Method for collaborative monitoring of power equipment insulation and decentralized control of transformer network segments

Abstract

The application relates to the technical field of power distribution network collaborative control, and particularly discloses a power equipment insulation collaborative monitoring and transformer network segment decentralized control method, which comprises the following steps: deploying a sensing unit on each phase of a transformer to synchronously collect real-time leakage currents, winding hotspot temperatures, bus reference phase voltages and neutral point displacement voltages of the alternating current power distribution network of each phase; performing filtering and purification on the real-time leakage currents of each phase to obtain actual leakage currents of each phase; and calculating a dielectric loss tangent value and a three-phase unbalance degree. According to the method, the insulation aging rate and the safety power boundary of each phase can be calculated in real time through multi-modal data collection and an electro-thermal coupled digital twin model in actual operation of the power distribution network, so that the defect that a traditional three-phase overall evaluation is insufficient in local deterioration perception is made up.

Description

Technical Field

[0001] This invention relates to the field of power distribution network collaborative control technology, specifically a method for collaborative monitoring of power equipment insulation and decentralized control of transformer network segments. Background Technology

[0002] AC distribution networks typically contain multiple transformer substations. In current distribution network operation and management, transformer monitoring is mostly limited to periodic offline insulation preventative tests, or relies solely on the temperature of a single overall winding for routine early warning. Meanwhile, power flow control and load regulation in distribution networks primarily focus on the three-phase transformer as a whole as the control object, mainly alleviating the operating pressure on heavily loaded transformers by adjusting transformer taps or disconnecting part of the load.

[0003] In real-world operating scenarios, transformers often face three-phase load imbalance. This imbalance can cause additional temperature rise in a specific phase winding, leading to accelerated aging of single-phase insulation. Current technologies lack real-time assessment methods for phase insulation status under electrothermal coupling conditions, making it difficult to accurately identify specific phases in the early stages of degradation.

[0004] When a risk of insulation degradation is detected in a phase, the existing three-phase overall control mechanism is unable to dynamically transfer the load of the degraded phase to its own healthy phase or a healthy phase in an adjacent transformer area. This results in the transformer windings with localized insulation degradation still having to bear a high load, increasing the risk of thermal breakdown of the equipment. Furthermore, conventional overall power flow regulation methods are unable to simultaneously suppress power quality problems such as neutral point displacement voltage rise caused by three-phase imbalance. Summary of the Invention

[0005] This invention aims to at least partially address one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a method for collaborative monitoring of power equipment insulation and decentralized control of transformer network segments, thereby improving the operational safety of distribution network equipment.

[0006] To achieve the above objectives, a first aspect of the present invention proposes a method for collaborative monitoring of insulation in power equipment and decentralized control of transformer network segments, comprising:

[0007] S1. Sensor units are deployed in each phase of the transformer to synchronously collect the real-time leakage current, winding hot spot temperature, bus reference phase voltage and neutral point displacement voltage of each phase. The real-time leakage current of each phase is filtered and purified to obtain the actual leakage current of each phase, and the dielectric loss tangent and three-phase unbalance are calculated.

[0008] S2. Construct a digital twin model of transformer electrothermal coupling, input the actual leakage current and winding hot spot temperature of each phase, and combine the additional temperature rise correction caused by three-phase imbalance to calculate the real-time insulation aging rate of each phase; with the real-time insulation aging rate not exceeding the limit as a constraint, solve the maximum safe apparent power threshold, active power upper limit and reactive power adjustment margin range of each phase, and divide each phase into insulation deterioration phase and bearing healthy phase based on the real-time insulation aging rate;

[0009] S3. Construct a topology diagram of the phase distribution network segment, and calculate the power flow acceptance score of the healthy phase based on the insulation status and remaining carrying capacity of each phase using a graph convolutional neural network, and generate the phase power flow transfer ratio.

[0010] S4. Implement active power-frequency droop control and independent reactive voltage droop control for each phase in the distribution network. Based on the phase power flow transfer ratio and various power constraint ranges, dynamically adjust the reactive power droop coefficients of the insulation-deteriorated phase and the receiving healthy phase to achieve power flow transfer. Simultaneously suppress the neutral point displacement voltage through zero-sequence reactive power injection to manage three-phase imbalance.

[0011] To achieve the above objectives, a second aspect of the present invention provides a power equipment insulation collaborative monitoring and transformer network segment distributed control system, comprising:

[0012] The multi-modal acquisition and purification calculation module is used to deploy sensing units in each phase of the transformer to synchronously acquire the real-time leakage current, winding hot spot temperature, bus reference phase voltage and neutral point displacement voltage of each phase, perform filtering and purification on the real-time leakage current of each phase to obtain the actual leakage current of each phase, and calculate the dielectric loss tangent and three-phase unbalance.

[0013] The twin-driven and power constraint solving module is used to construct a digital twin model of transformer electrothermal coupling. It takes the actual leakage current and winding hot spot temperature of each phase as input, and calculates the real-time insulation aging rate of each phase by combining the additional temperature rise correction caused by three-phase imbalance. With the real-time insulation aging rate not exceeding the limit as a constraint, it solves the maximum safe apparent power threshold, active power upper limit and reactive power adjustment margin range of each phase, and divides each phase into insulation deterioration phase and bearing healthy phase based on the real-time insulation aging rate.

[0014] The phase topology construction and path optimization module is used to construct the phase distribution network segment topology map. It calculates the power flow acceptance score of the healthy phase based on the insulation status and remaining carrying capacity of each phase using a graph convolutional neural network, and generates the phase power flow transfer ratio.

