An Active Method for Suppressing Voltage Exceedances in a Photovoltaic-Storage System in a Distribution Network

Abstract

This invention provides a method for proactively suppressing voltage exceedances in a photovoltaic-storage system within a power distribution network. By frequently collecting and fusing data on grid connection point electrical parameters, surrounding load characteristics, meteorological conditions, and historical electricity consumption behavior, it dynamically identifies the current load type and establishes a time-varying equivalent feeder impedance and load sensitivity model, thereby enabling multi-step prediction of future short-term voltage amplitudes. Based on voltage change trajectories and rates of change, it uses fuzzy membership discrimination to classify voltage trend states and dynamically adjusts the exceedance threshold based on the trend results, generating a control trigger signal. Through a reverse confirmation mechanism, it prevents false triggering or unnecessary adjustments, ultimately issuing adjustment commands to the photovoltaic-storage system, thus closing the loop to suppress voltage exceedance risks. This invention improves the reliability of prediction and the accuracy of control, achieving proactive, intelligent, and dynamic voltage safety management.

Description

Technical Field

[0001] This invention relates to the field of intelligent voltage control technology for distribution networks, and in particular to an active method for suppressing voltage over-limits in a photovoltaic-storage system in a distribution network. Background Technology

[0002] In the field of intelligent voltage control technology for distribution networks, especially with the integration of high proportions of renewable energy (such as photovoltaics and energy storage), the prediction and active control of voltage exceedances have become a key research focus. Traditional distribution network voltage control schemes mainly employ static threshold judgment technology, which compares the real-time collected bus voltage value with a preset threshold. When the voltage exceeds the limit, the regulating device is triggered to perform reactive power compensation or active power reduction. This type of method has certain practicality in scenarios with small load fluctuations and stable power output, but in actual distribution networks with large-scale grid connection of distributed energy and complex load structure changes, there are obvious problems of response lag and false triggering. Most current mainstream technologies rely on passive judgment based on current sampled values, failing to consider the dynamic evolution trends of power sources and loads, and lacking detailed differentiation of load types and quantitative modeling of operating condition sensitivity. For example, while some technologies shorten data latency by increasing sampling frequency, they still struggle to predict future voltage exceedance risks in advance. Furthermore, commonly used bus voltage regulation relies on statically or empirically set exceedance thresholds, making it impossible to flexibly adjust threshold boundaries according to the real-time operating environment. Some publicly available technologies attempt to introduce voltage change rate analysis, but have not formed a systematic predictive model linking active / reactive power output with voltage response trajectories, resulting in voltage exceedance control actions lagging behind the actual occurrence of risks. Currently known intelligent voltage regulation devices for distribution networks are generally suitable for urban public networks with relatively slow load fluctuations, but not for complex distribution scenarios with high proportions of distributed renewable energy integration and significant load structure fluctuations. Existing technologies lack precise classification and identification of constant power loads, constant current loads, and motor-type loads, and lack dynamic load sensitivity identification and parameter adaptation capabilities. This makes voltage prediction models susceptible to accuracy degradation due to network topology changes or environmental disturbances, thus affecting the timeliness and accuracy of control actions. Although some intelligent regulation devices have online parameter identification functions, they mostly adopt a single model structure, which cannot achieve advanced trend perception in complex mixed load scenarios, resulting in delays in identifying voltage limit exceedance risks. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, the present invention provides an active method for suppressing voltage over-limit of photovoltaic-storage systems in power distribution networks.

[0004] The technical solution of this invention is implemented as follows: A method for actively suppressing voltage exceedance in a photovoltaic-storage system in a power distribution network, comprising: S1: Acquire real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point, as well as the operating current and power factor of surrounding loads, and simultaneously collect meteorological information and historical electricity consumption behavior data as the basic input for dynamic load identification and voltage trend prediction. S2: Based on the real-time operating data and historical electricity consumption behavior, classify and identify the current load type, distinguish typical load characteristics such as constant power type, constant current type and motor type, and generate corresponding load feature labels for subsequent electrical parameter modeling and voltage response characteristic analysis. S3: Based on the load characteristic labels and the real-time collected voltage and current sequences, the equivalent feeder impedance and load sensitivity coefficient are identified online using a sliding window recursive algorithm, and a time-varying electrical parameter model reflecting the dynamic characteristics of the local network under the current operating conditions is constructed. S4: Combine the time-varying electrical parameter model with the current active / reactive power injection amount and input it into the voltage amplitude prediction model based on sensitivity coupling relationship to generate a voltage response trajectory prediction sequence for the next 1 to 3 time steps, which serves as the basis for over-limit risk assessment. S5: Based on the voltage response trajectory prediction sequence, a fuzzy membership function is used to classify its trend state, and three trend state categories are output: 'stable', 'rising and approaching the limit', or 'rapid rise', forming a quantifiable voltage change trend discrimination result; S6: Dynamically adjust the upper and lower boundaries of the voltage over-limit judgment threshold according to the trend discrimination result. When the predicted voltage enters the 'rising and approaching over-limit' region and the duration exceeds the set lag period, it is determined that the control trigger condition is met and an initial control trigger signal is generated. S7: After generating the control trigger signal, start the reverse confirmation logic to continuously monitor the voltage change trend of subsequent sampling points. If the voltage trend is detected to fall back and leave the critical region, cancel the issued control trigger signal to avoid unnecessary adjustment actions. S8: If the control trigger signal is not cancelled, the final confirmed control command is output to the power regulation unit to start reactive power support or active power reduction measures, thereby actively suppressing the risk of voltage exceeding the limit in the distribution network and completing a closed-loop control process.

[0005] The present invention provides an active method for suppressing voltage exceedance in a photovoltaic-storage system in a power distribution network, which has the following beneficial effects: (1) This invention significantly improves the foresight and accuracy of voltage control by introducing a voltage over-limit prediction mechanism based on dynamic load identification. Unlike fixed thresholds, which are prone to causing control signal lag or false triggering, this invention combines multi-source real-time data with load type identification technology to distinguish the electrical response characteristics of typical loads such as constant power, constant current, and motors online, and integrates meteorological information and historical electricity consumption behavior to enhance the ability to perceive operating conditions. On this basis, the equivalent feeder impedance and load sensitivity coefficient are updated recursively using a sliding window to realize time-varying modeling of local network electrical parameters, effectively overcoming the model mismatch problem caused by frequent adjustments to the distribution network topology. This dynamic modeling method enables voltage amplitude prediction to truly reflect the system response law under the current operating state. Compared with the traditional fixed sensitivity model, it greatly improves the correlation and robustness of short-term voltage trend prediction, providing a reliable basis for subsequent control decisions. (2) This invention constructs a voltage response trajectory prediction model based on time-varying sensitivity coupled with active / reactive injection, and innovatively uses a fuzzy membership function to classify the trend of the predicted sequence, dividing the voltage evolution process into three states: "stable", "rising and approaching the limit", and "rapid rise". Based on this, the upper and lower boundaries of the judgment threshold are dynamically adjusted, realizing the transformation from rigid criteria to flexible range. When the predicted voltage enters the "rising and approaching the limit" region and continues to exceed the set lag period, the system immediately generates a control trigger signal to start reactive power support or active power reduction actions, thereby completing active intervention before the limit is exceeded. At the same time, reverse confirmation logic is introduced. If subsequent sampling shows that the voltage trend falls back and leaves the critical range, the unexecuted control command is automatically canceled to avoid unnecessary equipment actions caused by instantaneous disturbances, significantly reducing control frequency and equipment wear. This mechanism ensures safety while taking into account economy and stability, effectively solving the technical contradiction of difficulty in balancing response speed and judgment accuracy in traditional methods. Attached Figure Description Figure 1 This is a flowchart of an active voltage over-limit suppression method for a photovoltaic-storage system in a power distribution network according to the present invention; Figure 2 This is a sub-flowchart of a method for actively suppressing voltage exceedance in a photovoltaic-storage system in a power distribution network according to the present invention; Figure 3 This is another sub-flowchart of the active voltage over-limit suppression method for a photovoltaic-storage system in a power distribution network according to the present invention. Detailed Implementation

