Parameter correction method of power system simulation model and electronic equipment

Abstract

The invention provides a parameter correction method of a power system simulation model and electronic equipment, relates to the technical field of power systems, and solves the technical problem of large parameter fluctuation in the power system parameter correction process in the prior art. The adopted scheme is as follows: calculating the sensitivity of parameters, and constructing a sensitive parameter set; taking the sensitive parameter set as a to-be-corrected parameter set, and correcting elements in the parameter set for the first time; and constructing an optimization result difference analysis module, selecting parameters needing secondary correction from the primary correction result, constructing a mapping relation model of a simulation track and a correction result based on an artificial intelligence method, injecting parameter data needing secondary correction into the model, and obtaining and outputting a correction result of the parameters subjected to secondary screening. According to the method, through closed-loop iterative optimization, a simulation track is highly matched with actual measurement data, and the accuracy and reliability of model parameters of the power system are guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a parameter correction method and electronic equipment for a power system simulation model. Background Art

[0002] This section is intended to provide background or context for embodiments of the present invention as recited in the claims. The description herein may include concepts that could be explored, but not necessarily concepts that have been previously conceived or explored. Therefore, unless otherwise indicated herein, the material described in this section is not prior art with respect to the specification and claims of this application and is not admitted to be prior art by inclusion in this section.

[0003] Power system simulation is a crucial tool for studying and optimizing power system operation. The accuracy of the results directly impacts decision-making, making timely and effective adjustment of simulation error parameters crucial. Currently, sensitivity identification-based methods primarily construct sensitive parameter sets by analyzing the sensitivity of a single or limited set of variable trajectories to the responses of various parameters. This approach offers advantages in initially revealing the impact of parameters on system dynamics, providing a basis for subsequent parameter correction.

[0004] However, due to the complex and volatile trajectories of variables in actual power systems, a single data source often struggles to fully capture all key dynamic characteristics, resulting in significant deficiencies in the coverage and representativeness of the constructed sensitive parameter sets. Furthermore, interference from noise and measurement errors can cause sensitivity identification results to deviate from actual conditions, limiting the accuracy and applicability of this method under complex operating conditions.

[0005] Based on the construction of sensitive parameter sets, conventional optimization algorithms (such as particle swarm optimization and genetic algorithms) are widely used in the parameter calibration process. These algorithms typically continuously adjust parameters by minimizing the error between simulated and measured values. Although these methods can theoretically achieve a gradual approximation of system states, they are prone to falling into local optimality in high-dimensional parameter spaces and multimodal objective functions. Furthermore, they are highly dependent on the initial parameter settings and the search range, which limits the stability and accuracy of the optimization results.

[0006] Kalman filtering methods based on recursive filtering and reinforcement learning methods that have emerged in recent years (such as Soft Actor-Critic, Deep Q-Network, and Deep Deterministic Policy Gradient) have been introduced into the field of parameter correction. These methods achieve high-precision parameter correction by more accurately capturing the dynamic characteristics of the system through adaptive updates and deep data mining techniques. However, these advanced algorithms also face many challenges: when dealing with highly nonlinear systems and complex synergies between parameters, the convergence speed and correction accuracy of recursive filtering methods are difficult to meet practical requirements; and reinforcement learning methods require a large amount of interaction data to train stable strategies. There are still huge challenges in designing reward functions that can truly reflect the dynamic characteristics of the system and effectively guide the learning of the intelligent agent. At the same time, low computational efficiency and large fluctuations in results also limit their widespread application.

[0007] Therefore, although advanced algorithms show great potential in theory, there are still problems in actual power system parameter correction, which requires further improvement and optimization. Summary of the Invention

[0008] In order to solve the technical problem of large parameter fluctuations during power system parameter correction in the prior art, the present invention aims to provide a parameter correction method and electronic equipment for a power system simulation model.

