A method for simplifying the dynamic model of a microgrid

Abstract

The present invention discloses a method for simplifying a dynamic model of a microgrid, belonging to the technical field of data processing and computer applications. The method includes obtaining a real-time operation data set of the microgrid and extracting a dynamic feature vector through hierarchical extraction of dynamic features; performing joint structure-parameter order reduction processing on the dynamic feature vector to generate a set of reduced-order parameters; performing multi-objective optimization decision-making based on the set of reduced-order parameters to generate optimized decision-making parameters; generating a simplified dynamic model according to the optimized decision-making parameters. Among them, the hierarchical extraction of dynamic features adopts a preset frequency band division rule, and the joint structure-parameter order reduction processing includes screening using a preset sensitivity threshold. The present invention adopts the hierarchical extraction of dynamic features and the joint structure-parameter order reduction technology, combined with multi-objective game optimization and error adaptive compensation mechanism, which can significantly improve the calculation efficiency while ensuring the dynamic simulation accuracy and meet the requirements of different microgrid operation conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and computer applications, and in particular to a method for simplifying a microgrid dynamic model. Background Art

[0002] Microgrid dynamic modeling is one of the core technologies for power system analysis and control, which reflects the real-time dynamic characteristics of the microgrid through mathematical simulation. In the prior art, the dynamic model usually needs to run in the full-order mode to ensure accuracy, but the high-dimensional characteristics of the full-order model will lead to excessive consumption of computing resources and it is difficult to meet the real-time decision-making requirements.

[0003] Currently, traditional order reduction methods mainly adopt balanced truncation method or modal analysis method to reduce the dimension by removing high-frequency dynamic modes or low-sensitivity parameters. There are also methods to extract features through a fixed time window and combine empirical thresholds to screen key variables. In addition, the prior art obtains the order reduction parameter set through single-objective optimization, such as only pursuing the minimum voltage error, without considering the comprehensive balance of computing efficiency and system dynamic response ability.

[0004] However, the traditional feature extraction has a fixed time window length and cannot adapt to the dynamic characteristic changes of different frequency bands, resulting in the misremoval of high-frequency key modes; the static sensitivity threshold is difficult to adapt to multi-condition scenarios, and the generalization ability of the order reduction parameter set is insufficient; the single-objective optimization process ignores key constraints such as topological response time, resulting in frequent decision failures of the model under complex conditions and relying on the full-order model for repeated reconstruction. Summary of the Invention

[0005] To solve the above problems, the present invention provides a method for simplifying a microgrid dynamic model, which adopts dynamic feature hierarchical extraction and structure-parameter joint order reduction technology, combines multi-objective game optimization and error adaptive compensation mechanism, and can significantly improve the computing efficiency while ensuring the dynamic simulation accuracy, and meet the requirements of different microgrid operating conditions.

[0006] The above object can be achieved by the following solutions:

[0007] A method for simplifying a microgrid dynamic model, including obtaining a real-time operation data set of the microgrid, and obtaining a dynamic feature vector through dynamic feature hierarchical extraction; performing structure-parameter joint order reduction processing on the dynamic feature vector to generate an order reduction parameter set; performing multi-objective optimization decision-making based on the order reduction parameter set to generate optimized decision-making parameters; generating a simplified dynamic model according to the optimized decision-making parameters; wherein, the dynamic feature hierarchical extraction adopts a preset frequency band division rule, and the structure-parameter joint order reduction processing includes screening by using a preset sensitivity threshold.

[0008] Optionally, the obtaining of the real-time operation data set of the microgrid and the extraction of the dynamic feature vector through dynamic feature hierarchical extraction include: collecting the voltage volatility and power change rate of the microgrid nodes to form an original feature matrix; intercepting the original feature matrix according to a preset sliding time window to generate a windowed feature matrix, where the length of the sliding time window is adjusted according to a preset window adjustment coefficient; performing spectrum feature extraction and non-steady state detection on the windowed feature matrix according to a preset frequency band division rule, and outputting a dynamic feature vector.

[0009] Optionally, the structural-parameter joint order reduction processing of the dynamic feature vector to generate a reduced-order parameter set includes: decomposing the dynamic feature vector into a fast-changing feature subset and a slow-changing feature subset; calculating the sensitivity of the variables in the fast-changing feature subset, and removing the variables in the fast-changing feature subset with sensitivity lower than a preset sensitivity threshold to generate a fast-changing reduced-order subset; adjusting the variables in the fast-changing reduced-order subset by using a preset dynamic compensation factor, and combining with the slow-changing feature subset to obtain a reduced-order parameter set.

[0010] Optionally, the sensitivity threshold includes: extracting the spectrum feature parameters of the dynamic feature vector; calculating the contribution degree index of the variable according to the spectrum feature parameters and a preset frequency band weight coefficient; generating a sensitivity threshold according to the contribution degree index; where the update period of the sensitivity threshold is synchronized with the adjustment period of the sliding time window.

[0011] Optionally, the method further includes: extracting the state variable comparison data between the reduced-order model and the full-order model from a preset historical operation log to construct a historical reduced-order error data set; performing cross-correlation on the historical reduced-order error data set and the spectrum feature parameters to calculate a gain coefficient; and correcting the dynamic compensation factor by using the gain coefficient.

[0012] Optionally, the execution of multi-objective optimization decision-making based on the reduced-order parameter set to generate optimization decision parameters includes: constructing an accuracy-efficiency game model, where the accuracy-efficiency game model uses the reduced-order parameter set as an input variable; the accuracy-efficiency game model performs a multi-stage Pareto optimization process and matches features with a preset typical working condition scenario library to generate preliminary decision parameters; and performing a three-level fault tolerance correction on the preliminary decision parameters to output optimization decision parameters.