[0006] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0007] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. like Figure 1 As shown, this invention provides an active method for suppressing voltage exceedances in a photovoltaic-storage system within a power distribution network, specifically including: S1: Acquire real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point, as well as the operating current and power factor of surrounding loads, and simultaneously collect meteorological information and historical electricity consumption behavior data as the basic input for dynamic load identification and voltage trend prediction. S2: Based on the real-time operating data and historical electricity consumption behavior, classify and identify the current load type, distinguish typical load characteristics such as constant power type, constant current type and motor type, and generate corresponding load feature labels for subsequent electrical parameter modeling and voltage response characteristic analysis. S3: Based on the load characteristic labels and the real-time collected voltage and current sequences, the equivalent feeder impedance and load sensitivity coefficient are identified online using a sliding window recursive algorithm, and a time-varying electrical parameter model reflecting the dynamic characteristics of the local network under the current operating conditions is constructed. S4: Combine the time-varying electrical parameter model with the current active / reactive power injection amount and input it into the voltage amplitude prediction model based on sensitivity coupling relationship to generate a voltage response trajectory prediction sequence for the next 1 to 3 time steps, which serves as the basis for over-limit risk assessment. S5: Based on the voltage response trajectory prediction sequence, a fuzzy membership function is used to classify its trend state, and three trend state categories are output: 'stable', 'rising and approaching the limit', or 'rapid rise', forming a quantifiable voltage change trend discrimination result; S6: Dynamically adjust the upper and lower boundaries of the voltage over-limit judgment threshold according to the trend discrimination result. When the predicted voltage enters the 'rising and approaching over-limit' region and the duration exceeds the set lag period, it is determined that the control trigger condition is met and an initial control trigger signal is generated. S7: After generating the control trigger signal, start the reverse confirmation logic to continuously monitor the voltage change trend of subsequent sampling points. If the voltage trend is detected to fall back and leave the critical region, cancel the issued control trigger signal to avoid unnecessary adjustment actions. S8: If the control trigger signal is not cancelled, the final confirmed control command is output to the power regulation unit to start reactive power support or active power reduction measures, thereby actively suppressing the risk of voltage exceeding the limit in the distribution network and completing a closed-loop control process.

[0008] Step S1: Acquire real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point, as well as the operating current and power factor of surrounding loads, and simultaneously collect meteorological information and historical electricity consumption behavior data as the basic input for dynamic load identification and voltage trend prediction. Specifically, this includes: S1.1: High-frequency sampling processing is performed on the voltage, current, active power and reactive power at the grid connection point of the photovoltaic-storage system to obtain a continuous time-series of real-time electrical parameters, which serves as the basic input data reflecting the current grid operation status; For the three-phase voltage, three-phase current, active power and reactive power signals at the grid connection point of the photovoltaic-storage system, a multi-channel synchronous sampling method is adopted (parameters: sampling frequency of 10-20kHz, preferably 12kHz, sampling resolution of not less than 16bit) to achieve high-precision parallel acquisition of the bus electrical quantities, so as to ensure that each quantity has strict timing consistency under the same time reference. Furthermore, the phase-locked loop synchronization algorithm (parameters: phase-locked bandwidth 5Hz, harmonic suppression filter order 4) is used to achieve phase synchronization locking of the AC signal at the grid connection point and obtain the phase reference data of the three-phase voltage and current, providing a unified phase reference for subsequent power calculation and dynamic characteristic analysis; Furthermore, the Fast Fourier Transform (FFT) algorithm (parameters: Hanning window with a length of 2048 points) is used to decompose the spectral components of the sampled sequence, remove stray high-frequency noise other than power frequency, and generate a pure power frequency fundamental component sequence to improve the stability of power calculation. Furthermore, based on the synchronously acquired effective values ​​of voltage and current, the instantaneous active power is calculated using the instantaneous power theory formula. With instantaneous reactive power The instantaneous power of each phase is obtained through Analogous calculation, where Instantaneous voltage The instantaneous current is used as the average value across the three phases to obtain the total system power, thus forming continuous time-sequential power data. Furthermore, by using a sliding window statistical method (window length 50ms, step size 10ms), short-time mean and variance sequences of voltage, current and power are generated to characterize the dynamic fluctuation characteristics of electrical quantities at a microscale. Through the above-mentioned high-frequency synchronous sampling, phase locking, spectrum cleaning, instantaneous power calculation and sliding window feature extraction processing methods, the original grid connection point sampling signal of the previous step is transformed into a continuous time-series standardized real-time electrical parameter sequence with noise suppression, so as to realize accurate and stable perception of the current grid operation status. For example, a three-channel voltage sensor and a three-channel current transformer are deployed at a 10kV distribution network connection point, with sampling parameters set to a frequency of 12kHz and a resolution of 16bit. The phase-locked loop bandwidth is set to 5Hz, and a fourth-order IIR low-pass filter is used to suppress harmonics. In the FFT calculation, a Hanning window of 2048 points is used to filter out frequency components other than 50Hz±1Hz. In the instantaneous power calculation, the three-phase instantaneous power is calculated according to the formula... , , Calculate separately, and apply the formula to the three-phase results. The average instantaneous active power of the system is obtained. The sliding window statistical results show that the average voltage is stable at around 9980V, the variance is below the required range, and the power fluctuation amplitude is significantly reduced, achieving a significant improvement in the stability and accuracy of real-time data. S1.2: Based on the real-time current and power factor measurements at the grid connection point, perform data acquisition on the operating status of surrounding loads, and generate an associated input set containing the power characteristics of the load side to support subsequent load type classification and feature modeling; Based on the real-time current and power factor measurements at the grid connection point, a three-phase synchronous sampling method (parameters: sampling frequency ≥ 1 kHz, sampling accuracy ≥ 0.2%) is used to achieve full acquisition of the current signals of the surrounding load branches. A phase synchronous calibration algorithm is used to perform time reference correction on the three-phase current waveform to eliminate the phase deviation caused by sampling delay. Furthermore, the instantaneous active power of each load branch is calculated using an instantaneous power calculation method (parameter: based on the product of the synchronous sampling current I(t) and the grid connection point voltage U(t)). With instantaneous reactive power The decomposition calculation is performed to obtain the power time series dataset for each load node; Furthermore, by using a power factor calculation method (parameter: using the arccos function to calculate the phase difference φ of the voltage and current waveforms), the power factor PF value of each load branch is estimated in real time, and a load-side power characteristic description vector containing characteristic parameters such as voltage offset, current amplitude, and phase difference is generated. Furthermore, a sliding time window statistical analysis method (window length = 1 min, step size = 10 s) is adopted to dynamically update the statistical quantities such as the mean, variance and peak value of the power characteristic description vector, and to obtain a statistical feature set that can reflect the stability and volatility of load operation. Furthermore, through feature normalization (parameter: Z-score standardization), features such as voltage offset, phase difference, and power fluctuation rate with different dimensions are transformed into dimensionless values ​​of a uniform scale. The features of each branch are bound into an associated input set through the primary key index (node ​​number + timestamp), forming structured data that can be directly connected to the load type identification model. Through the above sliding statistical analysis and normalization process, the real-time current and power factor measurement results of the previous step are transformed into multi-dimensional power characteristic parameters covering each surrounding load node, so as to achieve the expected technical effect of providing a continuous, stable and quantifiable set of related inputs for the subsequent load classification and feature modeling modules. For example, in a 10kV distribution network operation scenario, the grid connection point has four main load branches, two of which are constant power production equipment, one is a constant current lighting load, and one is a mixed load of motors. The sampling frequency is configured to 2048Hz, and the three-phase sampling accuracy is controlled within 0.1%. The three-phase current waveforms of each load are obtained through synchronous sampling. and the corresponding bus voltage waveform Instantaneous active power is calculated using instantaneous power decomposition. With instantaneous reactive power Where φ is the phase difference between voltage and current. A sliding time window of 60 seconds with a step size of 10 seconds is used to calculate the mean power factor and fluctuation rate of each branch. For example, the mean power factor (PF) of a constant power branch is... Volatility is lower than The average value of power (PF) in a certain motor branch is Volatility It also exhibits periodic peaks, a feature directly identified as a starting-impact load characteristic. After normalization, a correlation input set containing 12 parameters, including voltage offset (unit: V), phase difference (unit: rad), and volatility (dimensionless), is generated. Model verification shows a significant improvement in classification accuracy and an increase in trend prediction lead time of approximately 3 sampling periods, meeting the high-precision input requirements for dynamic load identification and voltage trend prediction. S1.3: Use meteorological monitoring terminals to synchronously acquire data on external environmental variables such as light intensity, ambient temperature and wind speed, and combine them with the distributed power generation output characteristic model to calculate the impact factor on photovoltaic output fluctuations, forming an input for environmental disturbance compensation. S1.4: Extract historical electricity consumption behavior data for the same period from the historical database, including typical daily load curves and electricity consumption pattern clustering labels. Based on the timestamp alignment strategy, perform spatiotemporal matching processing with the current real-time data to generate a composite input vector with behavioral prior information. S1.5: The above real-time electrical parameter sequence, load-side power characteristics, environmental disturbance compensation terms, and behavioral prior information are fused from multiple sources to generate a structured input matrix in a unified format, which serves as the joint input basis for the dynamic load identification module and the voltage amplitude prediction model.

