Training method and device of load composition identification model

Abstract

A training method and device of a load composition identification model, the method comprising: obtaining a comprehensive load model; performing sensitivity analysis on load composition of the comprehensive load model by using a Sobol index method to obtain a sensitivity coefficient of the load composition; obtaining a load composition identification error based on the sensitivity coefficient of the load composition; constructing a loss function based on the load composition identification error and a curve fitting error; and importing a sample data set into the load composition identification model for training until the loss function converges, thereby obtaining a trained load composition identification model. The method and device provided in the embodiments of the present application effectively enhance the stability and robustness of the identification result and have strong adaptability to changes in operating conditions, voltage disturbances and distributed power access.

Description

Technical Field

[0001] This invention relates to the field of large power grid safety and stability and simulation modeling technology, and more specifically, to a training method and apparatus for a load composition identification model. Background Technology

[0002] With the widespread application of new energy sources and power electronics technologies in power systems, system load characteristics are gradually exhibiting diversified, nonlinear, and dynamic features. The large-scale integration of distributed photovoltaic (PV) systems, energy storage systems, and power electronic loads has rendered the assumptions of traditional load models, which primarily rely on static impedance loads and induction motors, increasingly invalid. The voltage and current response characteristics of loads are influenced by a combination of factors, including climate, user behavior, power electronic control strategies, and fluctuations in distributed power source output. This leads to increased system power fluctuations and greater uncertainty in load composition, posing new challenges to power system dispatching, security analysis, and stability control.

[0003] In power system analysis and simulation, accurately obtaining the load composition at nodes is fundamental for power flow calculations, transient stability analysis, and voltage control. However, the dynamic changes in load composition during actual operation are difficult to measure directly, and traditional load identification methods typically rely on offline measurement data or experience-based parameter estimation. On the one hand, these methods struggle to reflect the complex load characteristics involving distributed power sources and power electronic devices; on the other hand, traditional model identification processes are highly dependent on data and cannot fully integrate historical operating information with real-time monitoring data, resulting in insufficient identification accuracy and poor real-time performance.

[0004] In recent years, data-driven methods have seen rapid development in the field of power system modeling and state estimation. Models such as deep neural networks (DNNs) and convolutional neural networks (CNNs) can automatically extract features from high-dimensional operating data, approximating complex nonlinear relationships and providing new approaches for load characteristic identification. However, purely data-driven methods still have shortcomings in generalization ability and physical interpretability. Especially when data distribution is uneven or the sample size is limited, the model is prone to overfitting, affecting the reliability of the identification results. Summary of the Invention

[0005] In view of this, the present invention proposes a training method and apparatus for a load composition identification model, which aims to solve one or more of the technical problems mentioned in the background section above.

[0006] In a first aspect, embodiments of the present invention provide a training method for a load composition identification model, the method comprising: obtaining a comprehensive load model; performing sensitivity analysis on the load composition of the comprehensive load model using the Sobol exponent method to obtain a sensitivity coefficient of the load composition; obtaining a load composition identification error based on the sensitivity coefficient of the load composition; constructing a loss function based on the load composition identification error and the curve fitting error; and importing a sample dataset into the load composition identification model for training until the loss function converges to obtain a completed load composition identification model.

[0007] Furthermore, the integrated load model is pre-obtained using the following method: setting the active power ratio parameters of the load composition as photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ; in, S PV_B This indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

[0008] Furthermore, the sensitivity analysis of the load composition of the integrated load model using the Sobol index method to obtain the sensitivity coefficient of the load composition includes: obtaining the sensitivity coefficient of the load composition using the following formula: ; in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

[0009] Furthermore, the load composition identification error is obtained based on the sensitivity coefficient of the load composition, including: obtaining the load composition identification error using the following formula: ; ; in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

[0010] Furthermore, the aforementioned λ k,t The expression is as follows: ; in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t - The error ratio between the actual load composition value and the value after 2 iterations.

[0011] Furthermore, the curve fitting error is obtained using the following formula: ; in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

[0012] Furthermore, the loss function constructed based on the load composition identification error and curve fitting error includes: ; in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

[0013] Secondly, embodiments of the present invention also provide a training apparatus for a load composition identification model. The apparatus includes: an acquisition unit for acquiring a comprehensive load model; a sensitivity analysis unit for performing sensitivity analysis on the load composition of the comprehensive load model using the Sobol exponent method to obtain a sensitivity coefficient for the load composition; a load composition identification error unit for obtaining a load composition identification error based on the sensitivity coefficients of the load composition; a loss function unit for constructing a loss function based on the load composition identification error and the curve fitting error; and a training unit for importing a sample dataset into the load composition identification model for training until the loss function converges, thereby obtaining a completed load composition identification model.

