Load identification method and system for user photovoltaic self-generation and self-use scene

Abstract

The invention provides a load identification method and system for a user photovoltaic self-generation and self-use scene, and the method comprises the steps: obtaining the electric parameter data of a to-be-identified user and an adjacent known photovoltaic user in a set time period, and the electric parameter data comprise a current effective value sequence, a power sequence and a harmonic sequence; detecting a load switching event of the user to be identified based on the electric parameter data, and judging whether photovoltaic access exists or not; in response to the existence of photovoltaic access, calculating the photovoltaic capacity of a to-be-identified user based on the electric parameter data and the photovoltaic capacity of an adjacent known photovoltaic user, and processing the power sequence and the harmonic sequence based on the photovoltaic capacity of the to-be-identified user so as to separate and obtain a pure load power sequence and a pure load harmonic sequence; determining steady-state characteristics corresponding to each load switching event based on the pure load power sequence, the pure load harmonic sequence and the current effective value sequence; and transferring the steady-state features to a pre-trained deep learning fusion network, and outputting a load identification result corresponding to each load switching event.

Description

Technical Field

[0001] This application relates to the field of power grid technology, specifically to a load identification method, system, electronic device, and storage medium for a user's self-consumption photovoltaic scenario. Background Technology

[0002] With the rapid popularization of distributed photovoltaic (PV) systems, some users have installed PV systems without declaration or registration, leading to problems such as inaccurate metering in distribution areas, distorted load models, abnormal line losses, and control failures. Existing PV detection methods mainly rely on macroscopic data such as voltage anomalies, reverse power flow, and low-frequency energy analysis, making it difficult to accurately identify hidden PV systems with small capacity and features concealed within the load. Furthermore, PV integration alters current waveforms and harmonic characteristics, making it difficult for traditional non-intrusive load monitoring (NILM) algorithms to accurately decompose load events in mixed scenarios.

[0003] Existing NILM methods typically assume that the system only contains user loads and do not consider the transient-steady-state aliasing problem caused by the superposition of photovoltaic power generation and loads. Once photovoltaics are connected, load events are masked by the steady-state ripple and harmonics of the inverter, causing feature extraction to fail and making it impossible to guarantee the accuracy of identification.

[0004] Therefore, given the above problems, a new method is needed that simultaneously possesses the capabilities of photovoltaic identification, photovoltaic stripping, and high-precision load identification. Summary of the Invention

[0005] This application provides a load identification method and system for user-generated and self-consumed photovoltaic scenarios, which solves the problem that traditional NILM cannot separate power generation, harmonics and load events when photovoltaics are installed privately, thus leading to misidentification and omission.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] Firstly, this application provides a load identification method for user-generated and self-consumed photovoltaic scenarios, the method comprising:

[0008] S1. Obtain electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period, wherein the electrical parameter data includes current RMS value sequence, power sequence and harmonic sequence;

[0009] S2. Based on electrical parameter data, detect the load switching events of the user to be identified and determine whether there is photovoltaic grid connection.

[0010] S3. In response to the determination that photovoltaic access exists, the photovoltaic capacity of the user to be identified is calculated based on the electrical parameter data and the photovoltaic capacity of the adjacent known photovoltaic users. Based on the photovoltaic capacity of the user to be identified, the power sequence and harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence.

[0011] S4. Based on the pure load power sequence, pure load harmonic sequence and current RMS value sequence, determine the steady-state characteristics corresponding to each load switching event. The steady-state characteristics include waveform diagrams and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

[0012] S5. Input the waveforms and multi-dimensional electrical features corresponding to each load switching event into a pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

[0013] Secondly, this application provides a load identification system for user-generated and self-consumed photovoltaic scenarios, the system comprising:

[0014] The data acquisition module is used to acquire electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period. The electrical parameter data includes current RMS value sequence, power sequence and harmonic sequence.

[0015] The event detection and photovoltaic identification module is used to detect load switching events of the user to be identified based on electrical parameter data, and to determine whether photovoltaic access exists;

[0016] The capacity estimation and photovoltaic stripping module is used to calculate the photovoltaic capacity of the user to be identified based on electrical parameter data and the photovoltaic capacity of adjacent known photovoltaic users in response to the determination of photovoltaic access. Based on the photovoltaic capacity of the user to be identified, the power sequence and harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence.

