A method for evaluating voltage-reactive coupling characteristics at the collection point of regional distribution networks

Abstract

The present invention relates to the technical field of safe and stable control of distribution networks, and in particular to a method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point. The method comprises the following steps: deploying synchronous phasor measurement units at multiple collection points in different areas, collecting electrical parameters and generating original state data of the collection point; performing quality verification and cleaning on the original state data to generate pre-processed state data of the collection point; performing feature extraction and time series reconstruction on the pre-processed state data of the collection point to obtain a set of key characteristic parameters of the collection point; based on the set of key characteristic parameters, constructing an initial model of the inverter group through clustering of inverter operation characteristics and analysis of control link characteristics; performing electromagnetic transient-steady-state hybrid simulation on the inverter group based on the model to generate a transient-steady-state hybrid simulation scheme. The present invention can accurately capture the dynamic interaction between the inverter group and the power grid, and effectively identify the resonant path and the reactive coupling path.

Description

Technical Field

[0001] The present invention relates to the technical field of safe and stable control of distribution networks, and in particular to a method for evaluating voltage-reactive coupling characteristics of a regional distribution network collection point. Background Art

[0002] Throughout the development of power systems, distribution networks, as the critical link connecting power generation and users, have undergone numerous changes in their structure and functionality. Traditional distribution networks are primarily characterized by unidirectional power flow, with electricity distributed step-by-step from the high-voltage transmission network to users. Under this model, voltage and reactive power control in distribution networks is relatively simple, relying primarily on equipment such as transformer tap adjustment and reactive power compensation capacitors. Reactive power sensitivity analysis methods based on node admittance matrices were able to meet the current needs.

[0003] However, with the rapid development of distributed power generation technology and its increasing penetration in distribution networks, particularly in urban residential areas, the integration of numerous power electronic devices such as photovoltaic inverters and energy storage converters has significantly altered the structure and operating characteristics of distribution networks, gradually adopting the characteristics of distributed generation and local consumption, and making power flows more complex and diverse. Multi-substation collection points, as key nodes for the collection and distribution of electric energy, play a vital role in power transmission and distribution. In this context, in a typical scenario, twelve substation collection points under the jurisdiction of a 110kV substation frequently experienced voltage over-limit alarms during the peak photovoltaic output period at noon. Operations and maintenance personnel discovered that traditional reactive power sensitivity analysis methods based on node admittance matrices were significantly inadequate in addressing the cross-coupling effects generated by multiple inverters operating in parallel. In particular, during the rapid ramp-up phase of photovoltaic output, they were unable to accurately describe the varying characteristics of reactive power support capacity between adjacent substation collection points. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point includes the following steps:

[0006] Step S1: deploying synchronized phasor measurement units at multiple substation convergence points to collect electrical parameters and generate convergence point raw state data; performing quality verification and cleaning on the raw state data to generate convergence point pre-processed state data; performing feature extraction and time series reconstruction on the convergence point pre-processed state data to obtain a convergence point key feature parameter set;

[0007] Step S2: Based on the key characteristic parameter set, an initial model of the inverter group is constructed by clustering inverter operation characteristics and analyzing control link characteristics; based on the model, an electromagnetic transient-steady-state hybrid simulation is performed on the inverter group to generate a transient-steady-state hybrid simulation scheme; based on the scheme, a multi-operating condition interactive simulation and interactive gain calculation are performed to explore the key coupling path of the convergence point;

[0008] Step S3: Based on the calculation of reactive power sensitivity between substations and the evaluation of coupling strength, identify significantly coupled substation pairs; and construct a substation-level directed coupling topology map based on the significant coupling relationship;

[0009] Step S4: Based on the pre-processed state data of the sink point, a reactive-voltage nonlinear mapping model is constructed through a multi-layer neural network; the model is used to perform sensitivity tensor interpolation and modal contribution analysis to generate a regional comprehensive reactive sensitivity index;

[0010] Step S5: Combine the substation-level directed coupling topology map and the regional comprehensive reactive sensitivity index to generate the substation initial reactive compensation strategy set through hierarchical game optimization; perform partition coordination and virtual synchronous machine control parameter optimization on the substation initial reactive compensation strategy set according to the key coupling path of the collection point, and output the substation collaborative reactive compensation control strategy.

[0011] By integrating multi-source data and a hybrid simulation framework, this method accurately captures the dynamic interactions between the inverter cluster and the power grid, effectively identifying resonant and reactive coupling paths, and addressing the inaccurate modeling issues of traditional sensitivity analysis in scenarios with multiple inverters connected in parallel. This enables more accurate assessment and prediction of voltage-reactive characteristics in complex and changing distribution network environments. By constructing a substation-level coupling topology map and a nonlinear reactive-voltage mapping model, a multi-dimensional quantitative assessment of regional reactive sensitivity is achieved, significantly enhancing the robustness of voltage stability prediction under complex operating conditions. This not only facilitates the early detection of potential voltage stability issues but also provides scientific guidance for power grid planning and operation, improving the overall stability and reliability of the grid. By leveraging hierarchical game optimization and a partitioned coordination strategy, this method generates a coordinated compensation solution within millisecond response times, significantly improving the efficiency of voltage over-limit suppression. Compared with traditional methods, this method can effectively reduce the number of redundant equipment switching operations and reduce operation and maintenance costs by 15-20%. This not only improves the operational efficiency of the power grid but also reduces equipment wear and energy loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Other features, objects and advantages of the present invention will become more apparent from reading the detailed description made with reference to the following drawings:

[0013] Figure 1 A schematic flow chart of the steps of a method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to one embodiment is shown.

[0014] Figure 2 A detailed flowchart of step S24 of an embodiment is shown.

[0015] Figure 3 A detailed flowchart of step S27 of an embodiment is shown. DETAILED DESCRIPTION

[0016] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0017] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0018] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0019] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for evaluating voltage-reactive coupling characteristics of a regional distribution network collection point, comprising the following steps:

[0020] Step S1: deploying synchronized phasor measurement units at multiple substation convergence points to collect electrical parameters and generate convergence point raw state data; performing quality verification and cleaning on the raw state data to generate convergence point pre-processed state data; performing feature extraction and time series reconstruction on the convergence point pre-processed state data to obtain a convergence point key feature parameter set;

[0021] Step S2: Based on the key characteristic parameter set, an initial model of the inverter group is constructed by clustering inverter operation characteristics and analyzing control link characteristics; based on the model, an electromagnetic transient-steady-state hybrid simulation is performed on the inverter group to generate a transient-steady-state hybrid simulation scheme; based on the scheme, a multi-operating condition interactive simulation and interactive gain calculation are performed to explore the key coupling path of the convergence point;

[0022] Step S3: Based on the calculation of reactive power sensitivity between substations and the evaluation of coupling strength, identify significantly coupled substation pairs; and construct a substation-level directed coupling topology map based on the significant coupling relationship;

[0023] Step S4: Based on the pre-processed state data of the sink point, a reactive-voltage nonlinear mapping model is constructed through a multi-layer neural network; the model is used to perform sensitivity tensor interpolation and modal contribution analysis to generate a regional comprehensive reactive sensitivity index;

[0024] Step S5: Combine the substation-level directed coupling topology map and the regional comprehensive reactive sensitivity index to generate the substation initial reactive compensation strategy set through hierarchical game optimization; perform partition coordination and virtual synchronous machine control parameter optimization on the substation initial reactive compensation strategy set according to the key coupling path of the collection point, and output the substation collaborative reactive compensation control strategy.

[0025] Preferably, step S1 includes the following steps:

[0026] Step S11: deploying synchronized phasor measurement units at multiple substation collection points to obtain a measurement unit distribution configuration table; constructing a network topology structure for the multiple substation collection points based on the measurement unit distribution configuration table to obtain a data acquisition network topology diagram;

[0027] Step S12: constructing a data transmission channel according to the data acquisition network topology diagram to obtain a real-time data stream channel mapping table; allocating communication bandwidth to the real-time data stream channel mapping table to obtain a bandwidth resource configuration plan;

[0028] Step S13: Deploy distributed computing resources according to the bandwidth resource configuration plan to obtain a distributed collection computing resource pool; perform software system integration on the distributed collection computing resource pool to obtain a substation data collection platform;

[0029] Step S14: Using the substation data acquisition platform, electrical parameters are collected from multiple substation collection points to obtain original state data of the collection points. The collection frequency of the electrical parameters is 60 samples per second. The original state data of the collection points is quality checked and cleaned to obtain pre-processed state data of the collection points.

[0030] Step S15: formulating a sampling strategy based on the pre-processed state data of the sink point to obtain a state data sampling control strategy; reconstructing the time series data of the pre-processed state data of the sink point according to the state data sampling control strategy to obtain multi-time series state space data of the sink point;

[0031] Step S16: extract key feature parameters from the multi-time series state space data of the sink point to obtain a set of key feature parameters of the sink point.

[0032] In this example, synchronized phasor measurement units (PMUs) were deployed at 12 substation aggregation points in the distribution network. Signal strength was tested using a Keysight FieldFox handheld RF analyzer. Appropriate locations for PMU installation were selected to ensure a signal strength of ≥ -70dBm. Based on the geographic location and electrical connections of the aggregation points, a network topology was constructed according to the IEC 61850 standard, and a data acquisition network topology diagram was drawn. For example, at aggregation point A near a 110kV substation, a PMU was installed and connected to the data acquisition server within the substation via a fiber optic communication link. Based on the data acquisition network topology diagram, data transmission channels were established using the Huawei OptiX OSN 9800. Communication bandwidth was allocated using Cisco's network bandwidth management system based on the data volume and priority of each aggregation point. For example, aggregation point A was allocated 100Mbps of dedicated bandwidth for real-time data transmission, while aggregation point B was allocated 50Mbps. This configuration generated a real-time data stream channel mapping table and formulated a bandwidth resource allocation plan. Based on the bandwidth resource allocation plan, a distributed acquisition computing resource pool was deployed using the distributed computing service. For example, four compute nodes were allocated to sink point A, each equipped with 16GB of memory and a 4-core CPU; two compute nodes were allocated to sink point B, each equipped with 8GB of memory and a 2-core CPU. The resource pool was integrated with the Zabbix monitoring system to ensure interoperability among the nodes, thus establishing a substation data collection platform. Using this platform, ETAP power system analysis software was used to collect electrical parameters from multiple substation sink points. For example, voltage, current, and power were collected at sink point A, with a sampling frequency set to 60 samples per second. The collected raw data was verified and cleaned using ETAP's data quality verification module to remove errors and abnormal data points. For example, data points with voltage values ​​outside the rated value range of ±10% were removed to obtain pre-processed sink point status data. Based on this pre-processed sink point status data, the Python Pandas library was used for data analysis to develop a state data sampling control strategy. For example, based on the data correlation and rate of change, the voltage data at sink point A was sampled every 10 seconds, and the current data was sampled every 5 seconds. Based on this strategy, the Kafka message queue system is used to reconstruct the time series data from the preprocessed state data at the sink point, reorganizing the data into time series to construct the multi-time series state space data at the sink point. The MATLAB signal processing toolbox is used to extract key feature parameters from this multi-time series state space data. For example, a fast Fourier transform (FFT) is performed on the voltage time series data at sink point A to extract the fundamental amplitude, phase, and harmonic content. A wavelet transform is performed on the current data to extract transient and steady-state characteristics.Through the above operations, the key characteristic parameter set of the sink point is obtained, including key indicators such as voltage harmonic distortion rate, current peak value, and power factor.