[0009] To solve the above technical problems, in a first aspect, according to some embodiments, the present invention provides a parameter correction method for a power system simulation model, comprising:

[0010] S1. Analyzing the sensitivity of parameters based on the variable type of the error to construct multiple sensitive parameter sets, specifically comprising: calculating the sensitivity of the parameters according to the variable type of different error parameters, normalizing the sensitivities, and constructing the sensitive parameter sets;

[0011] S2. Using the sensitive parameter set as the parameter set to be corrected, and performing a first correction on the elements in the parameter set, specifically comprising: updating the parameters using an adaptive inertia weight function and a dynamic acceleration constant strategy based on the difference between the time series trajectory output before and after each iteration and the target trajectory, thereby optimizing the parameters and obtaining different first correction results for the parameters;

[0012] S3, constructing an optimization result difference analysis module to select parameters that require secondary correction from the primary correction results;

[0013] S4. Construct a mapping relationship model between the simulation trajectory and the correction result based on an artificial intelligence method, inject the parameter data that needs secondary correction into the model, and obtain the correction result of the parameters screened out secondary.

[0014] Optionally, in some embodiments, calculating the sensitivity of the parameters according to the variable types of different error parameters specifically includes:

[0015] Apply an increment Δa to the parameter and calculate the overall change of the trajectory, as shown in Formula 1:

[0016]

[0017] Among them, Q1 represents the simulation trajectory with error before parameter correction, Q2 represents the simulation trajectory after the parameter increment is added, and S is the defined sensitivity value.

[0018] Optionally, in some embodiments, the normalizing the sensitivity specifically includes:

[0019] Formula 2 is used for normalization, as follows:

[0020]

[0021] Among them, n represents the label of the parameter, S n Indicates the sensitivity value of the parameter, S max Indicates the maximum sensitivity of all parameters in the set.

[0022] Optionally, in some embodiments, constructing the sensitive parameter set specifically includes:

[0023] According to the normalized results The sensitivity of all parameters is calculated and sorted according to the value of different variables, and the parameters ranked in the top k% are selected to construct the sensitive parameter set A of the variable. Y , Y represents the type of error variable, k is the custom threshold of sensitive parameters, and the larger the k, the greater the sensitivity of parameter selection.

[0024] Optionally, in some embodiments, the adaptive inertia weight function is as shown in Formula 3:

[0025]

[0026] Among them, w max and w min They represent the upper and lower limits of the inertia weight in the early and late stages respectively, a is the parameter for regulating the nonlinear attenuation rate, T is the maximum number of iterations, and t is the current number of iterations.

[0027] Optionally, in some embodiments, the strategy update parameters of the dynamic acceleration constant specifically include:

[0028] Formula 4 is used to adjust the proportion of individual and global search in different search stages, and the standard PSO fixed acceleration factor is dynamically adjusted as follows:

[0029]

[0030] It is used to realize the situation where c1 is large and c2 is small in the early stage of iteration, and c1 gradually decreases and c2 gradually increases in the later stage, making the particles more dependent on the global optimal solution.

[0031] Optionally, in some embodiments, the optimization of the parameters specifically includes:

[0032] A comprehensive evaluation function is used to achieve fine correction of system parameters, as shown in Formula 5:

[0033]

[0034] Among them, the first term in the formula 5 is used to measure the global error and calculate the particle x i The state at each time step t is the same as the reference trajectory x ref The error between (t) and the error influence of different stages is adjusted by the weight function w(t). The second term with the weight factor β is used to measure the smoothness of the trajectory and introduce a smoothness constraint to ensure that the motion trend of the particle matches the target trajectory.

[0035] By calculating the current F(x i ) and the historical optimal value F best In contrast, the global update rule is shown in Formula 6:

[0036] if F(x i )<F best -δ (14)

[0037] The global optimal solution is updated when the formula 6 is satisfied, where δ is the tolerance threshold, ensuring that the global optimal solution is updated only when there is a significant improvement. i ) is the objective function, which is used to measure the matching degree between the current particle trajectory and the reference trajectory, as well as the smoothness of the trajectory change.

[0038] Optionally, in some embodiments, the constructing of the optimization result difference analysis module to select parameters requiring secondary correction from the primary correction results specifically includes:

[0039] Set the screening index E, as shown in Formula 7:

[0040]

[0041] Where n represents the parameters in different sets A Y The number of times a appears in avgIndicates the average value of the optimization results of the parameter, a max Indicates the maximum value of the calculation result, a n Indicates the specific value of each optimization result of the parameter, and u% is used as the secondary screening threshold of the parameter;

[0042] If the calculated result of the screening index E of the current parameter is greater than u%, it is determined that the results of multiple optimizations of the current parameter are significantly different, and the parameter needs to be corrected twice.