[0013] Optionally, the execution of the multi-stage Pareto optimization process includes: constructing a first objective pair of voltage error rate and calculation time-consuming in the first stage; superimposing the topology change response time in the second stage to form a three-objective optimization space; applying a preset algorithm to obtain an iterative solution set in the three-objective optimization space; and matching features of the iterative solution set with a preset typical working condition scenario library to generate preliminary decision parameters.

[0014] Optionally, a three - level fault - tolerance correction is performed on the preliminary decision parameters, and the output optimized decision parameters include: performing a critical constraint verification on the preliminary decision parameters to generate a verification result, where the critical constraint verification includes an impedance matching test and a power conservation test; triggering residual compensation according to the verification result and performing the critical constraint verification again to generate a compensation result; generating a full - order model reconstruction instruction according to the compensation result.

[0015] Optionally, the construction of the typical operating condition scenario library includes: collecting benchmark model parameters under multiple operating modes to form an original scenario set; extracting feature fingerprints from the original scenario set to generate fingerprint coding vectors; binding the fingerprint coding vectors with the hash values of the blockchain storage module; performing pre - loading matching according to the similarity between the dynamic feature vectors and the fingerprint coding vectors during the real - time operation process.

[0016] Optionally, the method further includes: collecting the voltage volatility and power change rate output by the simplified dynamic model to obtain a predicted operation data set; using the predicted operation data set and the real - time operation data set to calculate an error index; updating the frequency band division rule and the frequency band weight coefficient according to the magnitude of the error index.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. Through dynamic feature hierarchical extraction and structure - parameter joint order reduction processing, the present invention significantly reduces the complexity of the model; while ensuring the simulation accuracy, it effectively reduces the calculation time consumption, and solves the problem of low calculation efficiency of traditional full - order models in high - dynamic - change scenarios.

[0019] 2. The present invention introduces a multi - objective optimization decision - making mechanism to optimize the game balance between accuracy and efficiency; through the matching of the Pareto optimization process and the typical operating condition scenario library, it can automatically generate highly adaptable optimal decision parameters under complex operating conditions, avoiding the loss of key dynamic characteristics caused by single - objective optimization.

[0020] 3. Based on the dynamic compensation factor correction technology of historical error data, the present invention enhances the robustness of the model; through the cross - time - scale error feedback mechanism, it can correct the order - reduction parameter deviation in real time, improve the adaptability of the simplified model to the non - steady - state conditions of the micro - grid, and solve the problem of long - term cumulative errors caused by static compensation in traditional methods.

[0021] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained through the structures pointed out in the specification, claims, and drawings. Brief Description of the Drawings

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0023] Figure 1 It is a schematic flow chart of a method for simplifying a microgrid dynamic model according to an embodiment of the present invention.

[0024] Figure 2 It is a schematic diagram of the three - band sensitivity distribution according to an embodiment of the present invention.

[0025] Figure 3 It is a schematic diagram of the Pareto optimization space according to an embodiment of the present invention.

[0026] Figure 4 It is a schematic diagram of the three - level fault - tolerant correction effect according to an embodiment of the present invention. Specific Embodiments

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0028] Refer to Figure 1 , an embodiment of the present invention proposes a method for simplifying a microgrid dynamic model. By using the dynamic feature hierarchical extraction and structure - parameter joint reduction technology, combined with the multi - objective game optimization and error adaptive compensation mechanism, it can significantly improve the calculation efficiency while ensuring the dynamic simulation accuracy and meet the requirements of different microgrid operating conditions.

[0029] The method of this embodiment specifically includes:

[0030] Obtain the real - time operation data set of the microgrid, and obtain the dynamic feature vector through dynamic feature hierarchical extraction;

[0031] Perform structure - parameter joint reduction processing on the dynamic feature vector to generate a reduced - order parameter set;

[0032] Execute multi - objective optimization decision - making based on the reduced - order parameter set to generate optimized decision parameters;

[0033] Generate a simplified dynamic model according to the optimized decision parameters;

[0034] Among them, the dynamic feature hierarchical extraction adopts a preset frequency band division rule, and the structure-parameter joint order reduction processing includes screening using a preset sensitivity threshold.

[0035] Specifically, frequency domain decomposition is performed on the real-time operation data through a preset frequency band division rule to distinguish the high-frequency transient components (such as the inverter switching frequency band) and the low-frequency steady-state components (such as the slow dynamics of power regulation) of the microgrid dynamic response characteristics. An adjustable sliding time window is used to adaptively adjust the sampling period. A short window is used in the high-frequency band to improve the time resolution, and a long window is used in the low-frequency band to enhance the feature stability. A dynamic generation model for the parameter sensitivity threshold is established, and the contribution index of each parameter is calculated based on the frequency band weight coefficient and historical error data. For high-frequency fast-changing parameters, key disturbance sources are retained through sensitivity threshold screening; for low-frequency slow-changing parameters, lossless retention is performed, and at the same time, a dynamic compensation factor is introduced to eliminate the energy loss caused by order reduction. Based on the accuracy-efficiency game model, a Pareto optimization space is constructed, and feature matching is performed in combination with the typical working condition scheme library. Through a three-level fault tolerance verification system, the convergence reliability of the model simplification process is ensured, and a positive feedback loop of parameter screening and decision optimization is realized. Through the fusion of data-driven and physical models, the technical framework limitation of the single-objective optimization of accuracy and efficiency in traditional order reduction methods is broken through, and the collaborative optimization of feature extraction, parameter order reduction, and decision optimization is realized.

