Microgrid cluster dimension reduction method and system based on scene self-adaption and topology maintenance

Abstract

The invention discloses a micro-grid cluster dimension reduction method and system based on scene self-adaption and topology maintenance. Aiming at different types of energy, constructing a steady-state characteristic index system and decoupling the steady-state characteristic index system into a plurality of groups of characteristics; the method comprises the following steps: clustering system time series data, constructing an operation state manifold diffusion matrix, clustering by using a fuzzy clustering method based on diffusion distance, and outputting an optimal scene division result; constructing a Wasserstein distance matrix by using the steady-state characteristic index of each type of energy, defining a scene objective function and solving an optimal index weight; and calculating a scene membership degree and a fusion index vector weight according to the real-time operation state, weighting the steady-state characteristic index system, and outputting a weighted characteristic matrix. And performing multi-scale topology analysis on the weighted feature matrix, constructing a topology stability constraint, and aggregating the power grid equipment by adopting a hierarchical clustering method based on the topology stability constraint. According to the technical scheme, the internal evolution rule of the operation state can be accurately captured, so that the aggregation result better meets the actual operation requirement.

Description

Technical Field

[0001] This application belongs to the field of distributed energy and energy storage model simplification technology, specifically involving a microgrid cluster dimensionality reduction method and system based on scenario adaptation and topology preservation. Background Technology

[0002] With the rapid development of distributed energy and new energy technologies, microgrid systems integrate various energy components, including gas turbines, energy storage devices, and photovoltaics, making system structures and operating conditions increasingly complex. To achieve precise optimization, it is usually necessary to model and make decisions for each physical component separately. This directly leads to an exponential expansion of the number of decision variables in the scheduling model with the scale of the components, resulting in a huge computational burden and severely restricting the real-time optimization and scheduling capabilities of large-scale microgrid clusters. In other words, there is a contradiction between refined modeling and the limited computing resources.

[0003] To alleviate the above contradictions, model dimensionality reduction and equivalent modeling techniques are considered key solutions. However, the following defects still exist in the existing technology: (1) Existing methods lack a unified standardized feature system to describe multiple types of energy components. This leads to poor comparability between devices, providing an inconsistent and unfair data foundation for subsequent clustering. (2) Scene division relies on traditional clustering algorithms (such as the patent application with publication number CN120893318A). Such methods cannot capture the real evolution law of microgrid operation data in high-dimensional space, resulting in scene division distortion and membership jumps at the boundary. (3) When measuring device similarity, static measurement criteria decoupled from the operation scenario are generally adopted (such as the dimensionality reduction projection method in the patent application with publication number CN120582125A). The process is static and does not consider the dynamic impact of different operation scenarios on the device aggregation criteria. This leads to the aggregation results failing to truly reflect the current scheduling needs and poor model adaptability. (4) The clustering process relies entirely on mathematical features and does not consider the actual physical connection relationship of devices in the power grid. This can easily lead to topological errors (such as generating electrically disconnected equivalent cells), making power flow calculations and stability analyses based on this unreliable and even threatening grid security. Summary of the Invention

[0004] This application aims to overcome the shortcomings of existing technologies and provide a microgrid cluster dimensionality reduction method and system based on scene adaptation and topology preservation. This method can adaptively aggregate devices based on their inherent electrical characteristics, thereby significantly reducing the computational complexity of the optimization model while ensuring model accuracy.

[0005] The first aspect of this application discloses a microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation, which adopts the following technical solution: For fully controlled energy, semi-controlled energy and uncontrollable energy, corresponding steady-state characteristic index systems are constructed respectively, and the steady-state characteristics of each type of energy are decoupled into multiple sets of characteristic indexes. Time-series data of microgrid cluster systems are collected and statistical features are extracted to construct an operational manifold diffusion matrix. The time-series data are clustered using a fuzzy clustering method based on diffusion distance, in which entropy regularization is introduced to smooth the cluster boundaries, a joint objective function is established and solved, and the optimal scenario partitioning result is output. Based on the steady-state characteristic index system of each type of energy, a Wasserstein distance matrix is ​​constructed; a scenario-based objective function is defined to solve and output the optimal index weights; the scenario membership degree is output according to the real-time operating status, the index vector weights are fused with the scenario membership degree, and then weighted with the steady-state characteristic index system to output the weighted feature matrix of each equipment type. Multi-scale topology analysis is performed on the weighted feature matrix to calculate the critical point of structural abrupt change and the topological stability interval; a hierarchical clustering method is constructed to aggregate real power grid equipment into multiple equivalent equipment and calculate the parameters of the equivalent equipment, and finally output the equivalent topology.

[0006] Furthermore, time-series data of the microgrid cluster system are collected and statistical features are extracted from it, including: The time-series data includes sampled data from all devices within the microgrid cluster. After time alignment, the data from devices of the same type are added point by point to output cluster behavior curves, including total load curve, total photovoltaic output curve, total gas turbine output curve, and average energy storage SOC curve. Statistical features are extracted from the cluster behavior curves and concatenated with time feature codes to generate a sample feature vector; a running manifold diffusion matrix is ​​constructed based on the feature vectors of all M samples.

[0007] Furthermore, the process of clustering the time-series data includes: The joint objective function consists of a diffusion distance term and an entropy regularization term. The diffusion distance term is represented as follows, for data points Calculate its distance to the cluster center The square of the diffusion distance, and the square of the membership degree. Multiply to obtain the sample To the scene Clustering driver terms; The diffusion distance term is obtained by doubly summing the clustering driving terms of all samples to all scenarios.

[0008] Furthermore, the joint objective function is solved using an alternating optimization strategy, including: Step 1: Fix the scene center of all scenes, and update the membership degree using the closed-form solution of the joint objective function; Step 2: Fix the membership degree of all samples, solve the Frescher mean using gradient descent, and update the scene center; Step 3: After completing Step 1 and Step 2, calculate the joint objective function value of the current iteration; if the change in the joint objective function value is less than the convergence threshold, terminate the iteration and output the final membership degree and scene center value; if the change in the joint objective function value is greater than or equal to the convergence threshold, return to Step 1 to continue the iteration.

[0009] Furthermore, after completing the iteration, the joint objective function is solved and output. Each scene center Scene membership matrix ; Use the elbow rule to select The optimal scenario.

