A method and system for modeling and parameter identification of differentiated typical transformer area comprehensive load

Abstract

The application provides a differential typical transformer area comprehensive load modeling and parameter identification method and system, the method comprises the following steps: based on the type of the transformer area, a differential transformer area load classification model system is constructed; based on the load classification model system, simulation experiments of various typical working conditions are carried out, the transient data of the key nodes of the transformer area are obtained, and the corresponding key parameter labels are calculated to form a training data set; based on the training data set, a convolutional neural network and a physical mechanism constraint hybrid model are trained to obtain a hybrid identification model capable of identifying the key parameters of the improved comprehensive load aggregation model containing new energy, and the hybrid identification model is used to identify the key parameters of the transformer area to be identified; the application realizes the differential modeling of different types of transformer area loads, constructs a multi-time scale dynamic representation model, the modeling object is various typical transformer areas, and finally a full-process closed-loop system covering transformer area modeling, parameter identification, model adaptation and model library management is formed.

Description

Technical Field

[0001] This invention relates to the field of distribution network load modeling technology, specifically to a method and system for comprehensive load modeling and parameter identification of differentiated typical transformer substations. Background Technology

[0002] As the last mile connecting the distribution network and users, the transformer substation is the core unit for the coordinated interaction of power generation, grid, load, and storage in the new distribution system. The accuracy of its load modeling directly determines the effectiveness of distribution network simulation analysis, fault handling, and optimization control. With the large-scale integration of new sources and loads such as high-proportion distributed renewable energy (photovoltaics, energy storage, etc.) and large-scale electric vehicle charging piles, the load structure of traditional residential, commercial, and public transformer substations has undergone a fundamental transformation, shifting from a single traditional load-dominated structure to a diversified hybrid structure of "renewable energy + traditional loads." This is driving the upgrade of transformer substations from "load-consuming" to "source-load interactive," while also bringing a series of technical challenges that urgently need to be solved in the comprehensive load modeling and parameter identification of transformer substations.

[0003] First, the load structures of different types of distribution substations vary significantly. High-proportion distributed photovoltaic (PV) substations exhibit "source" characteristics during the day, transmitting power back to the distribution network. Their transient response is dominated by inverter control strategies and low-voltage ride-through characteristics. High-proportion electric vehicle (EV) charging substations are centered on constant-power loads and are prone to grid disconnection under fault conditions. Their transient characteristics are closely related to the closed-loop control of the charging converter. Traditional residential and commercial substations are still mainly composed of ZIP static loads and induction motor loads, exhibiting typical electromechanical transient characteristics in their dynamic response. Traditional modeling methods do not consider the differences in load structures among different substations and use a uniform fixed model, resulting in a large deviation between the model and the actual substation characteristics, making it impossible to accurately depict the dynamic response patterns of various substations. Second, the access of new source loads causes the substation load to exhibit multi-timescale dynamic coupling characteristics. The μs-level switching transients of PV inverters, the ms-level electromechanical transients of induction motors, and the min-level characteristics of steady-state load operation are superimposed. Traditional quasi-steady-state modeling methods cannot take into account the dynamic behavior across all time scales and cannot accurately capture load transient fluctuations under fault, charge / discharge switching, and other scenarios. Furthermore, the deep coupling between new power electronic equipment and the control loop of traditional loads, the randomness of new energy output and the disorder of electric vehicle charging further exacerbate the nonlinear characteristics of the transformer area load, resulting in insufficient accuracy and poor adaptability of traditional parameter identification methods, making it difficult to achieve accurate parameter matching under different operating conditions.

[0004] Currently, traditional technical approaches in the field of transformer substation load modeling mainly include fixed structure mechanism modeling, traditional equivalent modeling, and single data-driven modeling. Fixed structure mechanism modeling, based on traditional ZIP load and induction motor models, while possessing a certain degree of physical interpretability, cannot adapt to the multi-various mixed load structures after the access of new sources and loads, and struggles to handle dynamic coupling problems across multiple time scales. Traditional equivalent modeling achieves reduced-order calculations by simplifying the load structure, improving computational efficiency, but carries the risk of "equivalence deviation" and fails to reflect the load differentiation characteristics of different transformer substations. Single data-driven modeling relies on transformer substation operating data for model training, requiring no in-depth understanding of equipment physical mechanisms, but has weak generalization ability, extremely high data quality requirements, and its parameter identification accuracy drops significantly under extreme conditions such as faults, making it difficult to meet engineering application needs. In summary, existing transformer substation load modeling technologies generally suffer from prominent problems such as poor load structure adaptability, insufficient coverage across multiple time scales, low parameter identification accuracy, and difficulty in addressing the needs of differentiated transformer substations, failing to support the refined modeling and management requirements of differentiated typical transformer substations including new energy sources.

[0005] Meanwhile, domestic and international research teams have conducted research on load modeling for distribution transformer areas containing new energy sources, focusing on key areas such as differentiated modeling, multi-timescale coupling processing, and optimization of parameter identification accuracy. International research institutions (such as Imperial College London and the Australian Electricity Research Institute) are exploring multi-agent-based equivalent modeling techniques for distribution transformer loads, achieving differentiated descriptions of different load types through partitioned equivalent modeling. However, this approach suffers from high model complexity, low computational efficiency, and difficulty in adapting to small-scale distribution transformer areas. Domestic research teams (such as Shanghai Jiao Tong University and South China University of Technology) are dedicated to developing hybrid modeling methods that integrate data-driven approaches with physical mechanisms, combining artificial intelligence algorithms to achieve rapid parameter identification. However, they still face bottlenecks such as insufficient adaptability to different distribution transformer load structures, poor parameter robustness under multiple operating conditions, and insufficient real-time performance of online identification. The current core bottleneck in the industry lies in how to overcome the limitations of traditional modeling techniques, achieve differentiated adaptation to different types of distribution transformer load structures, balance multi-timescale dynamic coupling characteristics and parameter identification accuracy, and simultaneously meet the efficiency requirements of online engineering applications, thus addressing the core pain points of insufficient differentiation, inadequate accuracy, and poor adaptability in traditional modeling. Summary of the Invention

[0006] To address the core challenges of existing technologies in modeling and identifying the comprehensive load of differentiated typical transformer substations under large-scale integration with high proportions of new energy and new loads, including poor structural adaptability, insufficient multi-timescale coupling processing, low parameter identification accuracy, and inadequate engineering application adaptability, this invention proposes a method for modeling and identifying the comprehensive load of differentiated typical transformer substations, comprising: Based on the type of transformer area, a differentiated transformer area load classification model system is constructed; Based on the load classification model system, simulation experiments of various typical working conditions are carried out to obtain transient data of key nodes in the transformer area. Based on the transient data and the detailed component model of the transformer area, calculations are performed to obtain the corresponding key parameter labels. The transient data and the corresponding key parameter labels constitute the training dataset. Based on the training dataset, a hybrid model of convolutional neural network and physical mechanism constraint is trained to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy sources. The hybrid identification model is then used to identify the key parameters of the transformer area to be identified. The improved integrated load aggregation model containing new energy sources is obtained by using a topology model that aggregates the load classification of distribution areas with the equivalent impedance of the upper-level power grid and multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources as a function of static voltage, dynamic voltage, and frequency.

