Power grid energy storage acceptance capability assessment method based on scene fingerprints and weighted migration

Abstract

The invention discloses a power grid energy storage acceptance capability evaluation method based on scene fingerprints and weighted migration, and the method comprises the steps: constructing an offline knowledge base, building massive power grid operation scenes, and carrying out the simulation of the massive power grid operation scenes, and evaluating the maximum energy storage acceptance capability of each operation scene; extracting a multi-dimensional structured feature vector of each operation scene as a scene fingerprint; the scene fingerprints of the target power grid are extracted, similarity matching is carried out on the scene fingerprints of the target power grid and the scene fingerprints in the off-line knowledge base, the energy storage acceptance capacity estimated value of the target power grid is determined, rapid and effective evaluation of the new power grid with scarce data is achieved, and a scientific basis is provided for energy storage planning; the problem that in the prior art, due to the fact that historical data are missing, the energy storage acceptance capacity cannot be evaluated is solved.

Description

Technical Field

[0001] This invention relates to the field of power system planning and operation, and specifically to a method for evaluating the grid energy storage acceptance capacity based on scenario fingerprinting and weighted migration. Background Technology

[0002] With the large-scale integration of new energy sources and the advancement of dual-carbon goals, energy storage systems are playing an increasingly important role in the power grid. They play an irreplaceable role in smoothing out new energy fluctuations, providing frequency regulation and peak shaving services, and delaying power grid investment. Accurately assessing the power grid's energy storage acceptance capacity, that is, the maximum energy storage capacity that the system can economically and efficiently accept under the premise of meeting the constraints of safe and stable operation, is a key prerequisite for realizing the scientific planning and optimized allocation of energy storage.

[0003] Traditional methods for assessing energy storage capacity primarily rely on detailed historical operating data and complex model-driven approaches such as time-series simulations, optimal power flow (OPF), stochastic programming, or robust optimization. These methods typically require a complete grid topology, detailed equipment parameters, and high-precision time-series data on load and renewable energy output. The assessment results are obtained through long-term dynamic simulations or complex mathematical programming solutions. However, in practical engineering applications, many scenarios, such as newly built industrial parks, remote microgrids, new urban areas, and island power grids, face a lack of historical operating data, sometimes even completely missing. In these areas, due to the lack of sufficient input data, traditional model-driven methods are difficult to implement or yield highly unreliable results, hindering effective energy storage planning and creating a vicious cycle of "no data → inability to assess → difficulty in planning."

[0004] To address the problem of data scarcity, some existing studies attempt to use simplified models or empirical formulas for estimation, such as empirical coefficient methods based on macroeconomic indicators like load levels and renewable energy penetration rates. However, these methods are too coarse, failing to fully consider the internal structure, safety constraints, and dynamic characteristics of the power grid, making it difficult to guarantee assessment accuracy. Furthermore, they lack physical mechanism support and are ill-suited to diverse power grid scenarios. Other studies explore using machine learning models such as neural networks and support vector machines to establish "input-output" mapping relationships, but these data-driven methods are highly dependent on the quality and quantity of training data. In newly developed regions, there is a lack of sufficient labeled samples for model training, leading to a prominent cold start problem and limiting their direct application.

[0005] In recent years, knowledge transfer technologies such as transfer learning, knowledge graphs, and case-based reasoning have begun to be introduced into the power system field, aiming to transfer knowledge from existing systems to new systems. However, existing technologies mostly focus on tasks such as fault diagnosis and load forecasting, and a dedicated transfer framework for assessing energy storage acceptance capacity is still lacking. General transfer learning methods often assume strong distributional consistency between the source and target domains or require complex adjustments to the model structure, making it difficult to directly adapt to the significantly different grid structures and operating modes in power systems. In addition, how to construct a high-fidelity, wide-coverage "scenario-capability" knowledge base and design a lightweight, fault-tolerant scenario matching mechanism remains a technical bottleneck that urgently needs to be addressed.

