Distributed power source access and control method in intelligent circuit breaker

Abstract

The application discloses a distributed power supply access and control method in an intelligent circuit breaker and particularly relates to the technical field of intelligent circuit breakers; the initial synchronization state of the distributed power supply is judged based on voltage, frequency and phase difference; after the synchronization condition is met, the frequency change rate, voltage amplitude and phase angle change of each node are collected, and an electrical diagram structure with dynamic weight is constructed; the node information is weighted and aggregated by using a diagram attention network, and a virtual inertia estimation value is output; the node output power is inversed based on a synchronous machine model, and the system load disturbance rate is estimated in combination with time difference; the inertia and the disturbance rate are input into a synchronization risk prediction model to generate a synchronization risk score value; when the score value is higher than a threshold, it is determined that there is a false synchronization risk, and grid connection is prohibited; if the score value is lower than the threshold, the circuit breaker is closed, and a ramp power injection is performed to realize smooth grid connection; the method effectively improves the dynamic judgment capability before the distributed power supply is connected to the grid and reduces the grid connection instability risk.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent circuit breakers, and in particular to a distributed power supply access and control method in an intelligent circuit breaker. Background Art

[0002] In smart circuit breakers, distributed power access and control refers to integrating distributed generation equipment, such as solar and wind power, into the power system and enabling real-time monitoring and management through smart circuit breakers. Smart circuit breakers not only provide traditional overload and short-circuit protection but also dynamically adjust the access status, power quality, and output power of distributed power sources, thereby ensuring safe and stable grid operation and improving power efficiency.

[0003] The existing technology has the following shortcomings:

[0004] Existing technologies typically determine synchronization timing when distributed power sources are reconnected to the grid by detecting voltage, frequency, and phase differences. However, in systems with extremely low inertia (such as microgrids composed solely of photovoltaic inverters), "false synchronization" can occur. Even if synchronization conditions are met, the lack of rotational inertia and sudden load changes at the moment of closing can lead to current surges, voltage collapse, and even phase lock failure, ultimately causing grid connection failure and triggering protective tripping. This problem is particularly severe in dynamic grid environments, exposing the technical shortcomings of existing synchronization determination methods, which ignore the system's dynamic response capabilities. Summary of the Invention

[0005] The object of the present invention is to provide a distributed power supply access and control method in an intelligent circuit breaker to address the deficiencies in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a distributed power supply access and control method in an intelligent circuit breaker, comprising:

[0007] Based on the voltage amplitude difference, frequency difference and phase difference, it is preliminarily determined that the distributed power generation meets the conditions for synchronous grid connection;

[0008] After the synchronization conditions are met, the frequency change rate, voltage amplitude and phase angle change data of the distributed power supply nodes are further collected to construct a graph structure containing the electrical topology relationship between the distributed power supply nodes;

[0009] Using a graph attention network to perform weighted aggregation on the node information in the graph structure, and outputting the estimated virtual inertia corresponding to each node;

[0010] Based on the synchronous machine model, the phase angle, frequency and voltage measurement data of multiple nodes in the system are combined to invert the node output power, and the system load disturbance rate is estimated through time difference method.

[0011] Based on the estimated value of virtual inertia and the system load disturbance rate, a synchronization risk prediction model is constructed to output a synchronization risk prediction score value;

[0012] If the synchronization risk prediction score is higher than the set threshold, it is determined that there is a risk of false synchronization and the grid connection is prohibited or delayed; otherwise, the intelligent circuit breaker is controlled to close to achieve smooth grid connection of distributed power sources.

[0013] Preferably, the preliminary determination that the distributed power source meets the synchronous grid connection conditions includes:

[0014] A frequency trend model within a sliding time window is established, and the frequency signals of the power grid and distributed generation are fitted to extract the frequency change rate and form an instantaneous synchronization trend vector.

[0015] The weighted phase error prediction method is used to input the phase difference change trend between adjacent nodes into the timing predictor to generate the future short-term phase offset prediction value;

[0016] A dynamic synchronization vector space model is constructed by integrating the predicted synchronization parameters. The voltage amplitude difference, frequency trend, and phase prediction error are integrated to construct a multidimensional vector. The normalized projection distance of the vector in the synchronization tolerance hyperplane is calculated to determine whether the system has entered the safe synchronization zone.

[0017] If the projection distance is less than the synchronization margin threshold, it is determined that the synchronization condition is met.