[0043] Optionally, in some embodiments, constructing a mapping relationship model between the simulation trajectory and the correction result based on an artificial intelligence method specifically includes:

[0044] Using the multi-head self-attention mechanism in Transformer, the time information X time By calculating the query, key, and value, we can effectively capture the information and use formula 8 to get:

[0045]

[0046] Among them, W Q 、W K 、W V is the weight matrix to be learned, and a large amount of node data is obtained by randomly perturbing the parameters;

[0047] The time difference method is introduced, as shown in Formula 9:

[0048]

[0049] Δx i (t) represents the first-order rate of change of the variable, which is used to describe the mutation point information, Δ 2 x i (t) is the second-order difference, which is used to reveal the acceleration characteristics;

[0050] After extracting multi-type variable time series data near the fault node, data processing is performed, a mapping relationship model between time series data information and parameter values ​​is trained, and the correction results of the parameters screened out twice are output;

[0051] The loss function used in the mapping relationship model training is shown in Formula 10:

[0052]

[0053] In a second aspect, an embodiment of the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any one of the methods described in the first aspect are implemented.

[0054] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0055] The above technical solution of the present invention has at least the following beneficial technical effects: the present invention proposes a parameter correction method for a power system simulation model, which realizes rapid identification and screening of key parameters by utilizing a classification sensitivity analysis method according to the different types of variables with errors in the power system, that is, calculating parameter sensitivity through the error trajectories of different variables, screening out a set of sensitive parameters, and performing a first correction using an optimization algorithm. During the correction process, the parameters are updated using a strategy of an adaptive inertia weight function and a dynamic acceleration constant, realizing nonlinear attenuation to achieve sufficient search in the early stage and rapid convergence in the later stage. Subsequently, a second screening is performed on the parameters with large fluctuations in the first correction results, and a mapping relationship between the parameter value and the trajectory change rate is further established to achieve a second correction. According to the above example simulation, it is shown that through closed-loop iterative optimization, the simulation trajectory is highly consistent with the actual measurement data, ensuring the accuracy and reliability of the power system model parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the conventional technology, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 This is a flow chart of a parameter correction method for a power system simulation model provided by an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of the parameter one-time correction process architecture provided by an embodiment of the present invention.

[0059] Figure 3 Schematic diagram of a secondary correction network architecture provided by an embodiment of the present invention.

[0060] Figure 4 This is a 3-machine 9-node system architecture diagram provided by an embodiment of the present invention.

[0061] Figure 5 This is a generator active power-time trajectory diagram before and after parameter verification provided by an embodiment of the present invention.

[0062] Figure 6 This is the trajectory fitting result after parameter correction provided by an embodiment of the present invention.

[0063] Figure 7This is a schematic block diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0065] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present invention.

[0066] If there are descriptions of "first," "second," etc. in the embodiments of this application, such descriptions are for descriptive purposes only and should not be understood as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include at least one of such features; the technical solutions of various embodiments may be combined with each other, based on the ability of ordinary technicians in this field to implement them.

[0067] It should be noted that the sequence numbers mentioned in this application do not necessarily mean that the steps must be strictly followed in the sequence numbers during the actual implementation process. The sequence numbers are used to distinguish each step for ease of explanation and to prevent confusion.

[0068] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0069] Power system simulation is a crucial tool for studying and optimizing power system operations. The accuracy of the results directly impacts decision-making, making timely and effective adjustment of simulation error parameters crucial. Variable trajectories in real power systems are complex and highly variable, making it difficult for a single data source to fully capture all key dynamic characteristics. This results in significant deficiencies in the coverage and representativeness of constructed sensitive parameter sets.

[0070] At present, in the current research on power system parameter calibration, the parameter calibration process is based on a certain error trajectory and extracts its sensitivity, and then seeks the parameter correction result through optimization calculation.

[0071] However, the sensitivity obtained based on a single error trajectory has significant limitations. This idea lacks the characteristic induction of the parameters' effects on the variable trajectory. During the iteration process, the algorithm may "blindly" adjust the parameters in order to minimize or maximize the objective function, and even produce extreme values, thereby obtaining results that deviate from physical or practical meaning. In addition, during the parameter adjustment process, the synergistic effect of the parameters on the correction results should be further considered, otherwise the iterative process of parameter correction is prone to oscillation.