[0017] The feature extraction module is used to determine the steady-state characteristics corresponding to each load switching event based on the pure load power sequence, pure load harmonic sequence, and current RMS value sequence. The steady-state characteristics include waveforms and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

[0018] The load identification module is used to input the waveforms and multi-dimensional electrical features corresponding to each load switching event into a pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

[0019] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the load identification method for user photovoltaic self-consumption scenarios of the first aspect.

[0020] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium including a computer program or instructions that, when executed, cause the load identification method for a user photovoltaic self-consumption scenario of the first aspect to be performed.

[0021] In this embodiment, by introducing the power sequences of adjacent known photovoltaic users as a reference benchmark and utilizing event time alignment and correlation analysis, the interference of photovoltaic power generation on the total electricity consumption characteristics can be effectively eliminated, thereby solving the inaccuracy problem caused by the distortion of "net load" characteristics in traditional non-intrusive load monitoring in photovoltaic access scenarios. Based on this, by estimating photovoltaic capacity and subtracting the photovoltaic power generation portion, pure load electrical characteristics are obtained. Combined with deep learning fusion identification of waveform diagrams and multi-dimensional numerical features, the accuracy of photovoltaic identification and the overall precision of subsequent load type identification are significantly improved.

[0022] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the load identification method for a user-generated and self-consumed photovoltaic scenario provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. At the same time, in the description of the embodiments of this application, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0026] Figure 1 This is a flowchart illustrating the load identification method for a user-generated and self-consumed photovoltaic scenario provided in an embodiment of this application.

[0027] The process for load identification in this user's self-consumption photovoltaic scenario is as follows:

[0028] Step S1: Obtain electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period. The electrical parameter data includes current RMS value sequence, power sequence and harmonic sequence.

[0029] That is, collecting power input parameter data at a frequency of seconds or higher at the user's power inlet, including: power sequence: P(t); current RMS value sequence: I rms (t); Current harmonic sequence H u (t)(2 to 9): H2(t)…H9(t).

[0030] In addition to collecting user load identification data, the system also collects similar characteristic data (PV capacity, power curves, harmonic curves, etc.) from adjacent known PV users to form reference templates for PV identification. It should also be noted that adjacent known PV users do not include electrical equipment, only PV-related data.

[0031] Step S2: Based on electrical parameter data, detect the load switching events of the user to be identified and determine whether there is photovoltaic access.

[0032] In this embodiment of the application, step S2 specifically includes the following steps:

[0033] Step S21: Based on the current RMS value sequence and harmonic sequence of the user to be identified, detect at least one load switching event and the corresponding event time.

[0034] In this embodiment of the application, step S21 can be specifically understood as:

[0035] Step S211: Based on the current RMS value sequence of the user to be identified, calculate the difference (|I0||0) between adjacent sampling times of the current RMS value sequence of the user to be identified. rms (t)-I rms (t-1)|), if the absolute value of the difference is greater than the first preset threshold ΔI th (that is, |I) rms (t)-I rms (t-1)|>ΔI th If ), then a sudden current event is determined to have occurred.

[0036] Step S212: Based on the harmonic sequence of the user to be identified, calculate the difference (|H|) between adjacent sampling times of the nth harmonic sequence. k (t)-H k (t-1)|), if the absolute value of the difference is greater than the second preset threshold ε h (that is, |H) k (t)-H k (t-1)|>ε hIf n = 1, then a harmonic transition event is determined to have occurred. The harmonic sequence of the user to be identified includes the Mth harmonic sequence, where M is a positive integer greater than or equal to 1, and n is a positive integer between 1 and M.

[0037] Step S213: When the current sudden change event and harmonic transition event are satisfied simultaneously, mark the event time as the load switching event.

[0038] It should be noted that this application may detect more than one load switching event and its corresponding event time. Specifically, the event set can be denoted as {T1, T2, ..., T...}. K}

[0039] Step S22: For each load switching event, based on the corresponding event time, trim the power sequences of the user to be identified and the adjacent known photovoltaic users, and then time-align and splice the trimmed power sequences.

