Key parameter analysis method of data-driven load model supporting credible simulation analysis

Abstract

The application discloses a kind of data-driven load model key parameter analysis methods of supporting credible simulation analysis.The application is based on the extraction of multiple types of load time-frequency domain characteristics, constructs mechanism type dynamic characteristic knowledge base, converts data-driven model into interpretable equivalent load model by approximate bayesian estimation and alternating direction multiplier method, and compares the accuracy loss of the dynamic characteristic response curve of both under the same simulation scene, to improve the simulation credibility of data-driven load model.The application realizes the mechanization reconstruction of data-driven load model.The equivalent load model keeps the data fitting accuracy while explicitly introducing the physical mechanism structure of load, so that the model result has clear physical meaning and engineering interpretability, and is more reasonable in load modeling expression.

Description

Technical Field

[0001] This invention relates to the field of load modeling technology, and specifically to a data-driven load model key parameter analysis method that supports reliable simulation analysis. Background Technology

[0002] Driven by dual carbon objectives, the power system is undergoing a transformation from a unidirectional power supply model to a multi-level interactive model involving power generation, grid, load, and storage. The large-scale grid connection of new loads and distributed renewable energy sources exhibits characteristics such as multi-condition time-varying and multi-level coupling. Traditional simulation analyses generally employ fixed-type, fixed-structure, and fixed-parameter load dynamic models using ZIP + induction motors, which cannot comprehensively reflect the dynamic characteristics of diverse loads. This has become a common core problem affecting the safety and stability analysis and risk prevention of new power systems. To meet the needs of power system safety and stability simulation analysis under the new circumstances, it is urgent to study new load dynamic characteristic modeling methods for power systems.

[0003] To address the modeling of load dynamic characteristics in the context of new power systems, novel load dynamic characteristic modeling methods, including distributed generation models and power electronic load models, have been developed both domestically and internationally. However, with the increasing variety, quantity, and operating conditions of new loads such as new energy sources and new grid-connected entities, the load dynamic characteristics of new power systems are becoming increasingly complex due to factors such as low-voltage tolerance characteristics on the load side, overshoot characteristics of control switching, high and low voltage ride-through and disconnection characteristics under faults, and the randomness, bidirectionality, and heterogeneity of multi-load power. Existing modeling methods for new power system load dynamic characteristics based on fixed models + statistical synthesis / overall measurement are still extensions and improvements of traditional methods and can no longer fully cope with the massive changes in nonlinear load dynamic characteristics. Therefore, it is urgent to study new theories for intelligent load dynamic characteristic modeling of new power systems. With the maturity of artificial intelligence technology and the improvement of multi-voltage level measurement systems for low-voltage switches, distribution areas, and feeders, data-driven load dynamic characteristic modeling methods based on machine learning, deep learning, and artificial intelligence have been developed both domestically and internationally. While data-driven modeling methods can cover multiple operating scenarios of new power systems, their inherent mechanisms are opaque and their physical interpretability is poor. Current AI interpretability is insufficient to meet the requirements of model transparency and practicality in simulation analysis and accident inversion when dealing with the dynamic characteristics of loads in new power systems. Furthermore, it is ill-suited to the requirements of multi-level power system safety and stability simulation analysis. Therefore, how to conduct research on interpretability enhancement techniques for data-driven load models, convert data-driven load models into equivalent load models, and overcome the technical bottleneck of poor simulation reliability of data-driven load models remains an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a data-driven load model key parameter analysis method that supports reliable simulation analysis.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A method for analyzing key parameters of a data-driven load model to support reliable simulation analysis includes the following steps: S1. Extract time-domain and frequency-domain features of multiple types of loads and construct a dynamic feature knowledge model base; S2. Based on the dynamic response characteristics of the data-driven load model, candidate mechanism model sets are matched using feature similarity and Bayesian estimation methods; S3. By introducing consistency constraint variables, the alternating direction multiplier method is used to jointly identify the mechanism parameters and residual parameters, and construct an equivalent load model; S4. Verify the accuracy loss of the equivalent load model to improve the reliability of simulation analysis.