[0072] Therefore, the present invention proposes a parameter correction method that iterates multiple times. By extracting error trajectories from multiple types of variables, the method then optimizes and corrects them based on each type of variable. For the correction results of parameter pairs based on different types of trajectories, parameters with significant differences are selected for secondary correction. Furthermore, their coupling relationships are considered to construct a deep learning mapping model. By intelligently extracting the characteristic relationship between trajectories and parameter values, the method effectively corrects all error-causing parameters.

[0073] The basic idea of ​​identifying and correcting the power system's differential parameters is as follows:

[0074] (1) Identification of parameters to be corrected and one-step optimization correction

[0075] Leveraging the vast amount of monitoring data from the power system, error trajectory data is extracted and classified to obtain error variables. All system parameters are preliminarily screened from the perspective of trajectory energy to identify parameters sensitive to error variables. These parameters are then used to construct multiple parameter sets to be corrected. Parameter selection thresholds are set, and an optimization algorithm is used to perform a single correction on each parameter set to be corrected, obtaining the parameter correction results after each correction.

[0076] (2) Secondary screening and regression mapping of parameters to be corrected

[0077] When optimizing multiple parameter sets to be calibrated, the optimization results obtained for different parameter sets may differ due to the constraints of the optimization conditions. By comparing the differences in the parameter calibration results, the discrepant parameters are selected and their correlation is further considered. By perturbing these parameters extensively, deep learning is used to obtain the mapping relationship between a large number of measurement trajectories and parameter values. By constructing a regression model, the parameter calibration results can be obtained by inputting the measurement trajectories.

[0078] Specifically, an embodiment of the present invention provides a parameter correction method for a power system simulation model, such as Figure 1 Shown, including:

[0079] S1. Analyzing the sensitivity of parameters based on the variable type of the error to construct multiple sensitive parameter sets, specifically comprising: calculating the sensitivity of the parameters according to the variable type of different error parameters, normalizing the sensitivities, and constructing the sensitive parameter sets;

[0080] S2. Using the sensitive parameter set as the parameter set to be corrected, and performing a first correction on the elements in the parameter set, specifically comprising: updating the parameters using an adaptive inertia weight function and a dynamic acceleration constant strategy based on the difference between the time series trajectory output before and after each iteration and the target trajectory, thereby optimizing the parameters and obtaining different first correction results for the parameters;

[0081] S3, constructing an optimization result difference analysis module to select parameters that require secondary correction from the primary correction results;

[0082] S4. Construct a mapping relationship model between the simulation trajectory and the correction result based on an artificial intelligence method, inject the parameter data that needs secondary correction into the model, and obtain the correction result of the parameters screened out secondary.

[0083] The following is a detailed description.

[0084] 1. Construct sensitive parameter sets based on different variable types:

[0085] There are many components in the power system, and it is unrealistic to directly correct all parameters. Therefore, the sensitivity of the parameters is calculated according to the different types of variables with errors. The sensitivity calculation method is to apply an increment Δa to the parameter and calculate the overall change of the trajectory:

[0086]

[0087] Among them, Q1 represents the simulation trajectory with error before parameter correction, Q2 represents the simulation trajectory after the parameter increment is added, and S is the defined sensitivity value. After calculating the sensitivity of all parameters, normalization processing is required to facilitate the comparison of the sensitivity of different parameters:

[0088]

[0089] Among them, n represents the label of the parameter, S n Indicates the sensitivity value of the parameter, S max Indicates the maximum sensitivity of all parameters in the set. The sensitivity of all parameters is calculated and sorted according to the value of different variables, and the parameters ranked in the top k% are selected to construct the sensitive parameter set A of the variable. Y , where Y represents the type of error variable.

[0090] 2. One-time calibration of parameters:

[0091] The sensitive parameter set serves as the parameter set to be corrected, and the elements within it undergo the next step of correction. This parameter correction requires an iterative optimization process based on the data changes at each node in the system. This process is based on the particle swarm optimization (PSO). Parameter optimization is achieved by comparing the differences between the time series trajectory output before and after each iteration and the target trajectory.