[0036] Optionally, the obtaining of the real-time operation data set of the microgrid and the dynamic feature vector obtained through dynamic feature hierarchical extraction include:

[0037] Collect the voltage volatility and power change rate of the microgrid nodes to form an original feature matrix;

[0038] Intercept the original feature matrix according to a preset sliding time window to generate a windowed feature matrix, where the length of the sliding time window is adjusted according to a preset window adjustment coefficient;

[0039] Perform spectral feature extraction and non-steady state detection on the windowed feature matrix according to a preset frequency band division rule, and output a dynamic feature vector.

[0040] Specifically, first deploy voltage sensors and power monitoring devices to collect the real-time voltage volatility and power change rate of each node of the microgrid. Assume that the microgrid contains n nodes, and the real-time voltage volatility and power change rate collected at each sampling time point t form an n×2-dimensional vector, and the data of T consecutive sampling time points are stacked in chronological order to form an n×2×T-dimensional original feature matrix . Subsequently, a sliding time window is used to segment and intercept the original feature matrix, and the initial time window length is , dynamically adjust the window length according to the window adjustment coefficient. When high-frequency fluctuations are detected, the time window is shortened to improve the time resolution. In the low-frequency band, the time window is enlarged to enhance the feature stability. Finally, the windowed feature matrix is obtained. Then, Perform multi-band spectrum sensing, extract the mean square error of the amplitude of the feature matrix in the preset frequency band as the spectrum parameter through discrete Fourier transform, and the preset frequency band is divided according to the preset frequency band rules, such as low frequency <1Hz, medium frequency 1-10Hz, and high frequency >10Hz. At the same time, perform non-stationary detection on the data in the window, and identify the non-stationary mutation moment by calculating the peak of the second-order derivative of the signal. Finally, the spectrum parameters and the non-stationary identification value are combined into a dynamic feature vector .

[0041] For example, for a microgrid system consisting of five photovoltaic nodes and three energy storage nodes, real-time voltage fluctuation and power change rate data are collected at a 1kHz sampling frequency for each node. The original feature matrix is transformed into a 5×2×10,000 matrix within 10 seconds. The initial time window, L, is set to 50 sampling points (0.05 seconds). When the sliding variance of the power change rate exceeds a threshold, the window is shortened to 25 sampling points with a window adjustment factor of 0.5. When the fluctuation is gentle, the window adjustment factor is adjusted to 1.2, extending the window to 60 sampling points. Spectrum sensing divides the window data into three frequency bands: low frequency (0-0.5Hz), medium frequency (0.5-5Hz), and high frequency (5-50Hz). The energy contribution of the voltage fluctuation within each frequency band is calculated. In non-steady-state detection, the second-order derivative of the power change rate within the window is calculated, and the threshold is set to 3 times the standard deviation of the historical steady-state data. When the second-order derivative of three consecutive sampling points exceeds the set threshold, it is marked as a non-steady-state event. Finally, a 28-dimensional dynamic feature vector consisting of 5 nodes × 3 frequency band energy coefficients + non-steady-state flag bits is generated.

[0042] Optionally, performing structure-parameter joint order reduction processing on the dynamic feature vector to generate a set of reduced-order parameters includes:

[0043] Decomposing the dynamic feature vector into a fast-changing feature subset and a slow-changing feature subset;

[0044] Calculating the sensitivity of the variables in the fast-changing feature subset, eliminating the variables in the fast-changing feature subset whose sensitivity is lower than a preset sensitivity threshold, and generating a fast-changing reduced-order subset;

[0045] The variables in the fast-varying reduced-order subset are adjusted using a preset dynamic compensation factor, and combined with the slow-varying feature subset to obtain a reduced-order parameter set.

[0046] Specifically, such as Figure 2As shown, first, based on the spectral decomposition characteristics of the dynamic feature vector, the vector is divided into a fast-changing feature subset and a slow-changing feature subset. Specifically, by presetting a frequency band division threshold , for example , the feature entries with high-frequency components greater than or equal to in the spectral feature parameters are classified into the fast-changing feature subset , and the entries corresponding to the low-frequency components are classified into the slow-changing feature subset . Subsequently, a sensitivity value evaluation method is used to calculate the sensitivity of each variable in the fast-changing feature subset to the system dynamic response :

[0047] ,

[0048] where is the maximum deviation of the node voltage. A sensitivity threshold is set. When , the variable is determined to be a minor disturbance source and is excluded to form a fast-changing reduced-order subset . For the fast-changing reduced-order subset , a dynamic compensation factor is introduced

[0049] ,

[0050] where is the fluctuation variance of the excluded variable, is the total fluctuation variance. Based on the dynamic compensation factor, the weights of adjacent frequency (1Hz) components in the fast-changing reduced-order subset are increased. Finally, the slow-changing feature subset is merged with the compensated fast-changing reduced-order subset to generate a reduced-order parameter set .

[0051] Exemplarily, the dynamic feature vector of a microgrid containing 2 wind turbines and 4 energy storage units includes 8 groups of frequency band energy coefficients. Set , where the 4 feature items corresponding to the high-frequency components (5Hz, 7Hz) of the wind turbine nodes are classified into the fast-changing feature subset , and the low-frequency components (0.2Hz, 1Hz) of the energy storage units form the slow-changing feature subset . By calculating the sensitivity , it is found that the sensitivity of the 0.2Hz component of the energy storage unit is 0.05, which is lower than the preset sensitivity threshold of 0.1, and this feature item is excluded. The of the excluded feature is 0.003, and the total variance is 0.012. The dynamic compensation factor , enhance the fast-changing reduced-order subset according to the dynamic compensation factor The weights of adjacent frequency (1Hz) components in the medium. The final reduced-order parameter set contains 6 core feature terms. The coupling effect of the fast dynamics caused by the pitch regulation of the fan and the slow fluctuation characteristics of the grid frequency is accurately distinguished through modal separation. The sensitivity threshold is dynamically adjusted to avoid the interference of high-frequency noise modes on the model accuracy, and the compensation factor effectively makes up for the energy loss caused by feature screening. The collaborative reduced-order processing of fast and slow-changing features significantly improves the adaptability of the model to the fault ride-through scenario of the doubly-fed fan.