[0006] Furthermore, step S1 specifically includes the following steps: S11. Obtain voltage, frequency, and power data for distributed new energy sources, charging stations / piles, and user-side energy storage in typical scenarios; S12. Screen the power-voltage / frequency cross-correlation time period and calculate the cross-correlation function using the generalized cross-correlation algorithm; S13. Extract time-domain and frequency-domain features based on wavelet transform to form time-frequency domain feature vectors of grid-connected components. Construct a load model feature library based on the time-frequency domain features of grid-connected components and label the corresponding simulation components.

[0007] Furthermore, the cross-correlation function in S12 is expressed as:

[0008]

[0009] In the formula, and Here is the filter function. yes The conjugate of ) It is voltage and power cross power spectral function, For generalized frequency domain weight components.

[0010] Furthermore, step S2 specifically includes the following steps: S21. Analyze the dynamic response characteristics of the data-driven load model and extract time-frequency domain features. ; S22. Normalize the feature dimensions and calculate the weighted Euclidean distance as the feature similarity; S23. Based on Bayesian estimation, calculate the posterior probability of matching the mechanistic model and output the candidate mechanistic model set. .

[0011] Furthermore, in S23, a set of candidate mechanism models is output. Represented as:

[0012] In the formula, M mat This is a set of mechanistic load models matched from the load model feature library; M L The minimal mechanistic model is sorted by matching probability. η This is the cumulative confidence threshold. p ( M i| F (s) ) represents the matching probability of the i-th mechanism model. Observe the feature vector for the data-driven load model.

[0013] Furthermore, step S3 specifically includes the following steps: S31. Output-based candidate mechanism model set M mat Establish an equivalent load model; S32. Identify the mechanism parameters and residual terms of the equivalent load model through joint optimization and define its objective function and constraints. S33. Introduce auxiliary variables to transform the optimization problem of the equivalent load model into a form with consistency constraints. Use the alternating direction multiplier method to iteratively optimize the equivalent load model and output the optimized equivalent load model.

[0014] Furthermore, in step S32, the objective function is:

[0015]

[0016]

[0017] in, It is the square norm 2; Output for data-driven load models; θ 0 represents the prior reference value for the mechanism parameter. r ( t ) is the residual term vector, g (•) represents the set of mechanistic models.M mat The mechanistic model structure obtained from the selection process; θ This represents the parameter vector of the load mechanism model. For mechanism parameter regularization, These are residual structure constraints used to limit the complexity of the residuals; λ θ , λ r These are the weighting coefficients.

[0018] Furthermore, the specific method for iteratively optimizing the equivalent load model using the alternating direction multiplier method in S33 is as follows: A1. Based on the physical constraints of the mechanism parameters, the mechanism parameters of the equivalent load model are updated under the condition of fixed residual terms and auxiliary variables. The optimization objective is expressed as:

[0019] In the formula, θ h+1 For the first h Mechanistic model parameters for +1 iteration; z h (t) is the th h Auxiliary variables for +1 iterations, Let h be the Lagrange multiplier for the h-th iteration. ρ As a penalty factor; A2. Based on steady-state residual suppression constraints, residual energy constraints, and frequency domain constraints, the residual term is updated under the condition of fixed mechanism structure output, fitting only the unexplained part of the mechanism physical parameters. The optimization objective is expressed as:

[0020] In the formula, For the first h +1 iteration residual term; A3. Update the consistency variables to ensure that the output of the mechanistic model is consistent with the residual compensation results. The update method is as follows:

[0021] In the formula, No. h Consistency variables for +1 iterations; A4. Check the convergence criterion. When the following conditions are met:

[0022] The iteration terminates when the time is right. In the formula, ε priThis is the allowable threshold for consistency constraint error, used to determine whether the deviation between the output of the mechanism model and the consistency variables is within the acceptable range for engineering. ε dual This is the allowable threshold for the magnitude of variable updates, used to determine whether a consistent variable has become stable in adjacent iterations; The parameters of the mechanism model in the h-th iteration are... These are the consistency variables for the h-th and h-1-th iterations, respectively.