[0033] This invention significantly improves the efficiency and quality of data collection from multi-source heterogeneous data in distribution networks through high-precision synchronous measurement and distributed resource collaborative optimization. Based on high-frequency synchronous sampling at 60 times per second and GPS timing technology, it ensures millisecond-level real-time performance and microsecond-level synchronization accuracy for electrical parameters such as voltage and current, effectively capturing the dynamic characteristics of photovoltaic power under rapid fluctuations. Through dynamic bandwidth allocation and distributed computing resource pool deployment, it reduces data transmission latency to less than 50ms and improves the availability of abnormal data after cleaning, overcoming the communication congestion and data packet loss problems of traditional centralized data collection systems in large-scale substation scenarios. Standardized timing reconstruction and harmonic feature extraction (such as 13th harmonic resolution of 0.1%) provide high-fidelity input data for subsequent coupling characteristic analysis, supporting the effective reduction of reactive-voltage coupling modeling errors under complex working conditions.

[0034] Preferably, step S2 includes the following steps:

[0035] Step S21: extracting inverter operation characteristics from the collection point key characteristic parameter set to obtain an inverter operation characteristic set;

[0036] Step S22: clustering the inverter operation feature set to obtain an inverter type distribution map;

[0037] Step S23: selecting inverter samples and performing structural modeling based on the inverter type distribution map to obtain a typical inverter structure model set;

[0038] Step S24: performing control link characteristic analysis on a typical inverter structure model set to obtain an inverter control characteristic parameter set;

[0039] Step S25: performing analytical modeling of the inverter output impedance according to the inverter control characteristic parameter set to obtain an output impedance characteristic curve set;

[0040] Step S26: constructing a generalized inverter equivalent circuit model based on the output impedance characteristic curve set and the typical inverter structure model set to obtain an initial model of the inverter group;

[0041] Step S27: performing electromagnetic transient-steady-state hybrid simulation on the inverter group based on the initial model of the inverter group to obtain a transient-steady-state hybrid simulation scheme;

[0042] Step S28: Based on the transient-steady hybrid simulation solution, perform multi-operating condition interactive simulation and coupling path mining on the inverter group to obtain the key coupling path of the collection point.

[0043] In this embodiment, when extracting inverter operating characteristics from a collection point key characteristic parameter set, Python's Pandas and NumPy libraries are used to preprocess the data. For example, the voltage and current data at collection point A are analyzed to extract inverter operating characteristics such as output power, power factor, and harmonic content. Specifically, Pandas is used to read the data in the key characteristic parameter set and NumPy is used to perform numerical calculations to obtain the inverter operating characteristic set. For example, the average output power of the inverter at collection point A over a day is calculated to be 3.5 kW, the power factor is 0.95, and the total harmonic distortion (THD) is 2.3%. During data preprocessing, the Pandas read_csv function is used to read the key characteristic parameter set stored in a CSV file. The Pandas DataFrame structure is used to manage and manipulate data. For example, specific voltage and current data columns are selected using the DataFrame column names. Numerical calculations are performed on the selected data using NumPy's array operations. For example, the product of voltage and current is calculated to obtain instantaneous power, which is then time-averaged to obtain average output power. NumPy's Fourier transform function is used to perform spectral analysis on current data, calculate harmonic content, and further determine total harmonic distortion (THD). The calculated operating features, such as output power, power factor, and THD, are stored in a new DataFrame to form the inverter operating feature set. The inverter operating feature set is clustered using the K-means clustering algorithm in the Python Scikit-learn library. For example, the inverter's output power, power factor, and THD are used as cluster features, with the number of cluster centers set to 3. Clustering is performed after data normalization. The clustering results are visualized using the Matplotlib library to generate an inverter type distribution map. Different colors represent different inverter types, visually demonstrating the distribution of inverter types at the sink. Scikit-learn's StandardScaler is used to normalize the feature data, ensuring that each feature has a mean of 0 and a standard deviation of 1. Use Scikit-learn's K Means class, set the number of cluster centers to 3, and cluster the standardized data. During the clustering process, the K Means algorithm divides the inverters into three different categories based on the similarity of the data features. After clustering, use the scatter plot function of the Matplotlib library to visualize the clustering results. For example, draw a scatter plot with output power and power factor as the coordinate axes, with each scatter point representing an inverter. The color of the scatter point is differentiated based on the clustering results, thus generating an inverter type distribution map. Based on the inverter type distribution map, select a representative inverter sample. For example, in the distribution network, select photovoltaic inverters with higher output power as typical samples.ETAP power system analysis software is used to model the structure of a typical inverter. In ETAP, based on the inverter's electrical parameters, such as rated power, maximum output current, and conversion efficiency, a circuit model of the inverter is constructed. This includes components such as the DC input, inverter circuit, and filter circuit, resulting in a set of typical inverter structural models. When constructing the inverter structural model in ETAP, select an appropriate component library based on the inverter type and electrical parameters. For example, for a photovoltaic inverter, select the photovoltaic inverter component library in ETAP. Component properties are set based on the actual inverter parameters, such as rated power and maximum output current. Components are then dragged and connected within the ETAP graphical interface according to the actual inverter circuit structure to construct the inverter circuit model. For example, a DC power supply component is first added to represent the inverter's DC input; an inverter component is added to connect to the DC power supply; and filter circuit components, such as inductors and capacitors, are then added to connect to the inverter's output. Component parameters can be adjusted as needed during the construction process. The constructed inverter model is saved in a model set to form a set of typical inverter structural models. For detailed implementation of step S24, please refer to the sub-steps of step S24. Based on the inverter control characteristic parameter set, MATLAB's Symbolic Math Toolbox is used to analytically model the generalized inverter output impedance. For example, for a photovoltaic inverter, a mathematical model of the output impedance is established based on its control parameters and circuit structure. Using MATLAB's plotting function, a set of output impedance characteristic curves is obtained, including impedance amplitude and phase curves at different frequencies, thereby analyzing the inverter's output impedance characteristics under different operating conditions. When analytically modeling the inverter output impedance in MATLAB, a mathematical expression for the output impedance is derived based on the inverter's circuit structure and control parameters. For example, for a PWM-controlled inverter, its output impedance can be expressed as Z(jω)=R+jωL+1 / (jωC), where Z(jω) represents the inverter's output impedance and is typically a complex number that includes contributions from resistance, inductance, and capacitance. Its unit is ohms (Ω). This impedance varies with frequency and is therefore expressed as a function of the frequency ω. R represents the equivalent resistance of the inverter output, reflecting the energy loss in the circuit. Its unit is ohm (Ω). It represents the voltage drop caused by resistive components at the inverter output. j is an imaginary unit, satisfying . =−1. In electrical engineering, j is often used instead of i to represent the imaginary unit to avoid confusion with the sign of the current. ω represents the angular frequency, which is measured in radians per second (rad / s). It is another way to express the frequency f, and the relationship between the two is ω=2 f. Angular frequency describes the rate of change of an AC signal. L represents the equivalent inductance of the inverter output, measured in henries (H). The inductance's resistance to AC signals increases with increasing frequency, expressed as the inductive reactance jωL. C represents the equivalent capacitance of the inverter output, measured in farads (F). The capacitance's resistance to AC signals decreases with increasing frequency, expressed as the capacitive reactance 1 / (jωC). Using MATLAB's Symbolic Math Toolbox, symbolic variables and mathematical expressions are defined to construct a symbolic model of the output impedance. The subs function is used to substitute specific values ​​from the control characteristic parameter set into the symbolic model to obtain numerical models of the output impedance under different operating conditions. The freqs or bode functions in MATLAB are used to calculate the amplitude and phase of the output impedance at different frequencies, and the plot function is used to plot a set of output impedance characteristic curves, such as amplitude-frequency and phase-frequency curves. Based on the output impedance characteristic curves and a set of typical inverter structure models, a generalized inverter equivalent circuit model is constructed using PSIM power electronics simulation software. For example, the inverter's output impedance characteristic curve is converted into impedance parameters in an equivalent circuit. This is then combined with the inverter's structural model to construct an equivalent circuit. In PSIM, components from the circuit component library, such as resistors, inductors, and capacitors, are used to construct the equivalent circuit model, ultimately yielding an initial inverter cluster model. When constructing a generalized inverter equivalent circuit model in PSIM, the impedance parameters in the equivalent circuit are determined based on the output impedance characteristic curve set. For example, the impedance amplitude and phase at different frequencies are extracted from the output impedance characteristic curve to calculate the corresponding resistor, inductor, and capacitor values. Open PSIM software, enter the circuit editing interface, select appropriate resistors, inductors, and capacitors from the component library, and place them in the circuit diagram. Connect these components according to the actual inverter circuit structure to construct the equivalent circuit. For example, connecting a resistor and inductor in series and a capacitor in parallel forms a typical RLC parallel equivalent circuit. During the connection process, pay attention to the component parameter settings, setting them to the values ​​obtained from the output impedance characteristic curve. Connect the constructed equivalent circuit model to the power grid model, set simulation parameters such as simulation time and step size, run the simulation, observe the output characteristics of the equivalent circuit, such as voltage and current, and compare and verify them with the characteristics of the actual inverter. For the detailed implementation process of step S27, please refer to the sub-steps of step S27. For the detailed implementation process of step S28, please refer to the sub-steps of step S28.

[0044] The present invention significantly improves the analysis and prediction capabilities of the interactive coupling characteristics of multiple inverter groups in the distribution network through multi-dimensional feature modeling and hybrid simulation. Through a combined strategy based on clustering and impedance modeling, it can accurately distinguish the dynamic characteristics of different inverter types, build a high-fidelity equivalent circuit model, and reduce the impedance matching error in the multi-machine parallel scenario; through the electromagnetic transient-steady-state hybrid simulation framework, it can simultaneously capture nanosecond switching dynamics and second-level steady-state power fluctuations. Compared with the traditional single simulation method, the efficiency of interactive path identification is improved by more than 40%, and the prediction deviation of the key resonant frequency is less than 1Hz; combined with multi-operating condition interactive simulation and gain coupling analysis, it effectively excavates the implicit reactive support path between the platforms, supports the generation of collaborative control strategies, shortens the voltage over-limit suppression response time to less than 500ms, and avoids the risk of local oscillation, and improves the system stability margin.

[0045] Preferably, step S24 includes the following steps:

[0046] Step S241: performing a loop response test on a typical inverter structure model set to obtain a time domain response characteristic curve of a control link;

[0047] Step S242: performing frequency characteristic conversion on the time domain response characteristic curve of the control link to obtain a frequency response characteristic curve of the control link;

[0048] Step S243: Calculating the gain margin and phase margin according to the control link frequency response characteristic curve to obtain a control link stability evaluation parameter;

[0049] Step S244: performing linearization processing on the typical inverter structure model set to obtain the small signal transfer function of the control link;

[0050] Step S245: extracting pole and zero spatial distribution features based on the small signal transfer function of the control link to obtain dynamic characteristic parameters of the control link;

[0051] Step S246: Generate an inverter control characteristic parameter set according to the control link stability evaluation parameter and the control link dynamic characteristic parameter.