[0009] Step S2: Based on the real-time operating data and historical electricity consumption behavior, the current load type is classified and identified, distinguishing typical load characteristics such as constant power, constant current, and motor types, and generating corresponding load feature labels for subsequent electrical parameter modeling and voltage response characteristic analysis. Specifically, this includes: S2.1: Based on the real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point obtained in S1, as well as the operating current and power factor of the surrounding loads, and combined with the synchronously collected meteorological information and historical electricity consumption behavior sequences, a multi-dimensional load feature vector is constructed, which includes instantaneous power fluctuation rate, voltage-current phase difference change trend, daily cycle similarity index, and temperature correlation coefficient, as the original input feature set for load type identification; Based on the real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point output by S1, as well as the operating current and power factor of the surrounding loads, a feature construction algorithm (parameters: sampling period Δt=200ms, acquisition window length W=60s) is used to calculate the instantaneous power fluctuation rate. The fluctuation rate is calculated by normalizing the standard deviation of the active power sequence to the mean within the sliding window, and is used to quantify the sensitive fluctuation characteristics of the load-side power within the sampling period. Based on the corresponding voltage and current synchronization sequence at the grid connection point, a phase difference estimation method (parameters: lock-in amplifier digital algorithm, fundamental frequency 50Hz) is used to extract the trend of voltage-current phase difference variation. Furthermore, by fitting the rate of change of phase difference over time through linear regression, the influence coefficient characterizing the load type on power factor stability under the current operating conditions is obtained; Based on the historical electricity consumption data provided by S1.4, the daily cycle similarity analysis method (parameter: standardized cross-correlation coefficient) is used to calculate the similarity between the current load curve and the typical daily curve, and obtain the daily cycle similarity index, which is used to characterize the degree of matching between the current load waveform and the stable operation mode. Based on the ambient temperature data collected by S1.3 and the photovoltaic power output characteristic model, the temperature sensitivity coefficient extraction method (parameter: linear correlation coefficient R) is used to evaluate the correlation between ambient temperature and load active power, and obtain the temperature correlation coefficient to reflect the influence of meteorological conditions on the response characteristics of load type. Through the above feature construction, the instantaneous power fluctuation rate, voltage-current phase difference change trend, daily cycle similarity index and temperature correlation coefficient are combined into a multi-dimensional load feature vector, and formatted to meet the input requirements of the classification model, thereby generating the original feature set for load type identification. For example, in a distribution network operation scenario with a high proportion of photovoltaic access, the average active power collected at the grid connection point within a 60-second window is 320 kW, with a standard deviation of 28 kW. The instantaneous power fluctuation rate is calculated as follows: =0.0875. The voltage-current phase difference slowly decreased from 0.95 radians to 0.92 radians. The rate of change of phase difference was obtained using linear regression. The current load curve has a correlation coefficient of 0.91 with the historical typical daily curve, indicating a high similarity. The ambient temperature is 32℃, and the temperature correlation coefficient is calculated to be 0.62, showing that temperature has a significant impact on the load. The original input feature vector is generated through feature combination. The input is then processed through a subsequent normalization and dimensionality reduction chain, ultimately achieving the availability and high information content of the feature set, laying the foundation for accurate classification of load types. S2.2: Perform normalization and outlier filtering on the multidimensional load feature vector, update the feature statistical distribution parameters using a sliding time window mechanism, and use principal component analysis algorithm to extract the first three principal component components to compress redundant information and retain more than 95% of the cumulative variance contribution rate, generating a dimensionality-reduced standardized feature matrix as a reliable input to the classification model. S2.3: Based on the standardized feature matrix, the pre-set lightweight ensemble classifier model is invoked to perform load type discrimination. This model is composed of random forest and support vector machine. During the training process, constant power type, constant current type and motor type load labeling data from the typical load sample library are introduced, and the probability distribution of each type is output to realize the preliminary classification decision of the current load characteristics. S2.4: Based on the classification probability distribution results, set a dynamic decision threshold: when the maximum category probability is greater than 0.85, directly determine the load feature label; if it is in the range of 0.6 to 0.85, combine the classification results of historical continuous periods to perform sliding consistency verification, avoid misjudgment due to transient disturbances, and finally generate a load feature label sequence with time stability. S2.5: Output the confirmed load characteristic tags to the next module and establish a mapping table between the tags and typical voltage response modes. The constant power type corresponds to strong reactive power sensitivity, the constant current type exhibits linear voltage drop characteristics, and the motor type reflects starting impact characteristics. This provides prior knowledge support for online identification of equivalent feeder impedance and load sensitivity coefficient in S3.