[0006] Therefore, there is a lack of existing technologies that can provide a rapid and reliable preliminary assessment of energy storage capacity under conditions of data scarcity, without relying on complete topology and high-dimensional time-series data. In particular, for newly built power grids that are still in the planning stage and only have macroscopic parameters, there is an urgent need for a new assessment method that breaks through data dependence, has strong generalization ability and self-learning characteristics, so as to solve the dilemma of difficulty in assessment without data and provide a scientific basis for the early layout and investment decision of energy storage facilities. Summary of the Invention

[0007] The purpose of this invention is to address the aforementioned problems by providing a grid energy storage acceptance capacity assessment method based on scenario fingerprinting and weighted migration. This method constructs a high-precision, wide-coverage offline knowledge base for "scenario-capacity" mapping and a lightweight scenario fingerprint extraction and weighted migration mechanism to achieve rapid and effective assessment of new grids with scarce data. This provides a scientific basis for energy storage planning and solves the problem in existing technologies where energy storage acceptance capacity assessment is impossible due to the lack of historical data.

[0008] The technical solution adopted in this invention is as follows: A method for assessing grid energy storage acceptance capacity based on scenario fingerprinting and weighted migration, the method comprising: An offline knowledge base is constructed, which includes a vector database and a relational database, as detailed below: A large number of power grid operation scenarios are established, and the maximum energy storage capacity of each operation scenario is evaluated through simulation. A multi-dimensional structured feature vector of each operation scenario is extracted as a scenario fingerprint. The vector database stores the scenario fingerprint, and the relational database is used to store scenario information. Online assessment, specifically including: The scene fingerprint of the target power grid is extracted, and the energy storage capacity of the target power grid is estimated by matching the scene fingerprint of the target power grid with the scene fingerprint in the offline knowledge base.

[0009] Furthermore, the establishment of the massive power grid operation scenario is as follows: Using standard testing systems combined with external time-series data as benchmark data, the Monte Carlo random sampling method is used to enhance the benchmark data, simulating uncertainties such as load forecasting deviations, fluctuations in renewable energy output, random equipment outages, and changes in network topology, thereby generating a massive number of power grid operation scenarios.

[0010] Furthermore, the standard testing system includes IEEE 14-node, IEEE 30-node, and IEEE 118-node testing systems; the external time-series data includes load data from the European Transmission Network Operators Alliance (ENTSO-E) and solar radiation and wind speed data from the NASA POWER database.

[0011] Furthermore, the simulation evaluation is as follows: For each operating scenario, the maximum energy storage capacity corresponding to each operating scenario is solved using the optimal power flow model, and this is used as the standard evaluation result.

[0012] Furthermore, the objective function of the optimal power flow model is to maximize the sum of the injected power of all candidate energy storage nodes, and the constraints include node power balance, line power flow, node voltage security, and generator output constraints.

[0013] Furthermore, the multidimensional structured feature vector in the scene fingerprint specifically includes core features, high-level features, and derived and statistical features; The core features include the maximum, minimum, and average values ​​of the total active load and the average value of the total reactive load; the maximum and average values ​​of the total output of new energy sources, the average value of the total output of conventional energy sources, and the power generation / load ratio; the penetration rate of new energy sources and the proportion of conventional energy regulation capacity; and the season, date type, and time period. The advanced features include the average load rate of critical lines, the proportion of heavily loaded lines, the average system voltage, the maximum voltage deviation, the number of voltage over-limit nodes, the wind and solar power output ratio, and the correlation coefficient between renewable energy output and load. The derived and statistical features include load peak-valley difference rate, daily load curve shape similarity, load change rate standard deviation, maximum value of new energy output change rate, and new energy prediction error simulation standard deviation.

[0014] Furthermore, the vector database is used to store the fingerprint of the running scenario and support similarity retrieval, and the relational database is used to store complete running scenario parameters, simulation evaluation results, and original snapshots.