[0023] The present invention has the following beneficial effects: This invention constructs an equivalent load model that includes mechanistic structure terms and residual compensation terms by synergistically analyzing load behavior from the perspectives of load dynamic response mechanism and data-driven modeling. This achieves a mechanistic reconstruction of the data-driven load model. While maintaining data fitting accuracy, this equivalent load model explicitly introduces the physical mechanism structure of the load, giving the model results clear physical meaning and engineering interpretability, making its load modeling expression more reasonable. This invention takes into account the differences between mechanistic models and data-driven models in terms of structural form, parameter scale, and dynamic characteristics. By introducing feature matching and similarity evaluation mechanisms, candidate mechanistic models are screened and combined, effectively avoiding the structural mismatch problem caused by direct parameter fitting. At the same time, consistency constraints and regularization are used to constrain the residual terms and prevent them from overfitting the measurement data, thereby ensuring a clearer division of responsibilities between the mechanistic structure and residual compensation. This invention further unifies the identification processes of mechanistic model parameters and residual parameters within the same optimization framework by introducing a joint identification method based on the alternating direction multiplier method. During the iteration process, the mechanistic parameters, residual variables, and constraint multipliers are updated alternately, achieving coordinated convergence of both. Compared to traditional step-by-step identification or post-correction methods, this method exhibits significant advantages in computational stability, convergence, and model consistency, resulting in a more stable and reliable equivalent load model suitable for power system analysis, simulation, and online applications. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the overall process of analyzing key parameters for the data-driven load model transformation of this invention.

[0025] Figure 2 This is a schematic diagram of a knowledge base for dynamic characteristics of multiple types of loads.

[0026] Figure 3 This is a schematic diagram of the joint identification algorithm for equivalent load models based on the alternating multiplier method. Detailed Implementation

[0027] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0028] A method for analyzing key parameters of a data-driven load model to support reliable simulation analysis, such as... Figure 1 As shown, it includes the following steps: S1. Combining simulation data and multi-business measurement data, extract time-domain and frequency-domain features of various load types such as distributed new energy, charging stations / piles, and user-side energy storage, and construct a dynamic feature knowledge model library, such as... Figure 2 As shown, specifically, it includes the following steps: S1.1 Obtain voltage, frequency, and power data of various grid-connected components such as distributed new energy, charging stations / piles, and user-side energy storage in typical scenarios in the simulation platform and measurement system; S1.2, Screening the power-voltage / frequency cross-correlation time period. With... U / f The decline / recovery led to P , Q The descent / recovery exhibits weakness, time delay, and multiplicity. A generalized cross-correlation algorithm is employed to detect... P , Q right U / f The time period with the strongest descent / recovery response is set as follows: U , f , P , Q High-precision sampling data that varies over time u ( t ), f ( t ), p ( t ), q ( t This includes voltage, frequency, active power, and reactive power signals. s u ( t ), s f ( t ), s p ( t ), s q ( t and noise signals n u( t ), n f ( t ), n p ( t ), n q ( t ): (1) The process of selecting strongly correlated time periods is illustrated using voltage and active power as examples. u ( t )and p ( t The cross-correlation function of () can be expressed as: (2) In the formula, G 12 ( ω )yes u ( t )and p ( t The cross-power spectrum function of ).

[0029] After filtering, the output cross-power spectrum of the voltage and power data is as follows: (3) In the formula: H 1 and H 2 represents the filter function; H* 2 ( ω )yes H 2( ω ) conjugate.

[0030] The generalized cross-correlation function of voltage and power can be expressed as: (4) (5) In the formula: For generalized frequency domain weight components.

