Distributed photovoltaic decomposition method and system based on intelligent electric meter data

Abstract

The invention discloses a distributed photovoltaic decomposition method and system based on intelligent electric meter data, and relates to the technical field of non-intrusive power load monitoring, and the method comprises the steps: constructing a data set, and carrying out the normalization processing of the data set, and obtaining the input data of model training, verification and testing; constructing a teacher neural network structure and a student neural network structure, and performing supervised training and verification; setting hyper-parameters of a knowledge distillation algorithm, calculating difference loss, and retraining to obtain a student photovoltaic power decomposition model; and inputting the total power signal data of the target load equipment into the student photovoltaic power decomposition model to obtain photovoltaic power generation equipment power data, and performing evaluation. According to the invention, high-precision analysis of the photovoltaic power generation power decomposed from the total power signal is realized. The teacher model can learn a complex nonlinear mapping relationship, and the student model learns the decomposition capability of the teacher model through a knowledge distillation process, so that the model complexity is greatly reduced while the higher precision is maintained.

Description

Technical Field

[0001] This invention relates to the field of non-intrusive power load monitoring technology, and in particular to a distributed photovoltaic decomposition method and system based on smart meter data. Background Technology

[0002] With the continuous development of smart grids and the ongoing promotion of intelligent sensing technology, power big data analysis has gradually become a research hotspot for experts and scholars in the power industry. Non-intrusive Load Monitoring (NILM) technology refers to installing only a sensor at the user's inlet to collect and analyze total inlet data to monitor the power consumption and operating status of each or every type of appliance in the household, thereby understanding the power consumption and usage patterns of each appliance. NILM technology is of great significance for optimizing grid planning, operation, and management, and for saving users energy and electricity costs. It is also one of the key technologies for promoting refined demand-side management and energy conservation and emission reduction.

[0003] Since its inception, Non-Intrusive Load Balancing (NILM) technology has received widespread attention, and various methods have been applied to it. Deep learning, a popular approach in recent years, has become a crucial tool for implementing NILM. Convolutional Neural Networks (CNNs) can effectively mine the internal features of various source and load devices. The Seq2point model, based on CNNs, transforms the NILM problem into a sequence-to-point model training process, obtaining the midpoint power value of a single source-load device window based on a continuously sliding total power window sequence, demonstrating good performance in load balancing. However, this model requires fine-tuning of over a million parameters and a large amount of labeled data for training, thus training and validation can only be performed in the cloud. This significant data transmission between the cloud and the edge increases communication costs and poses risks to user privacy and security. Therefore, to address these issues, there is an urgent need to establish a suitable lightweight neural network model that allows training, validation, and testing solely on edge devices. Summary of the Invention

[0004] In view of the problems existing in the decomposition of distributed photovoltaic power, this invention is proposed.

[0005] Therefore, the problem this invention aims to solve is training and validation in the cloud. However, the large amount of data transmission between the cloud and the edge increases communication costs.

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

[0007] In a first aspect, embodiments of the present invention provide a distributed photovoltaic decomposition method based on smart meter data, comprising,

[0008] Construct a dataset and normalize it to obtain input data for model training, validation, and testing;

[0009] Construct teacher and student neural network structures, and perform supervised training and validation;

[0010] Set the hyperparameters of the knowledge distillation algorithm, calculate the difference loss, and retrain to obtain the student photovoltaic power decomposition model;

[0011] Input the total power signal data of the target load equipment into the student photovoltaic power decomposition model to obtain the photovoltaic power generation equipment power data and evaluate it.

[0012] As a preferred embodiment of the distributed photovoltaic decomposition method based on smart meter data described in this invention, the supervised training and verification includes a column-to-point convolutional neural network model, as shown in the following equation:

[0013]

[0014] in, ε is W-dimensional Gaussian random noise.