[0015] Furthermore, the similarity matching specifically includes: The scene fingerprint of the target power grid is extracted, and then retrieved and matched through the offline knowledge base. That is, the weighted distance between the scene fingerprint of the target power grid and each scene fingerprint in the offline knowledge base is calculated to identify the set of most similar operating scenarios. The weighted distance calculation between the scene fingerprint of the target power grid and each scene fingerprint in the offline knowledge base is only performed by summing the feature values ​​of both sides in the dimensions of the derived and statistical features, and the K similar operating scenarios with the smallest weighted distance are selected.

[0016] Furthermore, the estimated energy storage capacity of the target power grid specifically includes: weighted fusion of the evaluation results of K similar operating scenarios based on similarity, converting the weighted distance into similarity, and using the similarity as the weight to perform a weighted average of the evaluation results of the K similar operating scenarios to obtain the estimated energy storage capacity of the target power grid.

[0017] Furthermore, the method also includes dynamic updating, which specifically includes: Once the target power grid obtains the measured evaluation results, the complete scenario fingerprint of the target power grid and the evaluation results are added as new records to the offline knowledge base, thereby enabling dynamic updates of the knowledge base.

[0018] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This invention presents a grid energy storage acceptance capacity assessment method based on scenario fingerprinting and weighted migration. By constructing a high-precision, wide-coverage offline knowledge base for "scenario-capacity" mapping and building a lightweight scenario fingerprint extraction and weighted migration mechanism, it enables rapid and effective assessment of new grids with scarce data, providing a scientific basis for energy storage planning. This solves the problem in existing technologies where energy storage acceptance capacity assessment is impossible due to the lack of historical data. Through a closed-loop process of assessment-feedback-update, the knowledge base is continuously optimized and the system performance becomes more accurate with use, significantly improving the long-term applicability and engineering value of the method. Attached Figure Description

[0019] Figure 1 This is the overall logical framework diagram of the grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration of the present invention; Figure 2 This is a flowchart of the operation of the grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings.

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0022] Example 1 This embodiment provides a grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration. It constructs an offline knowledge base based on a "scene-capacity" mapping to achieve efficient assessment of grid energy storage acceptance capacity. The core of this method lies in utilizing high-precision simulation and data-driven approaches, combined with advanced algorithms and models, to solve complex problems in grid energy storage acceptance capacity assessment. The entire technical framework includes four key technical components: data preparation and benchmark grid modeling, scene generation and data augmentation, high-precision simulation assessment, and feature extraction and structured data storage.

[0023] like Figure 2 As shown, the method in this embodiment includes two core processes: an offline knowledge base construction phase and an online evaluation and dynamic update phase. 1. The offline knowledge base construction phase includes: (1) Data preparation and benchmark power grid modeling: In order to build a simulation environment that is both universal and has reasonable physical complexity, this embodiment adopts the standard test system widely recognized in power system research as the benchmark model. These benchmark models provide complete power grid topology, component parameters and initial operating states, and are ideal templates for generating a large number of credible operating scenarios.

[0024] This embodiment primarily selects the IEEE 14-node, 30-node, and 118-node test systems as core research cases. These systems, published by the IEEE Electric Power Engineering Society, have become standard benchmarks for academic research in the global power field. These standard systems provide all the static model parameters required to build the knowledge base, including topology, line and transformer parameters, generator parameters, and load baseline values. In the data preparation stage, spatiotemporal time-series data were integrated by combining annual hourly total load data from the European Transmission Network Operators Alliance (ENTSO-E) and solar radiation and wind speed data provided by the NASA POWER database. After processing, these data can be converted into node-level photovoltaic and wind power output time-series data for subsequent analysis.

[0025] To overcome the limitations of single historical data, this embodiment uses the Monte Carlo random sampling method to enhance the baseline data, generating massive and diverse operating scenarios to comprehensively cover the potential operating states of the power grid. The data enhancement process is as follows: First, the acquired regional total load time-series data is scaled and distributed according to the initial load ratio of each node in the benchmark grid to form the daily load curve of each node. Simultaneously, photovoltaic or wind farms are connected to selected nodes, and corresponding renewable energy output data is allocated. Based on this, multiple Monte Carlo samplings are performed on the baseline operating point for each hour. A large number of instantaneous operating scenarios are generated by simulating random factors such as load forecast deviations, renewable energy output fluctuations, random equipment outages, and network topology changes. Finally, a set containing N independent scenarios is generated. Each operating scenario represents a specific power grid operating state, providing a sufficient sample basis for subsequent evaluation.