[0010] Furthermore, the Wasserstein distance matrix is ​​constructed, including: For any type of energy, the steady-state characteristic index system is organized into a device characteristic matrix. After normalizing the device characteristic matrix, the Wasserstein distance between any two devices in the matrix is ​​calculated, and the entropy regularization of the Sinkhorn algorithm is called to solve the problem. Combine the Wasserstein distances between all devices in the matrix to form a Wasserstein distance matrix; the matrix contains the... Line 1 Column elements That is, equipment and equipment The Wasserstein distance between them is defined as 0 for the diagonal elements.

[0011] Furthermore, the process of solving for the optimal index weights includes: Design a scenario-based objective function This indicates the maximum value including the inter-group equipment distance item minus the intra-group equipment distance item; The inter-group equipment distance item is The sum of weighted Wasserstein distances between all devices in the scenario, where the intra-group device distance term is... The sum of weighted Wasserstein distances for all devices within the group in the scene; where, The scenario-based objective function also includes an entropy regularization term for the weights. Optimize the solution of the scenario-based objective function and output the result. The optimal metric weight vector in this scenario is represented as: Final output The optimal index weight vector.

[0012] Furthermore, the steps for determining between-group and intra-group equipment include: An adaptive membership threshold is defined using the mean and standard deviation of memberships across all optimal scenarios. ; From the scene Select the one with a membership degree greater than The device calculates its average Wasserstein distance to the scene center, and the average Wasserstein distance and the corresponding scene... Characteristic radius ; For the scenario equipment in and equipment Wasserstein distance Then determine the device and equipment If it is an intra-group device, then it is an inter-group device.

[0013] Furthermore, the indicator vector weights are integrated with the scene membership, including: Identified based on real-time operating status The membership degree of the optimal scenario is expressed as: ; This Each membership degree and its corresponding The optimal index weight vectors of each index are multiplied separately and then weighted together to obtain the fused index weight vector. .

[0014] Furthermore, the fusion index weight vector is weighted with the steady-state characteristic index system, including: Multiple sets of feature indicators for each device in each type of energy are concatenated into a feature device vector. The feature device vector is then multiplied item by item with the fusion indicator weight vector of its respective scenario to output a weighted feature vector for each device. Stack the weighted feature vectors of all devices of the same device type row by row to obtain the weighted feature vector matrix.

[0015] Furthermore, the determination methods for the structural abrupt change critical point and the topologically stable interval include: The persistent homology method is used to perform multi-scale analysis on the weighted eigenvector matrix, and the Betti values ​​and Betti curves are output. Differentiating the Betti curve and identifying the local maximum of the absolute value of the derivative as the critical point of structural abrupt change, and setting the absolute value of the derivative below a preset threshold... The region is defined as the topologically stable interval.

[0016] Furthermore, the hierarchical clustering method includes: The score calculated based on the merge priority function determines the merge priority between any two device clusters. The device clusters with the highest merge priority score are selected for merging, and the cluster information is updated. The merging priority function is the product of feature similarity, electrical connection enhancement factor and topology penalty term; The feature similarity is the feature distance between two device clusters, calculated using the centroid connection method based on the weighted feature vector matrix. The electrical connection enhancement factor is the product of the physical connection relationship and the weighting coefficient plus 1. When there is a connection between two device clusters, the corresponding physical connection relationship is 1, otherwise it is 0.

[0017] The topology penalty term is a piecewise function. When two device clusters merge and cross a structural abrupt change threshold, the penalty term is 0; otherwise, it is 1.

[0018] Furthermore, the termination conditions for the hierarchical clustering include a primary criterion and alternative criters, including: The main criterion is that the hierarchical clustering is terminated when the topology penalty term between the two device clusters to be merged in the next step is 0. The alternative criterion is that hierarchical clustering is terminated when the number of clustering device clusters reaches the target value.

[0019] Furthermore, the equivalent topology is output, including: After the hierarchical clustering terminates, the output is... Each equipment cluster is considered as an equivalent device; parameters of similar devices in each energy type are aggregated based on the equivalent devices. In a microgrid cluster, an equivalent connection line is established between two equivalent devices that have at least one connection line. The impedance of the equivalent connection line is obtained by equivalent calculation using the original line impedances of all the original lines connecting the two equivalent devices, thus generating an equivalent topology.

[0020] The second aspect of this application discloses a microgrid cluster dimensionality reduction system based on scene adaptation and topology preservation, which implements the microgrid cluster dimensionality reduction method as described in the first aspect of this application. The system includes: Steady-state characteristic index construction module; used to construct corresponding steady-state characteristic index systems for fully controllable energy, semi-controllable energy and uncontrollable energy respectively, and decouple the steady-state characteristics of each type of energy into multiple sets of characteristic indices; The scenario segmentation module is used to collect time-series data of microgrid cluster systems and extract statistical features from them to construct the operating manifold diffusion matrix. It then clusters the time-series data using a fuzzy clustering method based on diffusion distance, where entropy regularization is introduced to smooth the cluster boundaries, and a joint objective function is established and solved to output the optimal scenario segmentation result. The equipment feature scenario coupling module is used to construct the Wasserstein distance matrix based on the steady-state feature index system of each type of energy; define the scenario-based objective function to solve and output the optimal index weights; output the scenario membership degree according to the real-time operating status, fuse the index vector weights with the scenario membership degree, and then weight them with the steady-state feature index system to output the weighted feature matrix of each equipment type. Clustering and dimensionality reduction module: used to perform multi-scale topology analysis on weighted feature matrices, calculate the critical point of structural abrupt change and the topological stability interval; construct hierarchical clustering method to aggregate real power grid equipment into multiple equivalent equipment and calculate the parameters of the equivalent equipment, and finally output the equivalent topology.

[0021] The beneficial effects of this application are that, compared with the prior art, this application: Through systematic technical improvements, significant enhancements have been achieved in the accuracy, adaptability, and physical realism of microgrid cluster equivalent modeling. Specifically: First, the constructed unified multi-dimensional feature index system solves the problem of inconsistent feature representations of heterogeneous devices, providing a standardized data foundation for subsequent analysis. Second, the scene partitioning method based on manifold diffusion and entropy regularization can more accurately capture the inherent evolutionary laws of operating states and ensure smooth and stable scene transitions. Third, the proposed scene-driven Wasserstein metric learning mechanism enables adaptive adjustment of device similarity evaluation criteria from static to dynamic, making the aggregation results more aligned with actual operational needs. Fourth, the introduced topological stability constraint method ensures that the key connectivity structure of the power grid is effectively maintained during clustering, fundamentally avoiding model unavailability issues caused by topological distortion. These improvements collectively ensure that the final equivalent model possesses both mathematical compactness and physical realism. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a microgrid cluster dimensionality reduction method based on scenario adaptation and topology preservation. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. The embodiments described in this application are merely some embodiments of this application, and not all embodiments. Based on the spirit of this application, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this application.