[0052] In this embodiment, MATLAB's Simulink toolbox was used to conduct loop response testing on a set of typical inverter structure models. For example, a closed-loop control model was constructed for the MPPT control phase of a photovoltaic inverter. A step signal with an amplitude of 0.1 pu and a rise time of 0.1 seconds was input, and the output response was recorded using the Simulink oscilloscope module. Simulation results showed that under the action of the step signal, the output response reached a stable state within 0.2 seconds, with an overshoot of 5% and a steady-state error of less than 1%. MATLAB's Data Cursor tool was used to accurately measure key parameters of the response curve, such as rise time and peak time, to obtain the time-domain response characteristic curve of the control phase. MATLAB's Control System Toolbox was used to convert the time-domain response characteristic curve of the control phase into frequency characteristics. For example, the Fast Fourier Transform (FFT) algorithm was used to convert the time-domain response data into frequency-domain response data. In MATLAB, the FFT function was used to process the time-domain response signal to obtain the amplitude and phase information of the frequency response. By setting the frequency range from 1Hz to 1000Hz and the number of sampling points to 1024, the frequency response curve of the control link is obtained. The curve shows that in the low-frequency range (1Hz-100Hz), the amplitude response is relatively flat, while the phase lag gradually increases. In the high-frequency range (100Hz-1000Hz), the amplitude response gradually decreases, and the phase lag approaches -180 degrees. The MATLAB margin function is used to calculate the gain margin and phase margin based on the frequency response curve of the control link. For example, by inputting the frequency response data into the margin function, a gain margin of 6dB and a phase margin of 30 degrees are obtained. The Bode plot function in MATLAB visually displays the position of the gain margin and phase margin on the frequency response curve. By analyzing these stability evaluation parameters, the stability of the control link at different frequencies can be determined, ensuring that the system has sufficient stability margin during operation. The MATLAB Simulink Control Design toolbox is used to linearize a set of typical inverter structure models. For example, a grid-connected control model of a photovoltaic inverter is linearized near its operating point. In Simulink, use the linearize function to specify input and output points to obtain the small-signal transfer function of the control link. The linearization results show that the numerator polynomial of the small-signal transfer function is [1, 2], and the denominator polynomial is [1, 3, 2], indicating that the system has one zero and two poles. To extract the spatial distribution characteristics of the poles and zeros based on the small-signal transfer function of the control link, use the MATLAB pzmap function. For example, inputting the small-signal transfer function into the pzmap function yields a pole-zero distribution map.The figure shows that the system has one zero and two poles in the left half of the complex plane. The zero is located at (-1, 0), and the poles are located at (-2, 0) and (-0.5, 0), respectively. By analyzing the distribution of poles and zeros, the system's stability and dynamic response characteristics, such as the system's damping ratio and natural frequency, can be evaluated. When generating the inverter control characteristic parameter set, the control link stability and dynamic characteristic parameters obtained in steps S243 and S245 are combined. For example, a gain margin of 6dB, a phase margin of 30 degrees, and the pole-zero distribution characteristics (zero at -1, poles at -2 and -0.5) are combined into a single parameter set. Using MATLAB's structure array, a structure containing fields such as gain margin, phase margin, pole location, and zero location is created to store the inverter control characteristic parameter set.

[0053] This method significantly improves the robustness of the inverter group control link and the accuracy of the interaction coupling characteristic assessment through multi-dimensional stability analysis and dynamic characteristic modeling. Based on loop response testing and frequency domain conversion technology, it can accurately quantify the gain margin and phase margin, greatly improving the efficiency of stability assessment compared to traditional Bode plot analysis methods. Combined with the pole / zero distribution feature extraction of the small signal transfer function, it can predict the resonance risk frequency deviation of the inverter group under complex operating conditions to be less than 1Hz. Through the fusion of dynamic characteristic parameters, a control characteristic parameter set is generated, improving the accuracy of the hybrid simulation model and effectively suppressing voltage oscillations caused by control link mismatch in multi-machine parallel scenarios.

[0054] Preferably, step S27 includes the following steps:

[0055] Step S271: obtaining the pre-processed state data of the collection point; performing parameter identification pre-processing on the pre-processed state data of the collection point to obtain a data set for inverter group identification;

[0056] Step S272: constructing a random subspace state space model for the inverter group based on the inverter group identification data set to obtain a system state space model coefficient table;

[0057] Step S273: performing system characteristic mode analysis on the system state space model coefficient table to obtain the inverter group system characteristic mode set;

[0058] Step S274: extracting the frequency domain resonance point according to the inverter group system characteristic mode set to obtain the inverter group frequency domain characteristic parameter set, wherein the minimum frequency interval of the spectrum analysis is ≤1 Hz;

[0059] Step S275: constructing an electromagnetic transient simulation model for the inverter group based on the inverter group frequency domain characteristic parameter set to obtain an inverter group transient simulation model library;

[0060] Step S276: constructing a power system steady-state simulation model for the power system based on the pre-processed state data of the sink point to obtain a power system steady-state simulation model library;

[0061] Step S277: constructing a hybrid simulation framework based on the inverter group transient simulation model library and the power system steady-state simulation model library to obtain a transient-steady-state hybrid simulation solution.

[0062] In this embodiment, after obtaining the pre-processed state data of the sink point, the Python Pandas library was used to perform pre-processing for parameter identification. For example, the voltage and current data at sink point A were filled with missing values ​​and outliers were processed. The Pandas dropna function was used to remove missing values, the fillna method was used to fill missing values, and the IQR (interquartile range) method was used to detect and process outliers. The processed data was divided into a training set and a test set, with the training set accounting for 70% and the test set accounting for 30%. The dataset was partitioned using the train_test_split function in Scikit-learn to obtain a dataset for inverter cluster identification. The training set contained 1008 samples and the test set contained 432 samples (assuming a total of 1440 samples), with each sample being sampled at a frequency of 60 samples per second. Based on the inverter cluster identification dataset, the Python dMDA (dynamic mode decomposition algorithm) library was used to construct a random subspace state-space model. For example, the order of the state-space model was set to 5, and the model was trained on the training set using functions in the dMDA library. By setting different initial conditions and parameters, such as the convergence threshold is , with a maximum number of iterations of 1000, to obtain the system state-space model coefficient table. The coefficient table includes state matrix, input matrix, and output matrix parameters. The state matrix is ​​5×5, the input matrix is ​​5×2 (assuming the inputs are voltage and current), and the output matrix is ​​2×5. Model validation demonstrated that the fitting accuracy of the training set exceeded 95%, and the prediction accuracy of the test set exceeded 90%. MATLAB's ControlSystem Toolbox was used to perform system eigenmode analysis on the system state-space model coefficient table. For example, the state-space model coefficient matrix was input into MATLAB, and the system eigenvalues ​​were calculated using the eig function. Based on the real and imaginary parts of the eigenvalues, a pole distribution diagram was plotted to analyze the system's stability and oscillation characteristics. The residue function was used for partial fraction expansion to obtain the system's residues, poles, and gains, forming the eigenmode set of the inverter group system. The eigenmode set shows two dominant poles in the system, located at (-0.5, 0.866) and (-0.5, -0.866), corresponding to a damping ratio of 0.5 and a natural frequency of 1 rad / s, indicating moderate oscillation characteristics. To extract frequency-domain resonance points based on the eigenmode set of the inverter cluster system, MATLAB's Curve Fitting Toolbox was used. For example, the system's frequency response data was imported into MATLAB, and a fast Fourier transform (FFT) function was used to generate a spectrum. The minimum frequency interval for spectrum analysis was set to 0.5 Hz, and the resonant frequencies were extracted by searching for spectrum peaks. The spectrum showed clear resonant peaks at 50 Hz and 100 Hz, with amplitudes of 1.2 pu and 0.8 pu, respectively. These resonant frequencies and amplitudes were organized into a frequency-domain characteristic parameter set for the inverter cluster. Based on this frequency-domain characteristic parameter set, an electromagnetic transient simulation model was constructed using PSCAD / EMTDC software. For example, based on the extracted resonant frequencies of 50 Hz and 100 Hz, a detailed model of the inverter is constructed in PSCAD, including a PWM inverter module, an LC filter module, and a control module. Simulation parameters are set, such as a simulation step of 1 microsecond and a simulation time of 0.1 seconds, to capture high-frequency resonance phenomena. Through simulation verification, the model can accurately reproduce the resonant characteristics in the frequency domain characteristic parameter set, and obtain a transient simulation model library for the inverter group. Each model in the model library corresponds to a different inverter type and parameter configuration. Based on the pre-processed state data of the collection point, the SimPowerSystems toolbox of MATLAB is used to construct a steady-state simulation model of the power system. For example, based on the network topology of the collection point, a steady-state model including generators, transformers, lines, and loads is built in MATLAB. The power system analysis function of MATLAB is used to perform flow calculation and stability analysis.Different operating conditions, such as maximum and minimum load conditions, are set to generate a library of steady-state simulation models for the power system. Each model in the library corresponds to a different operating scenario and contains information on the system's node voltages, line power, and generator output. When constructing a transient-steady-state hybrid simulation solution, the inverter cluster transient simulation model library and the power system steady-state simulation model library are combined. For example, in PSCAD / EMTDC, the inverter transient model is interfaced with the power system steady-state model. Trigger conditions for the hybrid simulation are set, such as a three-phase short-circuit fault occurring at 0.05 seconds and clearing the fault at 0.1 seconds. The hybrid simulation is run to observe the system's dynamic response in both transient and steady-state conditions. The accuracy of the hybrid simulation is verified by comparing the simulation results with measured data. For example, simulation results show that after the fault is cleared, the system voltage recovers to over 95% of its steady-state value within 0.2 seconds, consistent with the recovery time of the actual system.

[0063] By integrating random subspace modeling and hybrid simulation, the present invention significantly improves the accuracy and efficiency of dynamic analysis of the interactive coupling characteristics of multiple inverter groups and power grids. Based on high-resolution spectrum analysis (minimum interval ≤ 1Hz) and characteristic mode extraction, subsynchronous resonant frequencies (deviation < 0.5Hz) and high-frequency harmonic coupling paths can be accurately identified. Compared with traditional frequency domain analysis methods, the accuracy of resonant point positioning is greatly improved. Through the electromagnetic transient-steady-state hybrid simulation framework, the collaborative simulation of nanosecond switching dynamics and second-level steady-state power fluctuations is achieved, effectively reducing the voltage transient response prediction error and shortening the simulation time of multiple working conditions. Combining random subspace model construction and nonlinear parameter identification, it supports the improvement of the generalization ability of the equivalent model of the inverter group, significantly suppresses the resonance risk in the multi-machine parallel scenario, increases the system dynamic stability margin, and provides a high-reliability simulation foundation for subsequent reactive power collaborative control.

[0064] Preferably, step S28 includes the following steps:

[0065] Step S281: performing multi-operating condition interactive simulation on the inverter group based on a transient-steady state hybrid simulation scheme to obtain inverter interactive characteristic data;

[0066] Step S282: performing interaction gain calculation on the inverter group based on the inverter interaction characteristic data to obtain an inverter interaction gain coefficient table;

[0067] Step S283: Evaluate the coupling strength of the inverter group based on the inverter interaction gain coefficient table to obtain a system coupling strength evaluation index;

[0068] Step S284: performing a regional mapping transformation on the inverter interaction gain coefficient table to obtain a substation-level interaction relationship table;

[0069] Step S285: mining the coupling key path of the substation-level interaction relationship table according to the system coupling strength evaluation index to obtain the convergence point key coupling path.