[0026] (2) For each operating scenario, high-precision simulation evaluation is used to obtain the maximum energy storage capacity corresponding to each operating scenario; high-precision simulation evaluation is the computational core of knowledge base construction, used to evaluate the maximum energy storage capacity of each scenario. The standard answer, i.e., the maximum energy storage capacity, is calculated as follows: An energy storage capacity assessment model is defined, employing a mathematical model centered on Optimal Power Flow (OPF) for evaluation. The objective function is as follows: (1) In the formula, This refers to the set of all candidate energy storage access nodes in the system. Let be the injected power of the newly added energy storage at node k, with discharge being positive and charging being negative. The objective function aims to maximize the total discharge power of the newly added energy storage within the system, i.e., the system's energy storage capacity. The constraints of the objective function are as follows: Node power balance constraints: , ; Power flow constraints on the line: ; Node voltage safety constraints: ; Generator output constraints: , .

[0027] (3) Feature extraction and structured data storage: To transform the simulation results into a knowledge base that can be quickly retrieved, this embodiment constructs a digital feature fingerprint for each running scenario and establishes a dual-database storage structure. The specific process is as follows: Scene fingerprint feature extraction is performed for each scenario: To further support rapid retrieval and knowledge transfer, this embodiment proposes a multi-dimensional, hierarchical scene fingerprint feature extraction mechanism, which abstracts complex operating scenarios into structured feature vectors. The feature system includes three categories of indicators with a total of nine dimensions. The first category consists of core features, which are the most readily available indicators at the macro level. These form the foundational layer of the fingerprint and are essential features for the evaluation process. They are mandatory input data in the assessment of energy storage capacity in newly developed areas where data is scarce. These include: the maximum, minimum, and average values ​​of the system's total active load, and the average value of the total reactive load; the maximum and average values ​​of total renewable energy output, the average value of total conventional energy output, and the generation / load ratio; the penetration rate of renewable energy and the proportion of conventional energy regulation capacity; and spatiotemporal background information such as season, date type, and time period.

[0028] The second category consists of advanced features that reflect the internal operating status and safety constraints of the power grid. These include: the average load rate of critical lines and the proportion of heavily loaded lines, used to characterize the degree of network congestion; the system average voltage level, maximum voltage deviation, and the number of voltage-over-limit nodes, used to characterize voltage stability; and the wind-solar power output ratio and the correlation coefficient between renewable energy output and load curves, to quantify the spatiotemporal distribution characteristics of renewable energy and the intensity of anti-peak shaving.

[0029] The third category comprises derived and statistical features, which extract dynamic behavioral patterns from time-series data. These include: load peak-to-valley difference rate, similarity of daily load curve shape, and standard deviation of load change rate, used to describe load volatility and typical patterns; and the maximum value of the rate of change of renewable energy output and the standard deviation of simulation prediction error, reflecting the drastic fluctuations and uncertainty level of its output.

[0030] After scene fingerprint feature extraction, a structured knowledge base supported by a vector database and a relational database is constructed. The vector database stores the scene fingerprint composed of the aforementioned multi-dimensional features, supporting efficient similarity matching; the relational database stores complete scene parameters, simulation results, and original snapshots. In this way, high-dimensional and complex physical simulation results are transformed into machine-understandable and searchable "feature-capability" mapping relationships, laying a solid data foundation for knowledge transfer and rapid evaluation in the subsequent online assessment stage.