[0052] Optionally, the sensitivity threshold includes:

[0053] Extract the spectral feature parameters of the dynamic feature vector;

[0054] According to the spectral feature parameters and the preset band weight coefficients, calculate the contribution index of the variable;

[0055] Generate a sensitivity threshold according to the contribution index;

[0056] Among them, the update period of the sensitivity threshold is synchronized with the adjustment period of the sliding time window.

[0057] Specifically, first extract the energy ratio parameters of each frequency band from the dynamic feature vector , where k represents the frequency band index. Define the contribution index of a single variable :

[0058] ,

[0059] Among them is the preset band weight coefficient, represents the variable 's energy ratio in the k frequency band, is the total energy. For the sensitivity threshold , there is:

[0060] ,

[0061] Among them is the basic threshold coefficient, is the shape adjustment factor. When the variable contribution index increases, the sensitivity threshold exponentially decays. The above parameters 、 are all positively correlated with the current sliding time window length L and the window adjustment coefficient. When the time window is shortened, for the updated basic threshold coefficient , there is:

[0062] ,

[0063] Among them, is the original basic threshold coefficient, so as to ensure that the sensitivity threshold update period is completely synchronized with the time window adjustment.

[0064] Exemplarily, the dynamic characteristic vector of a three-phase inverter grid-connected node contains 3 frequency bands (0.5 - 2 Hz), (2 - 10 Hz), (10 - 50 Hz) energy proportions. Let the weight , calculate the contribution index as 0.425. At this time, the basic threshold , the current time window length is adjusted to , then the updated basic threshold of g , take the shape adjustment factor to obtain . At this time, if the sensitivity of a certain parameter is 0.16, which is less than the sensitivity threshold of 0.175, then this parameter will be eliminated. The contribution calculation based on the frequency band energy proportion can accurately reflect the dynamic influence differences of different variables in the key frequency bands. Through the anti-correlation dynamic threshold function, low-threshold protection of high-contribution parameters is realized, effectively avoiding the mis-elimination of important dynamic modes; the threshold parameters are updated in real time with the window adjustment, ensuring that the sensitivity criterion always matches the time resolution of the current analysis period. This adaptive mechanism in a microgrid containing multiple power sources of wind, light and energy storage can not only quickly reduce the slow dynamic redundant parameters of the energy storage system, but also completely retain the fast fluctuation characteristics of the photovoltaic output, ensuring the simulation accuracy of the dynamic model under variable time scales.

[0065] Optionally, the method further includes:

[0066] Extract the comparison data of the state variables of the reduced-order model and the full-order model from the preset historical operation log, and construct a historical reduced-order error data set;

[0067] Cross-correlate the historical reduced-order error data set with the spectral characteristic parameters, and calculate the gain coefficient;

[0068] Use the gain coefficient to correct the dynamic compensation factor.

[0069] Specifically, the construction of the historical reduced-order error data set needs to extract the comparison data of the state variables of the reduced-order model and the full-order model from the system historical operation log. Define the error term as the absolute percentage deviation of the voltage amplitude of the reduced-order model and the full-order model value on the same time section, and construct the error set according to the time stamp sequence . Construct an error feedback link. By associating the error data set with the spectral feature parameters of the dynamic feature vector, cross-time-scale feature matching is achieved. Specifically, extract the k main band energy weight coefficients in the dynamic feature vector at the current moment to form a band weight vector ; perform weighted moving average processing on the historical reduced-order error data set to generate a time-varying error memory sequence :

[0070] ,

[0071] where is the forgetting factor, with a range of 0.9 - 0.99, is the length of the time window. Establish the correlation matrix between the band weight vector and the error memory sequence , where ⊗ represents the Kronecker product, and output a k×m-dimensional correlation tensor. Perform bilinear pooling operation on the correlation tensor R to calculate the gain coefficient :

[0072] ,

[0073] where U and V are trainable parameter matrices for bilinear pooling operation, is the Sigmoid function, used to compress the gain coefficient to the interval (0, 1), and ⊙ represents element-wise multiplication. Apply the gain coefficient to the original dynamic compensation factor to obtain the corrected dynamic compensation factor :

[0074] ,

[0075] where is the current dynamic compensation factor, is the learning rate parameter, with a value range between 0.01 - 0.1. In this process, the cross-correlation between the spectral feature parameters and the historical errors can identify the model mismatch law in a specific frequency band. For example, when the errors in the high-frequency band continue to accumulate, the activation value of the corresponding channel in the correlation matrix increases, and the intensity of the dynamic compensation factor in this frequency band is increased through the gain coefficient to achieve error adaptive suppression.

[0076] Exemplarily, taking a microgrid with 3 groups of energy storage converters as an example, set the main band k = 3 (0.5 - 2Hz, 2 - 5Hz, 5 - 10Hz). When the average error in the 5 - 10Hz frequency band within 10 consecutive window periods, that is, the time-varying error memory sequence reaches 1.2%, the weight of this frequency band . After calculation by the correlation matrix, the corresponding channel is activated, generating a gain coefficient . Assume the original compensation factor , learning rate , the corrected compensation factor . By strengthening the high-frequency band compensation, the reduced-order model can still maintain the bus voltage simulation error below 0.5% when the photovoltaic output fluctuates rapidly. The dynamic association between error memory and spectral characteristics can adaptively locate the frequency-domain root cause of model mismatch. The bilinear pooling mechanism can decouple the coupling effects of errors in different frequency bands. The non-linear mapping of the gain coefficient avoids over-amplifying noise in single-band compensation, ensuring that the correction process of the compensation factor has frequency-domain selectivity and time-domain smoothness, and effectively improving the model accuracy margin in scenarios of drastic changes in wind and light output.