[0014] Furthermore, the integrated load model is pre-obtained using the following method: setting the active power ratio parameters of the load composition as photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ; in, S PV_B This indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

[0015] Furthermore, the sensitivity analysis unit is also used to: obtain the sensitivity coefficient of the load composition using the following formula: ; in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

[0016] Furthermore, the load composition identification error unit is also used to: obtain the load composition identification error using the following formula: ; ; in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

[0017] Furthermore, the aforementioned λ k,t The expression is as follows: ; in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t - The error ratio between the actual load composition value and the value after 2 iterations.

[0018] Furthermore, the curve fitting error is obtained using the following formula: ; in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

[0019] Furthermore, the loss function unit is also used for: ; in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

[0020] Thirdly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods provided in the above embodiments.

[0021] Fourthly, embodiments of the present invention also provide an electronic device, including: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the methods provided in the above embodiments.

[0022] The training method and apparatus for the load composition identification model provided in this invention employs the Sobol exponential method to perform sensitivity analysis on the load composition of the comprehensive load model, obtaining the sensitivity coefficient of the load composition. Based on the sensitivity coefficient, the load composition identification error is obtained. Based on the load composition identification error and the curve fitting error, a loss function is constructed. The sample dataset is then imported into the load composition identification model for training until the loss function converges, resulting in a fully trained load composition identification model. This overcomes the problems of insufficient identification accuracy, weak dynamic response capability, and poor model generalization caused by relying solely on offline parameter estimation or single data fitting in existing technologies. It fully integrates the advantages of both mechanism modeling and data-driven methods, effectively enhancing the stability and robustness of the identification results, and exhibiting strong adaptability to changes in operating conditions, voltage disturbances, and distributed power supply access. Attached Figure Description

[0023] Figure 1 An exemplary flowchart of a training method for a load composition identification model according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of the structure of the integrated load model according to an embodiment of the present invention is shown; Figure 3An exemplary flowchart illustrating the process of importing a sample dataset into a load composition identification model for training, according to an embodiment of the present invention, is shown. Figure 4 A schematic diagram of the EPRI-36 node system is shown. Figure 5 A data graph of a load-based sensitivity index curve according to an embodiment of the present invention is shown; Figure 6 A data graph illustrating the dynamic load composition identification effect according to an embodiment of the present invention is shown; Figure 7 A data graph showing the fitting degree of the active power response trajectory according to an embodiment of the present invention is shown; Figure 8 A schematic diagram of the structure of a training device for a load composition identification model according to an embodiment of the present invention is shown. Detailed Implementation

[0024] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0025] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0026] Figure 1 An exemplary flowchart of a training method for a load composition identification model according to an embodiment of the present invention is shown.

[0027] like Figure 1 As shown, the method includes: Step S101: Obtain the integrated load model.

[0028] Furthermore, the comprehensive load model is obtained in advance using the following method: The active power ratio parameters are set as follows: load composition includes photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ; in, S PV_BThis indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

[0029] Figure 2 A schematic diagram of the integrated load model according to an embodiment of the present invention is shown. Figure 2 As shown, the integrated load model comprises three parts: feeders, substations, and loads. The equivalent load part includes a monopole photovoltaic model, a battery energy storage model, a three-phase induction motor, and static loads. The load composition is defined by the active power ratio parameters of these three types of loads: photovoltaic model loads, dynamic loads, and energy storage model loads. Then, the relationship between it and the active power injection amount at the model port is constructed.

[0030] Step S102: Use the Sobol index method to perform sensitivity analysis on the load composition of the integrated load model to obtain the sensitivity coefficient of the load composition.

[0031] Further, step S102 includes: The sensitivity coefficient of the load configuration is obtained using the following formula: ; in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

[0032] Specifically, the Sobol index method is used to perform sensitivity analysis on the load model composition, and the load composition is quantified by calculating the total sensitivity index. The Sobol index method quantitatively assesses the sensitivity of each parameter to the output result by analyzing the contribution of the input parameter set to the model output variance.