[0031] 2. The online assessment and dynamic update phase includes: (1) Target scenario fingerprint extraction: In order to achieve rapid assessment of the energy storage acceptance capacity of newly built or historically data-deficient areas, this embodiment performs scenario fingerprint extraction on the target power grid at this stage. This process fully considers the limitations of data availability in actual engineering and adopts a hierarchical feature input mechanism to ensure the feasibility and robustness of the assessment. The specific process is as follows: For target areas to be evaluated, such as newly built industrial parks, new urban areas, or remote microgrids, priority should be given to collecting core characteristics that are easily accessible from macro-planning parameters or operational reports. These characteristics constitute the necessary inputs for the evaluation, including: the maximum, minimum, and average values ​​of the system's total active load and the average value of the total reactive load; the maximum and average values ​​of the total output of new energy sources such as photovoltaic and wind power, the average value of the total output of conventional energy sources, and the ratio of power generation to load; the penetration rate of new energy sources and the proportion of conventional energy regulation capacity; and spatiotemporal background information such as season, date type, and typical time period. The above indicators can usually be obtained through project planning documents, load forecasting reports, or simple calculations, without the need for detailed topology or high-precision time-series data.

[0032] For advanced features and derived and statistical features, if the target area already has a certain data foundation, such as having completed preliminary power flow calculations or having some monitoring data, then additional indicators such as the average load rate of key lines, the number of voltage over-limit nodes, the wind and solar power output ratio, the load peak-valley difference rate, and the intensity of new energy anti-peak regulation are extracted to improve the evaluation accuracy. If the relevant data is not available, the corresponding feature dimension is marked as "∞" or the default value and is automatically ignored in the subsequent matching process, reflecting the system's fault tolerance and lightweight characteristics.

[0033] Finally, the extracted features are used to construct a structured target scene fingerprint vector, which serves as the starting point for knowledge transfer.

[0034] (2) Weighted similarity matching, specifically including: After constructing the target scene fingerprint, the process proceeds to the knowledge base retrieval and matching stage. This embodiment identifies the most similar set of historical scenes by calculating the weighted distance between the target scene fingerprint and the fingerprints of each scene in the offline knowledge base, thus achieving knowledge transfer from the known to the unknown.

[0035] The matching process employs an improved weighted Euclidean distance metric, as shown in the following equation: (2) In the formula, and The target scenario and the offline knowledge base are respectively The first scenario 3D eigenvalues The pre-defined feature weights reflect the importance of the feature to the energy storage capacity; the summation is performed only on dimensions where the feature values ​​of both sides are not "∞", effectively avoiding the problem of calculation failure caused by the missing of some features.

[0036] After calculation, sort by distance from smallest to largest, and select the top... The most similar scenarios are selected as the candidate set for subsequent weighted fusion evaluation.

[0037] (3) Knowledge transfer and integration assessment Based on the similarity matching results, this embodiment employs a weighted fusion strategy to transfer the evaluation results of candidate scenarios to the target power grid, generating the final estimated energy storage capacity. Specifically, the weighted distance of each candidate scenario is first calculated... Converted to a similarity index, as shown in the following formula: (3) Similarity The higher the similarity score, the closer the candidate scene is to the target scene, and the greater its contribution to the final conclusion should be. Subsequently, similarity is used as a weight to further refine the evaluation. Evaluation results of energy storage acceptance capacity for each candidate scenario The weighted average is then used to obtain the evaluation result of the target power grid, as shown in the following formula: (4) This fusion mechanism not only makes full use of information from multiple similar scenarios, but also reduces the impact of outliers through weight allocation, thereby improving the stability and reliability of the evaluation results.

[0038] (4) Knowledge base is updated synchronously: To achieve continuous evolution and self-learning capabilities of the evaluation system, this embodiment designs a closed-loop update mechanism for the knowledge base. Once the target power grid is operational and has accumulated sufficient historical data, high-precision simulation evaluations can be performed, such as OPF calculation based on measured data, to obtain a more accurate "true value" of energy storage capacity. The complete feature fingerprint of the actual operating scenario, including previously missing advanced and derived features, can be extracted and added to the existing typical scenario knowledge base along with the newly obtained evaluation results as a high-quality record. This process not only expands the coverage of the knowledge base, but also enhances its characterization capabilities for specific types of power grids, such as those dominated by industrial loads and those with high wind power penetration, and continuously improves the matching accuracy for similar scenarios in the future.