[0015] The phase-distributed droop and imbalance management module is used to implement active power-frequency droop control and independent reactive voltage droop control for each phase in the distribution network. Based on the phase power flow transfer ratio and various power constraint ranges, it dynamically adjusts the reactive power droop coefficients of the insulation-deteriorated phase and the receiving healthy phase to achieve power flow transfer; simultaneously, it suppresses the neutral point displacement voltage through zero-sequence reactive power injection to manage three-phase imbalance.

[0016] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described method for collaborative monitoring of power equipment insulation and decentralized control of transformer network segments.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0018] The method provided by this invention, through multimodal data acquisition and electrothermal coupled digital twin model, can calculate the insulation aging rate and safe power boundary of each phase in real time during actual operation of the distribution network, thus overcoming the deficiency of traditional three-phase overall assessment in terms of insufficient perception of local degradation. When the system identifies insulation degradation in a single phase, it uses a graph convolutional neural network to score the carrying capacity of the globally healthy phases within the network segment and generate a phase-by-phase transfer strategy. During actual control execution, the system guides the power flow of the distribution network to automatically and smoothly transfer from the degraded phase to the healthy phase with remaining capacity by dynamically adjusting the independent reactive power voltage droop coefficient of each phase, and simultaneously suppresses neutral point displacement voltage through zero-sequence reactive power injection. This scheme, while ensuring uninterrupted power supply, proactively and specifically reduces the operating load of the vulnerable phase windings, alleviates the additional damage caused by three-phase imbalance, slows down the overall aging process of equipment, and improves the operational safety of distribution network equipment. Attached Figure Description

[0019] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0020] Figure 1 This is a flowchart illustrating the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by the present invention;

[0021] Figure 2 This is a comparison of waveforms before and after real-time leakage current adaptive morphology-wavelet packet fusion filtering purification in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by this invention.

[0022] Figure 3This is a three-dimensional surface plot of the nonlinear response surface of insulation aging rate under transformer electrothermal coupling condition in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by the present invention;

[0023] Figure 4 This is a convergence curve of the multi-objective weighted reward function based on the near-end strategy optimization (PPO) algorithm in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by this invention;

[0024] Figure 5 This is a waveform comparison of the suppression of three-phase current mutation rate before and after applying adaptive virtual impedance during the power flow transfer transient process in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by the present invention.

[0025] Figure 6 This is a smooth transition trajectory curve of the reactive power droop coefficient slope based on insulation transient vulnerability constraints in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by the present invention.

[0026] Figure 7 This is a Pareto front distribution diagram of the marginal cost game between active heat dissipation intervention and electrical power flow transfer in the power equipment insulation collaborative monitoring and transformer network segment decentralized control method provided by the present invention;

[0027] Figure 8 This is a schematic diagram illustrating the implementation of the power equipment insulation collaborative monitoring and transformer network segment distributed control system provided by the present invention;

[0028] Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0029] Embodiments of the present invention are described in detail below, examples of which are illustrated 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 intended to explain the present invention, and should not be construed as limiting the present invention.

[0030] The following description, with reference to the accompanying drawings, describes a method, system, and electronic device for collaborative monitoring of insulation in power equipment and decentralized control of transformer network segments according to embodiments of the present invention.

[0031] Example 1:

[0032] Figure 1 This is a flowchart illustrating a method for collaborative monitoring of power equipment insulation and decentralized control of transformer network segments according to an embodiment of the present invention. This method is applied to an AC distribution network containing multiple adjacent transformer substation segments.

[0033] In modern urban and industrial power distribution networks, with the diversification of load types and the increase in single-phase high-power nonlinear loads, transformers operate under complex conditions of three-phase imbalance and dynamic heavy load for extended periods. Traditional transformer control and protection strategies are mostly based on passive responses using overall three-phase parameters, making it difficult to accurately identify early degradation trends in single-phase insulation and to implement precise local power flow transfer when specific phases deteriorate.

[0034] To address the aforementioned technical bottlenecks, the method disclosed in this embodiment constructs a complete technical closed loop encompassing multimodal feature sensing, electrothermal twin diagnosis, topology path optimization, and distributed collaborative control. Specifically, the method provided in this embodiment includes the following steps:

[0035] Step S1: Multimodal data acquisition and feature purification calculation.

[0036] In actual operation of the distribution network, the evolution of transformer insulation status is directly reflected in minute leakage currents, localized temperature rises in windings, and fluctuations in bus voltage. Therefore, the core of step S1 is to deploy sensing units in each phase of the transformer to synchronously collect the real-time leakage current of each phase. The winding hot spot temperature, bus reference phase voltage, and neutral point displacement voltage of the AC distribution network affect the real-time leakage current of each phase. Perform filtering and purification to obtain the actual leakage current of each phase. And calculate the dielectric loss tangent and the three-phase imbalance.

[0037] For example, the specific deployment of the sensing unit is as follows: a high-frequency broadband Rogowski coil is installed on the grounding wire of the transformer bushing end screen to acquire microampere-level leakage current signals in a contactless manner; fluorescent fiber optic temperature probes are pre-embedded at the estimated hot spot locations of different phases inside the transformer windings; and the bus voltage signal is acquired using the voltage transformer on the low-voltage side of the transformer. To ensure strict alignment of data in all dimensions on the time axis, the bottom-level data acquisition unit uses the power frequency zero-crossing point of the A-phase voltage as the global time reference, triggering all analog-to-digital converters to sample synchronously. The sampling frequency can be set to the Hertz level to meet the requirements of high-frequency transient capture.