[0033] Assuming the load model uses a function O = f ( θ ) indicates that, in the formula θ =( θ 1, θ 2,L, θ d ) represents the model parameters. θ i ( i =1,2,L, d The distribution follows a uniform distribution on [0,1]. .like f 2 ( θ If it is integrable, then... f ( θ The following decomposition is performed: The influence of all input parameters and individual parameters on the model output of the load model is represented by the total output variance and partial variance, respectively: Furthermore, considering the interactions among multiple parameters, their degree of influence on the model output is defined as follows: Therefore, parameters θ i The sensitivity index to the model's dynamic response can be expressed as: In the formula, V -i Indicates except θ i The partial variances corresponding to the other parameters besides these. S T,i Indicates parameters θ i When the value changes, the overall impact of this parameter and its interaction with other parameters on the model's dynamic response is measured.

[0034] Composition of various load types in the comprehensive load model Corresponding to the parameter set in the previous formula θ The sensitivity coefficient is obtained by performing a softmax transformation on the total sensitivity index composed of various loads. β k As shown in the following formula: This coefficient characterizes the magnitude of the impact of fluctuations in various load compositions on the error of the model's output response curve. Load Composition Sensitivity coefficient β k The larger the value, the greater the fitting error between the model response curve and the actual response curve caused by fluctuations in its value. At the same time, β k Introduced as a weighting term into the subsequent loss function design, this enhances the contribution of highly sensitive parameters to the loss value, thereby guiding the model to fit its corresponding features more accurately.

[0035] The above embodiments utilize the Sobol index method to perform sensitivity analysis on model parameters, screen key factors affecting load characteristics, improve the interpretability and screening efficiency of key characteristic parameters, and ensure the rationality of the model structure.

[0036] Step S103: Based on the sensitivity coefficient of the load composition, obtain the load composition identification error.

[0037] Further, step S103 includes: The load configuration identification error is obtained using the following formula: ; ; in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

[0038] Specifically, The L1 norm is defined by the above formula. The error will cause the parameters that have a greater impact on the response curve to converge to more accurate prediction results, and accelerate the learning process of parameters with slow convergence speed.

[0039] Furthermore, λ k,t The expression is as follows: ; in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t - The error ratio between the actual load composition value and the value after 2 iterations.

[0040] Specifically, the proportionality coefficient K Used to guarantee weight λ k,t The sum of them is K This means scaling the weights under the same standard. This is to characterize the load composition. θ k Convergence speed in model training is defined as... ω k,t-1 for θ k The error ratio between the actual load composition value and the actual load composition value at iterations t-1 and t-2 is expressed as follows: As can be seen from the above two equations, the load with a slower convergence speed has a larger corresponding weight value in order to speed up its convergence speed.

[0041] Step S104: Construct a loss function based on the load composition identification error and curve fitting error.

[0042] Furthermore, the curve fitting error is obtained using the following formula: ; in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

[0043] Specifically, the response curve error is introduced. L curve The accuracy of load identification is improved by minimizing the fitting error of the power response curve.

[0044] Further, step S104 includes: ; in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

[0045] The above embodiments establish a load composition identification model based on a deep neural network, and integrate the identification error and the response curve fitting error in the loss function, taking into account both identification accuracy and response consistency, so as to achieve accurate fitting of the dynamic characteristics of the load during the model training process.

[0046] Step S105: Import the sample dataset into the load composition identification model for training until the loss function converges, and obtain the completed load composition identification model.

[0047] Specifically, the load composition identification model is obtained in advance using the following method: A fused CNN-LSTM deep learning model structure, consisting of concatenated Conv1D convolutional layers and LSTM temporal layers, was built using the TensorFlow / Keras framework. Network parameters were initialized and optimizer settings were configured. Details are shown in Table 1.

[0048] Table 1

[0049] The sample dataset is imported into the identification model for training, resulting in the load composition identification model. Figure 3 An exemplary flowchart illustrating the process of importing a sample dataset into a load composition identification model for training, according to an embodiment of the present invention, is shown. Figure 3 As shown, the sample set is normalized and divided into an 80% training set and a 20% test set for model training and validation, respectively. The training set is input into the network for offline training, extracting local pattern features and long-term dependencies from the data until the loss shows a convergent trend. The trained load composition identification model is saved so that real-time cross-sectional measurement data can be input into the trained load composition identification model to output the various load compositions under the target node online. .

[0050] The above embodiment uses the Sobol exponent method to perform sensitivity analysis on the load composition of the integrated load model, obtaining the sensitivity coefficient of the load composition. Based on the sensitivity coefficient of the load composition, the load composition identification error is obtained. Based on the load composition identification error and the curve fitting error, a loss function is constructed. The sample dataset is then imported into the load composition identification model for training until the loss function converges, resulting in a fully trained load composition identification model. This overcomes the problems of insufficient identification accuracy, weak dynamic response capability, and poor model generalization caused by relying solely on offline parameter estimation or single data fitting in existing technologies. It fully integrates the advantages of both mechanism modeling and data-driven methods, effectively enhancing the stability and robustness of the identification results, and exhibiting strong adaptability to changes in operating conditions, voltage disturbances, and distributed power supply access.