[0040] It is understandable that for each event point T i Both sets of data include photovoltaic (PV) data and data related to appliance operation, which are not convenient for subsequent determination of whether PV grid connection exists. Therefore, these data are pruned, and the remaining data unrelated to the event are spliced ​​together. Similarly, adjacent known PV users are pruned for the same time frame to facilitate subsequent determination of whether PV grid connection exists. The remaining data segments after pruning all event-related data from these two user power sequences are then plotted along a unified time axis (i.e., based on event time T). i To align with the reference points, they are reconnected. This splicing creates a new sequence that eliminates all local disturbances caused by load switching.

[0041] Furthermore, the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic users are respectively denoted as P. u (t) and P k (t).

[0042] Step S23: Normalize the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic users, and perform correlation analysis on the two normalized power sequences. Determine whether the user to be identified has photovoltaic access based on the analysis results.

[0043] In step S23, the spliced ​​power sequences of the user to be identified and the spliced ​​power sequences of adjacent known photovoltaic users are normalized, which can be specifically expressed as follows:

[0044]

[0045] Among them, P u (t) represents the spliced ​​power sequence of the user to be identified at time t, max(|P u|) is the maximum absolute value of the user's power sequence. This normalization operation scales the power sequence of the user to be identified to the range of [-1, 1]. P k (t) represents the spliced ​​power sequence of adjacent known photovoltaic users at time t, max(|P k ∣) is the maximum absolute value of the user's power sequence. This normalization operation also scales the power sequences of adjacent known users to the range of [-1, 1].

[0046] In step S23, correlation analysis is performed on the two normalized power sequences. The analysis results are used to determine whether the user to be identified has access to photovoltaic power. Specifically, this can be understood as:

[0047] Pearson correlation coefficient or cosine similarity (i.e., ρ = corr(P)) is used. u ′ (t),P k ′ (t))), calculate the correlation coefficient ρ between the two power sequences;

[0048] If the correlation coefficient is greater than or equal to the preset correlation threshold, it is determined that the user to be identified has photovoltaic access.

[0049] Among them, when ρ≥η (η is the correlation threshold, and the environmental factors such as light and radiation are basically the same in the same transformer area, that is, the coefficient threshold is set to 0.9), it is determined that the current user has photovoltaic access.

[0050] Step S3: In response to the determination that photovoltaic access exists, the photovoltaic capacity of the user to be identified is calculated based on the electrical parameter data and the photovoltaic capacity of the adjacent known photovoltaic users. Based on the photovoltaic capacity of the user to be identified, the power sequence and harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence.

[0051] In this embodiment of the application, step S3, which calculates the photovoltaic capacity of the user to be identified based on electrical parameter data and the photovoltaic capacity of adjacent known photovoltaic users, includes:

[0052] S31. Based on the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic users, determine the peak power ratio between the two power sequences;

[0053] S32, Photovoltaic capacity C based on adjacent known photovoltaic users k Based on the ratio of peak power to peak power, determine the photovoltaic capacity C of the user to be identified. u .

[0054] Specifically, it can be done through the following formula:

[0055]

[0056] Where max(P) u ) and max(P k The values ​​are the peak values ​​of the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic users, respectively.

[0057] Furthermore, in step S3, based on the photovoltaic capacity of the user to be identified, the power sequence and harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence, which can be understood as:

[0058] Based on the normalized spliced ​​power sequence P of the user to be identified k ′ (t), calculate the photovoltaic power output sequence of the user to be identified: P vu (t)=P k ′ (t)×C u Then calculate the pure load power sequence: P load (t)=P u (t)-P vu (t).

[0059] Then, through the formula Calculate the photovoltaic harmonic sequence H of the user to be identified vu (t), where H k (t) represents the harmonic components of adjacent known photovoltaic users, H vu (t) represents the photovoltaic harmonic components of the user to be identified, and then the pure load harmonic sequence is calculated: H load (t)=H u (t)-H vu (t).

