candidate identification threshold, the electric vehicle charging identification of the current load switching event is terminated.

7. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, S6 specifically includes: The subsequent electrical data of the branch where the candidate electric vehicle charging event is located is divided into subsequent analysis windows, and the degree of change in operating status is determined based on the power and current changes between adjacent subsequent analysis windows. When the first preset number of subsequent analysis windows meet the stable state determination conditions and no new load switching events are detected, the candidate electric vehicle charging event is determined to have entered a stable operating state, and the first preset number of subsequent analysis windows are included in the set of effective windows for the stable phase. When the number of windows in the set of effective windows in the stable phase is less than the second preset number, the subsequent stable operation windows are acquired until the second preset number of consecutive effective stable operation windows are formed; when the consecutive effective stable operation windows are not formed within the preset longest tracking time, the candidate electric vehicle charging event is determined to be in a state of insufficient evidence.

8. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, S7 specifically includes: Based on the pre-event stability window corresponding to the candidate electric vehicle charging event, and according to the incremental active power, incremental current RMS value, power factor change and steady-state incremental harmonic characteristics of each continuous effective stable operation window, a steady-state power-harmonic joint feature vector corresponding to each continuous effective stable operation window is constructed respectively. The steady-state power-harmonic joint feature vectors are matched with the electric vehicle stable charging feature template to determine the average steady-state template matching degree, and the continuous steady-state power-harmonic stability is determined based on the power change and steady-state incremental harmonic feature change between multiple continuous effective stable operation windows. Evidence for the stable phase is determined based on the average steady-state template matching degree and the continuous steady-state power-harmonic stability. When both satisfy the corresponding stable phase confirmation conditions, the candidate electric vehicle charging event is confirmed, and the final charging identification confidence is determined based on the stable phase evidence and the initial confidence of the startup phase; otherwise, the candidate electric vehicle charging event is rejected.

9. The method for managing and controlling electric vehicle charging at home based on non-intrusive load monitoring according to claim 1, characterized in that, S8 specifically includes: A first control threshold and a second control threshold are set, wherein the second control threshold is higher than the first control threshold; When the final charging identification confidence level is lower than the first control threshold, the corresponding branch continues to be monitored; When the final charging identification confidence reaches the first control threshold and is lower than the second control threshold, a charging alarm is sent, and if the stable charging characteristics of the electric vehicle are still detected after a preset response time, a disconnection command is sent to the corresponding controllable circuit breaker. When the final charging identification confidence level reaches the second control threshold, a charging alarm is sent and a disconnection command is sent to the corresponding controllable circuit breaker.

10. A non-intrusive load monitoring-based in-home charging management and control system for electric vehicles, characterized in that, include: The model and template building module is used to obtain a labeled load sample set of electric vehicle charging load and non-electric vehicle load, and to build a load classification model for the start-up phase, a feature template for the electric vehicle start-up phase, and a feature template for the electric vehicle stable charging. The electrical data acquisition module is used to collect electrical data of the user's power supply lines and establish a mapping relationship between the branch to be managed and the corresponding controllable circuit breaker. The load switching event detection module is used to preprocess the electrical data and detect load switching events, determine the time of occurrence of the load switching event and the branch where the event is located, and obtain the stable window before the event, the transient window after the event and the initial running window after the event. The startup phase feature extraction module is used to construct the startup phase event feature vector of the current load switching event based on the pre-event stable window, the startup transient window, and the post-event initial running window. The candidate event determination module is used to determine the initial confidence level of the startup phase based on the identification result of the startup phase event feature vector by the startup phase load classification model and the matching result of the startup phase event feature vector and the electric vehicle startup phase feature template, and to determine the load switching event that meets the candidate identification condition as a candidate electric vehicle charging event. The stable operation tracking module is used to continuously acquire subsequent electrical data of the branch where the candidate electric vehicle charging event is located, track the change of the candidate electric vehicle charging event from the start-up state to the stable operation state, and acquire a preset number of continuous and effective stable operation windows. The stable phase confirmation module is used to construct corresponding steady-state power-harmonic joint feature vectors based on each of the continuous and effective stable operation windows, determine stable phase evidence based on the matching degree between each of the steady-state power-harmonic joint feature vectors and the electric vehicle stable charging feature template, and the stability degree between each of the steady-state power-harmonic joint feature vectors, and confirm or deny the candidate electric vehicle charging event based on the stable phase evidence; when confirming the candidate electric vehicle charging event, the final charging identification confidence level is determined based on the stable phase evidence and the initial confidence level of the startup phase. The hierarchical control module is used to implement hierarchical control on the corresponding branch based on the final charging identification confidence level and the branch where the candidate electric vehicle charging event is located; The control result verification module is used to verify the control results based on the status of the controllable circuit breaker after control and the stable charging characteristics of electric vehicles in the corresponding branch.