[0010] like Figure 2As shown, step S3 involves: based on the load characteristic labels and the real-time acquired voltage and current sequences, using a sliding window recursive algorithm to identify the equivalent feeder impedance and load sensitivity coefficient online, and constructing a time-varying electrical parameter model reflecting the dynamic characteristics of the local network under the current operating conditions. Specifically, this includes: S3.1: Based on the load feature labels generated in the previous steps and the real-time collected grid connection point voltage and current sequences, construct a multivariate sampling matrix within the sliding time window, and perform synchronous alignment processing on the three-phase voltage RMS value, three-phase current RMS value and active / reactive power at each time step to form a time series dataset for parameter identification. Based on the load characteristic tags confirmed in the previous steps and the real-time collected three-phase voltage and three-phase current sequences of the photovoltaic-storage system grid connection point, a multivariate sampling matrix is ​​constructed using a sliding time window to achieve unified modeling input of time-series parameters. A data synchronization alignment algorithm (parameters: sampling window length ΔT, timestamp accuracy 1 ms) is adopted to achieve synchronous matching of the effective values ​​of three-phase voltage and three-phase current at each time step, ensuring that the source, grid and load data maintain phase consistency within the same sampling period; Furthermore, the aligned voltage data is processed using the power calculation formula. Current data Execute instantaneous active power and instantaneous inefficiency The calculation, in which The active power factor, To obtain the active / reactive power value at each time step, the reactive power factor is used as the active / reactive power factor, and this value is used as a supplementary column vector of the sampling matrix. Furthermore, a sliding window update mechanism (parameters: window length N=30, update step size M=1) is adopted to recursively expand the sampling matrix, realize the dynamic update of the matrix column vector, and retain the latest N time steps of multivariate data for identification of equivalent feeder impedance and sensitivity coefficient; Furthermore, a missing value interpolation algorithm (parameter: linear interpolation mode) is used to perform data integrity repair on the sampling matrix within the window, avoiding inconsistencies in matrix dimensions caused by missed sensor sampling, thus ensuring the stability of the subsequent modeling process; By performing matrix normalization (parameters: zero mean, unit variance), the dimensional differences of each physical quantity are eliminated, ensuring the numerical stability of subsequent recursive estimations and achieving the output of a time-series dataset with unified dimensions. The above method transforms the load feature labels and multi-source real-time electrical parameters from the previous step into a structured time-series dataset that can be used for parameter identification, thereby achieving the expected technical effect of data consistency and integrity. For example, in a distribution network scenario with a high proportion of photovoltaic (PV) integration, the sampling frequency of the three-phase voltage at the grid connection point of the PV-storage system is set to 1 kHz, and the sampling frequency of the three-phase current is also set to 1 kHz. The sliding time window length ΔT is set to 300 ms, corresponding to N=300 sampling points. Using the above alignment algorithm, the voltage and current values ​​of each phase are synchronously matched with a timestamp accuracy of 1 ms to calculate the instantaneous active power and reactive power. The instantaneous sampling point... =230 V, =15 A, power factor =0.95, then the instantaneous active power Through formula The calculated result is 3.2775 kW. Similarly, if =0.312, then reactive power through The result is 1.0776 kvar. The active / reactive power values ​​of each of the three phases are assembled into a matrix column vector and updated recursively through a sliding window to form a multiphase parameter matrix containing the past 300 ms. After matrix normalization, this matrix is ​​stably input into the subsequent equivalent impedance and load sensitivity identification stages. The output time-series dataset showed no frame drops or numerical abrupt changes within 15 minutes of continuous operation, significantly improving the stability and accuracy of parameter identification. S3.2: Based on the time series dataset and the load type indicated by the load feature label, select the corresponding electrical response model structure: for constant power loads, adopt the PQ node modeling method; for constant current loads, adopt the current amplitude normalization modeling method; and for motor loads, introduce the transient reactance correction factor to establish a local network equivalent circuit topology that adapts to different types of loads. S3.3: The least squares recursive algorithm (RLS) is used to fit the voltage deviation and current injection in multiple consecutive sampling periods within the sliding window, calculate the linear mapping relationship between the voltage change and the injected current between the beginning and end of the feeder, and obtain the dynamically updated equivalent feeder impedance parameter sequence, which is used as the core component of the time-varying electrical parameter model. Based on the sliding time window time series dataset generated by S3.1 and S3.2 and the selected electrical response model structure, the least squares recursive algorithm (RLS) is used (parameter: the forgetting factor λ is determined according to the load type change rate, and the initial covariance matrix P is set to a multiple of the identity matrix to ensure parameter identifiability) to achieve linear fitting estimation of voltage deviation and injected current within the sliding window; Furthermore, through the incremental update mechanism of the RLS algorithm, a linear error model is constructed in each new sampling period using the parameter vector estimated in the previous step and the current sample data, the gain vector K(t) is calculated, and this gain is used to adjust the impedance parameter vector. Online corrections are performed to ensure the stability of parameter convergence under load power jumps and distributed power output disturbances. Furthermore, the equivalent impedance parameter of the feeder is calculated using the following formula. : in This is a vector representation of the voltage changes at the beginning and end of the feeder within adjacent time steps within a sliding window. This is in vector form of the corresponding injection current change, ensuring that the data has undergone filtering preprocessing to suppress high-frequency noise; Furthermore, a time-series-based set of impedance parameters is formed through real-time updates of RLS parameter estimation. It is stored in the time-varying electrical parameter cache structure for use in the load sensitivity coefficient calculation in S3.4; Furthermore, by performing outlier detection and linear interpolation correction on the impedance parameter set, a smooth parameter sequence after removing outlier samples is output to ensure the stability and physical rationality of the model input; Through the RLS online algorithm described above, the sliding window fitting results are transformed into a highly time-efficient feeder equivalent impedance parameter sequence, enabling real-time plotting and modeling of the dynamic characteristics of the local network. For example, in a 10kV distribution branch under high-proportion photovoltaic access conditions, the sliding window length is configured as 20 sampling periods with a period length of 50ms. The forgetting factor λ is set to 0.98 to adapt to the rapid load fluctuation characteristics, and the initial covariance matrix P is set to 100 times the identity matrix to enhance the convergence speed in the early stage of disturbance. During operation, the measured voltage change ranges from 0.5V to 3V, and the current change ranges from 0.01A to 0.15A. These values ​​are substituted into the RLS incremental update formula to calculate the gain vector K(t) and update the impedance parameters online. The obtained equivalent feeder impedance smoothly adjusted from 1.2Ω to 1.35Ω under changes in photovoltaic output, with the parameter fluctuation range controlled within ±0.02Ω. After outlier removal and interpolation correction, the output impedance parameter sequence maintained continuity and physical rationality under different operating conditions, providing stable topology support for subsequent load sensitivity calculation and voltage prediction. S3.4: Based on the equivalent feeder impedance parameter sequence and the reactive power response characteristics of the current load type, calculate the voltage amplitude change rate caused by the unit reactive power change, and generate the load sensitivity coefficient; this coefficient characterizes the degree of influence of a specific load on voltage fluctuation at the current operating point, and constitutes the key variable of voltage sensitivity coupling relationship; S3.5: Bind the equivalent feeder impedance and load sensitivity coefficient according to the timestamp, combine with the load feature tags to form a time-varying electrical parameter model database with spatiotemporal correlation, and eliminate parameter jump noise through low-pass filtering to output a standardized parameter vector that can be used for voltage amplitude prediction model calling, thus completing the modeling closed loop of local network dynamic characteristics.

[0011] like Figure 3 As shown, step S4 involves combining the time-varying electrical parameter model with the current active / reactive power injection, inputting it into a voltage amplitude prediction model based on sensitivity coupling, and generating a voltage response trajectory prediction sequence for the next 1 to 3 time steps, which serves as the basis for exceeding limit risk assessment. Specifically, this includes: S4.1: Based on the time-varying electrical parameter model output by S3, extract the equivalent feeder impedance and load sensitivity coefficient under the current operating conditions, and use them as input parameters for network topology and load characteristics for voltage amplitude prediction, so as to reflect the actual electrical characteristics of the local distribution network under dynamic operating conditions. S4.2: Time-align the real-time active and reactive power data of the photovoltaic-storage system grid connection point collected by S1, and match them with the equivalent feeder impedance and load sensitivity coefficient obtained by S4.1 to form a joint input vector containing the source-grid-load coupling relationship, which is used to model the driving voltage response characteristics. S4.3: Using a voltage amplitude prediction model established based on nodal voltage sensitivity theory, calculate the weight matrix of the influence of active / reactive power changes on the bus voltage amplitude, where the sensitivity coefficient is defined as... U / P and U / Q is expressed by the first-order Taylor series approximation of the voltage change, forming a sensitivity coupling relationship mapping function; S4.4: Substitute the joint input vector generated in S4.2 into the sensitivity coupling relationship mapping function constructed in S4.3, perform dynamic voltage response calculation, obtain the voltage impact increment of the current active / reactive injection on the next 1 to 3 control cycles, and superimpose it on the current measured voltage value to generate a multi-step voltage response trajectory prediction sequence. Based on the joint input vector generated in S4.2 and the sensitivity coupling mapping function constructed in S4.3, matrix operations are used (parameter: equivalent feeder impedance matrix). Load sensitivity coefficient vector The calculation of multi-step voltage response increments is achieved by using the current active power change ΔP and reactive power change ΔQ. Furthermore, a recursive prediction algorithm is used (parameters: prediction step size k = 1~3, sensitivity weight matrix). This allows for the estimation of the dynamic voltage response under source-grid-load coupling conditions at each prediction step size, and the obtaining of the incremental sequence ΔU(k) ​​of the voltage amplitude over the next k cycles. Furthermore, vector superposition operation is employed (parameter: measured voltage at the current moment). The voltage trajectory is gradually constructed by predicting the increment ΔU(k), and a multi-step prediction vector containing the voltage amplitude sequence for the next 1 to 3 control cycles is generated. ; Furthermore, through boundary constraint processing (parameter: rated voltage) Tolerable deviation range This restricts the predicted voltage sequence to the physically feasible region, eliminates the abnormally large offset caused by mathematical extrapolation, and obtains the constrained and corrected voltage response trajectory. The following formula is used to calculate the multi-step voltage amplitude prediction: in, The voltage amplitude at the k-th prediction step. This is the current measured voltage value. This is the sensitivity weight matrix. and These represent the changes in active and reactive power, respectively. The algorithm described above transforms the joint input vector and sensitivity function result from the previous step into future multi-cycle voltage prediction data, thereby achieving the expected technical effect of proactive over-limit risk identification. For example, in a distribution network scenario with a high proportion of photovoltaic power, let the current measured voltage at the grid connection point be... =10.5kV, equivalent feeder impedance matrix =[0.15+j0.35]Ω, load sensitivity coefficient vector =[0.012,0.025], active power change ΔP=0.85MW, reactive power change ΔQ=0.42MVar, prediction step size k=1,2,3, sensitivity weight matrix Calculated from the node voltage sensitivity model =[[0.008,0.003],[0.009,0.0025],[0.0105,0.002]]. Substituting the above parameters into the formula, we calculate ΔU(1)=0.00936kV, ΔU(2)=0.01025kV, and ΔU(3)=0.01194kV. After vector superposition, the predicted voltage values ​​for the next three cycles are 10.50936kV, 10.51025kV, and 10.51194kV, respectively. After constraint correction, all predicted points are within the allowable range of ±0.5kV of the rated voltage, and the corresponding voltage response trajectory is classified as "rising and approaching the limit" in the trend analysis stage. In this embodiment, the prediction accuracy is significantly improved, and the system can identify potential limit risks two control cycles in advance, thus possessing proactive prevention and control capabilities. S4.5: Normalize and verify the rationality of the voltage response trajectory prediction sequence output by S4.4, remove abnormal prediction points that exceed the physical feasible region, and encapsulate it into a structured data format as an effective input for the next step of trend state division and limit risk assessment.