[0051] Example 1 The method provided in the embodiments of the present invention was implemented and tested in the EPRI-36 node system. Figure 4 A schematic diagram of the EPRI-36 node system is shown. Figure 4 As shown, all static load node models in the system are replaced with integrated load models, with the integrated load model under node 3 serving as the parameters to be identified. The battery storage capacity, photovoltaic installed capacity, and equivalent transformer in the model are set to 100MVA, 300MVA, and 500MVA, respectively. The remaining model parameters are set to the median values ​​of the adjustment range before the key load nodes are selected.

[0052] The algorithm process will be explained in detail below with specific implementation examples.

[0053] First, small-amplitude power disturbances are randomly applied to load nodes other than the target node. Power response data is generated by repeatedly calling the executable files for power flow calculation and transient stability calculation using a Python program interface. Before each simulation, load model parameters and disturbance parameters are randomly generated within a given interval. Each simulation lasts 5 seconds, with a sampling period of 0.01 seconds, until 10,000 samples are generated.

[0054] To effectively assess the impact of various load configurations on the dynamic response trajectory, 5000 sets of configuration parameter values ​​were sampled from the dataset, and the active power response corresponding to each set of sampled values ​​was obtained through a system simulation model. The overall sensitivity index of each load configuration to the power response curve was calculated using a formula, such as... Figure 5 As shown.

[0055] In Python 3.10, a neural network fusing CNN and BiLSTM was built using the improved loss function of this invention, and then pre-generated training set data was imported for training. A sample was randomly selected from the test set and input into the model to calculate the error between the identified value and the true value of the three types of loads, as shown in Table 2.

[0056] Table 2 Load identification and actual values ​​output by the model

[0057] To further verify the effectiveness and stability of the proposed method in the load composition identification task, 50 sets of samples were randomly selected from the test set and input into the trained model to solve for dynamic load composition. The identification results of each set were then concatenated sequentially, as follows: Figure 6 As shown.

[0058] To test the effectiveness of the improved training mechanism with the improved loss function in terms of power response curve fitting accuracy, the actual and identified values ​​in Table 2 were input into PSASP to generate dynamic response curves of active power, and compared with the actual response curves. Figure 7 As shown.

[0059] Figure 8 A schematic diagram of the structure of a training device for a load composition identification model according to an embodiment of the present invention is shown.

[0060] like Figure 8 As shown, the device includes: Acquisition unit 801 is used to acquire the comprehensive load model; Sensitivity analysis unit 802 is used to perform sensitivity analysis on the load composition of the comprehensive load model using the Sobol exponent method to obtain the sensitivity coefficient of the load composition. The load configuration identification error unit 803 is used to obtain the load configuration identification error based on the sensitivity coefficient of the load configuration. Loss function unit 804 is used to construct a loss function based on the load composition identification error and curve fitting error; Training unit 805 is used to import the sample dataset into the load composition identification model for training until the loss function converges, thus obtaining the completed load composition identification model.

[0061] Furthermore, the comprehensive load model is obtained in advance using the following method: The active power ratio parameters are set as follows: load composition includes photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ; in, S PV_B This indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

[0062] Furthermore, the sensitivity analysis unit 802 is also used for: The sensitivity coefficient of the load configuration is obtained using the following formula: ; in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

[0063] Furthermore, the load-based error identification unit 803 is also used for: The load configuration identification error is obtained using the following formula: ; ; in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

[0064] Furthermore, λ k,t The expression is as follows: ; in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t- The error ratio between the actual load composition value and the value after 2 iterations.

[0065] Furthermore, the curve fitting error is obtained using the following formula: ; in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

[0066] Furthermore, the loss function unit 804 is also used for: ; in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

[0067] The above embodiment uses the Sobol exponent method to perform sensitivity analysis on the load composition of the integrated load model, obtaining the sensitivity coefficient of the load composition. Based on the sensitivity coefficient of the load composition, the load composition identification error is obtained. Based on the load composition identification error and the curve fitting error, a loss function is constructed. The sample dataset is then imported into the load composition identification model for training until the loss function converges, resulting in a fully trained load composition identification model. This overcomes the problems of insufficient identification accuracy, weak dynamic response capability, and poor model generalization caused by relying solely on offline parameter estimation or single data fitting in existing technologies. It fully integrates the advantages of both mechanism modeling and data-driven methods, effectively enhancing the stability and robustness of the identification results, and exhibiting strong adaptability to changes in operating conditions, voltage disturbances, and distributed power supply access.