[0018] Preferably, constructing a graph structure including electrical topological relationships between distributed power supply nodes includes:

[0019] The wide-area synchronous measurement unit collects real-time frequency change rate, voltage amplitude, and phase angle change data of multiple distributed power supply nodes, and performs time alignment processing on the data in the form of node state vectors;

[0020] According to the electrical connection relationship, conductance / susceptance parameters and active power exchange behavior between distributed power nodes, a directed graph structure with weighted edges is defined. The weight of each edge in the graph is jointly determined by the dynamic power flow and admittance amplitude between nodes.

[0021] Preferably, the outputting of the estimated virtual inertia values ​​corresponding to each node includes:

[0022] The frequency change rate, voltage amplitude and voltage phase angle of each node in the constructed dynamic graph structure are standardized and encoded to form a unified input node feature matrix;

[0023] A multi-head graph attention network is constructed based on the node features and edge weights, and each attention head calculates the weighted attention coefficient between adjacent nodes;

[0024] By aggregating the feature vectors of each node and its neighboring nodes and combining them with the corresponding attention weights, the updated node state representation is calculated.

[0025] The updated node representation is input into a multilayer perceptron, and the virtual inertia estimate corresponding to the node is output based on regression, and supervised learning optimization is performed using historical measured inertia samples.

[0026] Preferably, estimating the system load disturbance rate by time difference includes:

[0027] Based on the frequency, voltage amplitude and phase angle measurement data of multiple nodes, a dynamic equation model of a second-order synchronous generator is established for each node, wherein the model uses the power angle as the state variable and the electromagnetic power as the control variable;

[0028] The electrical output power of each node is calculated by combining the line admittance and instantaneous phase angle difference between nodes. The estimated active power output at the current moment is obtained by multiplying the node voltage by the adjacent node voltage, and then by the admittance modulus and cosine term.

[0029] A synchronous disturbance screening mechanism is introduced. When the node phase angle change rate exceeds the set threshold, the dynamic inversion mode is activated to refine and re-estimate the actual output power of the node.

[0030] Output the equivalent active power output of each node at the current moment.

[0031] Preferably, a time series data structure is constructed based on the active output power of the nodes at multiple moments;

[0032] The symmetric difference method is used to numerically differentiate the output power time series of each node and calculate the power change rate of the node;

[0033] The disturbance rate of each node is weighted according to its access capacity and historical disturbance sensitivity to form a weighted average disturbance rate of the entire system.

[0034] Preferably, a confidence assessment is performed on the disturbance rate calculation result based on the node voltage stability, sampling noise level and frequency jitter amplitude;

[0035] The confidence is used as the node weight coefficient and is combined with the access capacity factor to construct the node disturbance weight matrix;

[0036] The confidence-weighted system load disturbance rate is obtained by multiplying the disturbance rates of all nodes by their corresponding weights and summing them up.

[0037] Preferably, the step of constructing a synchronous risk prediction model and outputting a synchronous risk score value includes:

[0038] Obtain the estimated virtual inertia value of each distributed power generation node and the system load disturbance rate as input feature vectors;

[0039] Construct a two-dimensional risk feature space with the inertia value as one axis and the disturbance rate as the other axis to form a state point distribution diagram;

[0040] Based on the predefined risk distribution boundary function, calculate the relative distance between the current feature point and the risk boundary in space;

[0041] The distance is normalized and a synchronous risk prediction score value is output.

[0042] Preferably, the synchronization risk prediction score value is compared with a set threshold; when the synchronization risk prediction score value is higher than the threshold, the system determines that there is a false synchronization risk, prohibits the circuit breaker from closing, and does not participate in active grid connection; if the synchronization risk prediction score value is lower than or equal to the threshold, the system controls the intelligent circuit breaker to close, and maintains a slow-changing ramp power injection curve for a period of time after closing to achieve smooth grid connection.

[0043] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0044] 1. By introducing virtual inertia estimation, system load disturbance rate analysis, and a synchronization risk prediction and scoring mechanism, this invention can comprehensively assess the system's dynamic response capability and grid connection risk before distributed power generation is connected to the grid, effectively identifying hidden dangers such as false synchronization that are difficult to detect using traditional criteria. By constructing a multidimensional feature space and introducing advanced algorithms such as graph neural networks and support vector machines, the accuracy and predictability of synchronization judgments are improved, achieving a transition from "static threshold control" to "dynamic intelligent decision-making."