[0060] Step S4: Based on the pure load power sequence, pure load harmonic sequence, and current RMS value sequence, determine the steady-state characteristics corresponding to each load switching event. The steady-state characteristics include waveform diagrams and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

[0061] In this embodiment of the application, step S4 specifically includes the following steps:

[0062] Step S411: For each load switching event, based on the corresponding event time and the current effective value sequence, determine the steady-state time t1 before the event and the steady-state time t2 after the event.

[0063] In other words, the point where the event occurs is the starting point for judgment. M cycles are traced back from the moment the event occurred to analyze the changes in the current waveform. During this period, the effective value of the current (I) is calculated. rms If I rmsIf the current waveform stops rising or falling and instead tends to stabilize, it indicates that the current waveform has entered a steady state, and this moment can be determined as the pre-event steady-state moment. Furthermore, after the event, the current waveform is searched forward for M periods to analyze whether the current is stable. During this period, the effective value sequence I of the current is calculated. rms Observe its changing trend, if I rms If the price no longer rises or falls continuously, but instead remains stable, then it can be determined as the steady state after the event.

[0064] For example, suppose a load switching event occurs at t = 10 seconds. At this moment, the current waveform changes instantaneously. Searching back M cycles from t = 10 seconds (the point of event occurrence), let's say M = 5. Looking back 5 cycles, we obtain the effective current value (I) over a period of time. rms (sequence). For example: I rms = [5.1, 5.0, 5.0, 4.9, 4.8], the effective value of the current gradually decreases, but the rate of decrease becomes smaller and smaller. If the effective value of the current no longer rises or falls significantly during this period, but remains relatively stable, it can be determined that the steady-state moment before the event occurred at some point before t = 10 seconds (e.g., t = 9 seconds).

[0065] Step S412: Extract the power values ​​corresponding to the steady-state time before the event and the steady-state time after the event from the pure load power sequence, and obtain the steady-state power characteristic data by subtraction.

[0066] That is, P load (t2) minus P load (t1).

[0067] Step S413: Extract the power values ​​corresponding to the steady-state time before the event and the steady-state time after the event from the pure load harmonic sequence, and obtain the steady-state harmonic characteristic data by subtraction.

[0068] That is, H load (t2) minus H load (t1).

[0069] Step S414: Based on steady-state power characteristic data and steady-state harmonic characteristic data, determine the steady-state characteristics corresponding to the load switching event. The steady-state characteristics include waveform diagrams and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

[0070] Among them, the multidimensional electrical characteristics include active power, reactive power, current distortion rate, third current harmonic, fifth current harmonic, seventh current harmonic, and fundamental current.

[0071] The waveform representing the degree of steady-state characteristic fluctuation can be understood as converting one-dimensional steady-state power characteristic data into a standardized two-dimensional grayscale image, preparing for subsequent image recognition models. First, the steady-state power characteristic data is normalized to its maximum absolute value to eliminate the influence of load power magnitude on the image amplitude, allowing the model to focus on learning the shape characteristics of the power curve changes. Then, image generation and size standardization are performed. The normalized single-cycle current waveform is mapped to a 64x64 pixel image with a white background. Each point is mapped to coordinates on the canvas and drawn using black. To obtain a clearer image, linear interpolation can be performed between adjacent points to draw line segments, instead of just drawing discrete points. Finally, a standard-sized grayscale image is obtained, clearly showing the unique net load power curve shape after load normalization, which can be used as input for the subsequent image recognition network.

[0072] Step S5: Input the waveform and multi-dimensional electrical features corresponding to each load switching event into the pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

[0073] In this embodiment, the pre-trained deep learning fusion network includes an image processing sub-network, a feature processing sub-network, and an adaptive fusion module.

[0074] The image processing subnetwork is used to perform convolution processing on the input waveform representing the degree of fluctuation of steady-state features, and output the probability distribution of the first load category.

[0075] The image processing sub-network employs a lightweight CNN structure for parallel extraction and fusion of multi-scale features to process steady-state images. At the beginning of the network, multiple convolutional kernels of different sizes are used in parallel for convolution, and the resulting feature maps are then concatenated. This allows the network to simultaneously capture subtle local distortions and the overall shape and contour of active and reactive power changes in the power curve.

[0076] The network ultimately outputs a load category probability distribution P based on the steady-state power curve image through fully connected layers and softmax layers. s =[P s 1, P s 2, ..., P s k].