[0039] Example 2 The following describes the specific implementation process of this invention in detail, using the evaluation process of a newly built industrial park under typical summer operating conditions as an example. Figure 1 As shown in the figure, the grid energy storage acceptance capacity assessment method based on scenario fingerprinting and weighted migration proposed in this embodiment includes: S1, Offline construction of a knowledge base for typical scenarios. This step is completed in advance in the background to provide a database of "standard answers" for online assessment.

[0040] S11, Data Preparation and Baseline Power Grid Modeling The IEEE 30-node test system was selected as the benchmark power grid model to obtain its complete topology, line impedance, transformer turns ratio, generator output upper and lower limits, and initial load values ​​of each node.

[0041] Integrating external spatiotemporal data: Obtaining hourly total load time-series data for a typical year from the ENTSO-E public platform; utilizing the NASA POWER database to obtain solar radiation intensity and wind speed data for the same region; and generating corresponding node-level renewable energy output time-series sequences through photovoltaic / wind power conversion models.

[0042] S12, Scene Generation and Data Augmentation The annual total load curve is allocated according to the initial power ratio of each load node in the IEEE 30-node system to form the daily load curve of each node; photovoltaic power plants are connected to nodes 3, 8, and 15, and wind farms are connected to nodes 1 and 11, and corresponding new energy output data are allocated.

[0043] The Monte Carlo random sampling method is used to superimpose random disturbances onto the basic operating curve: simulating load forecasting errors (±10%), fluctuations in renewable energy output (±15%), N-1 outages on critical lines, and adjustments to main transformer tap changes, among other uncertainties. This ultimately generates a set of numerous independent instantaneous scenarios, comprehensively covering the potential operating conditions of the power grid under different weather conditions, load levels, and network states.

[0044] S13, High-precision simulation evaluation For each generated scenario, an optimal power flow (OPF) mathematical model is established and solved to obtain the maximum energy storage capacity (unit: MW) corresponding to each scenario, which serves as the standard evaluation result for that scenario.

[0045] S14, Feature Extraction and Structured Data Storage For each completed simulation scenario, a structured scenario fingerprint feature vector is extracted. This scenario fingerprint adopts a hierarchical design to ensure that it can be effectively used under different data completeness conditions.

[0046] The feature vector contains three categories of indicators totaling 15 dimensions: The core features, specifically including nine dimensions, are all mandatory to extract: these features are necessary conditions for evaluation, can be obtained using only macro reports or simple calculations, and form the cornerstone of the fingerprint; including: Load levels: System maximum load (89.5MW), minimum load (62.3MW), average load (75.1MW), average reactive load (32.4MVar); Power output: average total output of new energy sources (28.7MW), average output of conventional energy sources (50.2MW), power generation / load ratio (1.05); Structural proportions: New energy penetration rate (32.1%), conventional energy regulation capacity ratio (45.0%). Spatiotemporal context: season (summer), date type (weekday), time period (peak).

[0047] Advanced features, specifically including 4 dimensions, are recommended for extraction: reflecting the internal state of the power grid, which requires simple power flow calculations. These include: average load factor of critical lines (68.2%), proportion of heavily loaded lines (15%), system average voltage (1.02 pu), and number of voltage-overrunning nodes (0).

[0048] Derivative and statistical features, specifically including two dimensions, are optional extraction: mining patterns from time series data; including: load peak-valley difference rate (30.4%), and the intensity of new energy anti-peak shaving (correlation coefficient -0.45).

[0049] The complete 15-dimensional fingerprint of each scenario, the corresponding energy storage capacity assessment result (e.g., 48.6MW), and the source grid identifier IEEE30 are stored in a vector-relational collaborative database to complete the knowledge base construction. The vector database uses core features as the primary key for indexing, ensuring effective matching even if the target scenario only provides core features.