[0012] Step S5: Based on the voltage response trajectory prediction sequence, a fuzzy membership function is used to classify its trend state, outputting three trend state categories: 'stable', 'rising towards the limit', or 'rapid rise', forming a quantifiable voltage change trend discrimination result. Specifically, this includes: S5.1: Obtain the voltage response trajectory prediction sequence for the next 1 to 3 time steps generated by S4 as input data; construct the voltage change rate vector within the sliding window based on the prediction sequence, and use the numerical differentiation method to calculate the voltage difference slope between adjacent time points in order to extract the voltage dynamic evolution characteristics and obtain the 'voltage change rate feature quantity' that characterizes the system voltage change rate. S5.2: Based on the voltage change rate feature and the current predicted voltage amplitude, construct a two-dimensional input feature space, where the horizontal axis is the predicted voltage amplitude and the vertical axis is the voltage change rate feature; according to the operating experience of the distribution network, set the initial distribution intervals for three typical trends, namely 'stable' (low change rate, voltage within the normal range), 'rising and approaching the limit' (medium positive change rate, voltage close to the upper limit threshold), and 'rapid rise' (high positive change rate, voltage approaching or exceeding the upper limit), forming the prior knowledge basis for fuzzy classification; Based on the voltage change rate characteristic obtained from S5.1 and the current predicted voltage amplitude, a feature space construction method is adopted (parameter: change rate characteristic). Predicted voltage This enables the mapping of trend inputs with different dimensions to a unified two-dimensional feature representation domain; Furthermore, through feature normalization processing method (parameter: maximum value) Minimum value Upper limit of rate of change This achieves dimensionless elimination of data from all dimensions and yields normalized predicted voltage amplitudes. With normalized rate of change characteristic Two sets of results; Furthermore, through a two-dimensional Cartesian coordinate mapping method (parameters: , ),Will As the horizontal axis, As the vertical axis, a two-dimensional input feature space matrix containing the coordinates of each prediction time point is generated. This is used to support subsequent fuzzy set modeling and trend state classification; Furthermore, based on the power distribution network operation experience parameter setting method (parameter: rated voltage) Voltage upper limit Trend change rate threshold This method delineates the initial distribution intervals for three typical trends in the feature space: a stable region when the voltage amplitude is within a normal range and the rate of change is low; a rising and approaching-limit region when the voltage is close to the upper limit threshold and the rate of change is moderate; and a rapidly rising region when the voltage is close to or exceeds the upper limit and the rate of change is high. The resulting interval boundary dataset is then obtained. ; Through empirical distribution parameters and The coordinate point comparison process maps the normalization result of the previous step to the initial position of the trend category, thereby realizing the construction of the prior knowledge base for fuzzy classification. For example, at a certain grid connection point, the rated voltage is set to 10.5kV, the upper limit threshold is set to 10.8kV, the maximum predicted voltage in the predicted sequence is 10.85kV, the minimum predicted voltage is 10.4kV, and the maximum rate of change is 0.15kV / cycle. The predicted voltage amplitude is then normalized. Where U is the current predicted voltage amplitude. The rate of change is normalized: Where ΔV is the current predicted rate of change of voltage. As the horizontal axis Using the vertical axis as the ordinate, a set of prediction points is formed in a two-dimensional coordinate system. According to empirical definition, the range of the 'stationary' region is set to... ∈[0,0.8] and ≤0.2; The range of the 'rising trend approaching the limit' zone is set to ∈(0.8,0.95] and ∈(0.2,0.5]; the 'rapid rise' region is set to >0.95 and >0.5. In this scenario, the coordinate point (0.92, 0.35) will be classified into the 'rising trend approaching the limit' zone, providing the initialization input for the subsequent S5.3 fuzzy membership function parameters, and ultimately achieving a stable mapping of the trend state prior knowledge; S5.3: For the aforementioned two-dimensional input feature space, define three corresponding fuzzy sets: , , Trapezoidal or Gaussian membership functions were designed to mathematically model each trend state. Based on historical measured data and simulation samples, the parameters of this set of membership functions were calibrated offline to ensure that they can accurately reflect the voltage response characteristics under different load and power supply conditions, and output a standardized membership function model. Based on the two-dimensional input feature space constructed by S5.2, the membership function model of three types of trend states is established by using fuzzy set theory to ensure the analyzability of the mathematical expression and the physical correspondence of different trend states. For the 'stable' trend state, a trapezoidal membership function (parameters: starting point a, plateau interval [b,c], ending point d) is used to achieve fuzzy representation of low voltage change rate and amplitude within the normal range. By optimizing the sensitivity of the plateau interval length, the model has higher recognition robustness for slow change processes. For the trend of 'rising and approaching the limit', a Gaussian membership function (parameters: mean m, upper and lower limits σ) is used to achieve a centralized representation of the voltage amplitude with a moderate positive rate of change and close to the upper limit threshold. The distribution width is controlled by adjusting the σ parameter to accurately reflect the fluctuation range of this trend under different load conditions. For the 'rapid rise' trend, a trapezoidal and Gaussian mixture membership function is used (parameters: mixture ratio λ, Gaussian mean). Standard deviation Trapezoid starting point and the end point The system models the critical state with high change rate and voltage amplitude approaching or exceeding the upper limit by weighted combination, and introduces an overshoot suppression factor in the process of mixing ratio λ tuning to avoid misjudging instantaneous spike signals. Using historical measured data and simulation samples based on node voltage sensitivity, the above membership function parameters are calibrated offline using the minimum mean square error (MSE) criterion. The specific calculation formula is as follows: in, This is the actual output value of the membership function. The ideal label value for the corresponding trend state. The total number of samples is denoted as MSE. By adjusting the parameter set of each membership function, the MSE is minimized, resulting in a standardized membership function model that maintains high prediction accuracy under different load and power output conditions. Through offline calibration and optimization of the parameters of the trapezoidal, Gaussian and mixed membership functions, the two-dimensional input feature space is mapped into a standardized membership function model of three quantifiable trend states, thereby improving the robustness and versatility of the trend state discrimination module. For example, in a distribution network scenario with a high proportion of photovoltaic access, the predicted voltage amplitude range on the horizontal axis of the input feature space is 1.00 pu~1.08 pu, and the voltage change rate range on the vertical axis is 0.0% / s~2.5% / s. The trapezoidal membership function parameters for the 'stable' state are configured as a=1.00, b=1.02, c=1.04, d=1.05; the mean m=1.06 and standard deviation σ=0.01 of the Gaussian membership function for the 'rising and approaching limit' state; and the Gaussian part of the 'rapid rise' mixed membership function. =1.075, =0.005, trapezoidal part =1.07, =1.08, mixing ratio λ=0.6. During offline calibration, the total number of samples N=5000, and the initial MSE was calculated to be 0.0048 using the minimum mean square error formula. After parameter iteration optimization, the MSE was significantly reduced to 0.0012. The optimized model demonstrates in real-time trend discrimination that it can sensitively respond to rapid voltage rises caused by load changes, while maintaining stable discrimination during slow load changes, avoiding false triggering of control signals, and achieving a significant improvement in prediction accuracy and operational stability. S5.4: Input the real-time updated predicted voltage amplitude and voltage change rate feature into the calibrated membership function model, execute the fuzzy inference process, calculate the membership value of the current voltage trend in the three fuzzy sets; based on the principle of maximum membership, select the category with the highest membership as the current trend state discrimination result, generate 'trend state label', and realize the conversion from continuous electrical quantity to discrete control semantics; S5.5: Perform time-series consistency filtering on the generated trend state labels, and use a sliding majority voting mechanism with a length of 3 to smooth the continuously output state label sequence to suppress instantaneous misjudgments caused by measurement noise or model fluctuations; output the final stable trend state category result, and use it as the logical basis for dynamically adjusting the over-limit judgment threshold in S6 to complete the closed-loop output of trend recognition.