[0024] As an embodiment of this application, a specific implementation method for a microgrid cluster steady-state model dimensionality reduction method based on electrical characteristics is disclosed. The execution flow of the method embodiment is as follows: Figure 1 .

[0025] S1: As one implementation method of this embodiment, a multi-dimensional steady-state characteristic index system is constructed.

[0026] To address the challenges of clustering caused by inconsistent dimensions and chaotic time scales of electrical characteristic indicators for microgrid equipment in existing technologies, this application proposes a unified equipment characteristic indicator system oriented towards steady-state analysis. This system constructs scalarized feature vectors with clear physical meaning and consistent data structures for fully controllable, semi-controllable, and uncontrollable energy sources, providing a solid data foundation for subsequent model dimensionality reduction using a unified advanced clustering algorithm.

[0027] In distributed energy systems within microgrids, fully controlled energy refers to traditional power generation equipment and energy storage systems that can precisely regulate output power, such as gas / diesel generators and energy storage batteries; semi-controlled energy refers to energy that is affected by external conditions but can be partially regulated through certain control methods, such as small-scale wind / hydropower, which is usually used in conjunction with energy storage systems; and uncontrollable energy refers to energy that is completely dependent on natural conditions and cannot be effectively controlled, typically including photovoltaic power generation and wind power generation.

[0028] 1.1: In a further implementation, a steady-state characteristic index system for fully controllable energy sources (gas turbines) is constructed. This system decouples the steady-state characteristics of the gas turbine into four groups of indexes with clearly defined physical meanings: steady-state power characteristics, economic characteristics, regulation characteristics, and steady-state environment and constraint characteristics.

[0029] Indicator Group 1: Steady-State Power Characteristic Vector ; ; This set of indicators describes the core power output capability and efficiency boundary of a gas turbine, among which Rated steady-state output power (MW); The minimum steady-state operating power (MW); The steady-state efficiency (%) is the efficiency under rated operating conditions. The steady-state efficiency (%) is the efficiency at minimum output. For steady-state load adjustment range, , dimensionless parameter.

[0030] Indicator Group Two: Steady-State Economic Characteristic Vector ; ; This set of indicators characterizes the steady-state cost and energy consumption characteristics of gas turbines from an economic perspective. , , These are the coefficients of the quadratic, linear, and constant terms of the quadratic polynomial cost function, which refers to the power generation cost per unit time. The units of the three coefficients are yuan / MW2·h, yuan / MW·h, and $ / h, respectively. They include no-load costs but do not include start-up and shutdown costs. For the unit in the typical load range (referring to The weighted average steady-state efficiency (%) of ) The average steady-state heat rate is (kJ / kWh).

[0031] Indicator Group 3: Steady-State Regulation Characteristic Vector ; ; This set of indicators quantifies the transfer capability and partial load performance of gas turbines between different steady-state operating points, among which The rate of ascent ( ); downhill climbing rate ( ); The minimum steady-state load percentage (%) is calculated as follows: ; The optimal steady-state economic operating point (MW); The steady-state efficiency (%) is given at 50% of the rated power.

[0032] Indicator Group Four: Steady-State Environment and Constraint Characteristic Vector ; ; This set of indicators aims to characterize the impact of the external environment on the steady-state performance of gas turbines and their inherent physical and environmental constraints. Temperature derating rate ( The effect of ambient temperature changes on steady-state output is a key inherent characteristic for measuring the environmental adaptability of gas turbines. For altitude reduction rate ( This reflects the impact of altitude on its output. ( )and These are the emission coefficients of nitrogen oxides and carbon dioxide under rated steady-state operating conditions, respectively. The unit cooling water requirement of the combined cycle unit ( ).

[0033] 1.2: In a further implementation, for semi-controlled energy sources (represented by battery energy storage systems BESS), their steady-state characteristics are decoupled into four groups of indicators with clear physical meanings: capacity characteristics, economic characteristics, regulation characteristics, and constraint characteristics.

[0034] Indicator Group 1: Capacity Characteristic Vector ; ; This set of indicators describes the core energy capacity and efficiency boundaries of energy storage systems, among which Rated energy capacity (MWh); Maximum state of charge (%); The minimum state of charge (%); The maximum depth of discharge (%) is determined by... definition; The round-trip efficiency (%) is the efficiency of a complete charge-discharge cycle. This indicator system aims to describe the steady-state characteristics of the device under its current health condition. Therefore, the state of charge is calculated based on the current available capacity, rather than the capacity at the beginning of the lifespan, thus implicitly including the impact of battery aging and degradation.

[0035] Indicator Group Two: Economic Characteristic Vector ; ; This set of indicators characterizes the cost characteristics of energy storage systems from an economic perspective, among which... The annualized value of the investment cost per unit power ( ); The annualized value of the unit energy investment cost ( ); Annual fixed maintenance costs ( ); Unit cycle degradation cost ( ); The marginal operating cost of an energy storage system can be expressed as follows: ,in The average electricity purchase price during the energy storage charging period.

[0036] Indicator Group 3: Adjustment Characteristic Vector ; ; This set of indicators quantifies the power exchange capability and efficiency characteristics of energy storage systems, among which Rated charging power (MW); Rated discharge power (MW); Charging efficiency at 50% of rated power (%); The discharge efficiency is 50% of the rated power (%). This represents the monthly average self-discharge rate (%). 50% power is a very typical and representative part-load operating condition. In actual grid applications, energy storage systems rarely operate at 100% full power continuously; they spend more time adjusting within a part-load range. Adding the "50% rated power efficiency" indicator aims to use a simple, standardized scalar value to capture and differentiate the key performance differences of different energy storage systems under non-full-load conditions.

[0037] Indicator Group 4: Constraint Characteristic Vector ; ; This set of indicators aims to characterize the operational lifespan and environmental constraints of energy storage systems. The total designed cycle life (cycles, referring to the number of cycles at standard temperature and 80% depth of discharge); For the designed calendar lifespan (years); The highest operating ambient temperature ( ); Minimum operating ambient temperature ( ); The health status threshold (%) at the end of life.