[0007] Preferably, the construction of a differentiated load classification model system for transformer substations based on their type includes: For each type of transformer substation, set the basic parameters of the substation based on its type and build a simulation scenario; Based on the basic parameters and simulation scenarios of the transformer substations, refined component models are built according to the differences in load structure of the substations. These refined component models are then integrated according to the load ratio to form a substation-level comprehensive load model, which serves as a differentiated substation load classification model system. The types include: high-proportion photovoltaic areas, charging areas, and traditional areas; the simulation scenarios include: normal operation, voltage sag fault, power backfeed, and charging pile charging / discharging switching. The power backfeed occurs in high-proportion photovoltaic areas, and the charging pile charging / discharging switching occurs in charging areas.

[0008] Preferably, the integrated load model at the district level for the high-proportion photovoltaic power station area includes a photovoltaic inverter model and a ZIP hybrid load model; The photovoltaic inverter model adopts a three-phase pulse width modulation inverter model with maximum power point tracking control and low voltage ride-through technology; the ZIP hybrid load model adopts a hybrid model of constant impedance, constant current and constant power. The charging area includes a charging pile model and a constant impedance auxiliary load model; The charging pile model used is a conventional DC fast charging pile model.

[0009] Preferably, the simulation experiments conducted based on the load classification model system for various typical operating conditions to obtain transient data of key nodes in the transformer area include: Based on the load classification model system of various types of transformer substations, simulation experiments were carried out under various typical operating conditions. Transient data of key nodes on the low-voltage side of the transformer substation were collected using a high-precision power analyzer during the simulation experiment. The transient data were preprocessed using a combination of outlier removal, improved wavelet denoising, and min-max normalization.

[0010] Preferably, the active power expression of the improved integrated load aggregation model containing new energy sources is as follows:

[0011] The reactive power expression for the improved integrated load aggregation model incorporating new energy sources is as follows:

[0012] In the formula, This represents the active power in the improved integrated load aggregation model that includes new energy sources. This represents the reactive power in the improved integrated load aggregation model that includes new energy sources. This indicates the active power under rated operating conditions. This indicates the reactive power under rated operating conditions. Indicates voltage deviation. This represents the static coefficient of active voltage. Represents the static coefficient of reactive voltage. This represents the dynamic time constant of the active voltage. This represents the dynamic time constant of reactive voltage. Represents the static coefficient of active frequency. The static coefficient of reactive frequency is represented; the key parameters include the static coefficient of active voltage, the static coefficient of reactive voltage, the dynamic time constant of active voltage, the dynamic time constant of reactive voltage, the static coefficient of active frequency, and the static coefficient of reactive frequency.

[0013] Preferably, the loss function used when training the hybrid model of convolutional neural network and physical mechanism constraints is calculated as follows:

[0014] In the formula, For loss function, For the number of samples, For the first The measured power value of each sample For the first The model predicts the power value for each sample. To constrain the penalty coefficient, For the first One key parameter to be identified For the first The maximum constraint value of each key parameter For the first Minimum constraint values ​​for each key parameter; and It is calculated using parameters from the detailed model of the components in the transformer area, representing the first... Physical mechanism constraints on key parameters.

[0015] Preferably, after identifying the key parameters of the station area to be identified using the hybrid identification model, the method further includes: The identified key parameter combinations are substituted into the improved integrated load aggregation model containing new energy sources. The model is adapted by combining the load structure and dynamic characteristics of various types of transformer substations. The optimal improved integrated load aggregation model, key parameter configurations and characteristic laws are classified and archived according to transformer substation type and integrated into the differentiated typical transformer substation load structured model library.

[0016] Based on the same inventive concept, the present invention also provides a differentiated typical transformer area integrated load modeling and parameter identification system, including: a load model module, a data acquisition module and a parameter identification module; The load model module is used to construct a differentiated load classification model system for transformer substations based on their type. The data acquisition module is used to conduct simulation experiments of various typical working conditions based on the load classification model system, obtain transient data of key nodes in the transformer area, and perform calculations based on the transient data and the detailed component model of the transformer area to obtain corresponding key parameter labels. The transient data and the corresponding key parameter labels constitute a training dataset. The parameter identification module is used to train a hybrid model of convolutional neural network and physical mechanism constraint based on the training dataset, to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy, and to use the hybrid identification model to identify the key parameters of the transformer area to be identified. The improved integrated load aggregation model containing new energy sources is obtained by using a topology model that aggregates the load classification of distribution areas with the equivalent impedance of the upper-level power grid and multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources as a function of static voltage, dynamic voltage, and frequency.

[0017] Preferably, the load model module is specifically used for: For each type of transformer substation, set the basic parameters of the substation based on its type and build a simulation scenario; Based on the basic parameters and simulation scenarios of the transformer substations, refined component models are built according to the differences in load structure of the substations. These refined component models are then integrated according to the load ratio to form a substation-level comprehensive load model, which serves as a differentiated substation load classification model system. The types include: high-proportion photovoltaic areas, charging areas, and traditional areas; the simulation scenarios include: normal operation, voltage sag fault, power backfeed, and charging pile charging / discharging switching. The power backfeed occurs in high-proportion photovoltaic areas, and the charging pile charging / discharging switching occurs in charging areas.

[0018] Preferably, the integrated load model of the high-proportion photovoltaic power station area in the data acquisition module includes a photovoltaic inverter model and a ZIP hybrid load model; The photovoltaic inverter model adopts a three-phase pulse width modulation inverter model with maximum power point tracking control and low voltage ride-through technology; the ZIP hybrid load model adopts a hybrid model of constant impedance, constant current and constant power. The charging area includes a charging pile model and a constant impedance auxiliary load model; The charging pile model used is a conventional DC fast charging pile model.

[0019] Preferably, the data acquisition module performs simulation experiments on various typical operating conditions based on the load classification model system to obtain transient data of key nodes in the transformer area, including: Based on the load classification model system of various types of transformer substations, simulation experiments were carried out under various typical operating conditions. Transient data of key nodes on the low-voltage side of the transformer substation were collected using a high-precision power analyzer during the simulation experiment. The transient data were preprocessed using a combination of outlier removal, improved wavelet denoising, and min-max normalization.

[0020] Preferably, the active power expression of the improved integrated load aggregation model containing new energy sources in the parameter identification module is as follows:

[0021] The reactive power expression for the improved integrated load aggregation model incorporating new energy sources is as follows:

[0022] In the formula, This represents the active power in the improved integrated load aggregation model that includes new energy sources. This represents the reactive power in the improved integrated load aggregation model that includes new energy sources. This indicates the active power under rated operating conditions. This indicates the reactive power under rated operating conditions. Indicates voltage deviation. This represents the static coefficient of active voltage. Represents the static coefficient of reactive voltage. This represents the dynamic time constant of the active voltage. This represents the dynamic time constant of reactive voltage. Represents the static coefficient of active frequency. The static coefficient of reactive frequency is represented; the key parameters include the static coefficient of active voltage, the static coefficient of reactive voltage, the dynamic time constant of active voltage, the dynamic time constant of reactive voltage, the static coefficient of active frequency, and the static coefficient of reactive frequency.