[0092] In terms of parameter updates, while drawing on the fundamental principles of traditional PSO, we propose a strategy of adaptive inertia weights and dynamic acceleration constants to address the conflict between global search and local convergence requirements in complex systems. The design employs a nonlinear decay architecture and, combined with the time-series trajectory matching objective for specific problems, makes targeted improvements to the weight function. This allows the algorithm to more fully explore the search space in the early stages and more quickly lock onto high-quality local solutions in the later stages.

[0093]

[0094] where w max and w min They represent the upper and lower limits of the inertia weight in the early and late stages respectively, a is the parameter for regulating the nonlinear attenuation rate, and T is the maximum number of iterations. The Sigmoid function is introduced to achieve a smooth transition, thereby dynamically adjusting the search intensity at different stages to adapt to the different needs of the overall and local changes in the system's timing trajectory.

[0095] In order to adjust the proportion of individual and global search in different search stages, the standard PSO fixed acceleration factor is dynamically adjusted:

[0096]

[0097] This part can achieve that in the early stage of iteration, c1 is larger and c2 is smaller, which enhances the individual's search ability and avoids falling into the local optimum; in the later stage, c1 gradually decreases and c2 gradually increases, making the particles more dependent on the global optimal solution.

[0098] In order to achieve fine correction of the system state, a comprehensive evaluation function is introduced:

[0099]

[0100] Among them, the first one measures the global error and calculates the particle x i The state at each time step t is the same as the reference trajectory x ref The error between the two stages is adjusted by the weight function w(t); the second term (with weight factor β) measures the smoothness of the trajectory and introduces a smoothness constraint to ensure that the motion trend of the particle matches the target trajectory and avoid oscillation or violent fluctuations.

[0101] By real-time calculation of F(x i ) and the historical optimal value F best In contrast, the following global update rules are designed:

[0102] if F(x i )<F best -δ (24)

[0103] If the above equation is satisfied, the global optimal solution is updated, where δ is the tolerance threshold, ensuring that the global optimal solution is updated only when there is significant improvement, avoiding frequent updates caused by noise.

[0104] The outline process of parameter primary calibration is as follows: Figure 2 The sampling of trajectory data is consistent with the type of error trajectory, that is, the parameter set A to be corrected Y When optimizing, it is necessary to collect data on the Y variable at the fault node. By optimizing the elements in the set of parameters to be corrected based on different types of error variables, different correction results with different parameters are obtained.

[0105] 3. Secondary screening of parameters to be corrected:

[0106] Since the parameter set A to be corrected is obtained based on different types of error variables Y , the elements in each parameter set are different. In addition, since the variable data information extracted during the parameter optimization process is different, even if some parameters to be corrected are in multiple parameter sets A Y , but the optimization results will be quite different. These parameters have a more significant synergistic effect on multi-type variable trajectories. Therefore, a secondary screening of these parameters is required. Define the screening index E:

[0107]

[0108] Where n represents the parameters in different sets A Y The number of times a appears in avg Indicates the average value of the optimization results of the parameter, a max Indicates the maximum value of the calculation result, a n Indicates the specific value of each optimization result for this parameter. Define u% as the secondary screening threshold for the parameter. When the calculated E value for certain parameters is greater than u%, it indicates that the optimization results for these parameters vary significantly. Further corrections should be made to these parameters to account for their coupling effect on the error variable.

[0109] The correction results of the parameters that were not screened were retained, the retained results were averaged, and these parameters were replaced in the model before the second correction.

[0110] 4. Secondary parameter correction based on deep learning:

[0111] The parameters of the secondary screening will have different effects on multiple variables, so this type of parameters has a stronger coupling effect on the trajectory of the variables. The operating state of the power system is time-dependent, and the multi-head self-attention mechanism in the Transformer can integrate the time information X time Integration, by calculating the query, key and value, to achieve effective capture of information:

[0112]

[0113] Among them, W Q 、W K 、W V is the weight matrix to be learned. By randomly perturbing the parameters, a large amount of node data can be obtained. On this basis, the time difference method is introduced:

[0114]

[0115] Δx i (t) represents the first-order rate of change of the variable, which can be used to describe the mutation point information; Δ 2 x i (t) is the second-order difference, which reveals acceleration characteristics and helps distinguish between slow changes and sudden changes. After extracting multi-type variable time series data near the fault node, data processing is performed to construct a mapping relationship model between time series data information and parameter values, and model training is performed.