[0077] The feature processing subnetwork is used to perform gated attention processing on the input multidimensional electrical features and output the second load category probability distribution.

[0078] The feature processing subnetwork includes a gating unit, a self-attention layer, and a multilayer perceptron classifier.

[0079] The gating unit is used to process the input multidimensional electrical features, generate the corresponding multidimensional gating vector, and multiply the multidimensional gating vector with the multidimensional electrical features element by element to obtain the gating selected feature vector.

[0080] The self-attention layer is used to calculate the interrelationships between features based on the gated feature vector and generate an attention weight vector. The attention weight vector is then multiplied element-wise with the gated feature vector to obtain a weighted feature vector.

[0081] A multilayer perceptron classifier is used to classify weighted feature vectors and output a second probability distribution P. f =[P f 1, P f 2, ..., P f k].

[0082] The adaptive fusion module is used to extract the first feature vector V from the image processing subnetwork before the output probability distribution. s And the second feature vector V of the feature processing subnetwork before the output probability distribution. f The first feature vector V s With the second eigenvector V f The data is fused, and a dynamic weight vector Vconcat is generated based on the fusion result. The probability distributions of the first load category and the second load category are weighted and summed using the dynamic weight vector to obtain the final probability distribution.

[0083] That is, Vconcat is input into the neural network, which consists of two fully connected layers and ultimately outputs a two-dimensional weight vector: α = [α s ,α f ],α s +α f =1;

[0084] This mapping function can dynamically allocate the most suitable fusion weights based on the feature confidence level exhibited by the current sample in the three-stream model. For example, for a load with a very unique steady-state waveform, the meta-network will assign α... s Assign higher weights; the final probability distribution is obtained by weighting the three stream outputs according to their adaptive weights: P fina =α s *P s +α f *P f ;

[0085] The final identification result is the load category with the highest probability value.

[0086] In summary, in this embodiment, by introducing the power sequences of adjacent known photovoltaic users as a reference benchmark and utilizing event time alignment and correlation analysis, the interference of photovoltaic power generation on the total electricity consumption characteristics can be effectively eliminated, thereby solving the inaccuracy problem caused by the distortion of "net load" characteristics in traditional non-intrusive load monitoring in photovoltaic access scenarios. Based on this, by estimating photovoltaic capacity and subtracting the photovoltaic power generation portion, pure load electrical characteristics are obtained. Combined with deep learning fusion identification of waveform diagrams and multi-dimensional numerical features, the accuracy of photovoltaic identification and the overall precision of subsequent load type identification are significantly improved.

[0087] The above combination Figure 1 This paper describes the load identification method for user photovoltaic self-consumption scenarios provided in the embodiments of this application. The following describes the load identification system for implementing the user photovoltaic self-consumption scenarios provided in the embodiments of this application.

[0088] The system includes: a data acquisition module, an event detection and photovoltaic identification module, a capacity estimation and photovoltaic stripping module, a feature extraction module, and a load identification module, as detailed below.

[0089] The data acquisition module is used to acquire electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period. The electrical parameter data includes current RMS value sequence, power sequence and harmonic sequence.

[0090] The event detection and photovoltaic identification module is used to detect load switching events of the user to be identified based on electrical parameter data, and to determine whether photovoltaic access exists;

[0091] The capacity estimation and photovoltaic stripping module is used to calculate the photovoltaic capacity of the user to be identified based on electrical parameter data and the photovoltaic capacity of adjacent known photovoltaic users in response to the determination of photovoltaic access. Based on the photovoltaic capacity of the user to be identified, the power sequence and harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence.

[0092] The feature extraction module is used to determine the steady-state characteristics corresponding to each load switching event based on the pure load power sequence, pure load harmonic sequence, and current RMS value sequence. The steady-state characteristics include waveforms and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

[0093] The load identification module is used to input the waveforms and multi-dimensional electrical features corresponding to each load switching event into a pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

[0094] Furthermore, the specific implementation of the above system is basically similar to the method implementation, so the description is relatively simple. For relevant details, please refer to the description of the method implementation. Moreover, it should be noted that in the various modules of the system of this application, the components are logically divided according to the functions they are to perform. However, this application is not limited to this and can re-divide or combine the components as needed.