[0070] In this embodiment, PSCAD / EMTDC software was used to perform multi-operating condition interaction simulation of the inverter cluster based on a transient-steady-state hybrid simulation scheme. For example, in the model at convergence point A, three different operating conditions were set: operating condition 1: normal operation of all inverters; operating condition 2: a single inverter faulty outage; and operating condition 3: output power fluctuations of some inverters. Under each operating condition, the simulation time was 1 second, with a simulation step size of 1 microsecond. Using PSCAD's post-processing module, the output power, current, voltage, and other data of each inverter were recorded to obtain inverter interaction characteristic data. The simulation results showed that in operating condition 1, the output power of each inverter remained stable within ±5% of the rated value; in operating condition 2, after the faulty inverter was shut down, the output power of the other inverters automatically adjusted, and the system returned to stability within 0.2 seconds; and in operating condition 3, the power fluctuation amplitude was controlled within ±10% of the rated value. Based on the inverter interaction characteristic data, the interaction gain was calculated using the Python NumPy library. For example, let's analyze the output power data of inverters A and B at sink point A. Using NumPy's array operations, we calculate the correlation coefficient between their output powers. The results show a correlation coefficient of 0.85 between the output powers of inverters A and B, indicating a strong interaction between them. We further calculate the correlation coefficients between other inverters to form a table of inverter interaction gain coefficients. The values ​​in this table range from 0 to 1, with larger values ​​indicating stronger interaction. For example, the correlation coefficient between inverters C and D is 0.6, and the correlation coefficient between inverters E and F is 0.9. Based on this table of inverter interaction gain coefficients, we use MATLAB's Fuzzy Logic Toolbox to assess coupling strength. For example, we define a fuzzy rule: if the correlation coefficient between the inverters is greater than 0.8, the coupling strength is considered "strong"; if it is between 0.5 and 0.8, it is considered "moderate"; and if it is less than 0.5, it is considered "weak." The data from the interaction gain coefficient table was input into a MATLAB fuzzy inference system. After fuzzification, rule-based inference, and defuzzification, an evaluation index for system coupling strength was obtained. The evaluation results showed that within the inverter cluster at confluence point A, 30% of the inverter pairs had strong coupling strength, 50% had medium coupling strength, and 20% had weak coupling strength. The average coupling strength index for the entire system was 0.65, indicating that the overall system coupling strength was above average. The Python libraries Pandas and Geopandas were used to perform regional mapping on the inverter interaction gain coefficient table. For example, the inverter distribution area at confluence point A was divided into four sub-regions: Region 1, Region 2, Region 3, and Region 4. Pandas was used to organize the data in the interaction gain coefficient table, mapping the interaction gain values ​​of each inverter pair to their corresponding sub-region. Using Geopandas' geographic information capabilities, the data was combined with geographic coordinate information to generate a substation-level interaction relationship table.The table not only contains the interaction gain values ​​for inverter pairs but also their geographical locations. For example, inverters A and B are located in region 1, with an interaction gain of 0.85; inverters C and D are located in region 2, with an interaction gain of 0.6. Based on system coupling strength evaluation metrics, the Python library Network X is used to mine critical coupling paths within the substation-level interaction table. For example, an interaction network graph is constructed, with inverters as nodes and interaction gains as edge weights. Using Network X's shortest path algorithms, such as the Dijkstra algorithm, critical coupling paths at the sink point are mined. The mining results show that at sink point A, critical coupling paths are primarily concentrated in regions 1 and 2, with the interaction gains of inverter pairs along these paths all exceeding 0.7. By calculating the total interaction gain and coupling strength of these paths, the importance of these paths in the system is determined. For example, a critical path involving inverters A, B, and C has a total interaction gain of 2.3 and a coupling strength evaluation metric of 0.8, indicating that this path has a significant impact on the system's voltage stability and reactive power balance.

[0071] This invention significantly improves the identification accuracy of the coupling paths of multiple inverter groups in the distribution network and the efficiency of system stability assessment through multi-dimensional interaction gain analysis and dynamic path mining. Based on multi-operating condition hybrid simulation and interaction gain quantification, it can accurately capture the dynamic coupling strength of substations under different operating scenarios (error ≤ 5%), improving the accuracy of critical path positioning compared to traditional single-operating condition analysis methods. Through regional mapping transformation and indicator fusion, complex device-level interaction data is converted into a substation-level topological relationship table, supporting the real-time update and visualization of coupling strength assessment indicators (such as resonance risk level and reactive power support capacity). Combined with the critical path mining algorithm, it can quickly identify hidden resonant paths and reactive power interaction hotspots, effectively shortening the response time of the voltage over-limit suppression strategy and improving the system's dynamic stability margin.

[0072] Preferably, step S3 includes the following steps:

[0073] Step S31: obtaining a substation-level interaction relationship table; constructing a reactive power sensitivity calculation framework based on the substation-level interaction relationship table to obtain a substation-level reactive power sensitivity calculation framework;

[0074] Step S32: calibrating parameters of the area-level reactive sensitivity calculation framework to obtain a calibrated area-level reactive sensitivity calculation framework;

[0075] Step S33: Calculate the reactive power sensitivity coefficient between substations using the calibration substation-level reactive sensitivity calculation framework to obtain initial data of the substation coupling relationship;

[0076] Step S34: extracting statistical features of the sensitivity coefficients from the initial data of the substation coupling relationship to obtain substation-level sensitivity coefficient distribution features;

[0077] Step S35: threshold screening is performed on the initial data of the substation coupling relationship according to the distribution characteristics of the substation-level sensitivity coefficient to obtain an initial set of significantly coupled substation pairs;

[0078] Step S36: quantifying the directional strength of the initial set of significant coupling stations to obtain a station-level directional coupling strength table;

[0079] Step S37: performing coupling stability assessment based on the area-level directional coupling strength table to obtain an area-level coupling stability index set; performing weight correction on the area-level directional coupling strength table based on the area-level coupling stability index set to obtain a set of significantly coupled area pairs;

[0080] Step S38: constructing a substation-level coupling topology map based on the set of significantly coupled substation pairs.

[0081] It is particularly important that step S38 further includes the following steps:

[0082] Step S381: constructing an initial reactive coupling graph structure based on a set of significantly coupled substation pairs; visualizing the initial reactive coupling graph structure to obtain an initial substation coupling topology graph;

[0083] Step S382: constructing a time-varying spectrum analysis framework based on the initial diagram of the substation coupling topology to obtain a substation-level time-varying spectrum analysis solution;

[0084] Step S383: extracting the graph evolution pattern of the initial graph of the substation coupling topology according to the substation-level time-varying graph analysis scheme to obtain substation-level graph evolution feature data;

[0085] Step S384: Perform community detection on the substation-level graph evolution feature data to obtain an initial substation-level community division plan;

[0086] Step S385: Perform stability analysis and optimization on the initial plan for the substation-level community division to obtain a distribution network coupling sub-area division map; and construct a substation-level coupling topology map based on the distribution network coupling sub-area division map.

[0087] In this embodiment, after obtaining the substation-level interaction relationship table, Python's Pandas library is used to organize the data. For example, the substation-level interaction relationship table for convergence point A is preprocessed to remove duplicate and invalid data. Using Pandas' merge function, the interaction relationship table is merged with the substation basic information table (such as substation name, number, and geographic location) to construct a complete substation-level reactive power sensitivity calculation framework. For example, the "Substation A-Substation B" interaction gain of 0.85 in the interaction relationship table is associated with the basic information of Substations A and B, forming a calculation framework that includes fields such as substation name, number, geographic location, and interaction gain. To calibrate the parameters of the substation-level reactive power sensitivity calculation framework, MATLAB's SimPowerSystems toolbox and Python's Scikit-learn library are used. For example, a power system model for convergence point A is constructed in MATLAB, and power flow calculations are performed to obtain baseline reactive power and voltage data for each substation. This data is imported into Python, and the parameters in the calculation framework are calibrated using Scikit-learn's linear regression model. For example, a linear regression model is established using the reactive power change calculated in MATLAB as the dependent variable and the interaction gain and substation parameters in the calculation framework as the independent variables. The model coefficients are adjusted to minimize the mean squared error between the predicted and actual values. The calibrated calculation framework more accurately reflects the reactive power sensitivity relationship between substations. Using the calibrated substation-level reactive sensitivity calculation framework, ETAP Power System Analysis software is used to calculate the inter-substation reactive power sensitivity coefficient. For example, the network model of sink point A and the calibrated parameters are imported into ETAP. Different reactive power disturbance scenarios are set, such as adding 10 kVar reactive power to substation A and observing the voltage changes in other substations, such as substations B and C. Through multiple simulations, the relationship between the reactive power change and voltage change between substations is recorded, and the inter-substation reactive power sensitivity coefficient is calculated. For example, the calculated results show that the reactive power sensitivity coefficient of substation A to substation B is 0.05 pu / kVar, meaning that a 1 kVar increase in reactive power in substation A causes a 0.05 pu increase in voltage in substation B. These coefficients are organized into initial data for substation coupling relationships. Statistical features of the sensitivity coefficients are extracted from this initial data using Python's SciPy and Matplotlib libraries. For example, SciPy's statistical functions are used to calculate statistical features of the sensitivity coefficients, such as the mean, standard deviation, skewness, and kurtosis. Assuming the initial data for substation coupling relationships contains 30 sensitivity coefficients, the calculated mean is 0.03 pu / kVar, the standard deviation is 0.01 pu / kVar, the skewness is 0.5, and the kurtosis is 3.2. These statistical features reflect the central tendency, dispersion, symmetry, and peakedness of the sensitivity coefficients.Matplotlib is used to draw histograms and box plots to intuitively display the distribution characteristics of the sensitivity coefficients. For example, the histogram shows that the sensitivity coefficients are mainly concentrated between 0.02pu / kVar and 0.04pu / kVar, and the box plot shows that the median is 0.03pu / kVar and the interquartile range is 0.01pu / kVar, which helps to identify outliers and distribution patterns in the data. When performing threshold screening on the initial data of the substation coupling relationship according to the distribution characteristics of the substation-level sensitivity coefficients, Python's NumPy library is used. For example, according to the statistical characteristics obtained in step S34, the screening threshold is set to the mean plus or minus twice the standard deviation, that is, 0.03±2×0.01, and the range is 0.01pu / kVar to 0.05pu / kVar. Using the Boolean indexing function of NumPy, the substation pairs with sensitivity coefficients within this range in the initial data of the substation coupling relationship are screened out to obtain the initial set of significantly coupled substation pairs. For example, the initial set includes 15 substation pairs, all with sensitivity coefficients ranging from 0.01 to 0.05 pu / kVar. This indicates significant coupling between these pairs, significantly impacting system voltage stability. The directional strength of the initial set of significantly coupled substation pairs is quantified using the Python Network X library. For example, a directed graph is constructed, treating substations as nodes, coupling relationships as directed edges, and sensitivity coefficients as edge weights. Network X is used to calculate the directional strength of each directed edge, representing the degree of reactive power influence from one substation to another. For example, the directional strength of the directed edge from substation A to substation B is 0.05 pu / kVar, while the directional strength of the directed edge from substation B to substation A is 0.03 pu / kVar, indicating that substation A has a greater impact on the reactive power of substation B than the reverse direction. These directional strengths are organized into a substation-level directional coupling strength table, which contains not only substation pair information but also directional strength values ​​in both the forward and reverse directions. When evaluating coupling stability based on the substation-level directional coupling strength table, a combination of MATLAB's Control System Toolbox and Python's Statsmodels library is used. For example, in MATLAB, the stability margin analysis function in Control System Toolbox is used to calculate the gain margin and phase margin for each substation pair, resulting in a set of substation-level coupling stability indicators. Assuming the gain margin from substation A to substation B is 8dB and the phase margin is 40 degrees, this indicates good coupling stability in this direction. Statsmodels is used in Python for statistical analysis, and regression analysis of stability indicators is performed to identify key factors influencing stability.Based on the evaluation results, the weights in the substation-level directed coupling strength table are modified. For example, the weights of substation pairs with a gain margin less than 6dB are reduced by 20%, resulting in a set of significantly coupled substation pairs. Based on this set of significantly coupled substation pairs, the initial reactive coupling graph structure is constructed using Python's Network X library. For example, substations are used as nodes, significant coupling relationships are used as directed edges, and the edge weights are the coupling strength values. Using Network X's graph functionality, a directed graph object is created, nodes and edges are added, and the edge weight attributes are set. For example, if the coupling strength between substations A and B is 0.05 pu / kVar, a directed edge from A to B is added to the graph with a weight of 0.05. The initial reactive coupling graph structure is visualized using Gephi software. In Gephi, the graph data generated by Network X is imported, the node size is set according to the reactive power size of the substation, and the edge thickness is set according to the coupling strength, to obtain the initial substation coupling topology graph. The diagram intuitively illustrates the coupling relationship between substations, with the thickness and direction of the edges clearly visible. For example, the edge from substation A to substation B is significantly thicker than the edge from substation B to substation A, indicating that A has a greater impact on B's reactive power. Based on the initial substation coupling topology map, a time-varying spectral analysis framework was constructed using MATLAB's System Identification Toolbox. For example, a time window of 1 hour and a sliding step of 15 minutes were set, and feature extraction was performed on the initial substation coupling topology map within each time window. MATLAB's time-varying model estimation function was used to estimate the temporal variation of the coupling strength for each substation pair. For example, a time-varying analysis of the coupling strength between substations A and B was performed, yielding a daily curve showing how the coupling strength varied. The results showed that the coupling strength was higher in the morning and evening, and lower at noon and late at night. By integrating the time-varying features of all substation pairs, a substation-level time-varying spectral analysis scheme was developed, including time window settings, feature extraction methods, and time-varying model parameters. According to the time-varying graph analysis scheme at the substation level, the graph evolution pattern of the initial graph of the substation coupling topology is extracted using Python's DynNetTex library. For example, the extraction time of the evolution pattern is set to 24 hours a day, and the dynamic network analysis function of DynNetTex is used to extract the time-varying characteristics of the initial graph of the substation coupling topology, such as the frequency of appearance / disappearance of edges, changes in node degree centrality, etc. The extraction results show that the degree centrality of substation A increases significantly in the morning and evening, indicating that its coupling relationship is more active during these periods. These time-varying features are organized into substation-level graph evolution feature data, including changes in degree centrality and clustering coefficient of each substation. When performing community detection on the substation-level graph evolution feature data, Python's igraph library is used. For example, the Louvain community detection algorithm of igraph is used to divide the initial graph of the substation coupling topology into communities.The initial number of communities was set to 3, the algorithm resolution parameter was set to 1.0, and the Louvain algorithm was run to obtain an initial plan for the substation-level community division. The results show that the substation at convergence point A is divided into three communities: Community 1 includes Substations A and B, Community 2 includes Substations C and D, and Community 3 includes Substations E and F. Substations within each community have strong coupling relationships, while coupling between communities is weak. Using the community quality assessment function in igraph, the modularity of the division scheme was calculated to be 0.65, indicating that the division scheme has a good community structure. The stability analysis and optimization of the initial substation-level community division scheme were conducted using a combination of MATLAB's Control System Toolbox and Python's SciPy library. For example, in MATLAB, the stability margin analysis function in the Control System Toolbox was used to calculate internal coupling stability indicators for each community, such as gain margin and phase margin. Assuming that the gain margin of community 1 is 10 dB and the phase margin is 45 degrees, it indicates that the coupling relationship within this community is relatively stable. In Python, using SciPy's optimization function, we optimized the community division scheme and adjusted the community boundaries to maximize the stability index of each community. In the optimized community division scheme, substation G was added to Community 1 and substation D was removed from Community 2. The modularity increased to 0.70, and the stability index improved by an average of 15%. Based on the optimized community division scheme, a distribution network coupling sub-area division map was constructed and visualized using Gephi to obtain the final substation-level coupling topology map. Different colors in the map represent different coupling sub-areas, clearly demonstrating the coupling relationship and community structure between substations.