[0038] It is important to note that industrial sites are characterized by significant electromagnetic interference, inverter switching noise, and high-frequency harmonics. Directly collected real-time leakage current... Containing significant burrs and baseline drift, it must undergo rigorous purification. In step S1, the real-time leakage current of each phase is... The process of performing filtering and purifying and calculating the dielectric loss tangent includes the following sub-steps:

[0039] First, based on the real-time leakage current within the sliding time window... The signal variance adaptive matching structure element is used to measure the real-time leakage current of each phase. The time-domain filtered signal is obtained by independently performing morphological opening-closing cascaded filtering and closing-opening cascaded filtering, and taking the average of the two. In engineering implementation, the length of the sliding time window can be set to include an integer number of power frequency cycles. The signal variance reflects the severity of the current signal fluctuation. When the variance is large, the system automatically matches a large-scale flat structure element; when the variance is small, it matches a small-scale structure element. Morphological opening operations can reduce positive impulse noise in the signal, and closing operations can fill in negative impulse noise in the signal. By performing parallel processing of opening-closing cascaded and closing-opening cascaded filtering and taking the average, random shot noise can be suppressed to the greatest extent, while preserving the power frequency fundamental frequency of the leakage current and the true high-frequency partial discharge characteristics, and suppressing phase shift errors.

[0040] Secondly, wavelet packet decomposition is performed on the time-domain filtered signal, and the high-frequency noise subband coefficients are reconstructed after threshold shrinkage processing to obtain the actual leakage current of each phase. In practice, a discrete wavelet basis with good orthogonality and compact support can be used to decompose the signal into multiple layers, dividing it into different frequency band nodes. Since white noise is uniformly distributed across the entire frequency band, while the energy of the actual insulation leakage signal is concentrated in specific low-frequency and mid-frequency sub-bands, the coefficients of the high-frequency noise sub-band are extracted and zeroed or reduced using soft or hard thresholding functions. Then, the time-domain signal is restored using a wavelet packet reconstruction algorithm, ultimately obtaining the pure actual leakage current. .

[0041] like Figure 2 The figure shows a comparison of waveform data before and after real-time leakage current adaptive morphology and wavelet packet fusion filtering purification. The horizontal axis represents time in milliseconds, showing the data evolution over three power frequency cycles from 0 to 60 milliseconds. The vertical axis represents the leakage current amplitude in microamps.

[0042] The light blue original leakage current curve in the figure shows violent spike-like fluctuations. At a specific moment, it is affected by the superposition of high-frequency pulses, and its peak amplitude surges to nearly 800 microamps. The overall waveform also shows obvious baseline shift, which objectively reflects the complex electromagnetic interference characteristics at the transformer operating site.

[0043] The red purified leakage current curve effectively filters out the high-frequency spike noise in the light blue curve and corrects the baseline drift, restoring the smooth shape of the sinusoidal fundamental wave. Its steady-state amplitude is maintained at around 490 microamps, verifying the technical effect of the purification and calculation module in extracting real power frequency characteristics under poor signal-to-noise ratio.

[0044] Simultaneously, observing the red curve reveals that, due to the inherent delays in multi-layer wavelet reconstruction and morphological window calculation, the purified waveform exhibits a slight phase lag of approximately 0.5 milliseconds compared to the theoretical peak, and a weak residual random fluctuation of about 5 microamps remains on the curve profile. This indicates that while the current fusion filtering method significantly improves the signal-to-noise ratio, it still has room for further upgrades and optimizations in terms of extremely high-precision phase-frequency synchronization and deep noise suppression.

[0045] Finally, the actual leakage current is determined based on the bus reference phase voltage. Phasor decomposition is performed to obtain the actual resistive and capacitive current amplitudes to calculate the dielectric loss tangent, and temperature rise characteristics are integrated to calculate the insulation degradation degree of each phase. In the signal processing logic, an orthogonal coordinate system is established with the phase of the bus reference phase voltage as the reference vector. The actual leakage current is extracted using a fast Fourier transform. The fundamental component of the voltage is projected onto a real axis in phase with the voltage to obtain the actual resistive current amplitude, and onto an imaginary axis leading the voltage by 90 degrees to obtain the capacitive current amplitude. The dielectric loss tangent is the ratio of the resistive current amplitude to the capacitive current amplitude. This tangent directly reflects the severity of dipole polarization loss and conductivity loss within the insulating dielectric. Based on this, the system uses the information entropy weighting method to calculate the information entropy of the dielectric loss tangent and the winding hot spot temperature, assigns appropriate weights for weighted summation, and outputs a continuously quantified insulation degradation index for each phase, thereby accurately grasping the insulation health status.

[0046] Step S2: Solving boundary constraints driven by digital twin.

[0047] After obtaining accurate status data, step S2 aims to: construct a digital twin model of the transformer's electrothermal coupling, and input the actual leakage current of each phase. The winding hot spot temperature, combined with the additional temperature rise correction caused by the three-phase imbalance, is used to calculate the real-time insulation aging rate of each phase; with the real-time insulation aging rate not exceeding the limit as a constraint, the maximum safe apparent power threshold, active power upper limit and reactive power adjustment margin range of each phase are solved, and each phase is divided into insulation deterioration phase and bearing healthy phase based on the real-time insulation aging rate.

[0048] Optionally, in step S2, the calculation of the real-time insulation aging rate of each phase specifically includes the following calculation process:

[0049] The first step is to calculate the total losses of each phase, including dielectric losses and winding copper losses. Transformer operating losses are the direct heat source causing temperature rise. Winding copper losses are calculated by multiplying the square of the effective value of the current flowing through each phase winding in real time by the DC resistance of the winding; dielectric losses are calculated based on the product of the dielectric loss tangent obtained in step S1, the system voltage, and the equivalent capacitance of the insulation. Adding these two values ​​together yields the total heat generation power of the transformer on that specific phase.