[0116] As a mapping model, calibration needs to improve the matching between parameter predictions and actual values. Using the mean squared error (MSE) has a stronger penalty for significant errors, which can encourage the model to pay more attention to the numerical predictions of these key parameters:

[0117]

[0118] The construction of the mapping model is as follows Figure 3 As shown in the figure, by properly training the model, directly injecting the measured data of the characteristic variables into the model, the correction results of the parameters screened out by the secondary screening can be output. The secondary correction results of the parameters are also replaced in the model to obtain a simulation model with corrected parameters, thus achieving complete correction of the model parameters.

[0119] In order to illustrate the effects of the embodiments of the present invention, verification is performed using the following method.

[0120] (1) Example verification:

[0121] like Figure 4As shown in Figure 1, a case study was performed based on the IEEE standard 3-machine 9-bus system. In this system, due to the difference between the parameters set in the simulation and the actual parameter data of the system, there is an error between the variable trajectory obtained by simulation and the actual trajectory.

[0122] The data from the standard system is used as real data. Random perturbations are performed on selected parameters in the system. After long-term operation, the actual parameters of the simulated system change, which can lead to a loss of simulation accuracy. For the simulation, a fourth-order model is used for the generator and a second-order model for the excitation system. A fault occurs at node 7 (B7), lasts 0.1 seconds, and has a sampling time of 5 seconds.

[0123] The active power per unit value P of No. 1 synchronous machine is SM_1 and the active power per unit value P of No. 2 synchronous machine SM_2 For example, the error trajectory is as follows Figure 5 shown.

[0124] It can be clearly observed from the figure that there is a significant difference between the actual measurement results and the simulation results before parameter calibration.

[0125] The actual measurement of the power system can obtain five types of variables: node voltage U, phase angle θ, generator active power P, reactive power Q, and rotor angular velocity ω. Based on these five types of variables, the sensitivity of all parameters in the system is calculated, and the obtained sensitive parameter set is As shown in Table 1. Among them, SM n Indicates the generator and its serial number; M is the inertia time constant (s); X d and X q are the direct-axis and quadrature-axis reactance (pu), respectively; X d ' and X q ' are the direct-axis and quadrature-axis subtransient reactances (pu), respectively; T d0 ' and T q0 ' is the subtransient open-circuit time constant of the direct and quadrature axes (s).

[0126] Table 1 Sensitive parameter set

[0127]

[0128] The elements in the sensitive parameter sets obtained according to different variables are not consistent. These parameters are calibrated once, and the results are shown in Table 2.

[0129] Table 2 Parameters primary calibration results

[0130]

[0131]

[0132] Among them, the parameter SM1_X d SM2_X d '、SM3_X d There are obvious fluctuations in the calibration results of SM2_M and SM3_M based on different variables, so these parameters need to be calibrated twice.

[0133] These five parameters were perturbed numerous times, and the corresponding node data for each perturbation was obtained. The resulting time series data was differentially processed to obtain the trajectory change rate. Using the change rate as the data input, a mapping relationship between the data change rate and the parameter value was constructed. The training set was divided into 80% and the remaining set was used for validation. A dropout ratio of 30% was used for the network.

[0134] Once the model is well trained, the differential trajectory of the actual measured variables can be used as input to obtain the secondary calibration results of the dominant parameters. The first calibration results are combined with the second calibration results to obtain the complete calibration of the parameters.

[0135] The overall calibration results of the parameters are shown in Table 3.

[0136] Table 3. Parameter values ​​before and after correction

[0137]

[0138] The parameter correction results are generally good, and the simulation results after parameter correction are compared, such as Figure 6 As shown in the figure, the effectiveness of the parameter correction results can be intuitively seen from the figure.

[0139] (2) Effect analysis:

[0140] 1. Selection of Leading Parameters: This method obtains multiple sets of sensitive parameters based on the sensitivity indices calculated for different types of variables and performs a calibration on each parameter. This effectively avoids the problem of insufficient leading parameter identification in traditional methods.