[0031] By selecting an appropriate weighting function, "sudden jumps" in signal and noise can be effectively suppressed. Cross-correlation functions are calculated point-by-point to construct the UP cross-correlation plane. Criteria for determining cross-correlation segments are proposed. Dynamic time bending, a time series similarity method, is used to identify and remove spurious correlation segments, ultimately selecting the truly strongly correlated time segments.

[0032] S1.3. Based on wavelet transform and other time-frequency domain algorithms, extract the time-frequency domain features of the load dynamic response characteristics. In the time domain, extract features such as rise / fall slope, peak-to-valley ratio, response delay time, and response duration; in the frequency domain, extract features such as bandwidth, resonant frequency, resonant peak value, and low-frequency gain. The load dynamic characteristic feature vector composed of time and frequency domains is as follows: (6) in: (7) (8) In the formula, For time domain features, In order, the rising / falling slope, peak-to-valley ratio, response delay time, and response duration are: For frequency domain characteristics, This is used as the bandwidth, resonant frequency, resonant peak value, and low-frequency gain; S1.4. Construct a load model feature library based on the time-frequency domain characteristics of grid-connected components and label the corresponding simulation components. Each model in the mechanism model library satisfies: (9) In the formula: M k For the first k Each mechanism sub-model; M is the set of load mechanism models; F (k) For the first k Feature vectors of each mechanism sub-model; f(k) m For the first k The first mechanistic sub-model m One characteristic.

[0033] S2. Based on the dynamic response characteristics of the data-driven load model, the mechanism model set is matched and parameters are initialized through feature similarity matching. Specific details include: S2.1 Analyze the time-frequency domain characteristics of the load output power response characteristics for the data-driven load dynamic characteristic model: (10) S2.2, Characteristic Dimension Normalization: (11) Define feature weight assignment: (12) In the formula: w is the feature weight vector; ω m For the first m The weights of each feature.

[0034] S2.3, The method for calculating the feature similarity measure between the data-driven load model and the mechanism load model based on weighted Euclidean distance is as follows: (13) in: (14) In the formula: S k Similarity of features between data-driven load models and mechanistic load models; W This is the feature weight diagonal matrix.

[0035] Mechanism models can be matched from the load feature library using similarity. This step is a preliminary screening based on deterministic similarity, which helps to narrow the search space for subsequent Bayesian estimation, eliminate obviously unreasonable mechanism structures, and ensure that Bayesian estimation calculation is stable and controllable.

[0036] (15) In the formula: M0 is the initial mechanism model set; δ This is the similarity threshold.

[0037] S2.4. Based on S2.1-S2.3, and combined with Bayesian estimation theory, the mechanism sub-model M in the load model feature library... k The Bayesian posterior probability that matches the data-driven load model is: (16) (17) In the formula: To match the posterior probability of the mechanism sub-model; This is a priori distribution determined by combining load type, operating conditions, or historical statistical information; Representation and Mechanism Sub-model M k The likelihood function related to the output error; Observation feature vectors for data-driven load models; This represents the feature extraction error.

[0038] Based on this, the load model feature library is calculated one by one. k The posterior probability vector of the mechanistic model: (18) In the formula: P is the posterior probability vector of the mechanism model in the feature library of the load model.

[0039] posterior probability ranking for mechanistic model matching: (19) In the formula: p ( Mi| F (s) ), p ( M j| F (s) ) are respectively the first i , j The probability of matching the mechanism model.

[0040] By progressively accumulating larger probability values, a set of candidate mechanism models is output using a confidence threshold criterion. (20) In the formula: M mat This is a set of mechanistic load models matched from the load model feature library; M L The minimal mechanistic model is sorted by matching probability. η This is the cumulative confidence threshold.