[0050] The second step involves using a third-order thermal network, inputting the total losses of each phase and the ambient temperature, and then solving for the updated winding hotspot temperatures by introducing an additional temperature rise correction term caused by three-phase imbalance. Within the twin model, the complex three-dimensional fluid heat transfer process inside the transformer is equivalent to a third-order lumped-parameter circuit model composed of thermal resistance and thermal capacity. The nodes represent the winding hotspots, the top oil layer, and the tank casing, respectively.

[0051] The total losses of each phase are injected into the winding nodes as a constant current source, and the ambient temperature is used as the grounding reference voltage. Specifically, when there is a severe three-phase imbalance in the AC distribution network, leakage flux will generate additional eddy current losses in the transformer structural components, leading to localized abnormal temperature rises. The system calculates this additional temperature rise correction term using the collected three-phase imbalance data and adds it to the fundamental thermodynamic equations, thereby solving for the updated hot spot temperatures of each phase winding in digital space with high fidelity.

[0052] The third step involves calculating the real-time insulation aging rate of each phase based on the Arrhenius thermal aging theory, by introducing a phase voltage offset correction term. The specific formula is as follows:

[0053] ;

[0054] In the formula, For real-time insulation aging rate, This is the pre-inflation factor for insulation aging. As the activation energy for insulation aging, Let be the ideal gas constant. This refers to the hot spot temperature of the phase winding. The reference phase voltage of the connected bus. This is the rated phase voltage amplitude. This is the voltage aging index.

[0055] This formula reveals the deep coupling relationship between the aging rate of insulation materials and temperature and voltage stress. The activation energy of insulation aging characterizes the energy barrier required for the thermal degradation or oxidation reaction of the insulating paper. When the hot spot temperature of the phase winding increases, the value of the exponential term increases sharply, leading to an exponential acceleration of the aging rate. Meanwhile, considering that the bus voltage may deviate due to reactive power shortage or excess in actual operation, a phase voltage deviation correction term introduced at the end of the formula corrects for the pure thermal aging rate with electrical stress. The higher the voltage, the greater the risk of partial discharge and the greater the rate of electrical tree growth in the insulation.

[0056] like Figure 3 A three-dimensional surface plot of the nonlinear response surface of insulation aging rate under transformer electrothermal coupling conditions, generated based on a digital twin model of transformer electrothermal coupling, is presented. In the figure, the horizontal axis represents the winding hot spot temperature in degrees Celsius, ranging from 60 to 140; the vertical axis represents the bus phase voltage in kilovolts, ranging from 9 to 11; and the vertical axis represents the insulation aging rate in relative values.

[0057] The overall surface morphology and color gradient characteristics reveal a significant nonlinear monotonically increasing trend in the response surface. Under rated conditions where the winding hot spot temperature is below 90 degrees Celsius and the bus phase voltage is close to 10 kV, the surface appears deep blue, and the insulation aging rate remains at a low baseline level.

[0058] As the temperature rises and the voltage deviation increases, the surface color gradually changes from blue to green and then to dark red, with the slope of the rise becoming increasingly steep. Especially under extreme electrothermal coupling conditions where the winding hot spot temperature reaches 140 degrees Celsius and the bus phase voltage rises to 11 kV, the relative value of the insulation aging rate surges to a peak value close to 8.

[0059] This data distribution objectively confirms the deep physical coupling between the thermal aging theory and the voltage offset correction term described in the specific implementation method, and proves that relying solely on thermodynamic parameters cannot accurately assess the health status of equipment. Introducing electrical stress characteristics can effectively improve the sensitivity to the local accelerated aging of single-phase insulation.

[0060] Meanwhile, it can be observed that in the high-stress boundary region of the surface near the highest temperature and highest voltage, there are slight grid-like undulations and local deformation data noise in the surface smoothness. This reflects that the current digital twin model driven by historical operating data has limitations in the coverage of extreme working condition sample features, which affects the boundary fitting accuracy to a certain extent. This indicates that there is still room for technological upgrades in the future expansion of the high-frequency transient extreme data sample library and the robustness optimization of nonlinear extrapolation algorithms.

[0061] Based on this, the controller performs inverse constraint solving. A maximum allowable aging rate limit for safe operation is set, and assuming that boundary conditions such as ambient temperature remain constant, a numerical approximation algorithm is used to deduce the maximum temperature rise that a phase can withstand under the current insulation condition, and further, the maximum allowable load current flowing through that phase. Thus, the system calculates the maximum safe apparent power threshold for each phase. Combined with the current power factor, this apparent power is decomposed to obtain the upper limit of active power and the reactive power adjustment margin range. Subsequently, based on whether the current insulation aging rate of each phase approaches or exceeds the safe threshold, the three phases are clearly divided into insulation-deteriorated phases that require load reduction and healthy phases with the potential to absorb load.

[0062] Step S3: Phase topology construction and graph neural network path optimization.

[0063] After identifying which phases need load transfer and their own limits, it is necessary to find the optimal power flow transfer path across the entire network. Step S3 specifically involves: constructing a phase-specific distribution network segment topology map, calculating the power flow transfer score of the healthy phases based on the insulation status and remaining carrying capacity of each phase using a graph convolutional neural network, and generating the phase-specific power flow transfer ratio.

[0064] Specifically, in step S3, the calculation of the power flow acceptance score of the healthy phase based on the insulation state and remaining carrying capacity of each phase using a graph convolutional neural network includes the following detailed data structure transformation and network inference process:

[0065] First, each transformer substation is abstracted as a node, and the tie lines are abstracted as edges. The insulation degradation degree and remaining carrying capacity of each phase corresponding to the node are extracted as node attributes, and the rated current carrying capacity and impedance of each phase corresponding to the tie line are extracted as edge attributes. Unlike traditional network topologies constructed solely based on three-phase average values, this embodiment decomposes the distribution network into a phase-separated topology structure containing three independent and parallel subgraphs: A, B, and C. Each transformer substation has a node mapping in each of the three subgraphs, and the node attribute matrix records in detail the real-time health status and power margin of each phase in that substation. The impedance attributes of the tie lines determine the magnitude of network losses that may occur during power flow transfer.