[0045] 2. This invention combines a ramped power injection control strategy after circuit breaker closure to significantly reduce the risk of current surge and voltage disturbances during grid connection, ensuring stable system frequency and voltage operation, preventing accidental activation of protective devices, and extending equipment life. The overall solution offers high security, adaptability, and scalability, making it particularly suitable for intelligent power distribution systems with high penetration of new energy and complex electrical structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0047] Figure 1 This is a mind map of the method of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0049] For examples, see Figure 1 As shown, the distributed power supply access and control method in the intelligent circuit breaker of this embodiment includes:

[0050] Based on the voltage amplitude difference, frequency difference and phase difference, it is preliminarily determined that the distributed power generation meets the conditions for synchronous grid connection;

[0051] After the synchronization conditions are met, the frequency change rate, voltage amplitude and phase angle change data of the distributed power supply nodes are further collected to construct a graph structure containing the electrical topology relationship between the distributed power supply nodes;

[0052] Using a graph attention network to perform weighted aggregation on the node information in the graph structure, and outputting the estimated virtual inertia corresponding to each node;

[0053] Based on the synchronous machine model, the phase angle, frequency and voltage measurement data of multiple nodes in the system are combined to invert the node output power, and the system load disturbance rate is estimated through time difference method.

[0054] Based on the estimated value of virtual inertia and the system load disturbance rate, a synchronization risk prediction model is constructed to output a synchronization risk prediction score value;

[0055] If the synchronization risk prediction score is higher than the set threshold, it is determined that there is a risk of false synchronization and the grid connection is prohibited or delayed; otherwise, the intelligent circuit breaker is controlled to close to achieve smooth grid connection of distributed power sources.

[0056] This paper addresses the issues of hysteresis, locality, and large static errors in traditional distributed power generation synchronization criteria before grid connection. It proposes a synchronization condition determination method that integrates predictive judgment with dynamic trend analysis. This method uses voltage, frequency, and phase as basic parameters and, through sliding window modeling, short-term trend prediction, and multidimensional feature space determination, comprehensively assesses the synchronization status between distributed power sources and the grid, improving the security and predictiveness of grid connection control.

[0057] During implementation, a fixed-width sliding time window is first set, for example, between 500 milliseconds and 1 second. The system continuously collects frequency measurement data from the grid and distributed generation within this window, and uses the least squares method to fit the first-order frequency trend curve.

[0058] Specifically, the frequency value of each sampling point and its corresponding timestamp form a sample pair. Using a linear fit, the first-order derivative of the frequency with respect to time is calculated and recorded as the frequency rate of change. This rate of change reflects the acceleration or deceleration of the power supply frequency relative to the grid and is an important indicator for determining whether synchronization will be achieved in the future.

[0059] Represented by a set of sampling points, this trend line can be expressed as "frequency equals a constant plus the frequency rate of change multiplied by time," meaning the frequency changes linearly over time. A negative frequency rate of change indicates that the system is trending toward grid frequency. A positive and large rate of change indicates a risk of frequency drift.

[0060] After acquiring real-time phase angle information, the system uses a first-order Kalman filter to perform state estimation and short-term future predictions on phase data to overcome the effects of sampling delays and grid disturbances in synchronous control. This Kalman filter model is based on a linear state-space equation, with a historical phase angle sequence as input and a predicted phase angle value for the next step as output.

[0061] At the same time, to improve the sensitivity and robustness of the prediction, voltage amplitude weighting is introduced into the filter update process. The more stable the voltage, the higher the credibility of the prediction value. When the voltage amplitude fluctuates violently, the phase prediction weight is automatically reduced, thereby suppressing the interference of non-steady state on synchronization determination.

[0062] The prediction error is defined as the difference between the current predicted phase angle value and the grid reference phase angle. The mean square process is performed over multiple consecutive time steps to form a sliding prediction error vector for subsequent synchronization judgment.

[0063] In this embodiment, to overcome the limitations of a single threshold determination, the present invention constructs a three-dimensional synchronization determination vector space. This space consists of three characteristic axes, corresponding to the voltage amplitude difference, the frequency change trend (i.e., the frequency change rate), and the predicted phase error.

[0064] At each time point, the current three parameter values ​​can be combined into a vector and projected onto a pre-set "synchronous safety hyperplane." This hyperplane can be constructed from synchronized successful samples from multiple simulations or measured data, serving as the tolerance boundary for the multi-dimensional criterion.

[0065] The Euclidean projection distance is used to determine the degree of deviation between the current vector and the hyperplane. If the projection distance is less than the set safety tolerance threshold, the current state is determined to be within the safe grid connection range; otherwise, it enters the waiting or dynamic adjustment state.

[0066] This method not only considers the deviation of the instantaneous state, but also takes into account the trend of state changes and the possibility of future offsets, thereby avoiding hasty grid connection before frequency drift or phase mutation is about to occur, significantly improving the robustness and predictability of synchronization decisions.