[0038] 1.3: In a further implementation, for uncontrollable energy sources with random power output (represented by photovoltaic power generation PV), their "quasi-steady-state" statistical characteristics are decoupled into four groups of indicators with clear physical meanings: power output characteristics, economic characteristics, prediction accuracy characteristics, and grid-friendly characteristics.

[0039] Index Group 1: Output Characteristic Vector ; ; This set of indicators describes the long-term output statistics of photovoltaic power plants, among which... The installed capacity (MWp) of the power plant; The annual average capacity factor (%); The average annual equivalent full-load hours (h); The coefficient of variation for the hourly output sequence is calculated by dividing the annual standard deviation of all hourly output data for the year by the average value of all hourly output data for the year. This is the average correlation coefficient with the output of other photovoltaic power plants in the region.

[0040] Indicator Group Two: Economic Characteristic Vector ; ; This set of indicators characterizes the cost and benefit characteristics of photovoltaic power plants from an economic perspective, among which... The annualized value of the unit capacity investment cost ( ); Annual fixed maintenance costs ( ); The marginal operating cost of photovoltaic power generation is typically close to zero but includes very little variable operation and maintenance cost. For grid connection price ( ); The annual average curtailment rate (%).

[0041] Indicator Group 3: Prediction Accuracy Characteristic Vector ; ; This set of indicators quantifies the prediction accuracy of photovoltaic power output at different time scales, among which The root mean square error (MW) of the day-ahead forecast is calculated as follows: , where N is the total number of statistical time points; The mean absolute percentage error (%) of the current day's forecast is calculated as follows: ,in This refers to the installed capacity; The forecast good rate (%) is the percentage of time points in the total statistical time points where the absolute value of the forecast error is less than a certain threshold (e.g., 15% of the installed capacity, a fixed value pre-set according to the power grid dispatching protocol or application scenario). and These represent the real-time root mean square error and the mean absolute percentage error, respectively.

[0042] Indicator Group 4: Grid Friendliness Characteristic Vector ; ; This set of indicators aims to characterize the support capability of photovoltaic power plants for the power grid, among which The power factor is adjustable within a certain range; Maximum reactive power output (Mvar); Maximum reactive power absorption (Mvar); The voltage regulation gain coefficient refers to the ratio of the change in reactive power output of the inverter to the change in the resulting terminal voltage at the grid connection point of the power plant. It is a key indicator for measuring its voltage support capability. The average total harmonic distortion (THD) of the grid-connected current is given by (%).

[0043] S2: As one implementation method of this embodiment, a scenario partitioning method based on the operating manifold diffusion matrix is ​​proposed for data clustering and state identification in microgrid cluster systems. The purpose is to extract representative "typical operating scenarios" from complex operating data and achieve accurate scenario partitioning.

[0044] In the actual operation of microgrid cluster systems, system status data (such as load and photovoltaic output) changes over time, forming complex operating trajectories. These data points are often not uniformly distributed, but rather concentrated around a certain low-dimensional manifold; this phenomenon is called manifold structure. However, conventional clustering methods mostly rely on Euclidean distance, which can only handle simple linear distances and cannot effectively capture the manifold structure features in the data. Based on the above considerations, this embodiment proposes a scene partitioning method based on the operating manifold diffusion matrix.

[0045] 2.1: In a further implementation, an operational manifold diffusion matrix is ​​constructed to capture the diffusion process of microgrid operational data on the inherent geometric manifold.

[0046] In a further specific implementation, the basic time window is first set to "day" according to scheduling requirements. The time-series data of all equipment in the cluster (load, photovoltaic, gas turbine, energy storage, etc.) are sampled hourly and aligned to 24 time points. The raw data of the same type of equipment are added point by point to obtain curves representing the overall behavior of the cluster, including the total load curve, the total photovoltaic output curve, the total gas turbine output curve, and the average energy storage SOC curve.

[0047] Furthermore, to reduce the dimensionality of the curve data, statistical features were extracted from the four curves respectively; Table 1 gives the specific scalars for the extracted statistical features.

[0048] Table 1 Specific Scalars for Statistical Feature Extraction

[0049] The 15 scalar features extracted from Table 1 are horizontally concatenated with the time feature codes, resulting in a 20-dimensional sample feature vector for each sample (representing one day). The time feature codes include seasonal features and weekday features; seasonal features are converted into 4-dimensional binary vectors using one-hot encoding (e.g., "Spring" is [1,0,0,0]), and weekday features are encoded using binary encoding, with weekdays encoded as 1 and non-weekdays as 0.

[0050] Furthermore, all M sample feature vectors are stacked into a feature matrix; among them, the 15 scalar features in each sample feature vector are Z-score standardized, while the temporal feature encoding remains unchanged.

[0051] In a further implementation, an affinity matrix is ​​constructed from the standardized sample matrix, where the scale parameter is the square of the median Euclidean distance between all sample pairs.

[0052] In affinity matrix Based on this, calculate its degree matrix. And normalize Finally, the diffusion matrix of the running state manifold, which characterizes the law of state evolution, is obtained. .

[0053] Furthermore, through eigenvalue decomposition of the diffusion matrix A diffusion distance is defined to capture the inherent set relationship of microgrid operating states on the data manifold, replacing the traditional Euclidean distance for scenario partitioning. In the calculation of the diffusion distance, the diffusion time step is treated as an attenuation parameter, and its range is determined through eigenvalue selection. Based on this, this application sets the diffusion time step to 3 to achieve a good balance between local and global information.

[0054] 2.2: As a further implementation method, based on fuzzy clustering based on diffusion distance, entropy regularization is introduced to smooth the transition of scene boundaries.

[0055] In the aforementioned fuzzy clustering based on diffusion distance, the membership degree (a value between 0 and 1) is used to describe the confidence level of a sample belonging to a certain typical scenario. However, in practical applications, it has been found that clustering methods based solely on diffusion distance are unstable when dividing the "fuzzy zones" (i.e., boundary scenarios) of different scenarios. Specifically, samples with similar physical characteristics exhibit unpredictable and drastic fluctuations in their membership degrees, a phenomenon known as the "oscillation problem." The reason for this oscillation problem is that in the state transition region, the intrinsic structure (manifold geometry) of the data is highly complex, and the diffusion distance is overly sensitive to this, causing small differences to be amplified into discontinuous jumps in membership degrees. This local instability, like a domino effect, disrupts the smoothness and reliability of all subsequent computational stages, ultimately making the entire system fragile.