[0066] Secondly, a phase-specific current carrying capacity index is defined to quantify the potential of a single-phase current transfer. Its calculation logic is the weighted sum of the ratio of the remaining carrying capacity of a node to the rated capacity of a single phase, and the ratio of the line current carrying capacity margin to the line rated current carrying capacity. The formula is expressed as:

[0067] ;

[0068] In the formula, The index represents the ability to absorb current trends. This is the node capacity weighting coefficient. This is the line margin weighting coefficient. For the remaining bearing capacity, This is the rated capacity for a single phase. The rated current carrying capacity of the phase line, This represents the current current in the current phase line.

[0069] This formula physically ensures that the node selected as the receiver not only has sufficient spare capacity in its transformer, but also that the feeder system connected to it has sufficient transmission bandwidth, thereby avoiding local thermal overload or large-area voltage drop during the power transfer process.

[0070] Subsequently, the topology of the phase-specific distribution network segment is processed using a two-layer graph convolutional neural network. The first layer aggregates the phase attribute information of neighboring nodes to generate hidden layer features, and the second layer combines global topology information to output the power flow carrying capacity score for each phase of each node. The propagation formula of the graph convolutional neural network is:

[0071] ;

[0072] In the formula, This is the hidden layer feature matrix output from the first layer. To add a self-loop phase topology adjacency matrix, This is the degree matrix corresponding to the adjacency matrix. This is a matrix of node attributes. This is the first layer of trainable weight matrix. It is a non-linear activation function.

[0073] Graph convolution is essentially a spatial information aggregation mechanism. Traditional fully connected neural networks cannot perceive the spatial physical connections of a distribution network, while the above formula uses an adjacency matrix to weighted sum the attribute features of adjacent transformer nodes. Adding self-loops preserves the node's own characteristic state; the inverse square root of the degree matrix is ​​used to symmetrically normalize the adjacency matrix, preventing feature vector scaling explosion or gradient vanishing during multi-layer network propagation. The first layer of graph convolution enables information interaction within local neighborhoods (such as adjacent transformer substations), extracting local supply and demand characteristics. The second layer further expands the receptive field, transmitting information to the global network. Finally, the network output layer scores each healthy phase through a fully connected mapping. Nodes with higher scores have better insulation, more remaining capacity, and lower path impedance. The system generates phase power flow transfer ratio commands based on the scores and their relative proportions, using normalization.

[0074] Step S4: Decentralized sagging collaborative control and imbalance management.

[0075] After obtaining the optimal power flow transfer ratio, this strategy must be physically executed through the underlying converter control logic. Step S4 is as follows: Implement active power-frequency droop control and independent reactive voltage droop control for each phase in the distribution network. Based on the phase-specific power flow transfer ratio and various power constraint ranges, dynamically adjust the reactive power droop coefficients of the insulation-deteriorated phase and the receiving healthy phase to achieve power flow transfer; simultaneously suppress the neutral point displacement voltage through zero-sequence reactive power injection to address three-phase imbalance.

[0076] Specifically, in step S4, the logic for dynamically adjusting the reactive power droop coefficient between the insulated deteriorated phase and the receiving healthy phase to achieve power flow transfer, and simultaneously suppressing neutral point displacement voltage through zero-sequence reactive power injection, includes the following steps:

[0077] In microgrids or distribution network segments with a high proportion of distributed power sources, converters typically employ droop control that simulates the external characteristics of traditional synchronous generators to achieve autonomous power allocation without the need for high-speed communication. In this embodiment, for the insulation-degraded phase, the system increases the reactive power droop coefficient of the corresponding phase, while using the upper limit of active power and the reactive power adjustment margin range of the corresponding phase obtained in step S2 as upper and lower limit constraints.

[0078] An increase in the droop factor manifests physically as a greater sensitivity of the converter output voltage to reactive power output, meaning the droop curve becomes steeper. When the reactive power droop factor of a phase with deteriorated insulation is intentionally increased, the corresponding output node voltage will decrease while maintaining the same reactive power support, causing the power flow that was originally directed to that phase to be "squeezed out." Simultaneously, using the upper limit of active power calculated by the digital twin as the limiting parameter of the inner current loop creates strict upper and lower limit constraints, effectively preventing transient impacts from causing the load of the deteriorated phase to rebound beyond the limit.

[0079] Accordingly, for the healthy phase, the system reduces the reactive power droop coefficient of the corresponding phase according to the phase-specific power flow transfer ratio. Reducing the droop coefficient makes the droop curve smoother, which means that the voltage of the healthy phase exhibits stronger support rigidity. In parallel-operated network segments, power flow will naturally flow spontaneously from the insulation-deteriorated phase with weaker voltage support (larger droop coefficient) to the healthy phase with stronger voltage support (smaller droop coefficient) along the distribution line. During this process, the transfer amount of power flow in each phase is strictly carried out according to the ratio allocated by the graph convolutional neural network.

[0080] Due to the implementation of the aforementioned asymmetrical phase-by-phase power flow transfer, asymmetry in three-phase voltage and current will inevitably occur, leading to a shift in the neutral point potential of the distribution network, affecting personal safety and equipment insulation. Therefore, the system compensates for this through the neutral line or the independent arms of a four-arm converter. The calculation method for zero-sequence reactive power injection to suppress neutral point displacement voltage is as follows: the phase compensation reactive power is equal to the product of the set unbalance compensation gain coefficient and the difference between the measured neutral point displacement voltage and the neutral point reference voltage. The specific formula is as follows:

[0081] ;

[0082] In the formula, To compensate for reactive power, For unbalanced compensation gain coefficient, This is the measured neutral point displacement voltage. This is the neutral point reference voltage.