[0023] Preferably, the loss function used by the parameter recognition module when training the hybrid model of convolutional neural network and physical mechanism constraints is calculated as follows:

[0024] In the formula, For loss function, For the number of samples, For the first The measured power value of each sample For the first The model predicts the power value for each sample. To constrain the penalty coefficient, For the first One key parameter to be identified For the first The maximum constraint value of each key parameter For the first Minimum constraint values ​​for each key parameter; and It is calculated using parameters from the detailed model of the components in the transformer area, representing the first... Physical mechanism constraints on key parameters.

[0025] Preferably, the system also includes a database module, which is used to substitute the identified key parameter combinations into the improved integrated load aggregation model containing new energy sources, combine the load structure and dynamic characteristics of various types of transformer substations, complete the model adaptation, and classify and archive the optimal improved integrated load aggregation model, key parameter configuration and characteristic laws according to transformer substation type, and integrate them into the differentiated typical transformer substation load structured model library.

[0026] This invention provides a method and system for differentiated typical transformer substation integrated load modeling and parameter identification. The method includes: constructing a differentiated transformer substation load classification model system based on the type of the transformer substation; conducting simulation experiments under various typical operating conditions based on the load classification model system to obtain transient data of key nodes in the transformer substation, and calculating corresponding key parameter labels based on the transient data and detailed component models of the transformer substation, using the transient data and corresponding key parameter labels to form a training dataset; training a hybrid model of convolutional neural network and physical mechanism constraints based on the training dataset to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy sources, and using the hybrid identification model to identify the key parameters of the transformer substation to be identified; wherein, the improved integrated load aggregation model containing new energy sources is obtained by aggregating the transformer substation load classification model using the equivalent impedance of the upper-level power grid plus the topology of multiple loads in parallel, and the key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources with voltage static, voltage dynamic and frequency changes. The main advantages of this invention lie in achieving differentiated and accurate modeling of loads in different types of distribution transformer areas, constructing a multi-timescale dynamic characterization model, and proposing a parameter intelligent identification method based on the fusion of AI and physical mechanisms. The modeling objects are various typical distribution transformer areas with a high proportion of renewable energy, aiming to balance modeling accuracy and engineering calculation efficiency. Ultimately, it forms a closed-loop system covering the entire process from distribution transformer area modeling, parameter identification, model adaptation to model library management. This invention can significantly improve the accuracy of dynamic characteristic analysis and parameter identification efficiency of loads in different distribution transformer areas, providing a standardized modeling tool for the refined management and control of distribution transformer areas with renewable energy. Through differentiated modeling, it can achieve coupled analysis of component-level characteristics and overall dynamics at the distribution transformer area level. Through AI-driven parameter identification and multi-condition adaptation technology, it achieves accurate characterization and efficient adaptation of transient and steady-state characteristics of various distribution transformer area loads. This effectively solves key technical problems such as voltage fluctuations, fault response deviations, and poor parameter adaptability caused by high-proportion renewable energy access, improving the safe and stable operation level and renewable energy absorption efficiency of distribution transformer areas with renewable energy. It provides core technical support for the planning and design, fault handling, and source-grid-load-storage coordinated control of distribution transformer units in new distribution systems. Attached Figure Description

[0027] Figure 1 A flowchart of a differentiated typical transformer substation integrated load modeling and parameter identification method provided by the present invention; Figure 2 This is a schematic diagram of a simulation case model of a 209-node system provided by the present invention; Figure 3 This is a schematic diagram of the improved NSLM aggregation model structure provided by the present invention; Figure 4 A flowchart illustrating an example of a differentiated typical transformer substation integrated load modeling and parameter identification method provided by the present invention; Figure 5 A schematic diagram illustrating the process of multi-condition simulation and high-quality dataset construction provided by this invention; Figure 6 A schematic diagram illustrating the implementation process of AI-driven intelligent identification of transformer area load parameters provided by the present invention; Figure 7 This invention provides a schematic diagram of a differentiated typical transformer area integrated load modeling and parameter identification system. Detailed Implementation

[0028] This invention proposes a method and system for comprehensive load modeling and parameter identification of differentiated typical transformer substations. Addressing the core pain points of comprehensive load modeling and parameter identification in differentiated typical transformer substations after the large-scale integration of high-proportion renewable energy and new load types, such as poor structural adaptability, insufficient multi-timescale coupling processing, low parameter identification accuracy, and insufficient engineering application adaptability, this invention provides a precise, efficient, and highly adaptable technical solution for comprehensive load modeling and parameter identification of transformer substations. By constructing a differentiated load model framework adapted to different transformer substation types (high-proportion photovoltaic substations, high-proportion electric vehicle charging substations, etc.), and integrating refined modeling of multiple component types, AI-driven parameter identification strategies, and multi-condition adaptation optimization technology, this invention achieves accurate characterization of the dynamic characteristics of transformer substations across multiple timescales, while simultaneously enabling rapid parameter identification and dynamic matching for transformer substations with different load structures. This invention can provide precise digital modeling support for differentiated typical distribution transformer areas containing new energy sources, and provide key technical tools for fault analysis, optimized control, and coordinated management and control of power generation, grid, load and storage at the distribution transformer area level. It promotes the transformation of distribution transformer area operation and management towards intelligent direction of "differentiated adaptation and refined regulation", helps the efficient consumption of high proportion of new energy sources at the distribution transformer area level, and comprehensively improves the safe and stable operation level and economic efficiency of distribution transformer area units in the new power distribution system.

[0029] This invention achieves differentiated load characterization of transformer substations by (1) constructing a differentiated transformer substation load classification model system, building a refined model at the component level based on the structural differences of high-proportion photovoltaic, electric vehicle charging and traditional transformer substations, integrating equipment such as ZIP static load, third-order induction motor, photovoltaic inverter with MPPT (maximum power point tracking) and LVRT (low voltage ride-through technology), and introducing load ratio adaptation coefficients; (2) proposing a multi-condition simulation and dataset construction method, covering typical faults and dynamic conditions, collecting transient data of key nodes, and constructing a standardized training dataset through combined preprocessing; (3) establishing an AI-driven parameter identification method, using a hybrid model of CNN and physical mechanism constraints, optimizing hyperparameters to achieve accurate parameter identification of the improved NSLM model (i.e., the improved integrated load aggregation model with new energy), and screening the optimal parameters through multi-index verification; (4) developing a model adaptation and model library upgrade system, refining the load characteristic correlation mechanism, classifying and archiving the optimal model and parameters, and forming a closed-loop management system that can be quickly called and dynamically updated. The innovation of this invention is reflected in three aspects: differentiated adaptation of transformer substations, collaborative modeling of multiple components, and precise AI identification. The core breakthrough is that it solves the problems of poor adaptability of traditional modeling and insufficient integration of new source and load technologies. It breaks the limitation of "one-size-fits-all" and realizes a closed loop of the whole process from component modeling, data processing, parameter identification to model library management, and builds a refined modeling and parameter identification system that adapts to different transformer substations and takes into account multiple time scales.