[0067] During the actual deployment process, this embodiment embeds the relevant algorithms in the embedded controller through the ARM main control chip, collects the frequency, voltage and phase angle data from the PMU in real time, and responds to the above judgment process in milliseconds through the sliding window cache and fast vector calculation module.

[0068] The actual measurement shows that under the typical photovoltaic inverter island operation state, the traditional Compared with the static threshold judgment method, this method can identify the potential frequency desynchronization trend about 200 to 300 milliseconds in advance and automatically suppress the grid-connected operation with potential false synchronization.

[0069] To further achieve high-precision assessment and proactive control of distributed generation grid connection risks in this invention, this embodiment provides a method for constructing a graph structure that integrates topology and dynamic state information, after initially determining that synchronization conditions are met. This method synchronously collects information such as the frequency change rate, voltage amplitude, and phase angle change of distributed generation nodes, and combines this information with the electrical connection relationships between nodes to establish a directed, weighted graph structure that dynamically reflects the system's operating status. This structure is then used for subsequent graph neural network analysis and inertia perturbation modeling.

[0070] During the implementation process, the synchronous measurement devices deployed on the distributed power supply side (such as PMU and WAMS terminals) collect real-time data including:

[0071] Node voltage amplitude;

[0072] Node voltage phase angle;

[0073] Nodal frequency and frequency change rate (calculated from the first-order derivative of the phase angle);

[0074] To improve the accuracy of subsequent graph calculations, all collected data must be strictly aligned based on the Global Positioning System (GPS) or IEEE 1588 time protocol to ensure that the state quantities of all nodes have a consistent time reference at the same moment.

[0075] In addition, the sampling frequency of all nodes is recommended to be no less than 50 Hz to ensure the accuracy of capturing dynamic behaviors.

[0076] After collecting node status, the direct connection between nodes is determined based on the microgrid's electrical configuration file (such as line parameters and switch status). If a wire, cable, or coupling switch connects node A to node B, a directed edge is added from A to B in the graph structure.

[0077] For each edge, its weight value is further calculated. In order to improve the expressiveness of the graph model, the present invention adopts an edge weight calculation method based on "dynamic power and susceptance joint weight":

[0078] First, calculate the instantaneous active power value output by node A to B, denoted as Pab;

[0079] Secondly, consult the system model and extract the admittance modulus Yab of the line connecting A and B;

[0080] Finally, the edge weights are calculated using the product normalization form, specifically:

[0081] The edge weight is equal to the power flow from node A to B multiplied by the admittance amplitude, and then normalized to a set range (for example, 0 to 1)

[0082] This construction method can simultaneously reflect:

[0083] the direction and strength of actual power interactions;

[0084] The transmission capacity and coupling tightness of the electrical path;

[0085] Compared with the conventional graph structure that uses "connection" as the judgment basis, this method is more dynamic and physically consistent in edge weight representation.

[0086] In order to further improve the adaptability of the graph structure to the system operating status, this embodiment introduces an "electrical similarity" function to quantify the degree of similarity between the states of the nodes.

[0087] This function is mainly based on two dimensions:

[0088] The voltage phase angle difference between adjacent nodes;

[0089] Frequency offset between adjacent nodes.

[0090] For each pair of adjacent nodes A and B, calculate the two differences respectively, and then construct the following normalized similarity value:

[0091] Electrical similarity = exponential decay function, where the input is the square of the phase angle difference plus the square of the frequency difference, and the result is between 0 and 1, with values ​​closer to 1 indicating greater synchronization

[0092] This value is used to modify the original edge weight to form the final edge weight:

[0093] Final edge weight = original edge weight × electrical similarity;

[0094] This operation means that when the electrical state deviates significantly (such as a sudden change in the frequency of a node or a phase angle drift), the edge weight will be reduced accordingly, reflecting the fact that the actual electrical coupling strength is weakened.

[0095] Ultimately, the system forms a dynamic graph structure with time sensitivity and state adaptability, which is used for graph neural network modeling and inertia / disturbance identification.

[0096] In the previous embodiment, a dynamic graph structure containing multi-dimensional state information such as node frequency, voltage phase angle, and amplitude has been constructed. To implement graph neural network processing, this embodiment first standardizes the node features to ensure uniformity of all dimensions.