[0015] As a preferred embodiment of the distributed photovoltaic decomposition method based on smart meter data described in this invention, the supervised training and verification further includes a loss function, as shown in the following equation:

[0016]

[0017] Where, θ p These are network parameters, where T is the total length of the input total power sequence, and Y is the total power sequence value. t:t+W-1 When inputting into the teacher and student neural network models, zero values ​​of length W-12 are added to both ends of the total power sequence;

[0018] During the supervised training of the student neural network model, the loss function is readjusted as follows:

[0019]

[0020] Here, KL is the divergence divergence between the feature outputs of the teacher model and the feature outputs of the student model.

[0021] As a preferred embodiment of the distributed photovoltaic decomposition method based on smart meter data described in this invention, the teacher neural network structure is shown in the following formula:

[0022]

[0023] Among them, P PV(t) represents the photovoltaic power generation at time t, and Z is the normalization factor used to ensure the correct relative magnitude of the power value. This represents an integral term related to time decay, used to simulate the characteristics of photovoltaic power changing over time, where S(τ) is the total power signal at time τ, α is the decay coefficient, and β is a scaling factor used to adjust the contribution of the summation term. i φ is the weight of the i-th feature; φ is a normalization function used to process the input feature x. i , where μ i and σ i These are features x i The mean and standard deviation are given, where N is the total number of features.

[0024] As a preferred embodiment of the distributed photovoltaic power decomposition method based on smart meter data described in this invention, the student photovoltaic power decomposition model is shown in the following equation:

[0025]

[0026] in, Let γ represent the photovoltaic power generation at time t in the student load decomposition model, γ be the adjustment factor, and G(i) be the information filtering function.

[0027] As a preferred embodiment of the distributed photovoltaic decomposition method based on smart meter data described in this invention, the evaluation includes the mean absolute error and the total signal error, calculated as follows:

[0028]

[0029]

[0030] MAE is used to calculate power prediction values. Compared with the actual measured value x t The average of the absolute errors between the two values, where T represents the total duration of the test period, and SAE is used to calculate the predicted power consumption during the test period. The ratio of the error between the true value r and the actual value r.

[0031] As a preferred embodiment of the distributed photovoltaic decomposition method based on smart meter data described in this invention, the evaluation further includes:

[0032] When MAE≤X and TSE≤Y, the load decomposition result is considered to meet the requirements, and the next step of data analysis or application stage is carried out. Otherwise, when MAE>X, it is judged that the average error is too large, and the algorithm parameters are optimized or the data quality is checked. When TSE>Y, it is judged that the total error exceeds the standard, and signal reconstruction or filtering processing is carried out.

[0033] When the requirements are not met, i.e., MAE>=X or TSE>=Y, if the model complexity is high, reduce the model complexity by reducing the number of layers or channels and retrain. If the distribution of training data differs greatly from that of test data, supplement the training data to cover more distributions and retrain. If the total power data quality is poor, perform signal preprocessing to reduce the impact of noise before inputting it into the model for retraining. If the decomposition algorithm has defects, try to improve the algorithm and adjust the loss function, optimizer, and training strategy.

[0034] Secondly, embodiments of the present invention provide a distributed photovoltaic decomposition system based on smart meter data, comprising:

[0035] The data construction module is used to build datasets and normalize them to obtain input data for model training, validation, and testing.

[0036] The network construction module is used to build teacher and student neural network structures, and to perform supervised training and validation.

[0037] The knowledge distillation module is used to set the hyperparameters of the knowledge distillation algorithm, calculate the difference loss, and retrain to obtain the student photovoltaic power decomposition model.

[0038] The evaluation module is used to input the total power signal data of the target load equipment into the student photovoltaic power decomposition model, obtain the power data of the photovoltaic power generation equipment, and evaluate whether the decomposition results meet the requirements.

[0039] The feedback optimization module improves and optimizes the model and data by optimizing model complexity, supplementing training data, data preprocessing, and adjusting algorithm hyperparameters when the decomposition results do not meet the requirements.

[0040] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described distributed photovoltaic decomposition method based on smart meter data.

[0041] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described distributed photovoltaic decomposition method based on smart meter data.