[0088] This invention significantly improves the accuracy of identifying reactive coupling relationships between distribution network substations and the efficiency of stability assessment through multi-level sensitivity modeling and dynamic coupling quantification. By extracting statistical features and screening with adaptive thresholds, it can accurately locate pairs of significantly coupled substations, improving screening accuracy compared to traditional methods. Directional strength quantification and weight correction mechanisms effectively reveal the dynamic correlation characteristics of the substation's reactive support capacity, improving the accuracy of interactive path visualization in the coupling topology. Combining stability index assessment with topology construction allows real-time tracking of system operating status changes, shortening the response time of the voltage support coordinated control strategy to less than 200ms, while reducing the number of redundant compensation device actions caused by misjudgment of coupling relationships by 30%.

[0089] Preferably, step S4 includes the following steps:

[0090] Step S41: obtaining the pre-processed state data of the convergence point; performing hierarchical clustering on the pre-processed state data of the convergence point to obtain a state data feature grouping set;

[0091] Step S42: constructing a multi-layer polynomial neural network structure based on the state data feature grouping set;

[0092] Step S43: performing parameter regularization on the multi-layer polynomial neural network structure to obtain an anti-overfitting structure parameter set;

[0093] Step S44: performing batch gradient descent training based on the anti-overfitting structure parameter set and the sink point preprocessing state data to obtain initial reactive power-voltage mapping parameters;

[0094] Step S45: performing Bayesian hyperparameter fine-tuning on the initial reactive power-voltage mapping parameters to obtain a fine-tuned reactive power-voltage mapping parameter set;

[0095] Step S46: performing model transient state integration on the fine-tuning reactive power-voltage mapping parameter set to obtain a reactive power-voltage mapping model;

[0096] Step S47: Based on the reactive power-voltage mapping model, a multi-dimensional sensitivity evaluation is performed on the multiple substation aggregation points to obtain a regional comprehensive reactive power sensitivity index.

[0097] In this embodiment, after obtaining the pre-processed state data of the collection point, hierarchical clustering is performed using Python's Scikit-learn library. For example, the pre-processed state data of the collection point A is analyzed, and the data contains voltage, current, and power factor features. Using Scikit-learn's AgglomerativeClustering class, the number of clusters is set to 3, the affinity is 'euclidean', and the link criterion is 'ward'. After the data is standardized, hierarchical clustering is performed. The clustering results show that the data is divided into three groups, of which the first group contains 20% of the data points, mainly concentrated in the low voltage area; the second group contains 50% of the data points, concentrated in the normal voltage area; and the third group contains 30% of the data points, concentrated in the high voltage area. The clustering results are saved as a state data feature grouping set. Based on the state data feature grouping set, a multi-layer polynomial neural network structure is constructed using the Keras library. For example, a neural network containing three hidden layers is constructed, and each hidden layer contains 128, 64, and 32 neurons. The number of neurons in the input layer is determined to be 10 based on the characteristic dimensions of the state data, and the number of neurons in the output layer is 1, used to predict voltage changes. A polynomial kernel function is used in the network to enhance the model's nonlinear expressiveness. The ReLU activation function is selected for the hidden layers to introduce nonlinearity, and a linear activation function is used in the output layer to output continuous values. When compiling the network, the Adam optimizer is used, and the mean squared error (MSE) is selected as the loss function to measure the difference between the predicted and actual values. Through the above operations, a multi-layer polynomial neural network structure is constructed that can capture the complex relationship between state data and voltage changes. To regularize the parameters of this multi-layer polynomial neural network structure, regularization functions in the Keras library are used. For example, L2 regularization is added to each hidden layer of the network, with a regularization coefficient of 0.01. When compiling the model, an early stopping callback function is added to stop training when the loss on the validation set stops improving for three consecutive epochs to prevent overfitting. In addition, a dropout layer is added after each hidden layer with a dropout rate of 0.2. This randomly discards the output of some neurons to enhance the model's generalization ability. The regularization method described above yields a set of structural parameters that resist overfitting. Based on this set of structural parameters and the preprocessed state data at the sink point, batch gradient descent training is performed using the Keras fit function. For example, the batch size is set to 32, with each epoch containing all samples from the training set. The training process is set to 100 epochs, with a learning rate of 0.001. During training, the model is validated using a validation set, which accounts for 20%.Training results show that the model loss on the training set gradually decreases, reaching below 0.02 by the 50th epoch. The loss on the validation set also reaches below 0.03 by the 50th epoch, demonstrating good model fitting. After training, the model weights are saved to obtain the initial reactive power-voltage mapping parameters. Bayesian hyperparameter fine-tuning of these initial reactive power-voltage mapping parameters is performed using the Scikit-learn BayesSearchCV class. For example, a hyperparameter search space is defined, including a learning rate (0.0001 to 0.01), a batch size (16 to 64), and a regularization coefficient (0.001 to 0.1). The number of Bayesian optimization iterations is set to 50, and the default f1 score is used as the evaluation metric. The Bayesian optimization algorithm is used to find the optimal hyperparameter combination within the search space. Optimization results show that the optimal learning rate is 0.002, the batch size is 32, and the regularization coefficient is 0.05. The model is retrained using these hyperparameters to obtain the fine-tuned reactive power-voltage mapping parameter set. When performing model transient state integration on the fine-tuned reactive-voltage mapping parameter set, use Keras's model integration function. For example, construct an integrated model containing five neural network models of the same structure, and initialize each model with the fine-tuned parameter set. During prediction, take the average of the outputs of the five models as the final prediction result. Through the above operation, the prediction error of a single model can be reduced. Use Keras's ModelCheckpoint callback function to save the optimal weights of each model during the training process. For the detailed implementation process of step S47, please refer to the sub-steps of step S47.

[0098] The present invention significantly improves the evaluation accuracy and model generalization capability of the reactive-voltage coupling characteristics of the distribution network by integrating data-driven modeling and adaptive optimization. Through the nonlinear mapping model constructed based on hierarchical clustering and multi-layer polynomial neural network, it can accurately capture the reactive-voltage dynamic response relationship in the high-penetration distributed power supply scenario, effectively reducing the sensitivity evaluation error; through regularization and Bayesian hyperparameter fine-tuning, it effectively suppresses the risk of model overfitting and enhances the robustness under different operating conditions. At the same time, batch training and transient state integration technology increase the model convergence speed by 30%, supporting millisecond-level updates of the regional comprehensive sensitivity index; combined with a multi-dimensional evaluation framework, it can quantify the modal contribution and condition number indicators of the system's reactive power regulation capability, provide dynamic priority sorting for the collaborative compensation strategy, improve the efficiency of voltage over-limit suppression, and reduce the number of redundant compensation equipment actions by 20%.

[0099] Preferably, step S47 includes the following steps:

[0100] Step S471: collecting historical operation data of multiple substations to obtain a historical operation data set of the substations; generating a reactive injection disturbance sequence based on the historical operation data set of the substations;

[0101] Step S472: constructing a voltage-reactive power response function for the reactive power injection disturbance sequence based on the reactive power-voltage mapping model to obtain a set of sink node sensitivity functions;

[0102] Step S473: performing radial basis interpolation based on the sink node sensitivity function set to obtain a reactive injection-voltage response surface; performing regression spline fitting on the reactive injection-voltage response surface to obtain a distribution system nonlinear characteristic surface;

[0103] Step S474: extracting typical operating points from the nonlinear characteristic surface of the power distribution system to obtain a data set of key operating points of the system; generating a Jacobian matrix based on the data set of key operating points of the system to obtain a discrete point reactive sensitivity table;

[0104] Step S475: performing tensor interpolation on the discrete point reactive sensitivity table to obtain a continuous reactive sensitivity table; performing matrix condition number evaluation based on the continuous reactive sensitivity table to obtain a sensitivity condition number evaluation index;

[0105] Step S476: performing singular value threshold analysis on the continuous reactive power sensitivity table to obtain sensitivity modal contribution;

[0106] Step S477: Evaluate the reactive power control capability of the power distribution system based on the sensitivity condition number evaluation index and the sensitivity modal contribution, and obtain a system reactive power control capability evaluation index;

[0107] Step S478: performing operation state space distance calculation on the system reactive power control capability evaluation index to obtain an operation point weight distribution map; performing weighted fusion on the system reactive power control capability evaluation index based on the operation point weight distribution map to obtain a regional comprehensive reactive power sensitivity index.