[0097] Specifically, each node contains the following three key features:

[0098] Real-time frequency rate of change (first derivative);

[0099] Voltage amplitude (V);

[0100] Voltage phase angle (θ);

[0101] The above three types of features are organized into vectors according to the sampling time and normalized to the interval [0,1] or [-1,1] to form a node feature tensor of uniform dimension, denoted as , where n is the number of nodes, d is the feature dimension, and R is a set of real numbers.

[0102] At the same time, the previously constructed graph structure with edge weights G = (V, E, A) is retained, where A∈ Represents the edge weight matrix, that is, each element in the adjacency matrix is ​​the dynamic coupling strength of the corresponding edge.

[0103] To fully capture the coupling relationship between different nodes in different state dimensions, this embodiment adopts a multi-head graph attention network (multi-head GAT), in which each attention head is responsible for extracting correlations in different directions.

[0104] For any adjacent nodes i and j, let xi and xj be their feature vectors respectively, and calculate the attention weight αij in the following form:

[0105] First, the node features are mapped separately through a shared linear transformation matrix, and then the mapping results of the two nodes are concatenated and sent to a single-layer perceptron and activated, and the output is a non-normalized attention score.

[0106] This process is expressed in words as follows:

[0107] Transform node features using a shared linear weight matrix;

[0108] Concatenate two node feature vectors;

[0109] The mutual influence scores are calculated using a linear mapping with parameters;

[0110] Use LeakyReLU for nonlinear activation;

[0111] The attention scores of all neighbor nodes are softmax normalized to obtain the normalized attention weights.

[0112] This attention mechanism can dynamically adjust the weight of information transmission according to the differences in node status, thereby more realistically simulating the influence intensity between different nodes in the power grid.

[0113] The final representation of each node is composed of the representations output by multiple attention heads. That is, by connecting multiple independent attention mechanisms in parallel, graph information of different scales is fused together, thereby improving feature expression ability and network stability.

[0114] In the feature aggregation stage, the system updates the feature representation of each node i to the weighted sum of its own features and the features of all neighboring nodes, where the weight is the above-mentioned attention coefficient.

[0115] To prevent vanishing gradients and feature drift during deep network training, this embodiment introduces a residual connection structure. Specifically, the input features are added to the aggregation results to form the final updated node representation. This structure ensures that the original features are not completely lost and enhances training stability and convergence speed.

[0116] After aggregation, the updated representations of all nodes are input into a two-layer fully connected neural network (MLP) to output the virtual inertia estimate of the current node.

[0117] The first layer uses the ReLU activation function to extract nonlinear features. The second layer directly regresses the output inertia value Hi, which is a continuous real value. The loss function of the entire network is the mean squared error (MSE), using historically measured or simulated inertia values ​​as supervision signals to minimize the deviation between the predicted value and the true value.

[0118] During the specific training process, the Adam optimizer is used, the initial learning rate is set to 1e-3, and batch normalization and early stopping mechanisms are used to prevent overfitting.

[0119] During the pre-grid synchronization assessment process, the system needs to dynamically determine the severity of load fluctuations between the distributed generation (DG) and the grid. Because changes in current, voltage, and frequency reflect the nature of power fluctuations in a microgrid, this embodiment uses an "equivalent synchronous machine" as a foundation and establishes a dynamic model based on power angle to express the functional relationship between node power and state variables.

[0120] For each distributed power generation node, the following simplified second-order synchronous machine equation is used for modeling:

[0121] The mechanical input power minus the electrical output power of the node is equal to the acceleration of the power angle multiplied by the inertia constant, plus the damping term;

[0122] The rate of change of the power angle is equal to the frequency;

[0123] The electrical output power is the product of the node voltage and the voltage of its adjacent node, multiplied by the cosine function of the admittance modulus of the two nodes and their phase angle difference.

[0124] Through the above relationships, electrical variables (such as voltage amplitude and phase angle) are linked to dynamic power output.

[0125] To achieve real-time modeling, the system collects the following data through synchronized measurement units (such as PMUs):

[0126] The current voltage amplitude of each node;

[0127] Phase angle (calculated by reference phase);

[0128] Nodal frequencies and their first derivatives.

[0129] After obtaining the connection topology and admittance parameters between nodes, the system can use the electrical flow calculation method to invert the electrical output power of each node in real time.

[0130] The calculation method is as follows:

[0131] For any two adjacent nodes A and B, let their voltage amplitudes be Va and Vb respectively, their admittance modulus be Yab, and their phase angle difference be θa minus θb;

[0132] Then the active output of node A to B is approximately Va multiplied by Vb, multiplied by the product of the modulus of Yab and the cosine of the phase angle difference;

[0133] The total output power of node A can be obtained by accumulating the power of all adjacent nodes.