[0042] The beneficial effects of this invention are as follows: a teacher-student network model based on knowledge distillation achieves high-precision analysis of photovoltaic power generation from total power signals. The teacher model can learn complex nonlinear mapping relationships, while the student model learns the decomposition ability of the teacher model through the knowledge distillation process, significantly reducing model complexity while maintaining high accuracy. A hierarchical training, validation, and testing strategy is adopted, making full use of data. The model parameters are trained using the training set, hyperparameters are adjusted using the validation set, and the final performance is evaluated using the test set, ensuring the effectiveness and robustness of the method. By setting reasonable thresholds for the mean absolute error (MAE) and total signal error (TSE), a strict evaluation standard for the decomposition results is established, which can quantitatively determine whether the decomposition results meet the needs of practical applications. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0044] Figure 1 This is a flowchart of a distributed photovoltaic decomposition method based on smart meter data.

[0045] Figure 2 This paper presents a teacher-based denoising autoencoder neural network structure for a distributed photovoltaic decomposition method based on smart meter data.

[0046] Figure 3 This paper presents a lightweight student sequence-to-point convolutional neural network structure for a distributed photovoltaic decomposition method based on smart meter data. Detailed Implementation

[0047] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0051] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0052] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0053] Example 1

[0054] Reference Figures 1-3 This is the first embodiment of the present invention, which provides a distributed photovoltaic decomposition method based on smart meter data, including:

[0055] S1: Construct a dataset and normalize it to obtain input data for model training, validation, and testing.

[0056] A typical power curve for photovoltaic power generation is constructed based on photovoltaic power generation data from publicly available datasets, and then randomly superimposed on the original total power data to construct a simulation dataset.

[0057] The dataset required for this method is constructed based on the total number of residential users and the typical working curve of photovoltaic power generation. The generated dataset is divided into training set, validation set and test set in a ratio of 3:1:1. The total number of users and photovoltaic power generation equipment data are normalized to obtain the input data for model training, validation and testing.

[0058] The dataset includes total power data with labels. Training, validation, and test data are obtained from the dataset. The sequence data used are normalized according to the following formula: power minus its mean and then divided by its standard deviation.

[0059]

[0060] Where, x t The power sequence refers to the actual measured values ​​of the power series, which includes the total power series and the load power series, and whose values ​​are obtained through equipment measurements. Let σ be the average value of the power series and σ be the standard deviation of the power series.

[0061] S2: Construct the teacher's neural network structure and the student's neural network structure, and perform supervised training and verification.

[0062] The teacher neural network structure of this method is constructed based on a denoising autoencoder. The teacher neural network structure is then trained and validated in a supervised manner using the processed dataset. The resulting network model is the teacher load decomposition model used to decompose photovoltaic power generation from the total power signal, as shown in the following equation:

[0063]

[0064] Among them, P PV (t) represents the photovoltaic power generation at time t, and Z is the normalization factor used to ensure the correct relative magnitude of the power value. This represents an integral term related to time decay, used to simulate the characteristics of photovoltaic power changing over time, where S(τ) is the total power signal at time τ, α is the decay coefficient, and β is a scaling factor used to adjust the contribution of the summation term. i φ is the weight of the i-th feature; φ is a normalization function used to process the input feature x. i , where μ i and σ i These are features x i The mean and standard deviation are given, where N is the total number of features.

[0065] The student neural network structure of this method is constructed, and supervised training and validation are performed on the processed dataset. The resulting network model is the student load decomposition model used to decompose photovoltaic power generation from the total power signal. The specific formula is shown below:

[0066]

[0067] in, Let γ represent the photovoltaic power generation at time t in the student load decomposition model, γ be the adjustment factor, and G(i) be the information filtering function.

[0068] The input to the teacher and student neural networks is represented as a sliding window Y of the total power sequence. t:t+W-1 Let t be time, W be the length of the sliding window for the total power sequence (W is an odd number), and the outputs of the two neural networks be the midpoint value of the power sequence window for the m-th photovoltaic device. The sequence of supervised training and validation processes is transformed into a point convolutional neural network model as shown in the following equation:

[0069]

[0070] in, ε is W-dimensional Gaussian random noise;

[0071] The loss function in the supervised training process is specifically expressed as:

[0072]

[0073] Where, θ p These are network parameters, where T is the total length of the input total power sequence, and Y is the total power sequence value. t:t+W-1 When inputting into the teacher and student neural network models, zero values ​​of length W-12 are added to both ends of the total power sequence.