[0056] To address the aforementioned issues, this application introduces a scenario membership entropy regularization term, which, together with the diffusion distance, guides the final clustering result. This preserves the effectiveness of the basic clustering method while resolving its instability in the boundary region.

[0057] In a further implementation, a joint objective function that integrates "distance accuracy" and "entropy smoothness" is constructed. , is represented as: ; In this joint objective function, the first term This is the diffusion distance term, which serves to guide the cluster centers (centers of each scene). Move towards the region with the densest data distribution. (Second item) This is the entropy regularization term, which affects the distribution of membership degrees of samples. Apply smoothing constraints to prevent abrupt changes at the boundaries. Table 2 below provides an explanation of the parameter notation in the joint objective function.

[0058] Table 2. Explanation of parameter symbols in the joint objective function

[0059] In this joint objective function, the diffusion distance term minimizes the sum of squared diffusion distances from all samples to the center of each scene, ensuring that the final typical scene found is the most representative state of the data aggregation. The entropy regularization term essentially minimizes the Shannon entropy of the membership distribution. Minimizing entropy is equivalent to encouraging a more "average" and "uncertain" membership distribution, thereby avoiding extreme either-or judgments and achieving a smooth transition at the boundary. Control the smoothness, when When the value is large, the "smoothness" objective has a high weight. The optimization result tends to produce a more average, more uncertain membership distribution with very smooth boundaries, but this may come at the cost of a slight sacrifice in clustering accuracy. When the data size is small, the "accuracy" objective is given high weight. The focus is more on clustering based on data distance, which may result in less smooth boundaries and abrupt changes.

[0060] 2.3: In a further implementation, an alternating optimization strategy is employed to address the optimization problem of the joint objective function. Specifically: Step 1: Fix the scene center and use closed-form solutions to efficiently update membership; Step 2: With fixed membership degrees, update the scene center by solving for FréchetMean using gradient descent on the nonlinear manifold structure of the data, ensuring that the update is consistent with the intrinsic geometry of the data; Step 3: Define clear and quantifiable outer layer convergence criteria (e.g., outer layer convergence threshold). After completing iteration steps 1 and 2, calculate the value of the joint objective function for the current iteration; when the change in the value of the joint objective function is less than... When the iteration terminates, output the final membership degree and scene center value. If the outer convergence criterion is not met, return to iteration step 1 to continue the outer iteration.

[0061] As an optional implementation method in this embodiment, typical operating scenarios of the microgrid are automatically learned based on historical data and optimized algorithms. This allows for rapid scenario identification of the current real-time operating status, meeting real-time decision-making requirements.

[0062] After steps 2.3 above, the joint objective function is solved and output as follows: Each scene center Scene membership matrix ; Each scenario center represents the most typical operating state of a microgrid cluster, and the scenario membership matrix represents the confidence level of each historical sample belonging to each scenario.

[0063] Determined by the elbow rule An optimal scenario; specifically: Define a range of candidate values ​​for the optimal number of scenes. For each candidate value, execute the optimization algorithm described in section 2.3 and record the converged joint objective function value. Plot the curve of the objective function value changing with the candidate values, and identify the candidate value corresponding to the elbow point as the optimal number of scenes. .

[0064] In a further optional implementation, the 5-fold cross-validation method is used to determine the entropy regularization coefficient based on the maximum value of the silhouette coefficient, and the eigenvalue decay criterion is used to determine the diffusion time step.

[0065] In a further specific implementation, the real-time operating status of the current microgrid cluster is input, and the corresponding typical scenario label (the scenario corresponding to the maximum membership degree) and membership vector are identified. The membership vector quantifies whether the current state belongs to each preset typical scenario (corresponding to...). The confidence (probability) of the optimal scenario.

[0066] S3: As one implementation method of this embodiment, the static device characteristic indicators constructed in S1 are combined with the dynamic operating scenarios identified in S2. The static capabilities of the device are adaptively evaluated and compared according to the dynamic scenarios to achieve high-precision dimensionality reduction.

[0067] To ensure the physical meaning of the comparison, this embodiment sets all subsequent processes to be performed on devices of the same type. That is, fully controllable devices are only compared with other fully controllable devices, and the same applies to the other two types. In describing the specific implementation method, this embodiment uses any type of device for explanation.

[0068] 3.1: In a further implementation, the various performance indicators of the equipment are transformed into a structure that can profoundly reflect their essential similarity, providing a basis for subsequent analysis operations.

[0069] Based on the steady-state characteristic index system for each type of energy constructed in S1, a device characteristic matrix is ​​formed. ,in 20 represents the number of devices, and 20 represents the dimension of the steady-state characteristic index system for each type of device. In this matrix, the feature vector for each device (i.e., each row) is normalized.

[0070] After normalization, the Wasserstein distance between any two devices in the matrix is ​​calculated, and the entropy regularization of the Sinkhorn algorithm is called for a stable solution. The distances between all devices of the same type are then combined into a complete distance matrix. The first in the matrix Line 1 Column elements That is, equipment and equipment Wasserstein distance between them ; This represents the Wasserstein distance between devices p and h on the z-th feature index. Diagonal elements are defined as 0.

[0071] 3.2: In a further implementation, a scenario-based objective function is defined according to the dynamic requirements of different scenarios, and a set of optimal index weights is automatically learned for each scenario to improve the accuracy of subsequent device similarity comparison.

[0072] In a specific implementation, a scenario-based objective function is designed. , is represented as: ; In this objective function, Let be the weight vector of the indicator to be optimized. The importance weight of the z-th feature indicator is given by the following condition: and . Let represent the Wasserstein distance between devices p and h on the z-th feature index. The overall representation is the weighted Wasserstein distance between devices p and h. The optimal number of scenarios determined in step S2. This represents the membership degree of a sample (or the current state) to the k-th scene (this value is calculated in step S2). In further explanation, during offline training... The value is taken from the membership degree of historical samples; in online applications, The value is taken from the membership degree of the real-time state. and This represents the set of device groups in the k-th scene view. Here is the entropy regularization coefficient. The negative entropy of the weights (i.e., the regularization term).

[0073] An optimal set of index weight vectors was obtained through optimization. This ensures that, in the current scenario, "devices within the same group" are as similar as possible in the feature space (small distance), and "devices between different groups" are as different as possible in the feature space (large distance).