[0134] This method does not require direct measurement of the power signal and only relies on the collected basic electrical parameters such as frequency and voltage. It is scalable and anti-interference.

[0135] When it is detected that the node phase angle change rate exceeds the set threshold (such as 5 degrees / second), the system automatically triggers the "high dynamic inversion mode". In this mode, damping correction and second-order derivative estimation terms are added to improve the accuracy of the estimation under disturbance conditions.

[0136] Based on the acquisition of the output power of all nodes, the system establishes a time series data structure to record the historical values ​​of the node output power at a fixed sampling interval.

[0137] The symmetric difference method is used, that is, taking the difference between the power values ​​at two adjacent time points and dividing it by the time interval, to obtain the instantaneous disturbance rate of the node. Mathematically expressed as:

[0138] The current disturbance rate is approximately equal to the power value at the next moment minus the power value at the previous moment, divided by the time interval between the two moments.

[0139] In order to avoid the influence of error accumulation and oscillation response, the differential operation adopts sliding window averaging processing.

[0140] Taking into account the differences in load contributions of different nodes in the system, this embodiment introduces a "disturbance weight matrix" for weighted averaging:

[0141] Each node is assigned a capacity weight, which represents the proportion of its rated power in the total system capacity;

[0142] A disturbance sensitivity factor is also introduced to indicate the possibility that the node frequently causes disturbances in historical operation;

[0143] Multiply the two factors together as the node weight coefficient and perform weighted summation on the disturbance rate of each node.

[0144] In addition, to improve the credibility of the results, the system also introduces a "state credibility factor", which includes:

[0145] Standard deviation of the voltage signal (reflecting stability);

[0146] Spectral distribution of sampling noise;

[0147] Current frequency fluctuation range.

[0148] The higher the credibility, the greater the weight of the corresponding disturbance rate in the weighting process; if the signal quality is poor, its disturbance rate contribution will be actively weakened.

[0149] The "global disturbance rate" output by the final system reflects the severity of the current load power change. The larger the value, the stronger the load dynamics and the lower the system's tolerance to grid-connected disturbances.

[0150] This metric is compared to the sync risk threshold:

[0151] If it is lower than the set value, the virtual inertia evaluation and circuit breaker grid connection operation are allowed to continue;

[0152] If the threshold is exceeded, the system will determine that there is a "disturbance-synchronous coupling risk", terminate the grid connection, and trigger backup power buffering or frequency regulation measures.

[0153] Through graph neural networks and synchronous machine modeling, we obtain the estimated virtual inertia (Hvirtual) of each distributed power generation node and the system's current load disturbance rate (dP / dt). This embodiment uses these as core inputs to construct the basic feature space for synchronization risk assessment.

[0154] To further enhance the trend perception capability of the model, the following additional input features are introduced:

[0155] The time rate of change of virtual inertia (ΔH / Δt) reflects the inertia response speed;

[0156] The time rate of change of the load disturbance rate (ΔdP / dt) reflects the trend of disturbance aggravation;

[0157] The standard deviation of system frequency fluctuation measures the current frequency stability.

[0158] All features are recorded with a uniform sampling period and preprocessed by standardization and normalization.

[0159] In the basic model, each pair of "inertia and disturbance rate" is mapped into a two-dimensional coordinate space, where:

[0160] The horizontal axis represents the estimated value of virtual inertia;

[0161] The vertical axis represents the disturbance rate value.

[0162] The system defines a risk distribution boundary region, which can be fitted through simulation data or set based on engineering experience. For example, low inertia and high disturbance areas are marked as high risk areas, and vice versa.

[0163] For any input feature point, calculate its normalized distance to the high-risk boundary, denoted as Drisk. The risk score Srisk is defined as:

[0164] When the point is in the low-risk zone, the score is close to 0;

[0165] As the point gradually approaches or crosses the boundary, the score gradually approaches 1;

[0166] If the point is inside the high-risk area, the score is 1, indicating an extremely high risk of synchronous instability.

[0167] This method realizes continuous risk expression based on spatial distance and is no longer limited to a single threshold judgment.

[0168] In order to improve the accuracy and generalization ability of the risk boundary, this embodiment introduces the support vector machine (SVM) as a discriminant model.

[0169] The training data consists of multiple combinations of inertia and disturbance rate features from historical simulations or field records, and is labeled to determine whether synchronization instability or frequency disturbances occurred after grid connection. The system uses a radial basis function (RBF) kernel to train the SVM and generate a nonlinear discriminant boundary.