[0095] In another aspect, the present invention provides an electronic device for implementing the load identification method for a user photovoltaic self-consumption scenario described above. This electronic device is not limited to a terminal device or server in a system. The electronic device includes, but is not limited to, a memory and a processor. The memory stores a computer program, and the processor is configured to execute the steps in any of the above method embodiments via the computer program.

[0096] In another aspect, the present invention provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional embodiments of the load identification method for user photovoltaic self-consumption scenarios described above. The computer program is configured to execute the steps in any of the above method embodiments during runtime.

[0097] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and still achieve the desired result. Furthermore, the specific order or sequential order shown in the drawings is not necessarily required to achieve the desired result; in some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0098] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A load identification method for a user-generated photovoltaic self-consumption scenario, characterized in that, include: S1. Obtain electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period, wherein the electrical parameter data includes current RMS value sequence, power sequence and harmonic sequence; S2. Based on the electrical parameter data, detect the load switching events of the user to be identified and determine whether there is photovoltaic access. S3. In response to the determination that photovoltaic access exists, the photovoltaic capacity of the user to be identified is calculated based on the electrical parameter data and the photovoltaic capacity of the adjacent known photovoltaic users. Based on the photovoltaic capacity of the user to be identified, the power sequence and the harmonic sequence are processed to separate the pure load power sequence and the pure load harmonic sequence. S4. Based on the pure load power sequence, the pure load harmonic sequence, and the current RMS value sequence, determine the steady-state characteristics corresponding to each load switching event, wherein the steady-state characteristics include waveform diagrams and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations. S5. Input the waveforms and multi-dimensional electrical features corresponding to each load switching event into a pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

2. The load identification method for user photovoltaic self-consumption scenarios according to claim 1, characterized in that, In step S2, based on the electrical parameter data, the load switching events of the user to be identified are detected, and it is determined whether photovoltaic grid connection exists, including: S21. Based on the current RMS value sequence and harmonic sequence of the user to be identified, detect at least one load switching event and the corresponding event time. S22. For each load switching event, based on the corresponding event time, the power sequences of the user to be identified and the adjacent known photovoltaic users are trimmed, and the trimmed power sequences are time-aligned and spliced ​​together. S23. Normalize the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic users, and perform correlation analysis on the two normalized power sequences. Determine whether the user to be identified has photovoltaic access based on the analysis results.

3. The load identification method for user photovoltaic self-consumption scenarios according to claim 2, characterized in that, Step S3, which calculates the photovoltaic capacity of the user to be identified based on the electrical parameter data and the photovoltaic capacity of the adjacent known photovoltaic users, includes: S31. Based on the spliced ​​power sequence of the user to be identified and the spliced ​​power sequence of the adjacent known photovoltaic user, determine the peak power ratio between the two power sequences; S32. Based on the photovoltaic capacity of the adjacent known photovoltaic users and the peak power ratio, determine the photovoltaic capacity of the user to be identified.

4. The load identification method for user photovoltaic self-consumption scenarios according to claim 2, characterized in that, In step S21, based on the current RMS value sequence and harmonic sequence of the user to be identified, at least one load switching event and its corresponding event time are detected, including: Based on the current effective value sequence of the user to be identified, calculate the difference between the current effective value sequence of the user to be identified at adjacent sampling times. If the absolute value of the difference is greater than a first preset threshold, it is determined that a current change event has occurred. Based on the harmonic sequence of the user to be identified, the difference between the nth harmonic sequence at adjacent sampling times is calculated. If the absolute value of the difference is greater than the second preset threshold, a harmonic transition event is determined to have occurred. The harmonic sequence of the user to be identified includes the Nth harmonic sequence, where N is a positive integer greater than or equal to 1, and n is a positive integer between 1 and N. The moment when both the current surge event and the harmonic transition event are simultaneously satisfied is marked as the event moment of the load switching event.