[0074] The noise reduction autoencoder network structure consists of an encoder structure and a decoder structure. First, the encoder network structure is obtained based on a two-layer convolutional neural network structure and a two-layer downsampling structure. Then, the latent features output by the encoder are subjected to two layers of upsampling and two layers of convolution to obtain the decoder network structure. Finally, the encoder and decoder structures are combined to obtain the noise reduction autoencoder network structure required by this method.

[0075] S3: Set the hyperparameters of the knowledge distillation algorithm, calculate the difference loss, and retrain to obtain the student photovoltaic power decomposition model.

[0076] During supervised training and validation, network parameters are updated in a supervised manner. The input of the lightweight sequence-to-point convolutional neural network is the total power signal data window sequence of the target user load node, and the output of the network is the power value corresponding to the midpoint of the window sequence. The total power signal data window sequence is input into the teacher and student power generation decomposition models obtained after training. The output of the two load decomposition models is the midpoint value of the photovoltaic power generation equipment power window. As the input window of the network model slides continuously, the midpoint values ​​of the photovoltaic power generation equipment power window are combined in time sequence to form the power decomposition sequence of the photovoltaic power generation equipment.

[0077] The difference loss function between the teacher network model and the student network model is calculated from the feature outputs of the teacher network and the student network. During the supervised training of the student neural network model, the loss function is readjusted as follows:

[0078]

[0079] Here, KL is the divergence divergence between the feature outputs of the teacher model and the feature outputs of the student model.

[0080] S4: Input the total power signal data of the target load equipment into the student photovoltaic power decomposition model to obtain the photovoltaic power generation equipment power data and evaluate it.

[0081] The determination of the Mean Absolute Error (MAE) threshold X is based on the analysis of historical data and expert experience, combined with the application scenarios of load decomposition (such as power grid operation, industrial process control, etc.), comprehensively considering the impact of error on system performance, and setting a practical yet challenging X value. This value aims to balance the practicality of the algorithm with its accuracy requirements, ensuring that the algorithm is neither too lenient to meet practical needs, nor too strict to be difficult to implement.

[0082] Similarly, determining the Total Signal Error (TSE) threshold Y requires considering both the fault tolerance range and the overall system stability requirements in practical applications. The total signal error reflects the accumulation of all errors during load decomposition and is crucial for ensuring long-term stable system operation. The selection of the Y value must balance technical feasibility and the specific requirements of the application scenario.

[0083] When MAE≤X and TSE≤Y, the load decomposition result is considered to meet the requirements, and the next step of data analysis or application stage can be carried out. Otherwise, when MAE>X, it is judged that the average error is too large, and the algorithm parameters are optimized or the data quality is checked. When TSE>Y, it is judged that the total error exceeds the standard, and signal reconstruction or filtering processing is performed.

[0084] When the requirements are not met, if MAE>=X or TSE>=Y, and the model complexity is high, reduce the model complexity by reducing the number of layers or channels and retrain. If the training data distribution differs significantly from the test data distribution, supplement the training data to cover more distributions and retrain. If the total power data quality is poor, perform signal preprocessing, such as filtering and interpolation, to reduce the impact of noise before inputting it into the model for retraining. If the decomposition algorithm has defects, try to improve the algorithm, adjusting the loss function, optimizer, and training strategy.

[0085] To address excessive MAE, if the data quality is low, perform data cleaning or re-acquisition. If the algorithm parameters are not ideal, optimize the parameter selection using machine learning methods. If excessive TSE is due to noise, employ more advanced filtering techniques. If the signal reconstruction is inaccurate, try different signal processing methods.