[0013] Step S6: Dynamically adjust the upper and lower boundaries of the voltage over-limit judgment threshold based on the trend discrimination result. When the predicted voltage enters the 'rising and approaching over-limit' region and the duration exceeds the set lag period, it is determined that the control triggering condition is met, and an initial control triggering signal is generated. Specifically, this includes: S6.1: Based on the voltage change trend discrimination result output by S5, a dynamic threshold adjustment function is constructed. This function takes the fuzzy membership value of the 'rising and approaching limit' state as the input variable, and uses the piecewise linear mapping relationship to calculate the upper boundary offset of the voltage limit judgment threshold, generating a dynamic upper limit threshold that can be adaptively adjusted according to the trend intensity, so as to enhance the sensitivity to the gradual voltage rise process. S6.2: Combining the rated voltage level of the distribution network with the statistical characteristics of historical voltage fluctuations, determine the basic lower limit threshold, and based on the proportion of motor loads in the current load characteristic labels, perform negative compensation on the lower limit threshold through the sensitivity correction coefficient to calculate the dynamic lower boundary that reflects the reactive power demand of the load recovery, forming a bidirectional adjustable voltage judgment range. S6.3: Map the voltage response trajectory prediction sequence generated by S4 for the next 1 to 3 time steps one by one to the dynamic voltage determination interval constructed by S6.1 and S6.2, and determine whether there are two or more consecutive prediction points falling into the 'rising trend approaching the limit' region. If so, start the lag timer and record the duration of the trend state. Based on the voltage response trajectory prediction sequence for the next 1 to 3 time steps output by S4 and the dynamic voltage determination interval parameters generated by S6.1 and S6.2, an interval mapping algorithm is adopted (parameter: predicted voltage sequence). Dynamic upper limit threshold Dynamic lower threshold This enables the mapping of the determination interval for each prediction point; Furthermore, by using a judgment logic function (parameters: trend type = 'rising approaching limit', judgment interval mapping result set), the cross-matching of the predicted point trend state and the judgment interval is realized, generating a three-valued state matrix, where the state values ​​are defined as: 0 = normal interval, 1 = rising approaching limit interval, 2 = rapid rise interval, which are used for subsequent continuous point detection; Furthermore, using a sliding window continuity detection algorithm (parameters: window length L=3, state matrix sequence), the number of continuous prediction points falling into the 'rising approaching limit' region is statistically analyzed, and the continuity count value is obtained. The continuity criterion is calculated using the following formula: in As a continuous determination criterion, This represents the number of predicted points that consecutively fall into the target area. Furthermore, based on The logic and relationship between the judgment value and the trend status label are determined to generate a trend continuation trigger criterion. When the criterion value is true, the lag timer initialization function is called (parameter: start timestamp). =The control period corresponding to the current prediction point), start the trend duration recording logic, and bind the timer reference to the current trend detection session; Through the above interval mapping and continuous detection algorithm, the trend state information in the predicted sequence is transformed into trend duration timing data, realizing a closed loop for time dimension determination under the triggering condition. For example, the voltage response trajectory of a distribution network connection point predicted for the next three time steps is as follows: ={1.045pu, 1.052pu, 1.058pu}, dynamic upper limit threshold =1.050 pu, dynamic lower threshold =0.950 pu. An interval mapping algorithm is used. The first step compares each predicted value with the judgment interval to determine the interval classification: Pt1 = normal interval, Pt2 = rising trending towards exceeding the limit interval, Pt3 = rising trending towards exceeding the limit interval. The second step performs continuous sliding window detection with a window length of 3, counting the number of consecutive predicted points falling into the 'rising trending towards exceeding the limit' region. =2, calculate the continuity criterion according to the formula: determination If true, the third step is to start the lag timer and record the start timestamp. For the current control period T0, the trend duration will be compared with the preset minimum lag period in S6.4. In this scenario, the predicted voltage will remain in a critical state beyond the limit for two or more time steps. The recording effectiveness of the lag period timer significantly improves the accuracy of trigger condition determination and avoids malfunctions caused by instantaneous spikes. S6.4: Compare the time length recorded by the hysteresis timer with the preset minimum trigger hysteresis. When the duration is greater than or equal to the hysteresis, confirm that the voltage over-limit risk has consistent evolution, eliminate instantaneous disturbance interference, and determine that the control triggering condition is met. S6.5: After the triggering conditions described in S6.4 are met, an initial control trigger signal is generated and marked as pending confirmation. At the same time, auxiliary decision parameters such as predicted voltage peak value, expected over-limit time and recommended adjustment power level are attached as input basis for subsequent reverse confirmation logic and power adjustment module. Based on the voltage over-limit risk evolution consistency criterion confirmed in S6.4, the control signal generation module is invoked, and a parameterized data encapsulation method (parameters: trend status label, predicted voltage value, predicted timestamp) is adopted to realize the creation and structured encoding of the initial control trigger signal. Furthermore, the peak value of the predicted voltage is calculated by using a peak extraction algorithm (parameters: predicted voltage trajectory sequence, sensitivity coupling coefficient), and a numerical result representing the maximum voltage deviation is obtained as an auxiliary decision input. Furthermore, by using a time-based positioning algorithm (parameters: predicted trajectory slope sequence, boundary of the over-limit area), the expected over-limit moment is determined, and a timestamp index is generated to identify the control cycle position where the potential over-limit occurs; Furthermore, a power level recommendation function (parameters: sensitivity coupling matrix, predicted voltage limit) is used to calculate the required adjustment power level and generate corresponding active / reactive power adjustment recommendations, where the power level... The calculation formula is defined as follows: in, To predict voltage exceeding the limit, This refers to the voltage-power sensitivity coefficient for the corresponding type of load; By using a signal tagging management method, the initial control trigger signal is appended with a "pending confirmation" status identifier, and the predicted voltage peak, expected over-limit time, and recommended power level are encapsulated as a set of structured auxiliary parameters to achieve seamless data interface with the subsequent reverse confirmation logic and power regulation module. For example, in the real-time prediction of a distribution network connection point, the trend status label is "rising and approaching the limit," the maximum value of the predicted voltage trajectory sequence is 1.08 pu, the trajectory slope is 0.015 pu / cycle, and the sensitivity coupling coefficient is 32 V / Mvar. The peak voltage is directly calculated as 1.08 pu using a peak extraction algorithm, and the time positioning algorithm identifies the expected limit-over time as the 5th control cycle. In the power magnitude recommendation function... V, V / Mvar, substitute into the formula The recommended reactive power compensation level is 0.156 Mvar. The control signal generation module outputs an initial control trigger signal with a "pending confirmation" status, and passes the above-mentioned peak value, voltage over-limit prediction time, and regulation power level parameters to the reverse confirmation logic. In actual operation verification, the response speed when the signal triggers the power regulation action is significantly improved, and no unnecessary frequent control actions occur.