[0030] Example 1: A method for integrated load modeling and parameter identification of differentiated typical transformer substations, such as Figure 1 As shown, it includes: Step 1: Based on the type of transformer area, construct a differentiated transformer area load classification model system; Step 2: Based on the load classification model system, conduct simulation experiments for various typical working conditions to obtain transient data of key nodes in the transformer area. Then, calculate the corresponding key parameter labels based on the transient data and the detailed component model of the transformer area. The transient data and the corresponding key parameter labels constitute the training dataset. Step 3: Based on the training dataset, train a hybrid model of convolutional neural network and physical mechanism constraints to obtain a hybrid identification model that can identify the key parameters of the improved integrated load aggregation model containing new energy sources, and use the hybrid identification model to identify the key parameters of the transformer area to be identified. Among them, the improved integrated load aggregation model with new energy sources is obtained by aggregating the load classification model of the distribution area using the equivalent impedance of the upper-level power grid and the topology of multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model with new energy sources as voltage static, voltage dynamic and frequency changes.

[0031] The following specific example illustrates this method.

[0032] Based on the improved NSLM aggregation model constructed in this invention, a simulation verification system for a 209-node radial combined distribution network was built. The system topology is as follows: Figure 2 As shown, the system includes 5 main network nodes, 2 IEEE 33 node distribution subnets, and 2 IEEE 69 node distribution subnets, with node numbering rules as described above. In the system, the load nodes of the IEEE 33 / 69 subnets are replaced with... Figure 3 The improved NSLM aggregation model shown is configured with differentiated source-load ratios according to different transformer substation types (such as high-proportion photovoltaic substations, high-proportion charging substations, and traditional substations). Multi-scenario simulation verification is carried out to compare the effects of this method with traditional modeling methods, thus verifying the effectiveness and superiority of this method.

[0033] The flowchart of this differentiated typical transformer area integrated load modeling and parameter identification method example is as follows: Figure 4 As shown.

[0034] Step 1 specifically includes: 1-1: Basic parameters of the transformer area and the construction of the simulation scenario This embodiment selects two typical transformer substations as test objects, and the specific basic parameters are as follows: (1) Photovoltaic distribution area: 10kV / 400V standard distribution area, equipped with conventional capacity distribution transformer, the equivalent impedance of the upstream power grid is calculated according to the following formula: (1) In the formula, The equivalent impedance (Ω) of the upstream power grid. This represents the percentage of the transformer's short-circuit voltage. The system's rated voltage. This refers to the system short-circuit capacity (MVA). The photovoltaic modules are configured as conventional monocrystalline silicon modules, and their total installed capacity accounts for a relatively high proportion of the total load in the distribution area. The ZIP mixed load consists of conventional residential electrical equipment, accounting for a moderate proportion. "Moderate" here means 25%–40%, which is a conventional configuration without additional control; "relatively high" means 50%–80%, which has high penetration and requires voltage regulation / reactive power compensation.

[0035] When building the simulation scenario, typical operating conditions of the actual transformer substation are simulated, including normal operation, voltage sag fault, power backfeed (PV substation only), and charging pile charging / discharging switching (charging substation only). This covers the entire timescale from μs-level PV inverter switching transients and ms-level fault transients to min-level load steady-state operation, providing scenario support for subsequent data acquisition, modeling, and parameter identification. The input for this step is the basic parameters of the transformer substation (transformer parameters, load structure ratio, etc.), and the output is the standardized transformer substation simulation scenario and scenario parameter configuration.

[0036] (2) Charging area: A standard 10kV / 400V charging area, equipped with a distribution transformer suitable for the charging pile load. The equivalent impedance of the upstream power grid is calculated according to formula (1). Conventional DC fast charging piles are selected, and their total power accounts for a relatively high proportion of the total load of the charging area. The auxiliary loads are lighting and monitoring facilities, which adopt a constant impedance model and have a moderate proportion. Here, the total power of the charging piles accounts for a relatively high proportion of the total load of the charging area, specifically 60%–90%. The auxiliary load accounts for a moderate proportion, specifically 5%–15% of the total load of the charging area.

[0037] When building simulation scenarios, typical operating conditions of the actual operation of the transformer substation are simulated, including normal operation, voltage sag fault, power backfeed (PV substation only), and charging pile charging and discharging switching (charging substation only). This covers the entire time scale from μs-level PV inverter switching transients and ms-level fault transients to min-level load steady-state operation, providing scenario support for subsequent data acquisition, modeling, and parameter identification.

[0038] 1-2: Implementation of Differentiated Distribution Area Load Classification Modeling Based on the differences in load structure between the two transformer substations, refined component models are built separately and integrated to form a substation-level comprehensive load model. This step follows the simulation scenario and basic parameters of the substation output in the first step. The specific implementation process is as follows, and the final output is the substation-level comprehensive load model and model parameters for the photovoltaic substation and the charging substation, respectively.

[0039] (1) Photovoltaic area load modeling The integrated load model for the photovoltaic distribution area consists of a photovoltaic inverter model (70%) and a ZIP hybrid load model (30%). The specific implementation of each component model is as follows: 1) Photovoltaic inverter model: A three-phase PWM (Pulse Width Modulation) inverter model with MPPT control and LVRT characteristics is adopted. The MPPT control adopts the disturbance observation method with a tracking accuracy of ≤2%. The LVRT characteristics meet the requirements of GB / T 19964-2021. When the voltage drops, reactive power support is issued to the grid. The reactive power support current command is calculated according to the following formula: (2) In the formula, The reactive current command is (A). The actual voltage (V) at the grid connection point. Rated voltage, This is the reactive power support coefficient. This is the maximum output current of the inverter.

[0040] The inverter adopts a conventional three-phase two-level topology and is equipped with standard filter components and a PI controller to meet the actual operation response requirements of the project.

[0041] 2) ZIP Hybrid Load Model: This model employs a hybrid approach of constant impedance, constant current, and constant power, with its active power... and reactive power The expression is as follows: (3) (4) In the formula, , Rated voltage Rated active and reactive power at the specified levels; , , These are the active power weighting coefficients for constant impedance, constant current, and constant power loads, respectively, satisfying... + + =1; , , These are their reactive power weighting coefficients, satisfying... + + =1, and the weighting coefficient is reasonably selected based on the load characteristics of residents in the photovoltaic area.

[0042] (2) Load modeling of charging station area The comprehensive load model for the charging station area consists of a charging pile model (85%) and a constant impedance auxiliary load model (15%). The specific implementation of each component model is as follows: 1) Charging Pile Model: A conventional DC fast charging pile model is adopted, exhibiting constant power load characteristics under normal voltage, determined by the closed-loop control of the charging converter. In case of a fault, if the voltage drops below a set threshold, low-voltage disconnection is achieved, and the disconnection response time meets engineering requirements. The charging converter adopts a conventional topology and a dual closed-loop control strategy, with parameters configured according to engineering standards.

[0043] The characteristic expression for a constant power load is as follows: (5) In the formula, The rated active power of the charging pile, The low-voltage disconnection threshold is sgn, which is the sign function. ≥ When the charging station is in normal operation; when < At that time, the charging station went offline.