[0083] Through negative feedback closed-loop control, when the measured neutral point displacement voltage deviates from the reference zero potential, the system automatically calculates the zero-sequence reactive power command that needs to be injected into the grid and drives the underlying power electronic switching devices to output the corresponding zero-sequence compensation current. This forces the offset neutral point back to the equilibrium position at the space vector level, thus realizing the coordinated management of three-phase power quality under asymmetrical transfer conditions.

[0084] To enhance the robustness and foresight of the above technical solutions in complex and ever-changing real-world industrial environments, this embodiment further incorporates the following enhancement steps.

[0085] 1. Before step S4, there is also a phase insulation trend feedforward pre-coordination step, the purpose of which is to transform passive response into active defense.

[0086] Specifically, the system constructs a long short-term memory (LSTM) network phase prediction model. It takes as input historical time window data the insulation degradation degree of each phase, phase load, ambient temperature, and imbalance degree sequence data, and outputs the phase insulation degradation trend and phase power threshold change trend for a future set time period. The LSM network utilizes its unique forget gate, input gate, and output gate mechanisms to effectively capture the long-distance time dependence of temperature accumulation effects and insulation aging sequences.

[0087] When the system predicts that the maximum safe apparent power threshold of a certain phase will drop below a preset percentage threshold of the single-phase rated capacity, it can send a pre-coordination command to the corresponding healthy phase of the selected optimal receiving network segment before the time threshold, reserving power receiving margin, without waiting for an actual fault to occur. This is equivalent to diverting traffic in advance before congestion occurs, greatly improving the system's smoothness in responding to sudden heavy loads.

[0088] 2. In addition, the method also includes a phase-separated four-level flexible protection mechanism, which independently determines the protection level according to the calculated insulation degradation degree of each phase. There are a first threshold, a second threshold and a third threshold, and the first threshold is less than the second threshold, and the second threshold is less than the third threshold.

[0089] Level 1 Normal Operation: When the insulation degradation does not exceed the first threshold, it indicates that the equipment is in good health, the corresponding phase is operating at rated capacity, the reactive power droop coefficient is maintained at the benchmark value, and the system is in a normal economic dispatch state.

[0090] Level 2 Early Warning and Load Reduction: When the insulation degradation is between the first threshold and the second threshold, it indicates that the insulation has begun to show signs of early aging. At this time, the various power constraint intervals solved in step S2 are activated, and the reactive power droop coefficient is adjusted according to the preset first step length to limit the power inflow under heavy load conditions, thereby curbing the accelerated aging trend from the source.

[0091] Level 3 Cooperative Transfer Stage: When the insulation degradation degree is between the second threshold and the third threshold, it indicates that the insulation has entered a dangerous state. The phase power flow transfer ratio generation mechanism described in step S3 is immediately activated to forcibly intervene and reduce the load of the corresponding phase to below the safety threshold, ensuring that the transformer does not experience thermal breakdown.

[0092] Level 4 Impulse-Free Cut-off: When the insulation degradation exceeds the third threshold, it indicates that the phase is highly likely to experience insulation avalanche breakdown within a very short time. To avoid operational overvoltages caused by cutting off large loads, the load on the corresponding phase is first reduced to no-load using the generated phase power flow transfer ratio. After confirming no current and reaching the preset delay time, the single-phase disconnect device is triggered for cut-off. This soft cut-off method effectively protects other healthy phases on the same transformer from transient overvoltage impacts.

[0093] 3. To avoid potential over-limit risks in the scheduling instructions output by the artificial intelligence algorithm, step S3 includes a pre-strategy verification step after generating the phase power flow transfer ratio. Specifically, the transformer electrothermal coupling digital twin model is used to simulate and pre-analyze the phase load distribution and imbalance distribution of the entire network after applying the phase power flow transfer ratio.

[0094] This step is equivalent to conducting a virtual load switching test in digital space. The system verifies whether all loads connected to healthy phases exceed the corresponding phase's active power limit and maximum safe apparent power threshold, and simultaneously verifies whether the network-wide voltage imbalance and current imbalance after the transfer are within preset voltage imbalance limits and preset current imbalance limits, respectively.

[0095] If any check fails, it indicates that the current transfer strategy is too aggressive. The system iteratively adjusts the phase power flow transfer ratio of each phase by a preset step size until all constraints are met, ensuring that every instruction issued to the physical controller meets the safety constraints.

[0096] 4. The actual power distribution network structure, line impedance, and even load electricity consumption habits will drift slowly over time, and static control parameters are often difficult to maintain long-term optimality.

[0097] Therefore, the method also includes an online closed-loop self-optimization step for the control strategy. The system constructs a phase-dimension reinforcement learning environment, using the current insulation degradation, load rate, voltage deviation, and imbalance of each phase to form multi-dimensional state-space feature quantities. A multi-objective weighted reward function is constructed using insulation protection reward items, voltage quality reward items, load balancing reward items, and imbalance mitigation reward items. The specific formula for the reward function is:

[0098] ;

[0099] In the formula, To accumulate multi-objective reward values, , , , Each reward is assigned a weighting coefficient. An insulation protection bonus item characterizing the effect of suppressing insulation aging rate. Voltage quality reward items are used to characterize the degree of improvement in voltage deviation. The load balancing reward is a measure of the degree of load balancing among devices. An imbalance management reward item to characterize the degree of reduction in three-phase asymmetry.