[0170] At runtime, the current system input inertia and disturbance rate are combined and fed into the SVM model, which outputs the classification confidence level of its "distance from the high-risk boundary." This confidence level is mapped to a continuous risk score, which is weighted and fused with the score constructed in the previous section (Drisk) to form the final score (Stotal).

[0171] This method makes up for the problem of insufficient recognition ability of static boundaries for special disturbance patterns, and has the advantages of adaptive learning and data-driven.

[0172] In the distributed power grid connection control method of the present invention, after completing the synchronization risk prediction scoring, the system uses the score value as a decision variable and compares it with the set synchronization safety threshold to determine the closing strategy of the circuit breaker and the dynamic response mode of the grid connection process.

[0173] Specifically, the system sets a static or dynamic synchronization risk threshold, typically between 0.6 and 0.8, to demarcate "safe" and "risky" areas. The threshold can be dynamically adjusted based on system structure, power source type, load fluctuation characteristics, or grid stability level.

[0174] First, the system compares the current synchronization risk prediction score with the threshold:

[0175] If the score exceeds the threshold, the system determines that the current grid-connected operation carries a high risk of synchronization instability, particularly the possibility of "false synchronization." This occurs when static synchronization criteria (such as voltage, frequency, and phase difference) are met, but low system inertia or severe disturbances can cause frequency jumps, voltage collapse, or phase lock loss at the moment of closing. Therefore, the system prohibits the intelligent circuit breaker from closing, and the distributed generation (DG) maintains an islanded or standby state, preventing it from injecting active power into the main grid.

[0176] When the score value is lower than or equal to the threshold value, it indicates that the current system has acceptable synchronization stability, and the system triggers a circuit breaker closing command to allow the distributed power source to enter the grid-connected operation stage.

[0177] To prevent voltage fluctuations or power angle jumps caused by sudden power injection at the moment the circuit breaker is closed, the system further adopts a "slow-ramp power injection curve" control method. That is, a slow-start time window (for example, 2 to 5 seconds) is set after the circuit breaker is closed. The target active power Ptarget is segmented into each time slice, and the actual output power P(t) is gradually increased to form a power-time ramp curve. Its mathematical expression is:

[0178] The power output P(t) is equal to Ptarget multiplied by the current time t divided by the total startup duration T, and generally increases monotonically between the interval [0, Ptarget].

[0179] Through this slow-changing control strategy, not only is smooth power injection achieved, but the system's grid-connected current impact, frequency mutation, and relay protection malfunction probability are also effectively reduced, thereby improving the safety of grid-connected operations and the quality of power.

[0180] In summary, this embodiment combines synchronous risk prediction scoring with a ramped power injection mechanism, introduces dynamic adjustment measures before and after grid connection, and realizes intelligent risk control of the entire distributed power grid connection process.

[0181] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A distributed power supply access and control method in an intelligent circuit breaker, characterized by: include: Based on the voltage amplitude difference, frequency difference and phase difference, it is preliminarily determined that the distributed power generation meets the conditions for synchronous grid connection; After the synchronization conditions are met, the frequency change rate, voltage amplitude and phase angle change data of the distributed power supply nodes are further collected to construct a graph structure containing the electrical topology relationship between the distributed power supply nodes; Using a graph attention network to perform weighted aggregation on the node information in the graph structure, and outputting the estimated virtual inertia corresponding to each node; Based on the synchronous machine model, combined with the phase angle, frequency and voltage measurement data of multiple nodes in the microgrid system, the node output power is obtained by inversion, and the load disturbance rate of the microgrid system is estimated by time difference method. Based on the estimated value of virtual inertia and the load disturbance rate of the microgrid system, a synchronization risk prediction model is constructed to output a synchronization risk prediction score value; If the synchronization risk prediction score is higher than the set threshold, it is determined that there is a risk of false synchronization and the grid connection is prohibited or delayed; otherwise, the intelligent circuit breaker is controlled to close to achieve smooth grid connection of distributed power sources.

2. The distributed power supply access and control method in an intelligent circuit breaker according to claim 1, characterized in that: The preliminary determination that the distributed power source meets the synchronous grid connection conditions includes: A frequency trend model within a sliding time window is established, and the frequency signals of the power grid and distributed generation are fitted to extract the frequency change rate and form an instantaneous synchronization trend vector. The weighted phase error prediction method is used to input the phase difference change trend between adjacent nodes into the timing predictor to generate the future short-term phase offset prediction value; A dynamic synchronization vector space model is constructed by integrating the predicted synchronization parameters. The voltage amplitude difference, frequency trend, and phase prediction error are integrated to construct a multidimensional vector. The normalized projection distance of the vector in the synchronization tolerance hyperplane is calculated to determine whether the system has entered the safe synchronization zone. If the projection distance is less than the synchronization margin threshold, it is determined that the synchronization condition is met.