[0044] 2) Constant Impedance Auxiliary Load Model: Lighting and monitoring facilities use purely resistive loads, and their impedance values ​​are calculated using the following formula: (6) In the formula, Z The constant impedance value (Ω).U 0 represents the rated voltage. The rated active power of the auxiliary load can be calculated based on conventional load parameters, and its active and reactive power expressions are as follows: (7) (8) (3) Integration of the comprehensive model of the transformer area The individual component models are integrated according to load proportions, and a load proportion adaptation coefficient k is introduced to construct a comprehensive load model at the transformer substation level. The active and reactive power expressions of the comprehensive model are as follows: (9) (10) In the formula, , These are the total active and reactive power (kW, kvar) of the transformer area, respectively. k 1. k 2 represents the proportional adaptation coefficient for the two types of loads, satisfying... k 1+ k 2 = 1; P 1. Q 1 represents the active and reactive power of the first type of load (photovoltaic inverter / charging pile); P 2. Q 2 represents the active and reactive power of the second type of load (ZIP mixed load / constant impedance load).

[0045] Load ratio matching coefficient between photovoltaic power station area and charging power station area k 1. k 2. Set reasonable parameters according to the load structure ratio of each transformer area to meet the requirements. k 1+ k 2=1. The input for this step is the load structure ratio of the transformer area and the model parameters of each component output from step 1-1. The output is the comprehensive load model of the two transformer areas and the model expression (Formulas 9 and 10).

[0046] Step 2 specifically includes: Multi-condition simulation and high-quality dataset construction Based on the integrated load model of the transformer area output in steps 1-2, three typical operating conditions are designed, simulation experiments are carried out, transient data of key nodes are collected, and a standardized training dataset is constructed after preprocessing to provide data support for subsequent parameter identification. This step follows the output results of the second step. The specific implementation process is as follows, and the final output is a standardized training dataset (including training set, validation set, and test set).

[0047] (1) Working condition design Based on the characteristics of the two transformer substations, the following operating conditions are designed to cover scenarios such as normal operation, fault, and dynamic switching: 1) Operating Condition 1: Normal operating condition, the simulation duration covers the normal operating cycle, the photovoltaic output of the photovoltaic area changes naturally according to the solar irradiance, and the charging piles in the charging area are started in batches to simulate the normal operating state of the area.

[0048] 2) Operating Condition 2: Voltage Sag Fault Condition. The simulation duration covers the entire process from fault occurrence to recovery. At a set time, a single-phase ground fault occurs on the low-voltage side of the transformer area. The fault lasts for a set duration, and the voltage drops to a certain percentage of the rated value. After the fault is cleared, the voltage recovers to the rated value, simulating the transient response of the load under fault conditions.

[0049] 3) Operating Condition 3: Dynamic switching operating condition, the simulation duration covers the dynamic change cycle of the load, the photovoltaic area simulates the scenario of photovoltaic output recovery after a sudden drop, the charging area simulates the scenario of multiple charging piles starting and stopping at the same time, and simulates the dynamic switching state of the load.

[0050] (2) Data collection A high-precision power analyzer is used to collect transient data such as active power, reactive power, voltage, and current at key nodes on the low-voltage side of the transformer substation. The sampling frequency meets the requirements for wideband signal acquisition, and the sampling duration covers all operating conditions to ensure that the collected data includes dynamic data across the entire time scale. Multiple sets of parallel data are collected for each operating condition to avoid data randomness, and the range of collected data conforms to the normal operating range of the transformer substation.

[0051] (3) Data preprocessing This step uses the collected raw transient data as input. To eliminate noise interference, outliers, and dimensional influences in the raw data, and to ensure the accuracy of subsequent parameter identification, a combined preprocessing strategy of "outlier removal → improved wavelet denoising → min-max normalization" is adopted. Each step is progressive and closely connected, such as... Figure 5 As shown, the specific implementation steps and input / output connections are as follows: 1) Outlier Removal: Using the collected raw transient data such as voltage, current, and power as input, the 3σ criterion is applied to remove outlier data exceeding the range [μ-3σ, μ+3σ] caused by equipment errors and interference during data acquisition, ensuring data validity. The output of this step is the clean raw data after outlier removal, which serves as the input for subsequent wavelet denoising improvements. Where μ is the data mean and σ is the data standard deviation, calculated using the following formula: (11) (12) In the formula, For the i-th collected data, n This represents the total number of data items.

[0052] 2) Improved Wavelet Denoising: Using the clean raw data after outlier removal as input, and considering the wide bandwidth and multi-timescale characteristics of the transformer load signal, the Symlet wavelet basis function is selected. The signal is divided into 16 sub-bands through 6-level wavelet packet decomposition. An energy entropy weighted thresholding method is used to smooth the traditional hard thresholding function, avoiding the "pseudo-Gibbs effect" and effectively filtering out high-frequency noise. The output of this stage is the denoised stationary signal, which serves as the input for min-max normalization. The specific formula is as follows: Subband energy entropy calculation: (13) In the formula, For the first The energy percentage of samples in each subband.

[0053] Dynamic threshold calculation: (14) In the formula, For the first The noise standard deviation of each sub-band n For the number of data points, For the first Dynamic threshold of each sub-band.

[0054] Smoothing threshold function: (15) In the formula, w is the original wavelet coefficient, w' is the denoised wavelet coefficient, and λ is the dynamic threshold.

[0055] 3) Min-Max Normalization: Using the stationary signal after improved wavelet denoising as input, to eliminate the dimensional differences of different physical quantities (voltage, current, power) and avoid affecting the training effect of the subsequent AI identification model, the denoised data is normalized to the [0,1] interval. The output of this step is the normalized signal, which is integrated to construct the training set, validation set, and test set, serving as the core input for parameter identification in the fourth step. The formula is as follows: (16) In the formula, x represents the denoised data. , ...

[0056] After preprocessing, a standardized training dataset is constructed, divided into a training set (70%), a validation set (20%), and a test set (10%), for subsequent parameter identification. The inputs to this step are the comprehensive load model output from steps 1-2 and the simulation scenario output from step 1-1, and the output is the preprocessed standardized training dataset, which is directly used as the input data for step 3.

[0057] Step 3 specifically includes: AI-driven intelligent identification of transformer area load parameters A hybrid model combining CNN and physical mechanism constraints is used to intelligently identify key parameters of the improved NSLM aggregation model, achieving accurate matching of load parameters for different transformer areas. This step follows the standardized training dataset output in the third step. The specific implementation process is as follows: Figure 6 The final output is the optimal parameter combination for the two transformer substations and a parameter identification error report.

[0058] (1) Construction of the improved NSLM aggregation model Figure 3 This invention presents an improved version of the NSLM (Network Substation Integrated Load Aggregation Model), which serves as the core equivalent modeling unit for renewable energy substations. The model employs a topology of "equivalent impedance of the upstream power grid + parallel connection of multiple loads," integrating typical source-load components such as ZIP static loads, induction motors, reactive power compensation capacitors, Class A VSCs (photovoltaic / direct-drive wind turbines), Class B VSCs (general photovoltaic), and doubly-fed induction generators on the 400V low-voltage side. This allows for a unified characterization of the static and dynamic load characteristics of different types of substations (high-proportion photovoltaic substations, charging substations, and traditional substations), providing a fundamental model support for subsequent parameter identification and multi-scenario simulation.

[0059] Among them, Class A VSC (photovoltaic / direct-drive wind turbine): high-voltage / medium-voltage grid connection, large capacity, low voltage ride-through LVRT, strong dynamic response, and modeled according to centralized new energy standards. Class B VSC (General Photovoltaic): Low-voltage 380V grid connection, small capacity, no forced ride-through, simple response, modeled as ordinary distributed photovoltaic.