[0141] 2. Accuracy and representativeness of error-dominated parameter selection: Adaptive inertia weight and dynamic acceleration factor strategies are introduced to improve local convergence speed while ensuring global search capability. After preliminary correction, the values ​​of most parameters have tended to the true values, but some parameters (such as SM1_X d SM2_X d '、SM3_X d , SM2_M, and SM3_M) show significant fluctuations, indicating the existence of a multi-type error variable coupling problem, which requires further correction. Compared with traditional methods, this method can further consider the relationship between multiple error parameters.

[0142] 3. Accuracy of parameter calibration results: For parameters that fluctuate greatly during the primary calibration, a secondary calibration method based on deep learning is used. Utilizing a large amount of random perturbation data, the first-order and second-order differential features of the input data are processed to achieve an accurate mapping between the parameter values ​​and the actual system state. From the data, after the secondary calibration, the values ​​of each key parameter are relatively close to the actual data. For example, SM1_X d After correction, 1.188 (pu) is close to the actual value 1.200 (pu); SM2_M is corrected from the original large deviation to 35.001 (s), which is close to the actual value 36.250 (s).

[0143] 4. Validity of parameter calibration: Further comparison of the simulated trajectory after parameter correction shows that the fitting effect between the simulated trajectory and the actual measured trajectory is significantly improved, the trajectory fluctuation is reduced, and the overall error is reduced, which verifies the effectiveness of the correction method.

[0144] Therefore, this application proposes a parameter correction method for a power system simulation model. This method calculates parameter sensitivity using the error trajectories of different variables, screens out a set of sensitive parameters, and uses an optimization algorithm to perform a first correction. Subsequently, a second screening is performed on parameters with large fluctuations in the first correction results. A mapping relationship between parameter values ​​and trajectory change rates is further established to achieve a second correction. Simulations based on the above examples show that closed-loop iterative optimization ensures a high degree of consistency between the simulation trajectories and the actual measured data, ensuring the accuracy and reliability of the power system model parameters.

[0145] The present invention realizes rapid identification and screening of key parameters by utilizing a classified sensitivity analysis method based on the different types of variables with errors in the power system; and introduces adaptive inertia weights and dynamic acceleration factors into the algorithm to achieve sufficient initial search and rapid convergence in the later stage through nonlinear attenuation. In addition, a comprehensive evaluation function is introduced to comprehensively consider global error and trajectory smoothness, and a global update rule (including tolerance threshold) is formulated to ensure that the global optimal solution is updated only when there is significant improvement; for the sensitive parameter sets generated by different types of error variables, a custom screening index is designed, and parameters that appear multiple times in multiple parameter sets and have large differences in optimization results are further marked. For unmarked parameters, the average value is used as the correction result and replaced in the model before the secondary correction; the Transformer is combined with the time series difference feature to realize a method for fine correction of parameters under multiple time series data.

[0146] The embodiment of the present invention further provides an electronic device 700, such as Figure 7 As shown, it includes a memory 701, a processor 702, and a computer program stored in the memory 701 and executable on the processor. When the processor 702 executes the program, the steps of the method described in any one of the above embodiments are implemented.

[0147] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the above embodiments.

[0148] An embodiment of the present invention further provides a computer program product, including a computer program, which is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device performs the steps of the method described in any one of the above embodiments.

[0149] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0150] It should be understood that the above-described specific embodiments of the present invention are merely illustrative or illustrative of the principles of the present invention and do not constitute limitations of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included within the scope of protection of the present invention. In addition, the appended claims are intended to cover all variations and modifications that fall within the scope and metes and bounds of the appended claims, or equivalents thereof.

Claims

1. A parameter correction method for a power system simulation model, characterized in that: include: S1. Analyzing the sensitivity of parameters based on the variable type of the error to construct multiple sensitive parameter sets, specifically comprising: calculating the sensitivity of the parameters according to the variable type of different error parameters, normalizing the sensitivities, and constructing the sensitive parameter sets; S2. Using the sensitive parameter set as the parameter set to be corrected, and performing a first correction on the elements in the parameter set, specifically comprising: updating the parameters using an adaptive inertia weight function and a dynamic acceleration constant strategy based on the difference between the time series trajectory output before and after each iteration and the target trajectory, thereby optimizing the parameters and obtaining different first correction results for the parameters; S3, constructing an optimization result difference analysis module to select parameters that require secondary correction from the primary correction results; S4. Construct a mapping relationship model between the simulation trajectory and the correction result based on an artificial intelligence method, inject the parameter data that needs secondary correction into the model, and obtain the correction result of the parameters screened out secondary.