[0041] S3. By introducing consistency constraint variables, the mechanistic model output and residual compensation term are unified into the same optimization framework. The alternating direction multiplier method (ADMM) is used to jointly identify the mechanistic parameters and residual parameters, such as... Figure 3 As shown, an equivalent load model consisting of "mechanism structure + residual part" is constructed, as detailed below: S3.1 Establishing the equivalent load model expression: Based on the candidate mechanism model set of the output M mat The equivalent load model of "mechanism structure + residual part" is uniformly expressed as: (twenty one) In the formula: y ( t ) represents the load output vector (active / reactive power); x ( t ) represents the external excitation input vector (system voltage / frequency); g (•) represents the set of mechanistic models. M mat The mechanistic model structure obtained from the selection process; θ This represents the parameter vector of the load mechanism model. r ( t ) is the residual term vector.

[0042] S3.2 Identify the mechanistic parameters and residual terms of the equivalent load model through joint optimization. The objective function is defined as follows: (twenty two) in: (twenty three) (twenty four) In the formula: It is the square norm 2; Output (active / reactive power) for data-driven load models. θ 0 represents the prior reference value for the mechanism parameter; This is a regularization term for mechanistic parameters, used to limit the deviation of mechanistic parameters from their physical prior values, ensuring the physical rationality and stability of the mechanistic model parameters; These are residual structure constraints used to limit the complexity of the residuals; λ θ , λ r These are the weighting coefficients.

[0043] The constraints are as follows: To ensure that the mechanistic parameters remain within a physically reasonable range during the joint identification process and to avoid non-physical drift of the mechanistic model parameters due to the existence of residual compensation terms, the following physical constraints on the mechanistic parameters are established: (25) In the formula: θ min This represents the minimum allowable values ​​vector for the mechanism parameters. θ max This is the minimum allowable value vector for the mechanism parameters.

[0044] By suppressing the energy of the residual compensation term within the steady-state interval, the steady-state load characteristics are forced to be characterized by the mechanistic model, ensuring the physical consistency of the equivalent load model under steady-state conditions. The steady-state residual suppression constraint for the residual term is established as follows: (26) In the formula: T ss For steady-state time sets; ε ss This is the allowable threshold for steady-state residuals, and it is generally set to a minimum value.

[0045] In the frequency domain, low-frequency dynamics correspond to the dominant physical processes of the load, while high-frequency dynamics are mostly caused by random disturbances, nonlinear switching, and other factors. By limiting the spectral energy of the residuals within the dominant frequency band of the mechanistic model, we ensure that the dominant dynamics are characterized by the mechanistic model, while the residuals are only used to compensate for high-frequency or local dynamic deviations. Therefore, we establish frequency domain constraints on the residuals: (27) In the formula: R ( jω ) represents the residual term r ( tThe Fourier transform of the residual represents the energy distribution of the residual at different frequencies; ω c The upper limit of frequency is dominated by the mechanistic model; ε f This is the low-frequency residual energy threshold.

[0046] To limit the overall energy of the residual term and prevent the residual model from dominating the system response, and to ensure that the equivalent load model is based on the mechanistic structure, the upper bound constraint on the overall residual energy is as follows: (28) In the formula: ε r This is the upper limit of residual energy, used to limit the overall amplitude level of the residual.

[0047] By introducing the mechanism-residual orthogonality constraint, the residual compensation term is prevented from participating in the equivalent adjustment of the mechanism parameters, thus structurally separating the functions of the mechanism model and the residual model. The constraint is as follows: (29) In the formula: This indicates the direction of the response of the mechanistic model output to changes in each parameter.

[0048] S3.3 Given that ADMM can decompose the original coupled optimization problem into multiple independent subproblems that are related by consistency constraints, and achieve a globally consistent solution through multiplier iteration, ADMM can be used to structurally separate the mechanism parameter identification problem from the residual compensation problem, achieving alternating optimization while ensuring that the outputs of the two are consistent.

[0049] (1) To decouple the mechanism structure from the residual compensation, auxiliary variables are introduced to transform the optimization problem into a form with consistency constraints. The specific auxiliary variables are as follows: (30) In the formula: z ( t ) is an auxiliary variable.