[0060] An improved NSLM aggregation model is constructed based on the load characteristics of the two transformer areas, and its active and reactive power expressions are as follows: (17) (18) In the formula, , These are the active and reactive power (kW, kvar) under rated operating conditions, respectively. This represents the static coefficient of active voltage. Represents the static coefficient of reactive voltage; This represents the dynamic time constant of the active voltage. Represents the reactive voltage dynamic time constant (ms); Represents the static coefficient of active frequency. Represents the static coefficient of reactive frequency; , These are voltage and frequency deviations, respectively. ΔU = U - U 0, = - , =50Hz, U and These are the actual voltage and frequency, respectively. U 0 and (These are the rated voltage and frequency, respectively).

[0061] The key parameters that need to be identified are: , , , , , There are a total of 6 parameters.

[0062] (2) Construction of hybrid recognition model A hybrid recognition model combining CNN and physical mechanism constraints is constructed. The CNN model is used to extract data features, while the physical mechanism constraints are used to ensure the rationality of the parameters and avoid the parameters from exceeding the actual range of engineering.

[0063] The CNN model structure adopts conventional parameter configuration. The input layer corresponds to the preprocessed time series data and includes convolutional layers, pooling layers, fully connected layers, and an output layer. The output layer corresponds to the parameters to be identified. The physical mechanism constraints are combined with the actual parameters of the equipment in the distribution area to set a reasonable parameter constraint range to ensure that the parameters conform to the actual engineering situation.

[0064] Physical mechanism constraints: Based on the actual parameters of the equipment in the distribution area, set the parameter constraint range: The loss function uses mean squared error (MSE) combined with a physical mechanism constraint penalty term, as shown in the following formula: (19) In the formula, For loss function, For the number of samples, For the first The measured power value of each sample For the first The model predicted power value for each sample is calculated by substituting the key parameters identified by the model into Formula 17. To constrain the penalty coefficient, For the first One key parameter to be identified For the first The maximum constraint value of each key parameter For the first Minimum constraint values ​​for each key parameter; and It is calculated using parameters from the detailed model of the components in the transformer area, representing the first... Physical mechanism constraints on key parameters.

[0066] (3) Parameter identification, training and optimization The preprocessed training set is input into the hybrid identification model, and the adaptive gradient descent algorithm (Adam) is used to optimize the model hyperparameters. Reasonable learning rate, number of iterations and other parameters are set. During the training process, overfitting is controlled by validation set loss. After training, the parameter identification error is calculated using the test set to ensure that the error meets the requirements.

[0067] During training, the validation set loss is calculated every 100 iterations. Training is stopped when the validation set loss no longer decreases after 50 consecutive iterations to avoid overfitting. After training, the model is tested using a test set, and the parameter identification error is calculated. The error calculation formula is as follows: (20) In the formula, The average parameter identification error (%) These are the actual values ​​of the parameters (calculated from the detailed model of the component). The parameter values ​​obtained for identification.

[0068] like If the value is greater than 5%, adjust the model hyperparameters (learning rate, penalty coefficient). α ), retrain until the error requirement is met; if If the parameter value is ≤5%, then the optimal parameter combination is output, completing parameter identification. The inputs to this step are the standardized training dataset and the improved NSLM aggregation model structure output from step three; the output is the optimal parameter combination for the two transformer substations. , The six parameters (including parameter identification error report) serve as the core input for the subsequent step 4.

[0069] (4) Identification results After training and optimization, the key parameter identification results of the improved NSLM aggregation model for the two transformer areas are as follows: 1) Photovoltaic area: The key parameter identification results are within the constraints, and the average parameter identification error meets the requirement of ≤5%.

[0070] 2) Charging station area: The key parameter identification results are within the constraints, and the average parameter identification error meets the requirement of ≤5%.

[0071] After step 3, the following also includes: Step 4: Model Adaptation and Upgrade of Structured Model Library Substitute the optimal parameter combination output from step 3 into the improved NSLM aggregation model, combine the load structure and dynamic characteristics of the two transformer areas to complete the model adaptation, classify and archive the optimal aggregation model, parameter configuration and characteristic laws according to transformer area type, and integrate them into the differentiated typical transformer area load structured model library. This step follows the output results of step 4. The specific implementation process is as follows. The final output is the adapted transformer area aggregation model and the upgraded structured model library.

[0072] (1) Model adaptation verification: The adapted aggregated model is compared with the detailed model of the transformer area through simulation, and the simulation fit is calculated. The formula for calculating the fit is as follows: (twenty one) In the formula, The simulation fit rate (%) The first detailed model of the transformer area One active power simulation value, For the adapted aggregation model, the first One active power simulation value, This represents the number of simulation data sets.

[0073] Verification results: The simulation fit of both transformer areas is ≥95%, the adaptation effect is good, and it meets the requirements of engineering applications.

[0074] (2) Model Library Upgrade: A structured model library is constructed, categorized into "Photovoltaic Distribution Areas," "Charging Distribution Areas," and "Traditional Distribution Areas." Each category stores the corresponding aggregated model, parameter configurations, load structure ratios, dynamic characteristic patterns, and applicable operating conditions. A model index is established to enable rapid model retrieval and dynamic updates. The model library uses an SQLite database for storage and supports retrieval based on distribution area type, load ratio, parameter range, and other conditions, meeting the modeling needs of different scenarios in practical engineering applications. The input for this step is the adapted aggregated model, parameter configuration, and distribution area load characteristic patterns; the output is the upgraded structured model library, which also provides model support for the simulation verification in step 5.

[0075] Step 5: Simulation Verification and Effect Analysis Based on the comprehensive load model output from steps 1-2, the optimal parameter combination output from step 3, and the adapted aggregate model output from step 4, multi-scenario simulation verification is carried out to compare the effects of this method with traditional modeling methods, verifying the effectiveness and superiority of this method. This step inherits the output results of all previous steps. The specific verification process is as follows. The final output is a simulation verification report and effect comparison analysis results.

[0076] 5.1 Verification Scenario Design Three verification scenarios are designed, covering typical operating conditions such as broadband disturbances, fault transients, and dynamic switching, as detailed below: Scenario 1: Wideband Disturbance Scenario (High Proportion Photovoltaic Area): A high proportion photovoltaic area model is configured at node 10 of the IEEE33 subnet to simulate the μs-level high-frequency disturbance generated by the switching action of the photovoltaic inverter. At the same time, the slow load change is superimposed to verify the accuracy of the method in capturing and modeling signals of multiple time scales.

[0077] Scenario 2: Fault Transient Scenario (High Proportion Charging Area): A high proportion charging area model is configured at node 50 of the IEEE69 subnet to simulate a two-phase short-circuit fault on the low-voltage side, verifying the ability of this method to characterize fault transients and its parameter adaptability.

[0078] Scenario 3: Comparison of dynamic response of the whole system: The improved NSLM aggregation model and the traditional ZIP + induction motor fixed model are used to simulate the normal operation of the 209-node system, and the simulation accuracy and computational efficiency of the two are compared.

[0079] 5.2 Verification Indicators and Calculation Formulas The following three core indicators were selected to quantitatively verify the effectiveness of this method: 1) Simulation fit degree γ: Calculated according to formula (21), it reflects the degree of fit between the model and the actual characteristics of the transformer area. The higher the γ, the higher the modeling accuracy.