[0074] In a further implementation, since the objective function directly depends on the scene membership degree output by S2, the criteria for determining "devices within a group" and "devices between groups" are not globally fixed, but dynamically adapt to different scenes identified by S2. Therefore, this application proposes a dynamic determination method based on scene feature radius.

[0075] First, define an adaptive membership threshold. The threshold is calculated as follows: ; in, 1 represents the optimal total number of scenarios. The standard deviation of the membership values ​​for all scenarios. This is an adjustable coefficient. The threshold is defined as the weighted standard deviation of the mean membership degree of all scenarios. According to the constraint S2, the sum of the membership degrees of all scenarios for any sample equals 1, so the average value across K1 scenarios is 1 / K1.

[0076] For each scenario Select those with a membership degree greater than 1. All devices are analyzed, and the average Wasserstein distance from these devices to the scene center point is calculated. This average Wasserstein distance represents the corresponding scene center point. Characteristic radius And for equipment in and equipment ,like If the condition is met, the two devices are classified as "intra-group devices"; otherwise, they are classified as "inter-group devices".

[0077] In the calculation of membership degree prediction, for When the standard deviation is large, it indicates that the current state clearly belongs to one or several scenarios. In this case, the threshold should be increased to select only the most representative devices with very high membership, ensuring the purity of the "scene feature radius" calculation. Conversely, when the standard deviation is small, it indicates that the current state is relatively "fuzzy," with roughly the same membership to all scenarios. In this case, the threshold should be lowered to include more devices in the feature radius calculation, avoiding inaccurate radius estimation due to insufficient data.

[0078] In a further specific implementation, optimization algorithms (such as gradient descent, Adam, etc.) are used to solve for the objective function. Maximize the weight vector .

[0079] 3.3: In a further implementation, the most suitable equipment indicator weights are generated for different operating scenarios under the current operating state of the microgrid cluster, and weighted with corresponding features to finally output the final weighted feature matrix under the current operating state. This step adopts a two-stage mode combining offline training and online fusion to balance the accuracy and real-time performance of the model. Specifically: During the offline training phase, an optimal set of metric weight vectors is pre-calculated and stored for each typical scenario using historical data: For the S2 step determined Each of the optimal scenarios By setting its scene membership (Membership degree is 0 in other scenarios), and the optimization algorithm described in 3.2 is called to solve the objective function, and an optimal index weight vector is trained separately. Traverse all Each scenario is executed, ultimately generating and storing a scenario weight mapping table: ; This mapping table serves as the basis for decision-making in subsequent online applications.

[0080] In a further implementation, a set of fusion index weights that best match the current real-time operating status of the device is dynamically generated. Specifically: The S2 step receives information identified based on the real-time operating status. The membership degree of the optimal scenario is expressed as ; Based on the scene weight mapping table, this Each membership degree is related to the above The weight vectors corresponding to each scenario are multiplied together and then weighted. This is represented as: ; This is the final output weight vector of the fusion index, applied to the current moment. This represents the offline trained index weight vector corresponding to scenario k.

[0081] After calculating the fusion index weight vector adapted to the current scenario Then, the weights are applied to weight the steady-state characteristic index system constructed in step S1.

[0082] For example, when the energy equipment type in a certain scenario is a gas turbine, its equipment characteristics are: ; The feature vector of the gas turbine is 20-dimensional. These 20 features are then multiplied term-by-term by their corresponding fusion index weight vectors (Hadamard product). The resulting weighted feature vector of the gas turbine is then output. ; That is, the first under the equipment type of gas turbine One device.

[0083] All under this device type The weighted feature vectors of each device are stacked row-wise to obtain the final weighted feature matrix of this type. Each row in the matrix is ​​a weighted eigenvector of a single device.

[0084] The weighted feature matrix is ​​the output of S3 and also serves as the direct input for topology-aware clustering in S4. The same processing and calculation are performed on other device types to obtain their respective weighted feature matrices.

[0085] S4: As one implementation method of this embodiment, based on the above grouping according to electrical characteristics, a topology stability constraint is introduced. This constraint ensures that the physical connection structure of the power grid is not destructively altered when grouping devices according to electrical characteristics.

[0086] 4.1: In a further implementation, based on the weighted eigenvector matrix output by S3, the persistent cohomology method in topology data analysis is invoked to perform multi-scale topology analysis. The evolution process of the microgrid's topology is quantified by calculating the Betti quantization at different scales. Then, by analyzing the rate of change of the Betti curves, critical points of structural abrupt changes that must be avoided and safe topological stability intervals are identified, ultimately generating topological stability rules to constrain the clustering process. Specifically: By differentiating the Betti curve, the local maximum of the absolute value of the derivative is defined as the "critical point of structural change". Installing equipment at this point will cause a drastic topological change and must be avoided. The absolute value of the derivative is less than a preset threshold. The region is defined as the "topology stable region," within which device merging is safe.

[0087] 4.2: In a further implementation, a hierarchical clustering algorithm is constructed based on the weighted feature vector matrix and topological stability constraint rules to aggregate large-scale real power grid equipment into an "equivalent model" composed of multiple "equivalent devices", and all parameters of the equivalent model are calculated to finally form a simplified network model that can be used for power grid analysis.

[0088] 4.2.1: This implementation first designs a merging priority function. This is used to determine the merging priority of device clusters. The priority function is expressed as: ; in, Represents any two clusters and Merging priority scores between them; For clusters and cluster Feature similarity between them; For electrical connection enhancement factor, where, For clusters and cluster The physical connection between them The weighting factor (set to 1 in this application) provides a higher merging priority for physically connected clusters of devices. This is a topological penalty term.

[0089] In a further specific implementation, the feature similarity is calculated as follows: based on the weighted feature vector matrix, the centroid connection method is used to calculate the feature distance between two clusters, and the feature distance is mapped to feature similarity through the Gaussian kernel function.

[0090] In a further specific implementation, the physical connection relationship is defined as: if any device exists and equipment On the line If they are directly connected, then Otherwise, it is 0.

[0091] In a further specific implementation, the topology penalty term is defined as a piecewise function: By calculating merged clusters and Change in Betti number before and after and To determine whether the merger would lead to a sudden change in topology. For the change of connected components, This represents the change in the number of closed loops.

[0092] when or When (i.e., the merger will cross the structural mutation critical point defined in Section 4.1). A value of 0 prohibits the merge; otherwise, a value of 1 allows the merge. The threshold for the rate of change of the Betti curve defined in 4.1 ( (Two specific thresholds in the text).