[0077] Optionally, performing multi-objective optimization decision based on the set of reduced-order parameters to generate optimization decision parameters includes:

[0078] Constructing an accuracy-efficiency game model, where the accuracy-efficiency game model uses the set of reduced-order parameters as input variables;

[0079] The accuracy-efficiency game model performs a multi-stage Pareto optimization process and matches features with a preset typical operating condition scenario library to generate preliminary decision parameters;

[0080] Performing three-level fault tolerance correction on the preliminary decision parameters and outputting optimization decision parameters.

[0081] Specifically, as Figure 3 shown, first construct an accuracy-efficiency game model, map the key parameters in the set of reduced-order parameters to optimization input variables, and define the model voltage error rate index and the calculation time-consuming index . Secondly, perform multi-stage Pareto optimization. In the first stage, minimize and as the optimization direction to generate an initial Pareto solution set; in the second stage, superimpose the topology change response time index to form a three-dimensional objective space and adjust the dominance level of the solution set through adaptive weight adjustment. Finally, perform three-level fault tolerance correction on the Pareto front solution: in the first step, check whether the voltage error rate exceeds the preset residual threshold; if it exceeds, activate the steady-state residual compensation module and inject correction current into the key nodes; if it still does not meet the requirement after compensation, trigger the full-order model reconstruction and quickly match the reference model parameters through the typical operating condition scenario library. Among them, the matching logic of the typical operating condition scenario library is to calculate the cosine similarity of the feature fingerprints between the set of reduced-order parameters and the scenarios in the library, and select the scenario with the highest similarity and the lowest historical error as the reference benchmark to generate preliminary decision parameters. Perform three-level fault tolerance correction on the preliminary decision parameters and output optimization decision parameters.

[0082] Exemplarily, for a microgrid containing a wind-solar-storage hybrid system, the reduced-order parameter set contains 8 key variables. First, it is input into the accuracy-efficiency game model, and the target weights [0.6, 0.3, 0.1] are set according to the operating data of the recent 24 hours. The first-stage optimization generates 15 sets of Pareto solutions, and the optimal solution is at the balance point of an error rate of 2.1% and a time consumption of 0.8 seconds; in the second stage, the topological change response time constraint is superimposed After seconds, the optimization space shrinks to 5 sets of solutions. The scheme matching engine detects that the similarity between the current dynamic feature vector and the fingerprint code of the typhoon Tian Gao fluctuation condition reaches 0.82, and calls the benchmark parameters of this condition for cross-verification. During the three-level fault tolerance correction, it is detected that the voltage error rate of a certain energy storage node reaches 3.5%, and an injection compensation current of 0.3 is triggered. After correction, the error rate drops to 1.2%. At this time, the model successfully passes the impedance matching test, and a decision set containing 6 optimized parameters is output. The staged optimization process effectively balances various conflicting objectives. The typical condition matching mechanism quickly locks the optimal decision area, and the three-level fault tolerance system classifies and processes model deviations of different severities, ensuring the robustness of the decision parameters under complex conditions. By introducing the topological response time constraint, the sacrifice of the system's dynamic regulation ability due to simply pursuing accuracy is avoided, and the progressive processing of residual compensation and full-order reconstruction ensures that the model is still recoverable under fault conditions.

[0083] Optionally, the execution of the multi-stage Pareto optimization process includes:

[0084] In the first stage, construct the first objective pair of voltage error rate and calculation time consumption;

[0085] In the second stage, superimpose the topological change response time to form a three-objective optimization space;

[0086] Apply a preset algorithm to obtain an iterative solution set in the three-objective optimization space;

[0087] Match the features of the iterative solution set with a preset typical condition scheme library to generate preliminary decision parameters.

[0088] Specifically, in the construction stage of the first-stage objective pair, define the voltage error rate and the calculation time consumption optimization objectives. The calculation formula for the voltage error rate is:

[0089] ,

[0090] where is the simulated voltage value of the kth node, is the measured voltage value of the kth node, m is the number of key nodes, is the reference voltage. The calculation time consumption objective is defined as:

[0091] ,

[0092] In the formula, is the single-iteration calculation time. When initializing the population, n parameters in the reduced-order parameter set are used as chromosome coding, each parameter corresponding to a gene position, and the value range of the parameters is dynamically constrained by the historical data in the typical operating condition scenario library.

[0093] In the second stage, the three-objective expansion stage, the topological change response time index is introduced :

[0094] ,

[0095] is the current stabilization time after the topological change of the i-th branch. At this time, the objective space is expanded from two dimensions to three dimensions, and a three-objective optimization space is constructed to simultaneously optimize the voltage error rate, calculation time consumption, and topological change response time. The preset algorithm can be a non-dominated sorting algorithm, such as an improved non-dominated sorting genetic algorithm, to obtain an iterative solution set in the three-objective optimization space. By introducing an adaptive weight factor, the priorities of each objective are dynamically adjusted to ensure finding the Pareto optimal solution in a complex multi-objective environment. The iterative solution set contains multiple equilibrium solutions, achieving different trade-offs between the voltage error rate, calculation time consumption, and topological change response time. The iterative solution set is input into a preset scheme matching engine, and feature matching is performed in combination with the typical operating condition scenario library to generate preliminary decision parameters. The scheme matching engine screens out the optimal parameter combination by calculating the similarity between the iterative solution and the historical typical operating condition scenarios and outputs the preliminary decision parameters.