3. The distributed power supply access and control method in an intelligent circuit breaker according to claim 1, characterized in that: Constructing a graph structure containing the electrical topology relationship between distributed power generation nodes includes: The wide-area synchronous measurement unit collects real-time frequency change rate, voltage amplitude, and phase angle change data of multiple distributed power supply nodes, and performs time alignment processing on the data in the form of node state vectors; According to the electrical connection relationship, conductance / susceptance parameters and active power exchange behavior between distributed power nodes, a directed graph structure with weighted edges is defined. The weight of each edge in the graph is jointly determined by the dynamic power flow and admittance amplitude between nodes.

4. The distributed power supply access and control method in an intelligent circuit breaker according to claim 1, characterized in that: The output of the virtual inertia estimation value corresponding to each node includes: The frequency change rate, voltage amplitude and voltage phase angle of each node in the constructed dynamic graph structure are standardized and encoded to form a unified input node feature matrix; A multi-head graph attention network is constructed based on the node features and edge weights, and each attention head calculates the weighted attention coefficient between adjacent nodes; By aggregating the feature vectors of each node and its neighboring nodes and combining them with the corresponding attention weights, the updated node state representation is calculated. The updated node representation is input into a multilayer perceptron, and the virtual inertia estimate corresponding to the node is output based on regression, and supervised learning optimization is performed using historical measured inertia samples.

5. The distributed power supply access and control method in an intelligent circuit breaker according to claim 1, characterized in that: The time difference method for estimating the microgrid system load disturbance rate includes: Based on the frequency, voltage amplitude and phase angle measurement data of multiple nodes, a dynamic equation model of a second-order synchronous generator is established for each node, wherein the model uses the power angle as the state variable and the electromagnetic power as the control variable; The electrical output power of each node is calculated by combining the line admittance and instantaneous phase angle difference between nodes. The estimated active power output at the current moment is obtained by multiplying the node voltage by the adjacent node voltage, and then by the admittance modulus and cosine term. A synchronous disturbance screening mechanism is introduced. When the node phase angle change rate exceeds the set threshold, the dynamic inversion mode is activated to refine and re-estimate the actual output power of the node. Output the equivalent active power output of each node at the current moment.

6. The distributed power supply access and control method in the intelligent circuit breaker according to claim 5, characterized in that: Based on the active output power of nodes at multiple moments, a time series data structure is constructed; The symmetric difference method is used to numerically differentiate the output power time series of each node and calculate the power change rate of the node; The load disturbance rate of each node microgrid system is weighted according to its access capacity and historical disturbance sensitivity to form a weighted average value of the microgrid system load disturbance rate.

7. The distributed power supply access and control method in the intelligent circuit breaker according to claim 6, characterized in that: Conduct confidence assessment on the calculated results of the microgrid system load disturbance rate based on node voltage stability, sampling noise level, and frequency jitter amplitude; The confidence is used as the node weight coefficient and is combined with the access capacity factor to construct the node disturbance weight matrix; The load disturbance rate of the microgrid system of all nodes is multiplied by its corresponding weight and then summed up to obtain the confidence-weighted load disturbance rate of the microgrid system.

8. The distributed power supply access and control method in an intelligent circuit breaker according to claim 1, characterized in that: The constructing of a synchronous risk prediction model and outputting a synchronous risk score value comprises: Obtain the estimated virtual inertia value of each distributed power generation node and the microgrid system load disturbance rate as input feature vectors; A two-dimensional risk feature space is constructed, with the inertia value as one axis and the microgrid system load disturbance rate as the other axis, to form a state point distribution diagram; Based on the predefined risk distribution boundary function, calculate the relative distance between the current feature point and the risk boundary in space; The distance is normalized and a synchronous risk prediction score value is output.

9. The distributed power supply access and control method in the intelligent circuit breaker according to claim 8, characterized in that: The synchronization risk prediction score value is compared with the set threshold; when the synchronization risk prediction score value is higher than the threshold, the microgrid system determines that there is a false synchronization risk, prohibits the circuit breaker from closing, and does not participate in active grid connection; if the synchronization risk prediction score value is lower than or equal to the threshold, the microgrid system controls the intelligent circuit breaker to close and maintains a slowly varying ramp power injection curve for a period of time after closing to achieve smooth grid connection.

Images