[0093] 4.2.2: In a further implementation, a clustering algorithm is executed; during initial clustering, each device is treated as an independent cluster.

[0094] Calculate the merge priority score among all cluster pairs, select the cluster pair with the highest merge priority score for merging, and update the cluster information.

[0095] In one specific implementation, the termination conditions of the clustering algorithm include a primary criterion based on topological stability and alternative criters based on the number of clusters. Specifically: The main criterion is: the clustering algorithm terminates when the topological penalty term between the two clusters to be merged in the next step is 0. An alternative criterion is: if there are specific requirements for operation and management, it can also be set to stop when the number of clusters reaches the target value Y.

[0096] After the clustering iteration terminates, the algorithm outputs... Each device cluster is considered as an "equivalent device".

[0097] 4.2.3: Calculate the equivalent model parameters for each equivalent device output from 4.2.2, and generate the connection relationships between equivalent devices, i.e., the equivalent topology. Specifically: Since step S3 processes devices of the same type separately, all devices within each cluster are of the same type (fully controlled, partially controlled, or uncontrolled). During parameter aggregation, different aggregation methods are used based on the physical meaning of the parameters: For parameters representing "total quantity" such as rated power, capacity, and upper and lower limits of output, the parameter is summed directly for all devices within the cluster; For parameters that represent “intensity” or “ratio”, such as cost coefficient and efficiency, a weighted average method is used; the weights can be based on the rated power of each device.

[0098] The specific aggregation rules differ slightly depending on the type of equipment. For example, for fully controlled energy equipment (such as thermal power plants and gas turbines), the power range is calculated using a summation method, while the cost coefficient is calculated using a weighted average method; for semi-controlled equipment (such as energy storage), the capacity is calculated using a summation method, while the charging and discharging efficiency is calculated using a weighted average method. Specific settings can be customized based on actual needs.

[0099] 4.2.4: An equivalent connection line is established between two equivalent devices that have at least one connection line in the original power grid. The impedance of the equivalent connection line is obtained by equivalent calculation using the impedance of all the original lines connecting the two clusters.

[0100] The final equivalent inter-group topology is generated. The aggregated parameters, together with the equivalent inter-group topology, constitute the final output of this application.

[0101] As an embodiment of this application, a microgrid cluster dimensionality reduction system based on scene adaptation and topology preservation is disclosed. Employing the specific implementation method described above for microgrid cluster dimensionality reduction, the system includes: Steady-state characteristic index construction module; used to construct corresponding steady-state characteristic index systems for fully controllable energy, semi-controllable energy and uncontrollable energy respectively, and decouple the steady-state characteristics of each type of energy into multiple sets of characteristic indices; The scenario segmentation module is used to collect time-series data of microgrid cluster systems and extract statistical features from them to construct the operating manifold diffusion matrix. It then clusters the time-series data using a fuzzy clustering method based on diffusion distance, where entropy regularization is introduced to smooth the cluster boundaries, and a joint objective function is established and solved to output the optimal scenario segmentation result. The equipment feature scenario coupling module is used to construct the Wasserstein distance matrix based on the steady-state feature index system of each type of energy; define the scenario-based objective function to solve and output the optimal index weights; output the scenario membership degree according to the real-time operating status, fuse the index vector weights with the scenario membership degree, and then weight them with the steady-state feature index system to output the weighted feature matrix of each equipment type. Clustering and dimensionality reduction module: used to perform multi-scale topology analysis on weighted feature matrices, calculate the critical point of structural abrupt change and the topological stability interval; construct hierarchical clustering method to aggregate real power grid equipment into multiple equivalent equipment and calculate the parameters of the equivalent equipment, and finally output the equivalent topology.

[0102] As an embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it employs the specific implementation method described above for the microgrid cluster dimensionality reduction method.

[0103] As an embodiment of this application, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, it adopts the specific implementation method described above for the microgrid cluster dimensionality reduction method.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A microgrid cluster dimension reduction method based on scenario adaptation and topology preservation, characterized in that, Comprise: For full control type energy, semi-control type energy and uncontrollable type energy, corresponding steady-state characteristic index system is constructed, and the steady-state characteristics of each type of energy are decoupled into multiple groups of characteristic indexes; Collect the time series data of the microgrid cluster system and extract statistical features from it to construct the running state manifold diffusion matrix; the time series data is clustered by a fuzzy clustering method based on diffusion distance, wherein the entropy regularization is introduced to smooth the clustering boundary, a joint objective function is established and solved, and the optimal scenario division result is output; Based on the steady-state characteristic index system of each type of energy, a Wasserstein distance matrix is constructed; a scenario-based objective function is defined to solve the optimal index weight; according to the real-time running state, the scenario membership is output, the index vector weight and the scenario membership are fused, and then the steady-state characteristic index system is weighted to output the weighted characteristic matrix of each device type; Multi-scale topological analysis is performed on the weighted characteristic matrix to calculate the structural mutation critical point and topological stable interval; a hierarchical clustering method is constructed to aggregate the real devices of the power grid into multiple equivalent devices and calculate the equivalent device parameters, and finally the equivalent topology is output.

2. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, Collect the time series data of the microgrid cluster system and extract statistical features from it, including: The time series data includes the sampling data of all devices in the microgrid cluster, and after time alignment, the data of the same type of device is added point by point, and the cluster behavior curve is output, including the total load curve, the total photovoltaic output curve, the total gas turbine output curve and the average energy storage SOC curve; From the cluster behavior curve, statistical features are extracted, and after being spliced with time features, a sample feature vector is generated; based on the feature vectors of all M samples, a running state manifold diffusion matrix is constructed.

3. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The process of clustering the time series data includes: The joint objective function is composed of a diffusion distance term and an entropy regularization term. The composition of the diffusion distance term is expressed as, for a data point , the square of the diffusion distance to the cluster center is calculated, the square is multiplied by the membership , and the clustering driving term of the sample to the scene is obtained; Double summation of the clustering driving term from all samples to all scenarios is the diffusion distance term.

4. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The joint objective function is solved by an alternating optimization strategy, including: Step 1: Fix the scenario center of all scenarios, and update the membership by using the closed-form solution of the joint objective function; Step 2: Fix the membership of all samples, and update the scenario center by solving the Fréchet mean by gradient descent method; Step 3: After completing steps 1 and 2, calculate the joint objective function value of the current iteration; if the joint objective function value change value is less than the convergence threshold, terminate the iteration and output the final membership and scenario center value; if the joint objective function value change value is greater than or equal to the convergence threshold, return to step 1 for iteration.