[0096] Exemplarily, consider a microgrid system with 10 nodes, 2 photovoltaic power generation units, and 1 energy storage system. In the first stage, the voltage error rate and the calculation time are defined as optimization objectives. Through 50 iterations of the improved non-dominated sorting algorithm, 20 equilibrium solutions are obtained. These solutions achieve a balance between a voltage error rate of 0.5% to 1% and a calculation time of 0.1 to 0.5 seconds. In the second stage, the topology change response time is superimposed as the third objective to form a three-objective optimization space. The algorithm is further iterated 30 times to obtain 15 Pareto optimal solutions. Among them, the optimal solution has a voltage error rate of 0.8%, a calculation time of 0.25 seconds, and a topology change response time of 0.15 seconds. Finally, through the scheme matching engine, the parameter combination with the lowest voltage error rate and reasonable calculation time and topology response time is screened out and output as the preliminary decision-making parameters. Through phased optimization, it is possible to take into account model accuracy, calculation efficiency, and system dynamic response ability simultaneously, avoiding performance imbalance caused by a single optimization objective. The improved non-dominated sorting algorithm shows high efficiency and robustness in multi-objective optimization, ensuring the diversity and high quality of Pareto optimal solutions. At the same time, the scheme matching engine combines historical operating condition data to effectively improve the applicability and adaptability of the parameter combination. The finally output preliminary decision-making parameters significantly improve the simulation accuracy and calculation efficiency of the microgrid dynamic model in a complex and changeable environment.

[0097] Optionally, the three-level fault tolerance correction of the preliminary decision-making parameters and the output of the optimized decision-making parameters include:

[0098] Performing critical constraint verification on the preliminary decision-making parameters to generate a verification result, where the critical constraint verification includes impedance matching inspection and power conservation inspection;

[0099] Triggering residual compensation according to the verification result and performing the critical constraint verification again to generate a compensation result;

[0100] Generating a full-order model reconstruction instruction according to the compensation result.

[0101] Specifically, as Figure 4As shown, a three-level fault-tolerant correction operation is implemented for the preliminary decision parameters. First, critical constraint verification is performed on the preliminary decision parameters. This verification includes two criteria: impedance matching and power conservation. For the impedance matching test, the deviation rate between the theoretical equivalent impedance value of each connected node and the value estimated by the reduced-order model is calculated. If the deviation rate at any node exceeds a threshold, the test is considered a failure. The power conservation test verifies the residual between the total system input power and the sum of the load power and loss power. If the absolute value of the residual exceeds a set threshold, a failure flag is triggered. The residual threshold is set using a sliding window statistical method. The standard deviation of the historical residual data over a window period of T is extracted, and three times the standard deviation of the historical residual data is used as the trigger threshold. When the critical constraint verification result triggers the residual threshold, the steady-state residual compensation module is loaded. Its logic injects a corrected current vector into the impedance mismatch nodes based on the current list of nodes that have failed verification, and adjusts the virtual energy storage compensation power of the power residual nodes. The power compensation amount is generated by multiplying the residual by a compensation coefficient, which is dynamically adjusted based on the node type. If the power residual still exceeds the threshold after three consecutive compensation attempts, or if the single compensation amount exceeds the device's safety limit, the compensation result is considered a failure. If compensation fails, the full-order model reconstruction command is activated, calling the baseline model parameters from the typical operating condition solution library with the closest signature fingerprint, re-calculating the reduced-order parameters, and updating the decision parameters. During the reconstruction process, the solution library parameters are verified for authenticity using a hash chain to ensure they have not been tampered with.

[0102] For example, after applying preliminary decision parameters, a microgrid detected that the impedance deviation rate at node 5 consistently exceeded 2%, while the total power residual reached 5%. The system activated the steady-state residual compensation module, injecting a compensation current of 0.3 into node 5 and adding 2 kW of virtual compensation power to the bus. Monitoring revealed that the impedance deviation remained at 1.5% after two compensation cycles, triggering a full-order model reconstruction. Fingerprint matching was used to reconstruct the reduced-order model using typhoon-condition benchmark parameters. Ultimately, the impedance deviation rate at node 5 was reduced to 0.3%, and the power residual was less than 0.5%. The beneficial effects of this verification example are reflected in the hierarchical handling of different levels of model mismatch and the adoption of a progressive fault-tolerance strategy to ensure model reliability while minimizing the cost of system reconstruction. Impedance matching verification effectively mitigates equivalent model deviations caused by device parameter aging, dynamic residual compensation mitigates the computational burden of frequent reconstructions, and blockchain verification ensures the legitimacy of the reconstructed parameters. A multi-level parameter verification and compensation mechanism was established, accurately identifying the root cause of model mismatch through critical constraint verification, addressing transient parameter deviations using steady-state residual compensation, and addressing systemic model failures through full-order reconstruction. The technical effect is reflected in significantly improving the dynamic adaptability of the model, reducing the risk of false operation, and ensuring simulation reliability under extreme working conditions.

[0103] Optionally, the construction of the typical operating condition solution library includes:

[0104] Collect the benchmark model parameters under various operating modes in the typical operating condition scenario library to form an original scenario set;

[0105] Extract the feature fingerprints from the original scenario set to generate fingerprint coding vectors;

[0106] Bind the fingerprint coding vectors to the hash values of the preset blockchain storage module;

[0107] Perform preloading matching according to the similarity between the dynamic feature vectors and the fingerprint coding vectors during the real-time operation process.