[0014] Step S7: After generating the control trigger signal, the reverse confirmation logic is activated to continuously monitor the voltage change trend of subsequent sampling points. If the voltage trend is detected to have fallen back and left the critical region, the issued control trigger signal is revoked to avoid unnecessary adjustment actions. Specifically, this includes: S7.1: Based on the initial control trigger signal, start the reverse confirmation timing window, set the continuous monitoring duration (typically 3 to 5 control cycles), and pause the output of the final control command to the power regulation unit during this period to reserve time for trend re-evaluation and prevent immediate malfunctions. S7.2: Perform moving average filtering on the continuous voltage sampling sequence after entering the reverse confirmation window to eliminate high-frequency noise interference and obtain a smoothed voltage trajectory data sequence as input conditions for trend backtracking analysis. S7.3: Based on the smoothed voltage trajectory data sequence, calculate the voltage change slope between the current time and the trigger time, and compare it with a preset trend fall threshold. If the voltage change slope is less than the trend fall threshold, it is determined that the voltage shows a significant downward trend, forming a trend fall criterion. S7.4: Perform a logical AND operation between the trend decline criterion and whether the voltage amplitude has exited the 'rising trend approaching the limit' region. When both conditions are met, generate a control signal cancellation flag and perform an active cancellation operation on the initial control trigger signal. S7.5: Feed back the control signal cancellation flag to the control logic center. If the control signal cancellation flag = 1, clear the status record of this control event and prohibit the issuance of the final control command in step S8, thus completing an avoidance process for an unnecessary action. Upon receiving the control signal generated by S7.4, the flag is cancelled. Then, an event-driven logic parsing method (parameters: flag status = 1, event type = cancellation request) is used to trigger the parsing of the current control logic center's state. Furthermore, by using the control flow state clearing method (parameters: event identifier, state record index), the entry of the current control event in the state record buffer is deleted, and an idle state storage slot is obtained to prevent old events from affecting subsequent decisions. Furthermore, by using a permission-blocking algorithm (parameters: permission identifier = prohibit, target step = S8), the control command transmission channel is blocked, and a prohibition signal is generated to block the process of the execution link sending the final control command to the power regulation module; Furthermore, the consistency of the control logic is verified through a state consistency verification method (parameters: current state = idle, number of historical cancellations, and validity of cancellation flag), and consistency verification success data is generated to update the system operation log. By combining the processing of cancellation flag feedback and state clearing, the cancellation judgment result of the previous step is transformed into the data state of control link blocking, thereby achieving the expected technical effect of avoiding unnecessary adjustment actions. For example, in a photovoltaic-storage system with a rated voltage of 10kV in a distribution network, the reverse confirmation module receives... Given the input conditions, the event-driven parsing method matches the revocation request type to the flag bit, extracts the state entry with the corresponding control event index number 20240115_07, and removes it from the event buffer by calling the state clearing method. The permission masking algorithm applies a prohibition flag to the S8 execution channel, sets the execution permission parameter of the target node to 0 in the control message scheduling queue, and blocks the message framing and distribution process. The state consistency verification method retrieves the revocation event count from the past 30 days, finding it to be 12, confirming that all revocations are valid, and updates the timestamp and corresponding event ID in the operation log. In this scenario, the system successfully avoids reactive power support actions caused by instantaneous load disturbances, reduces the number of unplanned operations of the power regulation unit, and maintains the stability of the voltage control link.

[0015] Step S8: If the control trigger signal is not revoked, a final confirmed control command is output to the power regulation unit to initiate reactive power support or active power reduction measures, thereby actively suppressing the risk of voltage exceeding limits in the distribution network and completing one closed-loop control process. Specifically, this includes: S8.1: Based on the voltage trend monitoring results output by the reverse confirmation logic, obtain the status flag of the current control trigger signal. The status flag includes three categories: 'pending confirmation', 'cancelled', and 'pending execution', which serve as input conditions for determining whether to initiate power regulation action. The status flag is parsed using a state machine model to clarify the decision stage of the current control command. S8.2: For control trigger signals in the 'waiting to be executed' state, call the preset control strategy mapping table, and calculate the required reactive power compensation or active power reduction ratio based on the degree of voltage deviation at the grid connection point and the predicted trajectory slope; wherein, the reactive power compensation is determined by the product relationship between the voltage-reactive power sensitivity coefficient and the predicted voltage limit, so as to achieve precise matching of voltage regulation requirements. S8.3: Convert the calculated reactive power compensation or active power reduction ratio into standardized control command parameters to generate an active / reactive power setpoint sequence suitable for the photovoltaic-storage converter (PCS) to receive; encapsulate the setpoint sequence based on the MODBUS / TCP communication protocol to form a control message frame that can be parsed by the power regulation unit; The reactive power compensation quantity ΔQ and the active power reduction ratio ρP output from step S8.2 are used as input control quantities, and the control quantity is standardized using a control quantity standardization algorithm (parameter: rated capacity). Rated voltage This enables the mapping of power values ​​to the unit system required by the photovoltaic-storage converter (PCS); Furthermore, through active / reactive power separation and limiting methods (parameter: maximum reactive power output) Maximum active power reduction ratio This allows for the interception of control quantities exceeding the equipment's capabilities, resulting in an active power setpoint within the equipment's executable range. With reactive power setpoint ; Furthermore, through a quantization encoding algorithm (parameters: integer encoding step size Δ, floating-point to fixed-point scaling factor) ), to achieve and Perform fixed-length binary encoding and generate a structured sequence of active / reactive power setpoints. This ensures its anti-interference capability and resolvability during communication. Furthermore, the MODBUS / TCP protocol frame construction method (parameters: function code 0x10, register start address) is used. Number of registers ), to achieve The protocol is encapsulated, and a control message frame containing a message header, function code, data length, data area, and CRC check field is generated. This meets the analytical requirements of the power regulation unit; By combining control quantity standardization, amplitude limiting, encoding conversion and protocol encapsulation, the control quantity in the previous step is converted into a control message that conforms to the PCS communication specification, ensuring the accuracy of power regulation commands in the transmission and parsing stages, and realizing the rapid execution capability of the photovoltaic energy storage system under voltage over-limit conditions. For example, at the rated capacity = 5 MW, rated voltage In a 10 kV photovoltaic-storage system, a certain execution process yielded a reactive power compensation of ΔQ = 0.8 Mvar and an active power reduction ratio of ρP = 0.15. (Active power setpoint) With reactive power setpoint The calculation formula is: in, Approximately 4.25 MW, Approximately 0.08 (normalized to rated voltage). During the encoding conversion stage, Δ = 0.01 is selected. = 100, will and Convert these to decimal integers 425 and 8 respectively, and then represent them in 16-bit unsigned fixed-point format to obtain... = [0x01A9, 0x0008]. Protocol encapsulation stage settings. = 0x0100、 =2, encapsulation function code 0x10, forming The length is 13 bytes. Performance verification shows that the encapsulated message is correctly parsed and downloaded to the execution queue on the PCS side, and the instruction response time is significantly improved, meeting the real-time voltage adjustment requirements in scenarios where a high proportion of renewable energy is connected to the distribution network. S8.4: Send the control message frame to the power regulation unit of the photovoltaic storage system and start the response timeout timer; verify whether the control command has been successfully received and entered the execution queue by receiving the feedback confirmation signal from the PCS. If no feedback is received within the specified time, the command retransmission mechanism is executed, and retransmission is allowed up to two times. S8.5: After the control command takes effect, continuously collect voltage, current and actual output power data of the grid connection point to construct a closed-loop control effect evaluation sequence; calculate the voltage recovery time and regulation error integral index based on the evaluation sequence, and use it as the input basis for the next round of online optimization of model parameters to complete the complete control closed loop from perception to execution to evaluation.