5. The microgrid cluster dimension reduction method based on scenario adaptation and topology preservation according to claim 4, characterized in that, Upon completion of the iterations, the joint objective function solution output a scene center and a scene membership matrix ; The elbow rule is invoked to select out the optimal scenario.

6. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, Wasserstein distance matrix is constructed, including: For the steady-state characteristic index system of any type of energy, a device characteristic matrix is organized; after normalizing the device characteristic matrix, the Wasserstein distance between any two devices in the matrix is calculated, and the entropy regularization of Sinkhorn algorithm is called to solve; The Wasserstein distances between all devices in the matrix are combined into a Wasserstein distance matrix; the element in the i-th row and j-th column of the matrix is the Wasserstein distance between device i and device j, and the diagonal elements are defined as 0. ​​​​​ 7. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The process of solving the optimal index weight includes: Designing a scenario-based objective function represents the maximum value including the inter-group device distance term minus the intra-group device distance term; the inter-group device distance term is a sum of weighted Wasserstein distances of all inter-group devices under a scenario, the intra-group device distance term is a sum of weighted Wasserstein distances of all intra-group devices under a scenario; wherein, ; the scenario-based objective function further comprises an entropy regularization term of the weight optimizing the scenarioed objective function, outputting an optimal index weight vector under the scenario, denoted as: ; finally output an optimal index weight vector.

8. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 7, wherein, The determination steps of inter-group devices and intra-group devices include: An adaptive membership threshold is defined by the mean and standard deviation of the membership values of all optimal scenarios ; From the scene select the device whose membership is greater than and calculate its average Wasserstein distance to the scene center, the average Wasserstein distance and the corresponding scene feature radius ; For devices in a scene and devices and Wasserstein distance of devices , devices and devices are determined as in-group devices, otherwise as inter-group devices.​ 9. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The index vector weight is fused with the scene membership, including: Identified based on real-time operating status The membership degree of the optimal scenario is expressed as: ； This membership is multiplied by its corresponding optimal indicator weight vector, respectively, and then weighted to obtain the fusion indicator weight vector .

10. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The fused index weight vector is weighted with the steady-state characteristic index system, including: A plurality of groups of characteristic indexes of each device in each type of energy are spliced into a characteristic device vector, the characteristic device vector is multiplied with the fused index weight vector of the scene to which the characteristic device vector belongs item by item, and a weighted characteristic vector of each device is output; The weighted characteristic vectors of all devices under the same device type are stacked in rows to obtain a weighted characteristic vector matrix.

11. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The determination manner of the structural mutation critical point and the topological stable interval includes: The weighted characteristic vector matrix is subjected to multi-scale analysis by using a persistent homology method, and Betti values and a Betti curve are output; The derivative of the Betti curve is calculated, and a local maximum point of the absolute value of the derivative is determined as a structure mutation critical point , and a region with an absolute value of the derivative less than a preset threshold is determined as a topologically stable interval.

12. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The hierarchical clustering method includes: The merging priority between any two device clusters is determined based on a score calculated by using a merging priority function, a device cluster pair with the highest merging priority score is selected for merging, and cluster information is updated; The merging priority function is a product of a characteristic similarity, an electrical connection enhancement factor, and a topological penalty term; The characteristic similarity is a characteristic distance between two device clusters, and is calculated based on the weighted characteristic vector matrix by using a centroid connection method; The electrical connection enhancement factor is a product of a physical connection relationship and a weight coefficient, and is added by 1; when the two device clusters are connected by a device, the corresponding physical connection relationship is 1, otherwise, the corresponding physical connection relationship is 0; The topological penalty term is a piecewise function; when the two device clusters cross the structural mutation critical point after being merged, the penalty term takes a value of 0, otherwise, the penalty term takes a value of 1.

13. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 12, wherein, The termination condition of the hierarchical clustering includes a main criterion and an alternative criterion, including: The main criterion is that, when the topological penalty term between the two device clusters to be merged next takes a value of 0, the hierarchical clustering is terminated; The alternative criterion is that, when the number of the clustered device clusters reaches a target value, the hierarchical clustering is terminated.

14. The microgrid cluster dimensionality reduction method based on scene adaptation and topology preservation of claim 1, wherein, The equivalent topology is output, including: After the hierarchical clustering terminates, output a cluster of devices, each device in the cluster being considered as an equivalent device; aggregate parameters of devices of the same type in each cluster of energy based on the equivalent devices; An equivalent connection line is established between two equivalent devices in the micro-grid cluster that have at least one connection line, and an impedance of the equivalent connection line is obtained by equivalent calculation through all original line impedances connecting the two equivalent devices, and the equivalent topology is generated.

15. A microgrid cluster dimension reduction system based on scenario adaptation and topology preservation, performing the microgrid cluster dimension reduction method according to any one of claims 1-14, characterized in that, The system includes: A steady-state characteristic index construction module, configured to construct a corresponding steady-state characteristic index system for full-control energy, semi-control energy and uncontrollable energy respectively, and to decouple the steady-state characteristics of each type of energy into a plurality of groups of characteristic indexes; A scene division module, configured to collect time series data of the micro-grid cluster system and extract statistical features therefrom, to construct a running state manifold diffusion matrix, to cluster the time series data by using a fuzzy clustering method based on a diffusion distance, to introduce entropy regularization to smooth the clustering boundary, to establish and solve a joint objective function, and to output an optimal scene division result; A device characteristic scene coupling module, configured to construct a Wasserstein distance matrix based on the steady-state characteristic index system of each type of energy, to define a scene target function to solve and output an optimal index weight, and to output a scene membership according to a real-time running state, to fuse the index vector weight with the scene membership, and to weight the fused index vector weight with the steady-state characteristic index system to output a weighted characteristic matrix of each device type; The clustering dimension reduction module is used for carrying out multi-scale topological analysis on the weighted feature matrix, calculating structure mutation critical points and topological stable intervals, constructing a hierarchical clustering method, aggregating real power grid devices into multiple equivalent devices and calculating equivalent device parameters, and finally outputting an equivalent topology.

16. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program, when loaded into a processor, implements the micro-grid cluster dimension reduction method according to any one of claims 1-14.

17. A computer-readable storage medium, the computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by a processor, implements the micro-grid cluster dimension reduction method according to any one of claims 1-14.

Images