[0108] Specifically, first, collect the benchmark model parameters under various operating modes to form an original scenario set. The benchmark model parameters refer to the key parameters that can accurately describe the dynamic characteristics of the microgrid under different operating conditions. These parameters usually include information such as node voltage, power flow, and frequency response. The collection process can be achieved by deploying a sensor system and data acquisition devices to ensure sufficient data samples are obtained under different operating modes (such as full-load operation, partial-load operation, fault conditions, etc.). Secondly, extract the feature fingerprints from the original scenario set to generate fingerprint coding vectors. Feature fingerprint extraction refers to processing the benchmark model parameters through algorithms to extract a unique identifier that can represent this operating mode. The fingerprint coding vectors can be encoded using methods such as support vector machine (SVM) and deep learning to ensure that each feature fingerprint can accurately reflect a specific operating mode. For example, parameters such as voltage volatility, power change rate, and frequency response of each operating mode can be combined into a high-dimensional vector, and a fingerprint coding vector can be generated through a dimensionality reduction algorithm (such as principal component analysis PCA). Next, bind the fingerprint coding vectors to the hash values of the blockchain storage module. Blockchain technology provides an immutable data storage and management system. Each fingerprint coding vector is assigned a unique hash value, which is embedded in the blockchain to ensure the security and integrity of the data. Specifically, the data of the fingerprint coding vectors can be encrypted and stored in the blocks of the blockchain. Each block has a unique and sequentially related hash value to ensure the data cannot be tampered with. Finally, during the real-time operation process, perform preloading matching according to the similarity between the dynamic feature vectors and the fingerprint coding vectors. Preloading matching means that during actual operation, the dynamic feature vectors collected in real time are used to calculate the similarity with the fingerprint coding vectors stored in the blockchain, so as to quickly match the closest operating condition mode. The similarity calculation can use methods such as cosine similarity and Euclidean distance. For example, when the cosine similarity between the dynamically collected feature vectors and the stored fingerprint coding vectors is close to 1, it means that the two have a high similarity, and the corresponding benchmark model parameters can be quickly preloaded.

[0109] Exemplarily, assume there is a microgrid system with multiple operating condition modes, and the operating modes include normal operation, sudden load increase, fault recovery, etc. First, collect parameters such as voltage, power, and frequency under each operating mode to form an original reference model parameter set. Then, perform feature extraction on these parameter sets to generate corresponding fingerprint coding vectors. For example, obtain the feature vectors of each operating condition through PCA dimensionality reduction. Next, store these fingerprint coding vectors in the blockchain, and each vector corresponds to a hash value. During real-time operation, collect the dynamic feature vectors of the current operating condition, calculate their similarity with the fingerprint coding vectors stored in the blockchain, and select the most similar operating condition mode for preloading and matching of parameters. Thus, the construction process of the typical operating condition scheme library is completed. By combining the feature fingerprints with the blockchain, a secure and reliable typical operating condition scheme library is constructed, which can quickly match and preload the optimal reference model parameters. This method not only improves the efficiency of model update but also ensures the security of data, avoiding model errors caused by data tampering or loss. At the same time, the immutability of the blockchain ensures the long-term effectiveness of the operating condition mode, providing a solid foundation for the dynamic model optimization of the microgrid.

[0110] Optionally, the method further includes:

[0111] Collect the voltage volatility and power change rate output by the simplified dynamic model to obtain a predicted operation data set;

[0112] Use the predicted operation data set and the real-time operation data set to calculate an error index;

[0113] Update the frequency band division rule and the frequency band weight coefficient according to the magnitude of the error index.

[0114] Specifically, implementing the method includes the following steps: First, collect the voltage volatility and power change rate output by the simplified dynamic model to form a predicted operation data set. Synchronously obtain the actual operation data set of the microgrid, which includes the measured voltage volatility and power change rate. Align the predicted values with the actual values in time, calculate the voltage absolute error and power absolute error at each sampling point to obtain an error sequence. Define a sliding error statistical window with a window period of T, and calculate the window average error mean of the error sequence within the window period 、 and variance 、 . For the error index , there is:

[0115] ,

[0116] where 、 are the voltage and power error weight coefficients, where And According to the error index Compare with the preset error level standard, and trigger the update of the frequency band division rule. The update logic of the frequency band division rule is: when Exceed the threshold For the events that occur, when more than 3 consecutive window periods are exceeded, adjust the original number of frequency band divisions proportionally. Specifically, expand the original m frequency bands to m + 1 frequency bands, and the new frequency band boundary point is located at the center frequency of the original maximum error frequency band. The frequency band weight coefficient The update is determined by the proportion of the error index in each frequency band. For the new frequency band weight coefficient , there is:

[0117] ,

[0118] Among them, Is the old frequency band weight coefficient, Is the average error increment of the k-th frequency band in the most recent window period, Is the learning rate, and the value range is 0.1 - 0.3.

[0119] Exemplarily, a microgrid includes 6 photovoltaic nodes and 2 groups of energy storage systems. The preset initial frequency band division is three frequency bands: low frequency 0 - 2Hz, medium frequency 2 - 5Hz, and high frequency 5 - 10Hz. The predicted operation data after the acquisition model is simplified shows that the error index in the high frequency band continuously exceeds the threshold within 5 window periods, triggering the update of the frequency band division rule. The newly added segmentation point 7.5Hz splits the high frequency band into 5 - 7.5Hz and

[0120] 70% of the total error of the segment, so the weight coefficient of this sub - frequency band is adjusted from the initial 0.3 to 0.45. Verification shows that the updated frequency band division can more finely track the 7.8Hz frequency oscillation caused by the pitch control of the fan, and improve the model voltage prediction accuracy by about 40% under this working condition. By establishing a dynamic feedback mechanism between the error index and the frequency band parameters, the online identification of the frequency domain characteristics of the model error is realized; the multi - window error statistics avoid mis - adjustment triggered by single - time abnormal fluctuations; the adaptive update of the weight coefficient strengthens the feature extraction ability for high - error frequency bands, while ensuring the complete retention of the steady - state characteristics of the low - frequency band; the newly added frequency band division effectively captures the narrow - band resonance phenomenon missed by the traditional wide - band coverage.