[0050] Therefore, the original joint problem is equivalent to: (31) Furthermore, the augmented Lagrangian function is constructed as follows: (32) In the formula: μ ( t ) are Lagrange multipliers; ρ This is a penalty factor.

[0051] (2) The iterative update steps of the Alternating Direction Multiplier Method (ADMM) are as follows: Step 1: Mechanism Parameter Subproblem The load model mechanism parameters are updated under the condition of fixed residuals and auxiliary variables, and the parameters are updated only within the mechanism space. The optimization objective is: (33) In the formula: θ h+1 For the first h Mechanistic model parameters for +1 iteration; z h (t) is the th h Auxiliary variables for +1 iterations.

[0052] The constraints are the physical constraints of the mechanism parameters shown in equation ().

[0053] Step 2: Residual Compensation Subproblem The residual terms are updated under the condition of fixed mechanistic structure output, fitting only the part not explained by the mechanistic model. The optimization objective is: (34) The constraints are the steady-state residual suppression constraint, residual energy constraint, and frequency domain restriction constraint shown in equations ()-().

[0054] Step 3: Update Consistency Variables To ensure consistency between the mechanistic model output and the residual compensation result, the update formula is as follows: (35) Step 4: Multiplier Update The explanatory boundary of the automatic adjustment mechanism / residuals, and the update formula are as follows: (36) Step 5: Convergence Criterion The convergence criterion for the algorithm is: (37) In the formula: ε pri This is the allowable threshold for consistency constraint error, used to determine whether the deviation between the output of the mechanism model and the consistency variables is within the acceptable range for engineering. ε dual This is the allowable threshold for variable update magnitude, used to determine whether a consistent variable has stabilized in adjacent iterations.

[0055] S3.4 outputs an equivalent load model consisting of "mechanism model + residual component": (38) The mechanism-based residual joint identification method based on the alternating direction multiplier method has the following process: Figure 2As shown, the process begins by inputting load response measurement data and a set of candidate mechanism models. An equivalent load model is then constructed, including mechanism structure terms and residual terms, with consistency constraints introduced to limit the degrees of freedom of the residual terms. Subsequently, an augmented Lagrangian function is constructed, and the alternating direction multiplier method is used for iterative solution. During each iteration, the mechanism model parameters, residual variables, consistency auxiliary variables, and Lagrangian multipliers are updated sequentially. After each iteration, it is determined whether the convergence conditions of the original residuals and dual residuals are satisfied. When the convergence criterion is met, the final equivalent load model is output, achieving the mechanistic transformation of the data-driven load model.

[0056] S4. Based on the equivalent load model output in S3, verify the accuracy loss of the converted load model in the same simulation environment to improve the reliability of the load model simulation analysis.

[0057]

[0058] In the formula: ε is the conversion accuracy loss between the data-driven load model and the equivalent load model.

[0059] 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.

[0060] 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 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] 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 1The steps of the function specified in one or more boxes.