[0108] In this embodiment, when collecting historical operation data for multiple substations, Python's Pandas library is used to read historical data stored in CSV files. For example, the historical operation data set of substation A contains data such as voltage, current, power factor, etc. for the past year, with a sampling frequency of once an hour. Use Pandas' read_csv function to load the data and perform preprocessing, including removing missing values ​​and outliers. Generate a reactive injection disturbance sequence based on the historical operation data set. For example, select reactive power change data within a day to construct a disturbance sequence containing 24 time points. Use the NumPy library to generate a random disturbance sequence with a disturbance amplitude within the range of ±0.1pu as the initial value of the reactive injection disturbance sequence. When constructing the voltage-reactive response function of the reactive injection disturbance sequence based on the reactive-voltage mapping model, use MATLAB's Curve Fitting Toolbox. For example, import the reactive injection disturbance sequence and the corresponding voltage response data into MATLAB. Use the polynomial fitting function in the Curve Fitting Toolbox and select the cubic polynomial as the fitting model to obtain the node sensitivity function set of the substation. Fitting results show that a cubic polynomial can well fit the voltage-reactive power response relationship, with a coefficient of determination (R²) exceeding 0.95. Each function in the function set corresponds to a sink node and describes the voltage response characteristics of that node to reactive power injection. The Python SciPy library was used to perform radial basis interpolation based on the sink node sensitivity function set. For example, the SciPy Rbf function was used to perform radial basis interpolation with reactive power injection as input and voltage response as output. The radial basis function was set to a Gaussian function with a smoothing parameter of 0.1 to obtain the reactive power injection-voltage response surface. The SciPy UnivariateSpline function was used to fit the response surface using a regression spline with a smoothing factor of 1.0, resulting in the distribution system nonlinear characteristic surface. The fitting results show that the regression spline accurately captures the nonlinear characteristics of the voltage-reactive power response, with a mean square error (MSE) of less than 0.01 pu. The Statistics and Machine Learning Toolbox in MATLAB was used to extract typical operating points from the distribution system nonlinear characteristic surface. For example, using the k-means clustering algorithm, the operating points on the characteristic surface are divided into three typical operating regions: low voltage, normal voltage, and high voltage. The number of clusters is set to three, and the clustering algorithm is run to obtain the center point of each region as the system's key operating point dataset. Based on this key operating point dataset, the Jacobian matrix is ​​generated using MATLAB's Symbolic Math Toolbox. For each key operating point, the partial derivative of reactive power with respect to voltage is calculated, and the Jacobian matrix is ​​constructed to obtain a discrete-point reactive sensitivity table.For example, the Jacobian matrix at a key operating point shows that the reactive power sensitivity to voltage is 0.05 pu / kVar. The Python TensorFlow library is used to perform tensor interpolation on the discrete-point reactive power sensitivity table. For example, the discrete-point reactive power sensitivity table is represented as a two-dimensional tensor and bilinearly interpolated using TensorFlow's interpolate function to obtain a continuous reactive power sensitivity table. The interpolation grid step size is set to 0.01 pu to ensure smoothness and accuracy of the interpolation result. Based on the continuous reactive power sensitivity table, the matrix condition number is evaluated using NumPy's linalg.cond function. The calculated condition number is 15.2, indicating that the system reactive power sensitivity matrix is ​​not ill-conditioned and the numerical calculation is stable. The condition number is used as a sensitivity condition number evaluation metric. The SignalProcessing Toolbox in MATLAB is used to perform singular value threshold analysis on the continuous reactive power sensitivity table. For example, the svd function is used to calculate the singular value decomposition of the continuous reactive power sensitivity table, obtaining the singular value spectrum. The singular value threshold is set to 0.01, and singular values ​​below the threshold are treated as noise and filtered out. The analysis results show that the first three singular values ​​account for over 95% of the total singular values, indicating that the modes corresponding to these three singular values ​​contribute to the system's primary modes. The contribution of each mode is calculated to obtain the sensitivity modal contribution. For example, the contribution of the first mode is 60%, the second is 30%, and the third is 5%, indicating that the system's reactive power control capability is primarily dominated by the first two modes. When evaluating the reactive power control capability of the distribution system based on the sensitivity condition number evaluation metric and the sensitivity modal contribution, the Python Matplotlib library is used for visualization analysis. For example, a bar chart of the condition number evaluation metric and the modal contribution is plotted to intuitively display the key factors of the system's reactive power control capability. Based on the evaluation results, the system's reactive power control capability evaluation index is calculated. For example, combining the condition number evaluation metric and the modal contribution yields a system reactive power control capability score of 85 (out of 100), indicating that the system has good reactive power control capability. MATLAB's Distances Toolbox was used to calculate the operating state space distance for the system's reactive power control capability evaluation indicators. For example, the ideal operating state of the system was defined as a point with a condition number evaluation index of 10 and a modal contribution of 100%. The Euclidean distance between the current evaluation indicator and the ideal state was calculated. The resulting distance value was 7.2, indicating the degree of deviation between the current operating state and the ideal state. Based on the operating point weight distribution diagram, a weighted fusion of the system's reactive power control capability evaluation indicators was performed. Matplotlib was used to plot the operating point weight distribution diagram, assigning weights based on distance values, with smaller distances giving larger weights. For example, the weights were assigned as 1 / distance to obtain a weight-normalized distribution diagram.Based on the weight distribution diagram, the system reactive power control capability evaluation indicators were weighted and fused, and the regional comprehensive reactive power sensitivity index was obtained as 0.85 (normalized value), indicating that the reactive power sensitivity of the region is at a relatively high level overall.

[0109] The present invention significantly improves the accuracy of reactive sensitivity assessment of distribution networks and the comprehensiveness of system control capability analysis through the fusion of multi-dimensional nonlinear modeling and dynamic weighting. Through reactive disturbance injection driven by historical data and radial basis interpolation fitting, it can accurately construct a nonlinear voltage-reactive response surface, effectively reduce the sensitivity modeling error, and effectively overcome the scenario limitations of traditional linearization methods; through the collaborative analysis of Jacobian matrix and tensor interpolation, it realizes the high-fidelity extension of discrete sensitivity to continuous space, supports the quantification of modal contribution of key operating points, and combines condition number evaluation and singular value decomposition. The prediction deviation of system reactive control stability index is less than 5%; relying on state space distance calculation and dynamic weight allocation, it can generate regional comprehensive sensitivity index within seconds, adapt to the rapid fluctuation scenario of photovoltaic output, improve the response accuracy of voltage collaborative control strategy, and reduce the action frequency of redundant compensation equipment.

[0110] Preferably, step S5 includes the following steps:

[0111] Step S51: performing coupling subgraph feature clustering on the substation-level coupling topology graph to obtain a substation coupling sub-region intensity distribution table;

[0112] Step S52: performing hierarchical game optimization based on the substation coupling sub-region intensity distribution table and the regional comprehensive reactive sensitivity index to obtain a substation double-layer grid-equipment optimization framework;

[0113] Step S53: constructing a mathematical model for the substation's two-layer grid-equipment optimization framework to obtain a substation's reactive power collaborative optimization mathematical model. The constructed mathematical model is a two-layer multi-objective optimization model, wherein the two-layer multi-objective refers to minimizing the voltage deviation of the regional distribution network and minimizing the total input capacity of the reactive power compensation equipment.

[0114] Step S54: performing distributed parallel solution on the reactive power collaborative optimization mathematical model of the substation area to obtain a set of non-inferior solutions for the substation area multi-objective;

[0115] Step S55: performing fuzzy membership weight aggregation on the substation multi-objective non-inferior solution set to obtain the substation initial reactive power compensation strategy set;

[0116] Step S56: performing partition coordination optimization on the initial reactive power compensation strategy set of the substation according to the key coupling path of the collection point to obtain the coordinated reactive power compensation control strategy of the substation.

[0117] It is particularly important that step 56 further includes the following steps:

[0118] Step S561: constructing a partition impedance matching coordination mechanism based on the key coupling path of the collection point; performing regional decomposition of the initial reactive power compensation strategy set of the substation area based on the partition impedance matching coordination mechanism to obtain a multi-region optimization sub-problem set;

[0119] Step S562: Solve the multi-region optimization sub-problem set in parallel to obtain a regional reactive power optimization feasible solution set; perform boundary constraint verification on the regional reactive power optimization feasible solution set to obtain the substation boundary coupling relaxation parameter;

[0120] Step S563: iteratively modify the feasible solution set of regional reactive power optimization according to the substation boundary coupling relaxation parameter to obtain the substation coordinated reactive power compensation solution set;

[0121] Step S564: PV output is predicted for multiple grid-connected points to obtain short-term PV output prediction data; time series load scenarios are generated based on the short-term PV output prediction data to obtain multi-period dynamic load scenario data;

[0122] Step S565: Performing dynamic timing planning based on the multi-period dynamic load scenario data and the area coordinated reactive compensation scheme set to obtain an initial reactive equipment switching timing scheme;

[0123] Step S566: Perform ramp constraints and minimum time interval screening on the initial reactive equipment switching sequence plan to obtain a reactive compensation device switching sequence table;

[0124] Step S567: constructing a virtual synchronous machine control framework according to the reactive power compensation device switching timing table to obtain an initial virtual synchronous machine control parameter set;

[0125] Step S568: Obtain an inverter group transient simulation model library; use the inverter group transient simulation model library to perform parameter sensitivity verification on the initial virtual synchronous machine control parameter set to obtain the substation coordinated reactive power compensation control strategy.

[0126] In this embodiment, Gephi software is used when clustering the coupling subgraph features of the substation-level coupling topology graph. For example, the substation-level coupling topology graph is imported into Gephi, and Gephi's modularity algorithm is used for community detection. The resolution parameter is set to 1.0, and after running the algorithm, the substation is divided into three coupling sub-areas. The substation coupling strength within each sub-area is high, while the coupling strength between sub-areas is low. The clustering results are exported to obtain a substation coupling sub-area strength distribution table, which contains the sub-areas to which each substation belongs and their coupling strength values. For example, substation A and substation B belong to sub-area 1 with a coupling strength of 0.8; substation C and substation D belong to sub-area 2 with a coupling strength of 0.6; substation E and substation F belong to sub-area 3 with a coupling strength of 0.7. When performing hierarchical game optimization based on the substation coupling sub-area strength distribution table and the regional comprehensive reactive sensitivity index, Gambit software is used. For example, each coupling sub-area is regarded as a player, and the player's objective function is reactive power optimization within the sub-area, and the constraints include voltage upper and lower limits and reactive compensation capacity limits. The payoff function of the game is set as the reactive power optimization effect of the sub-region, for example, minimizing voltage deviation and reactive power compensation capacity. Using Gambit to model and solve the game, a two-layer grid-device optimization framework for the substation area is obtained. This framework includes the optimal reactive power compensation strategy for each sub-region and the coordination strategy between sub-regions. For example, the optimal compensation capacity for sub-region 1 is 50 kV, for sub-region 2 is 30 kV, and for sub-region 3 is 40 kV.