[0121] It should be noted that for the formulas mentioned above, through the principle of dimensional consistency and mathematical standardization means (such as normalization, conversion of dimensionless parameters, or unification of unit systems), physical quantities of different attributes can be translated into unitless standard values or superposable parameters of the same dimension, thereby eliminating the interference of different dimensions on the operation logic and enabling the formulas to have mathematical operation rationality and objective law adaptability while retaining the characteristics of the original data distribution. This is a conventional technical means and will not be elaborated here. The electrical connections between the above-mentioned various units do not necessarily represent direct connections of the circuits. Indirect connection methods can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above-mentioned are only exemplary embodiments of the present invention and cannot be used to limit the scope of the present invention.

[0122] That is, any equivalent changes and modifications made in accordance with the teachings of the present invention still fall within the scope covered by the present invention. Those skilled in the art will easily think of other implementation schemes of the present invention after considering the specification and the disclosure of the practical truth. This application aims to cover any variations, uses, or adaptive changes of the present invention, and these variations, uses, or adaptive changes follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not recorded in the present invention.

Claims

1. A method for simplifying the dynamic model of a microgrid, characterized in that, The method further includes: Obtaining a real-time operation data set of the microgrid and obtaining a dynamic feature vector through hierarchical extraction of dynamic features; wherein, obtaining the dynamic feature vector includes: collecting the voltage volatility and power change rate of the microgrid nodes to form an original feature matrix; intercepting the original feature matrix according to a preset sliding time window to generate a windowed feature matrix, wherein the length of the sliding time window is adjusted according to a preset window adjustment coefficient; performing spectral feature extraction and non-steady state detection on the windowed feature matrix according to a preset frequency band division rule, and outputting a dynamic feature vector; Performing structure-parameter joint order reduction processing on the dynamic feature vector to generate a set of reduced-order parameters; wherein, generating the set of reduced-order parameters includes: decomposing the dynamic feature vector into a fast-changing feature subset and a slow-changing feature subset; calculating the sensitivity of the variables in the fast-changing feature subset, removing the variables in the fast-changing feature subset with sensitivity lower than a preset sensitivity threshold to generate a fast-changing reduced-order subset; adjusting the variables in the fast-changing reduced-order subset by using a preset dynamic compensation factor, and combining with the slow-changing feature subset to obtain a set of reduced-order parameters; Performing multi-objective optimization decision-making based on the set of reduced-order parameters to generate optimized decision-making parameters; wherein, generating the optimized decision-making parameters includes: constructing an accuracy-efficiency game model, and using the set of reduced-order parameters as an input variable for the accuracy-efficiency game model; the accuracy-efficiency game model performs a multi-stage Pareto optimization process and performs feature matching with a preset typical working condition scenario library to generate preliminary decision-making parameters; performing three-level fault tolerance correction on the preliminary decision-making parameters and outputting optimized decision-making parameters; Generating a simplified dynamic model according to the optimized decision-making parameters; Wherein, the hierarchical extraction of dynamic features adopts a preset frequency band division rule, and the structure-parameter joint order reduction processing includes screening by using a preset sensitivity threshold.

2. A method for simplifying a microgrid dynamic model according to claim 1, characterized in that, The sensitivity threshold includes: Extracting spectral feature parameters of the dynamic feature vector; Calculating a contribution index of a variable according to the spectral feature parameters and a preset frequency band weight coefficient; Generating a sensitivity threshold according to the contribution index; Wherein, the update period of the sensitivity threshold is synchronized with the adjustment period of the sliding time window.

3. A method for simplifying a microgrid dynamic model according to claim 2, characterized in that The method further includes: Extracting state variable comparison data between the reduced-order model and the full-order model from a preset historical operation log to construct a historical reduced-order error data set; Performing cross-correlation on the historical reduced-order error data set and the spectral feature parameters to calculate a gain coefficient; Correcting the dynamic compensation factor by using the gain coefficient.

4. A method for simplifying a microgrid dynamic model according to claim 1, characterized in that Performing a multi-stage Pareto optimization process includes: Constructing a first objective pair of voltage error rate and calculation time-consuming in the first stage; Superimposing the topology change response time in the second stage to form a three-objective optimization space; Applying a preset algorithm to obtain an iterative solution set in the three-objective optimization space; Performing feature matching on the iterative solution set and a preset typical working condition scenario library to generate preliminary decision-making parameters.

5. A method for simplifying a microgrid dynamic model according to claim 4, characterized in that, The performing three-level fault tolerance correction on the preliminary decision-making parameters and outputting optimized decision-making parameters includes: Perform critical constraint verification on the preliminary decision parameters to generate a verification result, where the critical constraint verification includes impedance matching inspection and power conservation inspection; Trigger residual compensation according to the verification result and perform the critical constraint verification again to generate a compensation result; Generate a full-order model reconstruction instruction according to the compensation result.

6. A method for simplifying a microgrid dynamic model according to claim 1, characterized in that, The construction of the typical operating condition scenario library includes: Collect the benchmark model parameters under multiple operating modes to form an original scenario set; Extract the feature fingerprints of the original scenario set to generate a fingerprint coding vector; Bind the fingerprint coding vector to the hash value of the blockchain storage module; Perform preloading matching according to the similarity between the dynamic feature vector and the fingerprint coding vector during the real-time operation process.

7. A method for simplifying a microgrid dynamic model according to claim 6, characterized in that, The method further includes: Collect the voltage volatility and power change rate output by the simplified dynamic model to obtain a predicted operation data set; Calculate an error index by using the predicted operation data set and the real-time operation data set; Update the frequency band division rule and the frequency band weight coefficient according to the magnitude of the error index.

Images