[0062] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0063] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for analyzing key parameters of a data-driven load model to support reliable simulation analysis, characterized in that, Includes the following steps: S1. Extract time-domain and frequency-domain features of multiple types of loads and construct a dynamic feature knowledge model base, which includes the following steps: S11. Obtain voltage, frequency, and power data for distributed new energy sources, charging stations / piles, and user-side energy storage in typical scenarios; S12. Screen the power-voltage / frequency cross-correlation time period and calculate the cross-correlation function using the generalized cross-correlation algorithm; S13. Extract time-domain and frequency-domain features based on wavelet transform to construct time-frequency domain feature vectors for grid-connected components. Construct a load model feature library based on the time-frequency domain features of grid-connected components and label the corresponding simulation components. S2. Based on the dynamic response characteristics of the data-driven load model, candidate mechanism model sets are matched using feature similarity and Bayesian estimation methods. This includes the following steps: S21. Analyze the dynamic response characteristics of the data-driven load model and extract time-frequency domain features. ; S22. Normalize the feature dimensions and calculate the weighted Euclidean distance as the feature similarity; S23. Based on Bayesian estimation, calculate the posterior probability of matching the mechanistic model and output the candidate mechanistic model set. ; S3. By introducing consistency constraint variables and using the alternating direction multiplier method, the mechanistic parameters and residual parameters are jointly identified, and an equivalent load model is constructed. This includes the following steps: S31. Output-based candidate mechanism model set M mat Establish an equivalent load model; S32. Identify the mechanistic parameters and residual terms of the equivalent load model through joint optimization, and define its objective function and constraints. The objective function is: in, It is the square norm 2; For data-driven load model output, The external stimulus input vector, For time, For time The upper limit of the summation; θ 0 represents the prior reference value for the mechanism parameter. r ( t ) is the residual term vector, g (•) represents the set of mechanistic models. M mat The mechanistic model structure obtained from the selection process; θ This represents the parameter vector of the load mechanism model. For mechanism parameter regularization, These are residual structure constraints used to limit the complexity of the residuals; λ θ , λ r These are the weighting coefficients; S33. Introduce auxiliary variables to transform the optimization problem of the equivalent load model into a form with consistency constraints, and use the alternating direction multiplier method to iteratively optimize the equivalent load model, and output the optimized equivalent load model. S4. Verify the accuracy loss of the equivalent load model to improve the reliability of simulation analysis.

2. The method for analyzing key parameters of a data-driven load model supporting reliable simulation analysis according to claim 1, characterized in that, The cross-correlation function in S12 is expressed as follows: In the formula, and Here is the filter function. yes conjugate, It is voltage and power cross power spectral function, These are generalized frequency domain weight components; This is a kernel function used to convert the power spectral density in the frequency domain. and weight Map back to the time domain using the inverse Fourier transform.

3. The method for analyzing key parameters of a data-driven load model supporting reliable simulation analysis according to claim 1, characterized in that, The set of candidate mechanism models is output in S23. Represented as: In the formula, M mat This is a set of mechanistic load models matched from the load model feature library; M 0 represents the initial mechanistic model set. M L The minimal mechanistic model is sorted by matching probability. η This is the cumulative confidence threshold. p ( M i | F (s) ) represents the matching probability of the i-th mechanism model. Observe the feature vector for the data-driven load model.

4. The method for analyzing key parameters of a data-driven load model supporting reliable simulation analysis according to claim 1, characterized in that, Specifically, in S33, the method of iteratively optimizing the equivalent load model using the alternating direction multiplier method is as follows: A1. Based on the physical constraints of the mechanism parameters, the mechanism parameters of the equivalent load model are updated under the condition of fixed residual terms and auxiliary variables. The optimization objective is expressed as: In the formula, θ h+1 For the first h Mechanistic model parameters for +1 iteration; z h (t) is the th h Auxiliary variables for the next iteration Let h be the Lagrange multiplier for the h-th iteration. ρ As a penalty factor; A2. Based on steady-state residual suppression constraints, residual energy constraints, and frequency domain constraints, the residual term is updated under the condition of fixed mechanism structure output, fitting only the unexplained part of the mechanism physical parameters. The optimization objective is expressed as: In the formula, For the first h The residual term of +1 iteration, As an auxiliary variable; A3. Update the auxiliary variables to ensure that the output of the mechanistic model is consistent with the residual compensation result. The update method is as follows: In the formula, No. h Auxiliary variables for +1 iterations; A4. Check the convergence criterion. When the following conditions are met: The iteration terminates when the time is reached; where, ε pri This is the allowable threshold for consistency constraint error, used to determine whether the deviation between the output of the mechanism model and the auxiliary variables is within the acceptable range for engineering purposes; ε dual This is the allowable threshold for the magnitude of variable updates, used to determine whether the auxiliary variable has stabilized in adjacent iterations; The parameters of the mechanism model in the h-th iteration are... Consistency variables for h-1 iterations.

Images