5. The load identification method for user photovoltaic self-consumption scenarios according to claim 2, characterized in that, Step S23 involves performing correlation analysis on the two normalized power sequences and determining whether the user to be identified has access to photovoltaic power based on the analysis results. The correlation coefficient between the two power sequences can be calculated using either the Pearson correlation coefficient or the cosine similarity. If the correlation coefficient is greater than or equal to a preset correlation threshold, it is determined that the user to be identified has photovoltaic access.

6. The load identification method for user photovoltaic self-consumption scenarios according to claim 1, characterized in that, In step S4, based on the pure load power sequence, the pure load harmonic sequence, and the current RMS value sequence, the steady-state characteristics corresponding to each load switching event are determined, including: For each load switching event, based on the corresponding event time and the current RMS value sequence, the steady-state time before the event and the steady-state time after the event are determined. The power values ​​corresponding to the steady-state time before the event and the steady-state time after the event are extracted from the pure load power sequence, and the steady-state power characteristic data are obtained by subtraction. The power values ​​corresponding to the steady-state time before the event and the steady-state time after the event are extracted from the pure load harmonic sequence, and the steady-state harmonic characteristic data are obtained by subtraction. Based on the steady-state power characteristic data and the steady-state harmonic characteristic data, the steady-state characteristics corresponding to the load switching event are determined, wherein the steady-state characteristics include waveform diagrams and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations.

7. The load identification method for user photovoltaic self-consumption scenarios according to claim 1 or 6, characterized in that, The pre-trained deep learning fusion network includes an image processing sub-network, a feature processing sub-network, and an adaptive fusion module; The image processing sub-network is used to perform convolution processing on the input waveform representing the degree of fluctuation of steady-state features, and output the first load category probability distribution. The feature processing subnetwork is used to perform gated attention processing on the input multidimensional electrical features and output a second load category probability distribution. The adaptive fusion module is used to extract the first feature vector of the image processing subnetwork before the output probability distribution and the second feature vector of the feature processing subnetwork before the output probability distribution, fuse the first feature vector and the second feature vector, generate a dynamic weight vector based on the fusion result, and use the dynamic weight vector to perform a weighted summation of the first load category probability distribution and the second load category probability distribution to obtain the final probability distribution.

8. The load identification method for user photovoltaic self-consumption scenarios according to claim 7, characterized in that, The feature processing subnetwork includes a gating unit, a self-attention layer, and a multilayer perceptron classifier; The gating unit is used to process the input multidimensional electrical features, generate a corresponding multidimensional gating vector, and multiply the multidimensional gating vector with the multidimensional electrical features element by element to obtain the gating selected feature vector. The self-attention layer is used to calculate the interrelationship between features based on the gated feature vector and generate an attention weight vector, and then multiply the attention weight vector element-wise with the gated feature vector to obtain a weighted feature vector. The multilayer perceptron classifier is used to classify the weighted feature vector and output the second probability distribution.

9. A load identification system for a user-generated photovoltaic self-consumption scenario, characterized in that, The system includes: The data acquisition module is used to acquire electrical parameter data of the user to be identified and adjacent known photovoltaic users within a set time period, wherein the electrical parameter data includes a current RMS value sequence, a power sequence, and a harmonic sequence; The event detection and photovoltaic identification module is used to detect the load switching events of the user to be identified based on the electrical parameter data, and to determine whether there is photovoltaic access. A capacity estimation and photovoltaic stripping module is used to, in response to the determination that photovoltaic access exists, calculate the photovoltaic capacity of the user to be identified based on the electrical parameter data and the photovoltaic capacity of the adjacent known photovoltaic users, and process the power sequence and the harmonic sequence based on the photovoltaic capacity of the user to be identified to separate the pure load power sequence and the pure load harmonic sequence. The feature extraction module is used to determine the steady-state characteristics corresponding to each load switching event based on the pure load power sequence, the pure load harmonic sequence, and the current RMS value sequence. The steady-state characteristics include waveforms and multi-dimensional electrical characteristics that characterize the degree of steady-state characteristic fluctuations. The load identification module is used to input the waveforms and multi-dimensional electrical features corresponding to each load switching event into a pre-trained deep learning fusion network, and output the load identification results corresponding to each load switching event.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the load identification method for user photovoltaic self-consumption scenarios as described in any one of claims 1-8.

Images