[0080] 2.) Parameter identification error δ: Calculated according to formula (20), it reflects the accuracy of parameter identification. The lower the δ, the better the identification effect.

[0081] 3) Simulation efficiency improvement rate η: The simulation time of this method is compared with that of the traditional modeling method. The calculation formula is as follows: (twenty two) In the formula, η is the simulation efficiency improvement rate (%). The simulation time required by traditional modeling methods The simulation time for this method is represented by η; the higher the η, the higher the simulation efficiency.

[0082] 5.3 Verification Results and Analysis After simulation verification, the verification results for each scenario are as follows: (1) Scenario 1 (wideband disturbance scenario): The simulation fitting degree of this method is ≥98%, the parameter identification error is ≤3.5%, and it can accurately capture the coupling characteristics of signals at multiple time scales; the traditional fixed model cannot adapt to the characteristics of multiple time scales, the fitting degree is only 82.3%, and the error is 12.7%.

[0083] (2) Scenario 2 (fault transient scenario): During the fault, the power fluctuation and recovery process of the transformer area modeled by this method are consistent with reality. The recovery time after fault clearing is ≤0.05s different from the measured data, and the fitting degree is ≥98.5%. The power deviation of the traditional fixed model is more than 10%, the recovery time deviation is 0.15s, and the fitting degree is only 81.5%.

[0084] (3) Scenario 3 (Comparison of dynamic response of the whole system): In the full-condition simulation of the 209-node system, the simulation efficiency of this method is more than 40% higher than that of the traditional method, and the simulation fitting degree is ≥97%, which is much higher than the 83% of the traditional method, which fully demonstrates the superiority of this method in balancing modeling accuracy and computational efficiency.

[0085] Implementation Conclusion This embodiment takes high-proportion photovoltaic power distribution areas and high-proportion electric vehicle charging power distribution areas as application objects, and fully implements the "a method and system for comprehensive load modeling and parameter identification of differentiated typical power distribution areas" proposed in this patent. It builds on the implementation results of all previous steps, and through differentiated load modeling, multi-condition data preprocessing, AI-driven parameter identification, model adaptation and model library upgrade, it realizes accurate modeling and parameter identification of loads of different types of power distribution areas, and finally forms a complete implementation closed loop.

[0086] The results show that the parameter identification error and simulation fitting degree of this method meet engineering requirements, and the simulation efficiency is significantly improved compared with traditional methods. It can accurately characterize the dynamic characteristics of distribution transformer areas across the entire time scale, effectively solving the problems of poor adaptability, low accuracy, and low efficiency of traditional modeling methods. This method can be directly applied to various typical distribution transformer areas containing new energy sources, providing a precise and efficient digital modeling tool for transformer area fault analysis, optimized control, and coordinated management of power generation, grid, load, and storage. It supports the safe and stable operation of distribution transformer units in new power distribution systems and the high-proportion consumption of new energy sources.

[0087] Example 2: Based on the same inventive concept, this invention also provides a differentiated typical transformer area integrated load modeling and parameter identification system, such as... Figure 7 As shown, it includes: Load model module, data acquisition module, and parameter identification module; The load model module is used to construct a differentiated load classification model system for transformer substations based on their type. The data acquisition module is used to conduct simulation experiments of various typical working conditions based on the load classification model system, obtain transient data of key nodes in the transformer area, and perform calculations based on the transient data and the detailed component model of the transformer area to obtain corresponding key parameter labels. The transient data and the corresponding key parameter labels constitute a training dataset. The parameter identification module is used to train a hybrid model of convolutional neural network and physical mechanism constraint based on the training dataset, to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy, and to use the hybrid identification model to identify the key parameters of the transformer area to be identified. The improved integrated load aggregation model containing new energy sources is obtained by using a topology model that aggregates the load classification of distribution areas with the equivalent impedance of the upper-level power grid and multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources as a function of static voltage, dynamic voltage, and frequency.

[0088] Preferably, the load model module is specifically used for: For each type of transformer substation, set the basic parameters of the substation based on its type and build a simulation scenario; Based on the basic parameters and simulation scenarios of the transformer substations, refined component models are built according to the differences in load structure of the substations. These refined component models are then integrated according to the load ratio to form a substation-level comprehensive load model, which serves as a differentiated substation load classification model system. The types include: high-proportion photovoltaic areas, charging areas, and traditional areas; the simulation scenarios include: normal operation, voltage sag fault, power backfeed, and charging pile charging / discharging switching. The power backfeed occurs in high-proportion photovoltaic areas, and the charging pile charging / discharging switching occurs in charging areas.

[0089] Preferably, the integrated load model of the high-proportion photovoltaic power station area in the data acquisition module includes a photovoltaic inverter model and a ZIP hybrid load model; The photovoltaic inverter model adopts a three-phase pulse width modulation inverter model with maximum power point tracking control and low voltage ride-through technology; the ZIP hybrid load model adopts a hybrid model of constant impedance, constant current and constant power. The charging area includes a charging pile model and a constant impedance auxiliary load model; The charging pile model used is a conventional DC fast charging pile model.

[0090] Preferably, the data acquisition module performs simulation experiments on various typical operating conditions based on the load classification model system to obtain transient data of key nodes in the transformer area, including: Based on the load classification model system of various types of transformer substations, simulation experiments were carried out under various typical operating conditions. Transient data of key nodes on the low-voltage side of the transformer substation were collected using a high-precision power analyzer during the simulation experiment. The transient data were preprocessed using a combination of outlier removal, improved wavelet denoising, and min-max normalization.

[0091] Preferably, the active power expression of the improved integrated load aggregation model containing new energy sources in the parameter identification module is as follows:

[0092] The reactive power expression for the improved integrated load aggregation model incorporating new energy sources is as follows:

[0093] In the formula, This represents the active power in the improved integrated load aggregation model that includes new energy sources. This represents the reactive power in the improved integrated load aggregation model that includes new energy sources. This indicates the active power under rated operating conditions. This indicates the reactive power under rated operating conditions. Indicates voltage deviation. Represents the static coefficient of active voltage. Represents the static coefficient of reactive voltage. This represents the dynamic time constant of the active voltage. This represents the dynamic time constant of reactive voltage. Represents the static coefficient of active frequency. The static coefficient of reactive frequency is represented; the key parameters include the static coefficient of active voltage, the static coefficient of reactive voltage, the dynamic time constant of active voltage, the dynamic time constant of reactive voltage, the static coefficient of active frequency, and the static coefficient of reactive frequency.

[0094] Preferably, the loss function used by the parameter recognition module when training the hybrid model of convolutional neural network and physical mechanism constraints is calculated as follows:

[0095] In the formula, For loss function, For the number of samples, For the first The measured power value of each sample. For the first The model predicts the power value for each sample. To constrain the penalty coefficient, For the first One key parameter to be identified For the first The maximum constraint value of each key parameter For the first Minimum constraint values ​​for each key parameter; and It is calculated using parameters from the detailed model of the components in the transformer area, representing the first... Physical mechanism constraints on key parameters.