[0016] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0017] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for actively suppressing voltage exceedances in a photovoltaic-storage system in a power distribution network, characterized in that, Includes the following steps: S1: Obtain real-time operation data of the photovoltaic and energy storage system grid connection point, and simultaneously collect meteorological information and historical electricity consumption behavior data; S2: Based on the real-time operating data and the historical electricity consumption behavior data, classify and identify the current load type, distinguish typical load characteristics, and generate corresponding load feature labels; S3: Based on the load characteristic labels and the real-time collected voltage and current sequences, the equivalent feeder impedance and load sensitivity coefficient are identified online using a sliding window recursive algorithm to construct a time-varying electrical parameter model; S4: Combine the time-varying electrical parameter model with the current active power injection and reactive power injection, and input it into the voltage amplitude prediction model based on sensitivity coupling relationship to generate a voltage response trajectory prediction sequence. S5: Based on the voltage response trajectory prediction sequence, perform trend state classification on it, and output three trend state categories including stable, rising and approaching the limit or rapidly rising, to form a voltage change trend discrimination result; S6: Dynamically adjust the upper and lower boundaries of the voltage over-limit judgment threshold according to the voltage change trend discrimination result. When the predicted voltage enters the rising and approaching over-limit region and the duration exceeds the set lag period, it is determined that the control trigger condition is met and an initial control trigger signal is generated. S7: After generating the control trigger signal, start the reverse confirmation logic to continuously monitor the voltage change trend of subsequent sampling points. If the voltage trend is detected to fall back and leave the critical region, the issued control trigger signal is cancelled.

2. The active voltage over-limit suppression method for a photovoltaic-storage system in a power distribution network according to claim 1, characterized in that, Following step S7, the following is also included: S8: If the control trigger signal is not cancelled, the final confirmed control command is output to the power regulation unit to start reactive power support or active power reduction measures and complete the closed-loop control process.

3. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 1, characterized in that, Step S1 specifically includes: High-frequency sampling processing is performed on the voltage, current, active power, and reactive power at the grid connection point of the photovoltaic-storage system to obtain a continuous time-series sequence of real-time electrical parameters; Based on the real-time current and power factor measurements at the grid connection point, data acquisition of the operating status of surrounding loads is performed to generate an associated input set containing the power characteristics of the load side; By synchronously acquiring external environmental variable data using meteorological monitoring terminals and combining them with the distributed power generation output characteristic model, the impact factor on photovoltaic output fluctuations is calculated, forming an input for environmental disturbance compensation. Historical electricity consumption data for the same period is extracted from the historical database and spatiotemporally matched with the current real-time data based on a timestamp alignment strategy to generate a composite input vector with prior behavioral information. The real-time electrical parameter sequence, the associated input set containing load-side power characteristics, the environmental disturbance compensation term, and the composite input vector with behavioral prior information are subjected to multi-source heterogeneous data fusion processing to generate a structured input matrix in a unified format.

4. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 3, characterized in that, In step S1, the real-time electrical parameters are obtained by synchronously sampling three-phase voltage, three-phase current, active power and reactive power with a sampling frequency of 10-20kHz and a resolution of not less than 16bit. The phase-locked loop synchronization and fast Fourier transform noise reduction are used, and a standardized time-series parameter sequence is formed by combining sliding window statistical processing.

5. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 1, characterized in that, Step S2 specifically includes: Based on the real-time voltage, current, active power, and reactive power data of the photovoltaic-storage system grid connection point, as well as the operating current and power factor of the surrounding loads, and combined with synchronously collected meteorological information and historical electricity consumption behavior sequences, a multi-dimensional load feature vector is constructed. Normalization and outlier filtering are performed on the multidimensional load feature vector. The feature statistical distribution parameters are updated using a sliding time window mechanism. The first three principal component components are extracted using a principal component analysis algorithm to generate a dimensionality-reduced standardized feature matrix. Based on the standardized feature matrix, a pre-set ensemble classifier model is invoked to perform load type discrimination. During the training process, constant power type, constant current type, and motor type load labeling data from the typical load sample library are introduced, and the probability distribution of each type is output. Based on the classification probability distribution results, a dynamic decision threshold is set. When the maximum class probability is greater than 0.85, the load feature label is directly determined. If it is in the range of 0.6 to 0.85, the sliding consistency check is performed by combining the classification results of historical continuous periods, and finally the load feature label sequence is generated.

6. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 5, characterized in that, The ensemble classifier model is composed of a combination of random forest and support vector machine.

7. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 1, characterized in that, Step S3 specifically includes: Based on the generated load feature labels and the real-time collected grid connection point voltage and current sequences, a multivariate sampling matrix within a sliding time window is constructed. The effective values ​​of three-phase voltage, effective values ​​of three-phase current, and active / reactive power at each time step are synchronously aligned to form a time-series dataset for parameter identification. Based on the time series dataset and the load type indicated by the load feature labels, select the corresponding electrical response model structure and establish a local network equivalent circuit topology that adapts to different types of loads; The least squares recursive algorithm is used to fit the voltage deviation and current injection in multiple consecutive sampling periods within the sliding window online, calculate the linear mapping relationship between voltage change and injection current between the beginning and end of the feeder, and obtain the dynamically updated equivalent feeder impedance parameter sequence. Based on the equivalent feeder impedance parameter sequence and the reactive power response characteristics of the current load type, the voltage amplitude change rate caused by a unit reactive power change is calculated, and a load sensitivity coefficient is generated. The equivalent feeder impedance and the load sensitivity coefficient are bound by timestamps, and a time-varying electrical parameter model database is formed by combining the load characteristic tags. The database is then processed by low-pass filtering to output a standardized parameter vector.

8. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 7, characterized in that, The corresponding electrical response model structure is as follows: for constant power loads, PQ node modeling is used; for constant current loads, current amplitude normalization modeling is used; and for motor loads, transient reactance correction factor is introduced.

9. The active voltage over-limit suppression method for a photovoltaic-storage system in a distribution network according to claim 1, characterized in that, Step S4 specifically includes: Based on the time-varying electrical parameter model of the output, the equivalent feeder impedance and load sensitivity coefficient under the current operating conditions are extracted and used as input parameters for network topology and load characteristics for voltage amplitude prediction. The real-time active and reactive power data of the collected photovoltaic-storage system grid connection point are time-aligned and matched with the equivalent feeder impedance and load sensitivity coefficient to form a joint input vector containing the source-grid-load coupling relationship. Using a voltage amplitude prediction model based on node voltage sensitivity theory, the influence weight matrix of active / reactive power change on bus voltage amplitude is calculated. The voltage change is expressed by a first-order Taylor series approximation, forming a sensitivity coupling relationship mapping function. Substitute the joint input vector into the sensitivity coupling relationship mapping function, perform dynamic voltage response calculation, obtain the voltage impact increment of active power injection and reactive power injection at the current moment on the next 1 to 3 control cycles, and superimpose it on the current measured voltage value to generate a voltage response trajectory prediction sequence. The voltage response trajectory prediction sequence is normalized and its rationality is verified, and then encapsulated into a structured data format.

10. A method for actively suppressing voltage exceedances in a photovoltaic-storage system in a power distribution network according to claim 1, characterized in that, The trend state is divided as follows: when the rate of change is low and the voltage amplitude is within the normal range, it is classified as a stable zone; when the rate of change is moderate and the voltage is close to the upper limit threshold, it is classified as an upward trend approaching the limit zone; and when the rate of change is high and the voltage is close to or exceeds the upper limit, it is classified as a rapid rise zone.

Images