[0068] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0069] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the training method for the load composition identification model provided in the above embodiments.

[0070] This invention also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the training method for the load composition identification model provided in the above embodiments.

[0071] The invention has been described with reference to a few embodiments. However, as will be known to those skilled in the art, and as defined in the appended claims, other embodiments besides those disclosed above fall equivalently within the scope of the invention.

[0072] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

[0073] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A training method for a load composition identification model, characterized in that, The method includes: Obtain the comprehensive load model; The Sobol index method was used to perform sensitivity analysis on the load composition of the integrated load model, and the sensitivity coefficient of the load composition was obtained. Based on the sensitivity coefficient of the load composition, the load composition identification error is obtained; A loss function is constructed based on the load composition identification error and curve fitting error. Import the sample dataset into the load composition identification model for training until the loss function converges, thus obtaining the completed load composition identification model.

2. The method according to claim 1, characterized in that, The comprehensive load model was obtained in advance using the following method: The active power ratio parameters are set as follows: load composition includes photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ； in, S PV_B This indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

3. The method according to claim 1, characterized in that, The sensitivity analysis of the load composition of the integrated load model using the Sobol index method yields the sensitivity coefficients of the load composition, including: The sensitivity coefficient of the load configuration is obtained using the following formula: ； in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

4. The method according to claim 1, characterized in that, The load composition identification error, obtained from the sensitivity coefficient based on load composition, includes: The load configuration identification error is obtained using the following formula: ； ； in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

5. The method according to claim 4, characterized in that, The λ k,t The expression is as follows: ； in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t - The error ratio between the actual load composition value and the value after 2 iterations.

6. The method according to claim 1, characterized in that, The curve fitting error is obtained using the following formula: ； in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

7. The method according to claim 1, characterized in that, The loss function is constructed based on the load composition identification error and curve fitting error, including: ； in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

8. A training device for a load composition identification model, characterized in that, The device includes: The acquisition unit is used to acquire the comprehensive load model; The sensitivity analysis unit is used to perform sensitivity analysis on the load composition of the integrated load model using the Sobol exponent method to obtain the sensitivity coefficient of the load composition. The load configuration identification error unit is used to obtain the load configuration identification error based on the sensitivity coefficient of the load configuration. The loss function unit is used to construct a loss function based on the load composition identification error and the curve fitting error; The training unit is used to import the sample dataset into the load composition identification model for training until the loss function converges, thus obtaining the completed load composition identification model.

9. The apparatus according to claim 8, characterized in that, The comprehensive load model was obtained in advance using the following method: The active power ratio parameters are set as follows: load composition includes photovoltaic model load, dynamic load, and energy storage model load. The relationship between the load composition and the active power injection at the model port is constructed as follows: ； in, S PV_B This indicates the total installed capacity of distributed photovoltaic power in the distribution network. S BESS_B This indicates the total installed capacity of power electronic interface loads in the distribution network. P bus This represents the total active power consumed, excluding transformer and line losses.

10. The apparatus according to claim 8, characterized in that, The sensitivity analysis unit is also used for: The sensitivity coefficient of the load configuration is obtained using the following formula: ； in, S T,k Indicates the parameters of the integrated load model θ k When the value changes, the overall impact of this parameter and its interaction with other parameters on the dynamic response of the integrated load model is considered.

11. The apparatus according to claim 8, characterized in that, The load constitutes an error identification unit and is also used for: The load configuration identification error is obtained using the following formula: ； ； in, λ k,t The weights set during model training are used to adjust the convergence speed of the load composition; β k The sensitivity coefficient is determined by the load.

12. The apparatus according to claim 11, characterized in that, The λ k,t The expression is as follows: ； in, t It refers to the model update cycle. T tem For temperature coefficient, K This is the proportionality coefficient. ω k,t-1 For load configuration in t -1 times and t - The error ratio between the actual load composition value and the value after 2 iterations.

13. The apparatus according to claim 8, characterized in that, The curve fitting error is obtained using the following formula: ； in, and Indicates the first i The samples are in the interval [ T E,i +1, T The identified and true values ​​of the power response curve.

14. The apparatus according to claim 8, characterized in that, The loss function unit is also used for: ； in, α These are parameters used to balance load identification error and power response fitting error. L p,t The load constitutes the identification error. L curve This represents the curve fitting error.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

16. An electronic device comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method according to any one of claims 1-7.

Images