[0096] Preferably, the system also includes a database module, which is used to substitute the identified key parameter combinations into the improved integrated load aggregation model containing new energy sources, combine the load structure and dynamic characteristics of various types of transformer substations, complete the model adaptation, and classify and archive the optimal improved integrated load aggregation model, key parameter configuration and characteristic laws according to transformer substation type, and integrate them into the differentiated typical transformer substation load structured model library.

[0097] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0099] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0100] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0101] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for integrated load modeling and parameter identification of differentiated typical transformer substations, characterized in that, include: Based on the type of transformer area, a differentiated transformer area load classification model system is constructed; Based on the load classification model system, simulation experiments of various typical working conditions are carried out to obtain transient data of key nodes in the transformer area. Based on the transient data and the detailed component model of the transformer area, calculations are performed to obtain the corresponding key parameter labels. The transient data and the corresponding key parameter labels constitute the training dataset. Based on the training dataset, a hybrid model of convolutional neural network and physical mechanism constraint is trained to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy sources. The hybrid identification model is then used to identify the key parameters of the transformer area to be identified. The improved integrated load aggregation model containing new energy sources is obtained by using a topology model that aggregates the load classification of distribution areas with the equivalent impedance of the upper-level power grid and multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources as a function of static voltage, dynamic voltage, and frequency.

2. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 1, characterized in that, The system of differentiated load classification models for transformer substations, based on their type, includes: For each type of transformer substation, set the basic parameters of the substation based on its type and build a simulation scenario; Based on the basic parameters and simulation scenarios of the transformer substations, refined component models are built according to the differences in load structure of the substations. These refined component models are then integrated according to the load ratio to form a substation-level comprehensive load model, which serves as a differentiated substation load classification model system. The types include: high-proportion photovoltaic areas, charging areas, and traditional areas; the simulation scenarios include: normal operation, voltage sag fault, power backfeed, and charging pile charging / discharging switching. The power backfeed occurs in high-proportion photovoltaic areas, and the charging pile charging / discharging switching occurs in charging areas.

3. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 2, characterized in that, The integrated load model for the high-proportion photovoltaic power distribution area includes a photovoltaic inverter model and a ZIP hybrid load model. The photovoltaic inverter model adopts a three-phase pulse width modulation inverter model with maximum power point tracking control and low voltage ride-through technology; the ZIP hybrid load model adopts a hybrid model of constant impedance, constant current and constant power. The charging area includes a charging pile model and a constant impedance auxiliary load model; The charging pile model used is a conventional DC fast charging pile model.

4. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 2, characterized in that, The simulation experiments conducted based on the load classification model system under various typical operating conditions yield transient data of key nodes in the transformer area, including: Based on the load classification model system of various types of transformer substations, simulation experiments were carried out under various typical operating conditions. Transient data of key nodes on the low-voltage side of the transformer substation were collected using a high-precision power analyzer during the simulation experiment. The transient data were preprocessed using a combination of outlier removal, improved wavelet denoising, and min-max normalization.

5. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 1, characterized in that, The active power expression for the improved integrated load aggregation model incorporating new energy sources is as follows: The reactive power expression for the improved integrated load aggregation model incorporating new energy sources is as follows: In the formula, This represents the active power in the improved integrated load aggregation model that includes new energy sources. This represents the reactive power in the improved integrated load aggregation model that includes new energy sources. This indicates the active power under rated operating conditions. This indicates the reactive power under rated operating conditions. Indicates voltage deviation. This represents the static coefficient of active voltage. Represents the static coefficient of reactive voltage. This represents the dynamic time constant of the active voltage. This represents the dynamic time constant of reactive voltage. Represents the static coefficient of active frequency. The static coefficient of reactive frequency is represented; the key parameters include the static coefficient of active voltage, the static coefficient of reactive voltage, the dynamic time constant of active voltage, the dynamic time constant of reactive voltage, the static coefficient of active frequency, and the static coefficient of reactive frequency.

6. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 5, characterized in that, The loss function used when training a hybrid model combining convolutional neural networks and physical constraints is calculated as follows: In the formula, For loss function, For the number of samples, For the first The measured power value of each sample For the first The model predicts the power value for each sample. To constrain the penalty coefficient, For the first One key parameter to be identified For the first The maximum constraint value of each key parameter For the first Minimum constraint values ​​for each key parameter; and It is calculated using parameters from the detailed model of the components in the transformer area, representing the first... Physical mechanism constraints on key parameters.

7. The method for integrated load modeling and parameter identification of differentiated typical transformer substations as described in claim 1, characterized in that, After identifying the key parameters of the station area to be identified using the hybrid identification model, the process further includes: The identified key parameter combinations are substituted into the improved integrated load aggregation model containing new energy sources. The model is adapted by combining the load structure and dynamic characteristics of various types of transformer substations. The optimal improved integrated load aggregation model, key parameter configurations and characteristic laws are classified and archived according to transformer substation type and integrated into the differentiated typical transformer substation load structured model library.

8. A system for integrated load modeling and parameter identification of differentiated typical transformer substations, characterized in that, include: Load model module, data acquisition module, and parameter identification module; The load model module is used to construct a differentiated load classification model system for transformer substations based on their type. The data acquisition module is used to conduct simulation experiments of various typical working conditions based on the load classification model system, obtain transient data of key nodes in the transformer area, and perform calculations based on the transient data and the detailed component model of the transformer area to obtain corresponding key parameter labels. The transient data and the corresponding key parameter labels constitute a training dataset. The parameter identification module is used to train a hybrid model of convolutional neural network and physical mechanism constraint based on the training dataset, to obtain a hybrid identification model that can identify key parameters of the improved integrated load aggregation model containing new energy, and to use the hybrid identification model to identify the key parameters of the transformer area to be identified. The improved integrated load aggregation model containing new energy sources is obtained by using a topology model that aggregates the load classification of distribution areas with the equivalent impedance of the upper-level power grid and multiple loads in parallel. The key parameters represent the response characteristics of the active and reactive power of the load in the improved integrated load aggregation model containing new energy sources as a function of static voltage, dynamic voltage, and frequency.

9. The differentiated typical transformer area integrated load modeling and parameter identification system as described in claim 8, characterized in that, The load model module is specifically used for: For each type of transformer substation, set the basic parameters of the substation based on its type and build a simulation scenario; Based on the basic parameters and simulation scenarios of the transformer substations, refined component models are built according to the differences in load structure of the substations. These refined component models are then integrated according to the load ratio to form a substation-level comprehensive load model, which serves as a differentiated substation load classification model system. The types include: high-proportion photovoltaic areas, charging areas, and traditional areas; the simulation scenarios include: normal operation, voltage sag fault, power backfeed, and charging pile charging / discharging switching. The power backfeed occurs in high-proportion photovoltaic areas, and the charging pile charging / discharging switching occurs in charging areas.

10. The differentiated typical transformer area integrated load modeling and parameter identification system as described in claim 9, characterized in that, The data acquisition module performs simulation experiments on various typical operating conditions based on the load classification model system to obtain transient data of key nodes in the transformer area, including: Based on the load classification model system of various types of transformer substations, simulation experiments were carried out under various typical operating conditions. Transient data of key nodes on the low-voltage side of the transformer substation were collected using a high-precision power analyzer during the simulation experiment. The transient data were preprocessed using a combination of outlier removal, improved wavelet denoising, and min-max normalization.

Images