[0127] When constructing the mathematical model for the substation's two-layer grid-device optimization framework, MATLAB's SymbolicMath Toolbox was used. For example, the optimization objective function for the upper grid layer is defined as minimizing the sum of the squares of all substation voltage deviations, expressed as: ,in, is the voltage of station i, is the reference voltage, and N is the number of all substations involved in the calculation of the sum of squared voltage deviations. The optimization objective function of the lower-level equipment is to minimize the reactive compensation capacity, which is expressed as: ,in, is the reactive compensation capacity of device z, and M is the number of all reactive compensation devices involved in the reactive compensation capacity minimization objective function. Using the Lagrange multiplier method, the upper and lower objective functions are coupled to obtain the mathematical model of reactive power collaborative optimization in the substation area: ,in is the Lagrange multiplier. Pyomo and Ipopt solvers are used to perform distributed parallel solutions to the reactive power collaborative optimization mathematical model of the substation area. For example, Pyomo is used to build the optimization model, define the variables, objective function, and constraints. Using Pyomo's distributed solution function, the model is decomposed into multiple subproblems, each of which corresponds to a coupled sub-region. The Ipopt solver is run on each subproblem, with the maximum number of iterations set to 1000 and the convergence tolerance set to . Inter-process communication is performed via MPI (Message Passing Interface) to coordinate the solution results of each subproblem, ultimately resulting in a set of multi-objective non-inferior solutions for the substation area, which includes multiple Pareto optimal solutions. For example, one solution in the set is: Sub-area 1 compensates 50 kVar, Sub-area 2 compensates 30 kVar, and Sub-area 3 compensates 40 kVar, with a corresponding total voltage deviation of 0.02 pu²; another solution is: Sub-area 1 compensates 40 kVar, Sub-area 2 compensates 25 kVar, and Sub-area 3 compensates 35 kVar, with a corresponding total voltage deviation of 0.03 pu². The Python Scikit-fuzzy library is used to aggregate fuzzy membership weights for the set of multi-objective non-inferior solutions for the substation area. For example, a fuzzy membership function is defined to fuzzify the performance of each non-inferior solution on each objective. Assuming that the optimization objectives include minimizing voltage deviation and minimizing reactive power compensation capacity, a Gaussian membership function is defined for each objective, with set center point and width parameters. For example, the membership function for voltage deviation has a center point of 0.01 pu² and a width of 0.005 pu²; the membership function for reactive power compensation capacity has a center point of 30 kVar and a width of 5 kVar. Using Scikit-fuzzy's fuzzy_or function, the membership of each non-inferior solution is aggregated to obtain a comprehensive membership value. Non-inferior solutions are ranked based on their comprehensive membership values, and the solution with the highest comprehensive membership is selected as the initial reactive power compensation strategy set for the substation. For example, if one non-inferior solution has a comprehensive membership of 0.85 and another has a comprehensive membership of 0.75, the solution with a comprehensive membership of 0.85 is selected as the recommended strategy, and the corresponding compensation strategy is 50 kV compensation for sub-area 1, 30 kV compensation for sub-area 2, and 40 kV compensation for sub-area 3. When constructing a partitioned impedance matching coordination mechanism based on the critical coupling path at the sink point, MATLAB's Control System Toolbox is used. For example, for the critical coupling path at sink point A, two major coupling regions were identified: Region 1 encompasses substations A and B, and Region 2 encompasses substations C and D. An impedance matching model was constructed for each region, with a target impedance set to 0.1 + j0.2Ω (based on system design specifications). The actual impedance was calculated using the MATLAB impedance function, and matching network parameters (such as inductance and capacitance) were adjusted to bring the actual impedance close to the target impedance. Based on this coordination mechanism, the initial reactive power compensation strategy for the substation was decomposed into two subproblems: Subproblem 1, for Region 1, targets a 50 kV compensation strategy; Subproblem 2, for Region 2, targets a 30 kV compensation strategy. Using the regional decomposition method, a multi-region optimization subproblem set was generated, each corresponding to the optimization requirements of a coupling region. The Python Dask distributed computing framework was used to parallelize the multi-region optimization subproblems.For example, subproblems 1 and 2 were assigned to different computational nodes for solution. Pyomo was used to construct an optimization model for each node, and the solution was performed using the Ipopt solver. The maximum number of iterations was set to 500, and the convergence tolerance was set to 1e-5. After the solution was completed, a feasible solution set for regional reactive power optimization was obtained: the feasible solution for region 1 was to compensate from 45 kV to 55 kV, and the feasible solution for region 2 was to compensate from 25 kV to 35 kV. Boundary constraint verification was performed using MATLAB's Simulink, simulating the voltage and power variations at the substation boundary. The verification results showed that when 55 kV was compensated in region 1, the voltage at the boundary exceeded the upper limit; when 25 kV was compensated in region 2, the reactive power was insufficient. Based on the verification results, the substation boundary coupling relaxation parameters were determined: the voltage relaxation parameter for region 1 was 0.02 pu, and the reactive power relaxation parameter for region 2 was 0.01 pu. The MATLAB fmincon function was used to iteratively modify the feasible solution set for regional reactive power optimization based on the substation boundary coupling relaxation parameters. For example, the feasible solution range for region 1 was revised to 45 kV to 53 kV (accounting for voltage relaxation parameters), and the feasible solution range for region 2 was revised to 27 kV to 35 kV (accounting for reactive power relaxation parameters). The initial iteration values ​​were set to 50 kV compensation for region 1 and 30 kV for region 2. Constrained optimization was performed using fmincon, with the objective function minimizing the voltage deviation. The constraints included the revised feasible solution range and the boundary relaxation parameters. After iterative convergence, a set of coordinated reactive power compensation schemes for the substations was obtained: 50 kV compensation for region 1 and 30 kV for region 2. This corresponds to a voltage deviation of 0.01 pu, and reactive power meets the required requirements. The Python Prophet library was used to forecast PV output for multiple substations. For example, historical PV output data for substation A (sampled hourly for the past year) was imported and combined with weather forecast data (such as solar irradiance and temperature). Time series modeling was performed using Prophet's default parameters to forecast PV output for the next 24 hours. The forecast results show that the photovoltaic output is between 0.6 and 0.9 pu during the day and 0 pu at night. Based on the forecast data, Python's SimPy library is used to generate time series load scenarios. The time step of the load scenario is set to 15 minutes, and multi-period dynamic load scenario data is generated. For example, the generated scenario data shows that from 10 am to 2 pm, the load demand is high and the photovoltaic output is sufficient; from dusk to night, the load demand decreases, but the photovoltaic output decreases. When performing time series dynamic planning based on multi-period dynamic load scenario data and the area coordinated reactive power compensation scheme set, MATLAB's Dynamic Programming Toolbox is used. For example, a day is divided into 24 periods, and each period corresponds to a load scenario.The state variable is defined as the reactive power compensation capacity for the current period, and the decision variable is the compensation adjustment for the next period. The objective function is to minimize the integral of the voltage deviation over the entire day, with constraints including upper and lower limits on the compensation capacity and the adjustment rate. Using a dynamic programming algorithm, an initial reactive power switching sequence is calculated: the compensation capacity in area 1 is increased to 55 kV at 10:00 AM and reduced to 50 kV at 4:00 PM; the compensation capacity in area 2 is increased to 35 kV at 7:00 PM and reduced to 25 kV at 11:00 PM. The Python Pandas library is used to filter the initial reactive power switching sequence for ramping constraints and minimum time intervals. For example, according to power grid operation regulations, the ramping constraint is set to a maximum compensation capacity change of 10 kV per hour, with a minimum time interval of 30 minutes. The rolling function in Pandas is used to check whether the switching sequence meets the constraints. The initial scheme's adjustment between 10:00 AM and 11:00 AM exceeded the ramping constraint, so it was adjusted to increase the compensation capacity in two phases, at 10:00 AM and 10:30 AM. The filtered reactive power compensation device switching schedule shows that the compensation capacity in Area 1 will be increased by 5 kVar at 10:00 AM and increased by another 5 kVar at 10:30 AM; it will be reduced by 5 kVar at 4:00 PM and reduced by another 5 kVar at 4:30 PM. MATLAB's Simulink was used to build the virtual synchronous machine control framework based on the reactive power compensation device switching schedule. For example, a virtual synchronous machine control model was constructed for the inverter cluster at point A, including virtual inertia control, droop control, and reactive power compensation control modules. Based on the switching schedule, the virtual inertia constant and droop coefficient were adjusted at specific times. For example, at 10:00 AM, the virtual inertia constant was adjusted from 0.1 Hz to 0.15 Hz, and the droop coefficient was adjusted from 0.02 pu / kVar to 0.03 pu / kVar to accommodate the increase in compensation capacity. Simulations verified that the adjusted control parameters effectively maintained system voltage stability, resulting in an initial set of virtual synchronous generator control parameters: a virtual inertia constant of 0.15 Hz, a droop coefficient of 0.03 pu / kVar, and a reactive power compensation gain of 0.5. After obtaining a library of transient simulation models for the inverter cluster, PSCAD / EMTDC software was used to perform parameter sensitivity checks on the initial set of virtual synchronous generator control parameters. For example, the transient simulation model of the inverter cluster at point A was selected, and simulations were performed with different virtual inertia constants (0.1 Hz, 0.15 Hz, and 0.2 Hz) and droop coefficients (0.02 pu / kVar, 0.03 pu / kVar, and 0.04 pu / kVar). Each simulation lasted 0.5 seconds, with a step size of 1 microsecond. The voltage and frequency responses of the system were recorded under different parameter settings.The simulation results were analyzed using MATLAB's Sensitivity Analysis Toolbox to calculate the sensitivity of each parameter to system performance indicators (such as voltage deviation and frequency deviation). The results showed that the sensitivity of the virtual inertia constant to frequency deviation was -0.05 pu / H, and the sensitivity of the droop coefficient to voltage deviation was 0.03 pu / (pu / kVar). Based on the sensitivity analysis, the control parameters were adjusted to obtain the final coordinated reactive power compensation control strategy for the substation area: the virtual inertia constant was set to 0.15 H and the droop coefficient was set to 0.03 pu / kVar, ensuring system stability and responsiveness under different operating conditions.

[0128] The present invention significantly improves the global coordination and dynamic response efficiency of the reactive compensation strategy of the distribution network through hierarchical game optimization and multi-objective collaboration. By clustering the characteristics of coupled sub-regions and aggregating fuzzy weights, it can accurately quantify the reactive interaction intensity between substations, and combined with a distributed parallel solution framework, effectively reduce the computational time of complex optimization problems; through a two-layer grid-equipment optimization model and a partition coordination strategy, it realizes the dynamic allocation of reactive resources across substations, effectively shortens the voltage over-limit suppression response time, and reduces the switching frequency of redundant compensation equipment, thereby improving the overall energy efficiency of the system; relying on the generation of non-inferior solution sets and the adaptation of key coupling paths, it can quickly generate robust compensation schemes in high-penetration new energy scenarios, and increase the system stability margin.