[0100] Subsequently, a proximal policy optimization algorithm is used to train the control policy network. The output reactive power droop correction ratio, phase power flow transfer rate, and imbalance compensation gain coefficient are used as the action space for optimization, forming a closed loop of control execution, data acquisition, and parameter update. By limiting the step size of policy updates, the proximal policy optimization algorithm avoids performance collapse during optimization in a complex continuous action space, enabling the control policy to self-evolve online and continuously approach the optimal Pareto front of multi-objective cooperation.

[0101] like Figure 4 The graph shows the convergence curve of the multi-objective weighted reward function based on the proximal policy optimization algorithm in the online closed-loop self-optimization step of the control strategy. The horizontal axis represents the number of training epochs (in units), and the vertical axis represents the cumulative multi-objective reward value (dimensionless).

[0102] In the graph, the light blue curve represents the original reward value for each training round, and the dark blue solid line represents the smoothed reward value after moving average processing. In the early training stage, from 0 to 400 training rounds, because the control policy network is in a random exploration state, the output action commands such as the reactive power droop coefficient correction ratio and power flow transfer rate have not yet been optimized, leading to voltage over-limit or accelerated insulation aging in the system, and the cumulative multi-objective reward value is at an extremely low level of around -20.

[0103] As the number of training rounds increased to 800, the dark blue smooth curve showed a significant and rapid upward trend, and the reward value quickly climbed to around 150. This indicates that the near-end policy optimization algorithm effectively found an action strategy that could balance insulation protection and load balancing by continuously updating the network parameters.

[0104] After 1500 training rounds, the smoothed reward curve gradually converged and stabilized in the range of around 180, confirming that the closed-loop self-optimization strategy can find a control scheme with better overall performance in long-term iteration.

[0105] Meanwhile, observing the light blue original reward curve in the figure reveals that even in the steady-state phase of the later convergence stage, the original data still exhibits random oscillations of nearly 20 units, and the dark blue smooth curve also shows slight local downward fluctuations between 1600 and 2000 rounds. This objectively reflects that in the complex operating environment of a real power distribution network, characterized by highly nonlinearity and time-varying parameters, the current near-end strategy optimization algorithm with fixed parameter configuration still exhibits certain exploration jitter when dealing with boundary conditions where multiple objectives are mutually constrained. This implies that in future practical engineering applications, there is still room for further technical upgrades in terms of algorithm exploration noise attenuation mechanisms and dynamic learning rate optimization.

[0106] 5. Finally, when a distributed energy storage system with phase-separated access is configured within the AC distribution network segment using this method, the phase-separated control module of the energy storage converter of the distributed energy storage system is incorporated into a unified distributed control system for the network segment. During the transient process of power flow transfer according to the phase-separated power flow transfer ratio, the distributed energy storage system is scheduled to synchronously participate in phase-separated active power buffering and reactive voltage support.

[0107] Distributed energy storage systems possess millisecond-level four-quadrant power response capabilities. When a phase with deteriorated insulation sheds its active load, the energy storage system can instantly switch to a charging state to absorb excess power. When a receiving phase suddenly experiences a heavy load, causing a local voltage dip, the energy storage system can immediately generate reactive power to support the voltage, greatly mitigating the transient impact on the main power grid during large-scale power flow transfers.

[0108] An analysis of the shortcomings of existing technologies reveals that traditional transformer control and insulation monitoring often operate as two independent subsystems, lacking effective data interaction and coordination mechanisms. Most existing centralized distribution network management systems rely on the assumption of three-phase balance for power flow allocation. When faced with the problem of accelerated local aging of single-phase insulation caused by three-phase imbalance, they lack sufficient monitoring granularity and spatial scheduling flexibility. When a phase experiences severe aging, conventional operations often only involve standard overall capacity reduction or power outages, severely impacting power supply reliability.

[0109] The technical solution disclosed in this embodiment, through the organic combination of sensor purification, twin evaluation, graph convolution path optimization, and decentralized independent droop control, forms a self-healing defense system with extremely high engineering practical value in the field of power distribution networks. Through real-time phase-by-phase independent control, the solution integrates insulation protection and power flow scheduling. It not only effectively enhances the system's ability to detect early-stage local insulation degradation in transformers, but also achieves precise local control without power interruption, actively and smoothly diverting overload power flows from the deteriorated phase winding to the remaining healthy nodes in the network. This mechanism alleviates overheating pressure on specific windings, extends the overall lifespan of the transformer, and effectively suppresses neutral point voltage deviation under unbalanced transfer conditions, comprehensively improving the resilience and power quality of urban and industrial power distribution networks under heavy loads and extreme weather conditions.

[0110] Example 2:

[0111] In the process of decentralized control and power flow transfer in transformer network segments, when the system determines that a phase has insulation degradation through a digital twin model and decides to transfer power flow to other healthy phases, the control system needs to change the output characteristics of the converter by adjusting the reactive power droop coefficient.

[0112] However, power electronic converters have extremely wide bandwidth response characteristics and extremely fast switching speeds. During the transient process of dynamically adjusting the droop coefficient, they are highly susceptible to inducing high-frequency voltage fluctuations, active power oscillations, and transient inrush current surges in local distribution lines. For insulation media in a healthy state, these transient electrical shocks are within their tolerance range; however, for windings with deteriorated insulation, microcracks, moisture points, or partial discharge channels may already exist internally, and the dielectric strength of the insulation media has been significantly reduced.

[0113] If no special protection is taken during the transient process of power flow transfer, the transient electrical stress generated by the switching of control parameters can easily induce a sharp deterioration of partial discharge, or even directly lead to irreversible breakdown of the insulating medium.