[0086] For all the results after the operation, recalculate MAE and TSE. If the conditions are met (MAE≤X and TSE≤Y), the processing ends and the application stage begins. If the conditions are not met, adjust the thresholds X and Y according to the actual situation or return to the second layer for more in-depth analysis and processing.

[0087] Furthermore, this embodiment also provides a distributed photovoltaic decomposition system based on smart meter data, including:

[0088] This embodiment also provides a computer device applicable to the distributed photovoltaic decomposition method based on smart meter data, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the distributed photovoltaic decomposition method based on smart meter data as proposed in the above embodiment.

[0089] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0090] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the distributed photovoltaic decomposition method based on smart meter data as proposed in the above embodiments.

[0091] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0092] Example 2

[0093] Reference Figure 2 and Figure 3 This is the second embodiment of the present invention, which provides a distributed photovoltaic decomposition method based on smart meter data. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0094] Taking a practical application scenario as an example, the specific steps for achieving non-invasive lightweight load decomposition using the method of this invention are as follows:

[0095] 1) Obtain the training, validation, and test data from the dataset and normalize the data;

[0096] Taking the UK-DALE public dataset [4] as an example, the data of Building 1 in the UK-DALE dataset is selected as the total dataset. The typical working curve of photovoltaic power generation is extracted, and the total port data of Building 1 is randomly superimposed to obtain the required total port and sub-item electrical appliance datasets. The ratio of training, validation and test datasets is set to 3:1:1. In the design of this invention, as shown in the figure, Figure 2 The student lightweight sequence shown is used to perform photovoltaic power decomposition under a point convolutional neural network structure.

[0097] 2) Initialize the parameters of the teacher noise reduction autoencoder neural network, input the normalized power data into the network structure for training and obtain the network model, and calculate the model feature output.

[0098] 3) Initialize the parameters of the student lightweight sequence-to-point neural network, input the normalized power data into the network structure for training and obtain the network model, and calculate the model feature output. Obtain the required student neural network model by knowledge distillation of the teacher neural network model and retraining. The student lightweight sequence-to-point neural network model occupies 0.11 megabytes of memory and 0.266 megabytes of floating-point operations, which is 125 times and 24 times lower than the classic sequence-to-point Seq2point model, which occupies 13.84 megabytes of memory and 6.42 megabytes of floating-point operations, respectively.

[0099] 4) Test the trained model by inputting the test aggregate power sequence of Building 1 into the trained model to decompose the photovoltaic power of Building 1.

[0100] In the examples of materials used in this study, the evaluation indicators for the load decomposition results include mean absolute error (MAE) and signal aggregate error (SAE), and their expressions are as follows:

[0101]

[0102]

[0103] Here, MAE is used to calculate power prediction values. (that is, the power decomposition value obtained by the method of this invention) and the actual measured value x t The average of the absolute errors between the two values, where T represents the total duration of the test period, and SAE is used to calculate the predicted power consumption during the test period. The ratio of the error between the true value r and the actual value r.

[0104] The MAE and SAE results of photovoltaic power generation decomposition using the method of the present invention are shown in Table 1.

[0105] Table 1 shows the MAE and SAE indices from the decomposition results.

[0106]

[0107] As can be seen from the above research materials, the method of the present invention can greatly reduce the number of parameters and computational load of the neural network model, and achieve the photovoltaic power generation decomposition effect quickly at the edge user end.

[0108] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A distributed photovoltaic decomposition method based on smart meter data, characterized in that: include, Construct a dataset and normalize it to obtain input data for model training, validation, and testing; Construct teacher and student neural network structures, and perform supervised training and validation; Set the hyperparameters of the knowledge distillation algorithm, calculate the difference loss, and retrain to obtain the student photovoltaic power decomposition model; Input the total power signal data of the target load equipment into the student photovoltaic power decomposition model to obtain the photovoltaic power generation equipment power data and evaluate it.

2. The distributed photovoltaic decomposition method based on smart meter data as described in claim 1, characterized in that: The supervised training and validation includes a column-to-point convolutional neural network model, as shown in the following equation: in, ε is W-dimensional Gaussian random noise.