[0050] S2, Online assessment of the energy storage capacity of newly built industrial parks. This step is for newly built parks and performs a real-time assessment. The specific process is as follows: S21, Target Scene Fingerprint Extraction When extracting the scene fingerprint of an industrial park under planning and construction, priority is given to ensuring the integrity of core features, while other features are selectively filled in based on data availability. The core features are all available and must be extracted: Maximum active load: 80 MW; Minimum active load: 56 MW (estimated by maximum load × (1 - peak-valley difference)); Average active load: 56 MW (estimated by maximum load × load factor); Average reactive load: 22.4 MVar (estimated based on a power factor of 0.95); Average total output of renewable energy: 16 MW (wind power installed capacity × capacity coefficient); Average output of conventional energy: 64 MW (balanced load); Generation / load ratio: 1.0; Renewable energy penetration rate: 20% (maximum wind power output / maximum load); Conventional energy regulation capacity share: 40% (estimated based on typical thermal power regulation capacity); Spatiotemporal context: Season = Summer, Date type = Weekday, Time period = Peak Advanced features, set as default values, include: average load factor of critical lines: ∞ (unknown); proportion of heavily loaded lines: ∞ (unknown); average system voltage: ∞ (unknown); number of voltage over-limit nodes: ∞ (unknown); derived and statistical features (some can be estimated); load peak-valley difference rate: 0.3 (typical industry value); new energy anti-peak shaving intensity: -0.3 (estimated based on wind power output characteristics).

[0051] Finally, construct the target fingerprint vector: [80,56,56,22.4,16,64,1.0,0.2,0.4,Summer,Workday,Peak,∞,∞,∞,∞,0.3,-0.3], a total of 18 dimensions, of which 9 core features are complete and the remaining 9 are partially missing; this method is designed to allow for the absence of high-level and derived features, and the distance is calculated only based on the core features during matching to ensure the feasibility of the evaluation.

[0052] S22, Weighted Similarity Matching Call the knowledge base and calculate The weighted Euclidean distance to each record in the database, weight The model is trained based on historical data using a random forest model. For example, the weights for renewable energy penetration rate are 0.18, maximum load is 0.15, and seasonality is 0.05. Advanced / derived features have low or zero weights and are automatically ignored when the target value is ∞. After sorting, the K=3 most similar scenarios with the smallest distance are selected. Scenario A: [82,58,0.71,18,0,110,Summer,Workday,0.22,0,65%,1,0.28,0.15,-0.32], Evaluation result 50MW, Distance 0.12, Similarity ; Scenario B: [78,54,0.69,15,0,130,Summer,Workday,0.18,0,70%,0,0.20,0.12,-0.28], Evaluation result 45MW, Distance 0.15, Similarity ; Scenario C: [85,60,0.72,20,1,100,Spring,Weekday,0.25,0,60%,2,0.30,0.18,-0.35], Evaluation result 55MW, Distance 0.25, Similarity ; S23, Knowledge Transfer and Integration Assessment The evaluation results of three similar scenes are fused using a similarity-weighted average algorithm: (5) The assessment results indicate that the newly built industrial park can accommodate an energy storage system capacity of approximately 50 MW under typical summer operating conditions.

[0053] S24, Knowledge Base Updated Simultaneously If the industrial park has accumulated complete historical operating data after one year of operation, and its energy storage capacity is reassessed through high-precision OPF simulation, the accurate result is 52 MW.

[0054] The complete fingerprint of this new scenario can then be used to extract more accurate network state features and the evaluation result of 52 MW as a new record and added to the typical scenario knowledge base.

[0055] This update enhances the knowledge base's coverage of scenarios such as "large industrial loads + moderate wind power penetration," improves the accuracy of future assessments of similar projects, and enables the system to learn and evolve continuously.