[0129] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0130] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point, characterized in that: The following steps are involved: Step S1: deploying synchronized phasor measurement units at the convergence points of multiple substations to collect electrical parameters and generate original state data of the convergence points; Performing quality verification and cleaning on the original state data to generate convergence point preprocessing state data; performing feature extraction and time series reconstruction on the convergence point preprocessing state data to obtain a convergence point key feature parameter set; Step S2: Based on the key characteristic parameter set, an initial inverter group model is constructed by clustering inverter operation characteristics and analyzing control link characteristics; Based on this model, electromagnetic transient-steady-state hybrid simulation is performed on the inverter group to generate a transient-steady-state hybrid simulation scheme; Based on this solution, through multi-operating condition interaction simulation and interaction gain calculation, the key coupling paths of the convergence point are excavated; Step S3: Based on the calculation of reactive power sensitivity between substations and the evaluation of coupling strength, identify significantly coupled substation pairs; and construct a substation-level directed coupling topology map based on the significant coupling relationship; Step S4: Based on the pre-processed state data of the sink point, a reactive-voltage nonlinear mapping model is constructed through a multi-layer neural network; the model is used to perform sensitivity tensor interpolation and modal contribution analysis to generate a regional comprehensive reactive sensitivity index; Step S5: Combine the substation-level directed coupling topology map and the regional comprehensive reactive sensitivity index to generate the substation initial reactive compensation strategy set through hierarchical game optimization; perform partition coordination and virtual synchronous machine control parameter optimization on the substation initial reactive compensation strategy set according to the key coupling path of the collection point, and output the substation collaborative reactive compensation control strategy.

2. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: deploying synchronized phasor measurement units at multiple substation collection points to obtain a measurement unit distribution configuration table; constructing a network topology structure for the multiple substation collection points based on the measurement unit distribution configuration table to obtain a data acquisition network topology diagram; Step S12: constructing a data transmission channel according to the data acquisition network topology diagram to obtain a real-time data stream channel mapping table; allocating communication bandwidth to the real-time data stream channel mapping table to obtain a bandwidth resource configuration plan; Step S13: Deploy distributed computing resources according to the bandwidth resource configuration plan to obtain a distributed collection computing resource pool; perform software system integration on the distributed collection computing resource pool to obtain a substation data collection platform; Step S14: Using the substation data acquisition platform, electrical parameters are collected from multiple substation collection points to obtain original state data of the collection points. The collection frequency of the electrical parameters is 60 samples per second. The original state data of the collection points is quality checked and cleaned to obtain pre-processed state data of the collection points. Step S15: formulating a sampling strategy based on the pre-processed state data of the sink point to obtain a state data sampling control strategy; reconstructing the time series data of the pre-processed state data of the sink point according to the state data sampling control strategy to obtain multi-time series state space data of the sink point; Step S16: extract key feature parameters from the multi-time series state space data of the sink point to obtain a set of key feature parameters of the sink point.

3. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 1, wherein: Step S2 includes the following steps: Step S21: extracting inverter operation characteristics from the collection point key characteristic parameter set to obtain an inverter operation characteristic set; Step S22: clustering the inverter operation feature set to obtain an inverter type distribution map; Step S23: selecting inverter samples and performing structural modeling based on the inverter type distribution map to obtain a typical inverter structure model set; Step S24: performing control link characteristic analysis on a typical inverter structure model set to obtain an inverter control characteristic parameter set; Step S25: performing analytical modeling of the inverter output impedance according to the inverter control characteristic parameter set to obtain an output impedance characteristic curve set; Step S26: constructing a generalized inverter equivalent circuit model based on the output impedance characteristic curve set and the typical inverter structure model set to obtain an initial model of the inverter group; Step S27: performing electromagnetic transient-steady-state hybrid simulation on the inverter group based on the initial model of the inverter group to obtain a transient-steady-state hybrid simulation scheme; Step S28: Based on the transient-steady hybrid simulation solution, perform multi-operating condition interactive simulation and coupling path mining on the inverter group to obtain the key coupling path of the collection point.

4. The method for evaluating voltage-reactive coupling characteristics of a regional distribution network collection point according to claim 3, characterized in that: Step S24 includes the following steps: Step S241: performing a loop response test on a typical inverter structure model set to obtain a time domain response characteristic curve of a control link; Step S242: performing frequency characteristic conversion on the time domain response characteristic curve of the control link to obtain a frequency response characteristic curve of the control link; Step S243: Calculating the gain margin and phase margin according to the control link frequency response characteristic curve to obtain a control link stability evaluation parameter; Step S244: performing linearization processing on the typical inverter structure model set to obtain the small signal transfer function of the control link; Step S245: extracting pole and zero spatial distribution features based on the small signal transfer function of the control link to obtain dynamic characteristic parameters of the control link; Step S246: Generate an inverter control characteristic parameter set according to the control link stability evaluation parameter and the control link dynamic characteristic parameter.

5. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 3, wherein: Step S27 includes the following steps: Step S271: obtaining the pre-processed state data of the collection point; performing parameter identification pre-processing on the pre-processed state data of the collection point to obtain a data set for inverter group identification; Step S272: constructing a random subspace state space model for the inverter group based on the inverter group identification data set to obtain a system state space model coefficient table; Step S273: performing system characteristic mode analysis on the system state space model coefficient table to obtain the inverter group system characteristic mode set; Step S274: extracting the frequency domain resonance point according to the inverter group system characteristic mode set to obtain the inverter group frequency domain characteristic parameter set, wherein the minimum frequency interval of the spectrum analysis is ≤1 Hz; Step S275: constructing an electromagnetic transient simulation model for the inverter group based on the inverter group frequency domain characteristic parameter set to obtain an inverter group transient simulation model library; Step S276: constructing a power system steady-state simulation model for the power system based on the pre-processed state data of the sink point to obtain a power system steady-state simulation model library; Step S277: constructing a hybrid simulation framework based on the inverter group transient simulation model library and the power system steady-state simulation model library to obtain a transient-steady-state hybrid simulation solution.

6. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 3, wherein: Step S28 includes the following steps: Step S281: performing multi-operating condition interactive simulation on the inverter group based on a transient-steady state hybrid simulation scheme to obtain inverter interactive characteristic data; Step S282: performing interaction gain calculation on the inverter group based on the inverter interaction characteristic data to obtain an inverter interaction gain coefficient table; Step S283: Evaluate the coupling strength of the inverter group based on the inverter interaction gain coefficient table to obtain a system coupling strength evaluation index; Step S284: performing a regional mapping transformation on the inverter interaction gain coefficient table to obtain a substation-level interaction relationship table; Step S285: mining the coupling key path of the substation-level interaction relationship table according to the system coupling strength evaluation index to obtain the convergence point key coupling path.

7. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 1, wherein: Step S3 includes the following steps: Step S31: obtaining a substation-level interaction relationship table; constructing a reactive power sensitivity calculation framework based on the substation-level interaction relationship table to obtain a substation-level reactive power sensitivity calculation framework; Step S32: calibrating parameters of the area-level reactive sensitivity calculation framework to obtain a calibrated area-level reactive sensitivity calculation framework; Step S33: Calculate the reactive power sensitivity coefficient between substations using the calibration substation-level reactive sensitivity calculation framework to obtain initial data of the substation coupling relationship; Step S34: extracting statistical features of the sensitivity coefficients from the initial data of the substation coupling relationship to obtain substation-level sensitivity coefficient distribution features; Step S35: threshold screening is performed on the initial data of the substation coupling relationship according to the distribution characteristics of the substation-level sensitivity coefficient to obtain an initial set of significantly coupled substation pairs; Step S36: quantifying the directional strength of the initial set of significant coupling stations to obtain a station-level directional coupling strength table; Step S37: performing coupling stability assessment based on the area-level directional coupling strength table to obtain an area-level coupling stability index set; performing weight correction on the area-level directional coupling strength table based on the area-level coupling stability index set to obtain a set of significantly coupled area pairs; Step S38: constructing a substation-level coupling topology map based on the set of significantly coupled substation pairs.

8. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 1, wherein: Step S4 includes the following steps: Step S41: obtaining the pre-processed state data of the convergence point; performing hierarchical clustering on the pre-processed state data of the convergence point to obtain a state data feature grouping set; Step S42: constructing a multi-layer polynomial neural network structure based on the state data feature grouping set; Step S43: performing parameter regularization on the multi-layer polynomial neural network structure to obtain an anti-overfitting structure parameter set; Step S44: performing batch gradient descent training based on the anti-overfitting structure parameter set and the sink point preprocessing state data to obtain initial reactive power-voltage mapping parameters; Step S45: performing Bayesian hyperparameter fine-tuning on the initial reactive power-voltage mapping parameters to obtain a fine-tuned reactive power-voltage mapping parameter set; Step S46: performing model transient state integration on the fine-tuning reactive power-voltage mapping parameter set to obtain a reactive power-voltage mapping model; Step S47: Based on the reactive power-voltage mapping model, a multi-dimensional sensitivity evaluation is performed on the multiple substation aggregation points to obtain a regional comprehensive reactive power sensitivity index.

9. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 8, characterized in that: Step S47 includes the following steps: Step S471: collecting historical operation data of multiple substations to obtain a historical operation data set of the substations; generating a reactive injection disturbance sequence based on the historical operation data set of the substations; Step S472: constructing a voltage-reactive power response function for the reactive power injection disturbance sequence based on the reactive power-voltage mapping model to obtain a set of sink node sensitivity functions; Step S473: performing radial basis interpolation based on the sink node sensitivity function set to obtain a reactive injection-voltage response surface; performing regression spline fitting on the reactive injection-voltage response surface to obtain a distribution system nonlinear characteristic surface; Step S474: extracting typical operating points from the nonlinear characteristic surface of the power distribution system to obtain a data set of key operating points of the system; generating a Jacobian matrix based on the data set of key operating points of the system to obtain a discrete point reactive sensitivity table; Step S475: performing tensor interpolation on the discrete point reactive sensitivity table to obtain a continuous reactive sensitivity table; performing matrix condition number evaluation based on the continuous reactive sensitivity table to obtain a sensitivity condition number evaluation index; Step S476: performing singular value threshold analysis on the continuous reactive power sensitivity table to obtain sensitivity modal contribution; Step S477: Evaluate the reactive power control capability of the power distribution system based on the sensitivity condition number evaluation index and the sensitivity modal contribution, and obtain a system reactive power control capability evaluation index; Step S478: performing operation state space distance calculation on the system reactive power control capability evaluation index to obtain an operation point weight distribution map; performing weighted fusion on the system reactive power control capability evaluation index based on the operation point weight distribution map to obtain a regional comprehensive reactive power sensitivity index.

10. The method for evaluating voltage-reactive coupling characteristics at a regional distribution network collection point according to claim 1, wherein: Step S5 includes the following steps: Step S51: performing coupling subgraph feature clustering on the substation-level coupling topology graph to obtain a substation coupling sub-region intensity distribution table; Step S52: performing hierarchical game optimization based on the substation coupling sub-region intensity distribution table and the regional comprehensive reactive sensitivity index to obtain a substation double-layer grid-equipment optimization framework; Step S53: constructing a mathematical model for the substation's two-layer grid-equipment optimization framework to obtain a substation's reactive power collaborative optimization mathematical model. The constructed mathematical model is a two-layer multi-objective optimization model, wherein the two-layer multi-objective refers to minimizing the voltage deviation of the regional distribution network and minimizing the total input capacity of the reactive power compensation equipment. Step S54: performing distributed parallel solution on the reactive power collaborative optimization mathematical model of the substation area to obtain a set of non-inferior solutions for the substation area multi-objective; Step S55: performing fuzzy membership weight aggregation on the substation multi-objective non-inferior solution set to obtain the substation initial reactive power compensation strategy set; Step S56: performing partition coordination optimization on the initial reactive power compensation strategy set of the substation according to the key coupling path of the collection point to obtain the coordinated reactive power compensation control strategy of the substation.

Images