2. The method according to claim 1, characterized in that The sensitivity of the parameters is calculated according to the variable types of different error parameters, specifically including: Apply an increment Δa to the parameter and calculate the overall change of the trajectory, as shown in Formula 1: Among them, Q1 represents the simulation trajectory with error before parameter correction, Q2 represents the simulation trajectory after the parameter increment is added, and S is the defined sensitivity value.

3. The method according to claim 1, characterized in that The normalizing process of the sensitivity specifically includes: Formula 2 is used for normalization, as follows: Among them, n represents the label of the parameter, S n Indicates the sensitivity value of the parameter, S max represents the maximum sensitivity of all parameters in the set; The constructing of the sensitive parameter set specifically includes: According to the normalized results The sensitivity of all parameters is calculated and sorted according to different variables, and the parameters ranked in the top k% are selected to construct the sensitive parameter set A of the variable. Y , Y represents the type of error variable, and k is a custom threshold for sensitive parameters.

4. The method according to claim 1, wherein The adaptive inertia weight function is shown in Formula 3: Among them, w max and w min They represent the upper and lower limits of the inertia weight in the early and late stages respectively, a is the parameter for regulating the nonlinear attenuation rate, T is the maximum number of iterations, and t is the current number of iterations.

5. The method according to claim 1, wherein The strategy update parameters of the dynamic acceleration constant specifically include: Formula 4 is used to adjust the proportion of individual and global search in different search stages, and the standard PSO fixed acceleration factor is dynamically adjusted as follows: It is used to realize the situation where c1 is large and c2 is small in the early stage of iteration, and c1 gradually decreases and c2 gradually increases in the later stage, making the particles more dependent on the global optimal solution.

6. The method according to claim 1, characterized in that The optimization of the parameters specifically includes: A comprehensive evaluation function is used to achieve fine correction of system parameters, as shown in Formula 5: Among them, the first term in the formula 5 is used to measure the global error and calculate the particle x i The state at each time step t is the same as the reference trajectory x ref The error between (t) and the error influence of different stages is adjusted by the weight function w(t). The second term with the weight factor β is used to measure the smoothness of the trajectory and introduce a smoothness constraint to ensure that the motion trend of the particle matches the target trajectory. By calculating the current F(x i ) and the historical optimal value F best In contrast, the global update rule is shown in Formula 6: if F(x i )<F best -δ (5) If the above formula 6 is satisfied, the global optimal solution is updated, where δ is the tolerance threshold, ensuring that the global optimal solution is updated when there is significant improvement, where F(x i ) is the objective function, which is used to measure the matching degree between the current particle trajectory and the reference trajectory, as well as the smoothness of the trajectory change.

7. The method according to claim 1, wherein The construction of the optimization result difference analysis module selects the parameters that need secondary correction from the primary correction results, specifically including: Set the screening index E, as shown in Formula 7: Where n represents the parameters in different sets A Y The number of times a appears in avg Indicates the average value of the optimization results of the parameter, a max Indicates the maximum value of the calculation result, a n Indicates the specific value of each optimization result of the parameter, and u% is used as the secondary screening threshold of the parameter; If the calculated result of the screening index E of the current parameter is greater than u%, it is determined that the results of multiple optimizations of the current parameter are significantly different, and the parameter needs to be corrected twice.

8. The method according to claim 1, characterized in that The mapping relationship model between the simulation trajectory and the correction result is constructed based on the artificial intelligence method, specifically including: Using the multi-head self-attention mechanism in Transformer, the time information X time By calculating the query, key, and value, we can effectively capture the information and use formula 8 to get: Among them, W Q 、W K 、W V is the weight matrix to be learned, and a large amount of node data is obtained by randomly perturbing the parameters; The time difference method is introduced, as shown in Formula 9: Δx i (t) represents the first-order rate of change of the variable, which is used to describe the mutation point information, Δ 2 x i (t) is the second-order difference, which is used to reveal the acceleration characteristics; After extracting multi-type variable time series data near the fault node, data processing is performed, a mapping relationship model between time series data information and parameter values ​​is trained, and the correction results of the parameters screened out twice are output; The loss function used in the mapping relationship model training is shown in Formula 10:

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images