[0056] This article uses specific embodiments to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method for evaluating the grid energy storage acceptance capacity based on scene fingerprinting and weighted migration, characterized in that, The method includes: Construct an offline knowledge base, which includes a vector database and a relational database, as detailed below: A large number of power grid operation scenarios are established, and the maximum energy storage capacity of each operation scenario is evaluated through simulation. A multi-dimensional structured feature vector of each operation scenario is extracted as a scenario fingerprint. The vector database stores the scenario fingerprint, and the relational database is used to store scenario information. Online assessment, specifically including: The scene fingerprint of the target power grid is extracted, and the energy storage capacity of the target power grid is estimated by matching the scene fingerprint of the target power grid with the scene fingerprint in the offline knowledge base.

2. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 1, characterized in that, The establishment of the massive power grid operation scenario is as follows: Using standard testing systems combined with external time-series data as benchmark data, the Monte Carlo random sampling method is used to enhance the benchmark data, simulating uncertainties such as load forecasting deviations, fluctuations in renewable energy output, random equipment outages, and changes in network topology, thereby generating a massive number of power grid operation scenarios.

3. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 2, characterized in that, The standard testing systems include IEEE 14-node, IEEE 30-node, and IEEE 118-node testing systems; the external time-series data include load data from the European Transmission Network Operators Alliance (ENTSO-E) and solar radiation and wind speed data from the NASA POWER database.

4. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 1, characterized in that, The simulation evaluation is as follows: For each operating scenario, the maximum energy storage capacity corresponding to each operating scenario is solved using the optimal power flow model, and this is used as the standard evaluation result.

5. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 4, characterized in that, The objective function of the optimal power flow model is to maximize the sum of the injected power of all candidate energy storage nodes. The constraints include node power balance, line power flow, node voltage safety, and generator output constraints.

6. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 1, characterized in that, The multidimensional structured feature vector in the scene fingerprint specifically includes core features, high-level features, and derived and statistical features; The core features include the maximum, minimum, and average values ​​of the total active load of the system, the average value of the total reactive load; the maximum and average values ​​of the total output of new energy sources, the average value of the total output of conventional energy sources, and the power generation / load ratio; New energy penetration rate; proportion of conventional energy regulation capacity; season, date type, and time period; The advanced features include the average load rate of critical lines, the proportion of heavily loaded lines, the average system voltage, the maximum voltage deviation, the number of voltage over-limit nodes, the wind and solar power output ratio, and the correlation coefficient between renewable energy output and load. The derived and statistical features include load peak-valley difference rate, daily load curve shape similarity, load change rate standard deviation, maximum value of new energy output change rate, and new energy prediction error simulation standard deviation.

7. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 6, characterized in that, The vector database is used to store the fingerprints of the operating scenario and support similarity retrieval, while the relational database is used to store complete operating scenario parameters, simulation evaluation results, and original snapshots.

8. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 7, characterized in that, The similarity matching specifically includes: The scene fingerprint of the target power grid is extracted, and then retrieved and matched through the offline knowledge base. That is, the weighted distance between the scene fingerprint of the target power grid and each scene fingerprint in the offline knowledge base is calculated to identify the set of most similar operating scenarios. The weighted distance calculation between the scene fingerprint of the target power grid and each scene fingerprint in the offline knowledge base is only performed by summing the feature values ​​of both sides in the dimensions of the derived and statistical features, and the K similar operating scenarios with the smallest weighted distance are selected.

9. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration as described in claim 8, characterized in that, The estimated energy storage capacity of the target power grid specifically includes: weighted fusion of the evaluation results of K similar operating scenarios based on similarity, converting the weighted distance into similarity, and using the similarity as the weight to perform a weighted average of the evaluation results of the K similar operating scenarios to obtain the estimated energy storage capacity of the target power grid.

10. The grid energy storage acceptance capacity assessment method based on scene fingerprinting and weighted migration according to claim 1, characterized in that, The method also includes dynamic updating, which specifically includes: Once the target power grid obtains the measured evaluation results, the complete scenario fingerprint of the target power grid and the evaluation results are added as new records to the offline knowledge base, thereby enabling dynamic updates of the knowledge base.

Images