3. The distributed photovoltaic decomposition method based on smart meter data as described in claim 2, characterized in that: The supervised training and validation also include a loss function, as shown in the following equation: Where, θ p These are network parameters, where T is the total length of the input total power sequence, and Y is the total power sequence value. t:t+W-1 When inputting into the teacher and student neural network models, zero values ​​of length [(W-1) / 2] are added to both ends of the total power sequence; During the supervised training of the student neural network model, the loss function is readjusted as follows: Here, KL is the divergence divergence between the feature outputs of the teacher model and the feature outputs of the student model.

4. The distributed photovoltaic decomposition method based on smart meter data as described in claim 3, characterized in that: The structure of the teacher's neural network is shown in the following formula: Among them, P PV (t) represents the photovoltaic power generation at time t, and Z is the normalization factor used to ensure the correct relative magnitude of the power value. This represents an integral term related to time decay, used to simulate the characteristics of photovoltaic power changing over time, where S(τ) is the total power signal at time τ, α is the decay coefficient, and β is a scaling factor used to adjust the contribution of the summation term. i φ is the weight of the i-th feature; φ is a normalization function used to process the input feature x. i , where μ i and σ i These are features x i The mean and standard deviation are given, where N is the total number of features.

5. The distributed photovoltaic decomposition method based on smart meter data as described in claim 4, characterized in that: The student photovoltaic power decomposition model is shown in the following equation: in, Let γ represent the photovoltaic power generation at time t in the student load decomposition model, γ be the adjustment factor, and G(i) be the information filtering function.

6. The distributed photovoltaic decomposition method based on smart meter data as described in claim 5, characterized in that: The evaluation includes the mean absolute error and the total signal error, calculated as follows: MAE is used to calculate power prediction values. Compared with the actual measured value x t The average of the absolute errors between the two values, where T represents the total duration of the test period, and SAE is used to calculate the predicted power consumption during the test period. The ratio of the error between the true value r and the actual value r.

7. The distributed photovoltaic decomposition method based on smart meter data as described in claim 6, characterized in that: The evaluation also includes, When MAE≤X and TSE≤Y, the load decomposition result is considered to meet the requirements, and the next step of data analysis or application stage is carried out. Otherwise, when MAE>X, it is judged that the average error is too large, and the algorithm parameters are optimized or the data quality is checked. When TSE>Y, it is judged that the total error exceeds the standard, and signal reconstruction or filtering processing is carried out. When the requirements are not met, i.e., MAE>=X or TSE>=Y, if the model complexity is high, reduce the model complexity by reducing the number of layers or channels and retrain. If the distribution of training data differs greatly from that of test data, supplement the training data to cover more distributions and retrain. If the total power data quality is poor, perform signal preprocessing to reduce the impact of noise before inputting it into the model for retraining. If the decomposition algorithm has defects, try to improve the algorithm and adjust the loss function, optimizer, and training strategy.

8. A distributed photovoltaic decomposition system based on smart meter data, based on the distributed photovoltaic decomposition method based on smart meter data according to any one of claims 1 to 7, characterized in that: include, The data construction module is used to build datasets and normalize them to obtain input data for model training, validation, and testing. The network construction module is used to build teacher and student neural network structures, and to perform supervised training and validation. The knowledge distillation module is used to set the hyperparameters of the knowledge distillation algorithm, calculate the difference loss, and retrain to obtain the student photovoltaic power decomposition model. The evaluation module is used to input the total power signal data of the target load equipment into the student photovoltaic power decomposition model, obtain the power data of the photovoltaic power generation equipment, and evaluate whether the decomposition results meet the requirements. The feedback optimization module improves and optimizes the model and data by optimizing model complexity, supplementing training data, data preprocessing, and adjusting algorithm hyperparameters when the decomposition results do not meet the requirements.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the distributed photovoltaic decomposition method based on smart meter data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the distributed photovoltaic decomposition method based on smart meter data as described in any one of claims 1 to 7.

Images