[0114] To effectively overcome this hidden danger, this embodiment introduces an adaptive virtual impedance buffer and variable slope droop trajectory planning step based on the insulation transient vulnerability on the steady-state control architecture of Embodiment 1, so as to ensure the safe and stable transfer process.

[0115] Specifically, for the transient process of dynamically adjusting the reactive power droop coefficient to achieve power flow transfer in step S4, the control system first needs to accurately assess the target's ability to withstand electrical shocks relative to degradation.

[0116] This assessment process no longer relies solely on the absolute value of insulation aging, but focuses on the rate of insulation deterioration and the activity of high-frequency discharge. The system is based on the rate of change of the dielectric loss tangent calculated in step S1 within a historical time window, and the actual leakage current. The energy proportion of the mid-to-high frequency noise subband is used to calculate and quantify the insulation transient vulnerability index corresponding to the phase insulation's shock resistance.

[0117] The physical basis of this step is that the absolute magnitude of the dielectric loss tangent reflects the overall degree of insulation degradation, while its rate of change within a historical time window reflects the acceleration of insulation degradation. If the rate of change is extremely high, it indicates that the insulation of that phase is in a rapid thermal degradation deterioration path and is extremely insensitive to external voltage fluctuations.

[0118] Meanwhile, the actual leakage current The high-frequency noise sub-band extracted after wavelet packet decomposition often contains nanosecond to microsecond-level pulse clusters generated by partial discharges within the insulation. Calculating the proportion of the energy of this high-frequency noise sub-band in the total energy of the full-band signal can objectively quantify the current activity level of partial discharges.

[0119] For example, the process of calculating the insulation transient vulnerability index based on the parameters solved in step S1 can be implemented through a specific weighted mathematical model. The system first extracts time series data of a set duration from the past as a historical time window, and uses the least squares method to perform linear fitting on the dielectric loss tangent value sequence within the time window. The slope of the fitted line is then extracted as the rate of change of the dielectric loss tangent value within the historical time window.

[0120] Subsequently, Passevar's theorem was used to calculate the actual leakage current. The energy proportion of the high-frequency noise subband is obtained by dividing the discrete sequence energy at the high-frequency nodes corresponding to the wavelet packet by the sum of the energies of all frequency band nodes. Furthermore, the system calculates the insulation transient vulnerability index using the following formula:

[0121] ;

[0122] In the formula, The corresponding insulation transient vulnerability index, The preset weighting coefficient for the rate of change of dielectric loss. The rate of change of the dielectric loss tangent value within the historical time window. The preset high-frequency energy weighting coefficient, The actual leakage current The energy percentage of the mid-to-high frequency noise subband.

[0123] Calculated insulation transient vulnerability index This is a continuous real value. The larger the value, the worse the insulation resistance of the transformer winding is, and the stronger the buffer protection needs to be applied during the transient process of power flow transfer.

[0124] It is important to note that after obtaining the insulation transient vulnerability index, the system needs to map its physical effects to the underlying control loop. To this end, the system constructs an adaptive virtual impedance model that includes virtual resistance and virtual inductance, forward mapping the insulation transient vulnerability index to virtual resistance command values ​​and virtual inductance command values, respectively. This adaptive virtual impedance model is then connected in series and superimposed into the control loop for reactive power droop control. The virtual resistance dampens active power oscillations during the transient transition process, while the virtual inductance suppresses the rate of current change during the transient transition process.

[0125] In the control architecture of power electronic converters, reactive voltage droop control is an outer-loop control, and its output serves as the reference command for inner-loop voltage and current control. The virtual impedance is not a physically existing resistor or reactor, but rather, by modifying the voltage reference command output of the outer loop, the external characteristics of the converter's ports are equivalent to those of a corresponding physical impedance connected in series. This equivalent impedance has minimal impact during steady-state operation, but during transient processes caused by parameter abrupt changes, it can rapidly alter the system's pole distribution and damping characteristics.

[0126] Optionally, the system will use the insulation transient vulnerability index As input variables, the corresponding virtual resistance and virtual inductance command values ​​are calculated in real time using a preset proportional or exponential mapping function. Subsequently, a voltage sag correction term is introduced at the output of the reactive voltage droop control loop, calculated using the following formula:

[0127] ;

[0128] In the formula, This is the inner loop voltage reference command after virtual impedance correction. This is the initial voltage reference command directly output by the outer loop of the reactive voltage droop control. The virtual resistance command value, This represents the actual output current of the corresponding phase of the converter. The virtual inductance command value, It is a time variable.

[0129] Through this adaptive virtual impedance model, the control system achieves proactive defense against transient electrical stress. During power redistribution caused by power flow shifts, resonant oscillations are highly likely to occur between the distribution lines and filter capacitors. The virtual resistance command value... The voltage reference command is dynamically reduced according to the amplitude of the current. Since it is in phase with the current, it is equivalent to adding a purely resistive power consumption link inside the control system, thereby greatly improving the damping ratio of the system and quickly absorbing and attenuating the active power oscillation energy.

[0130] On the other hand, transformer windings themselves possess distributed capacitance and longitudinal inductance, and sudden changes in external current can generate extremely high inter-turn voltage gradients along the windings. The virtual inductance command value... Specifically designed to address the rate of change of current, when the output current is detected to be attempting to rise or fall sharply, the derivative term generates a large corrective negative feedback, forcing the voltage reference command to drop. This suppresses the rate of change of current at its source, preventing high current change rates from generating destructive inter-turn overvoltages on vulnerable windings.

[0131] like Figure 5 The graph shows a comparison of the suppression of three-phase current surge rate before and after applying adaptive virtual impedance during the power flow transition transient process. The horizontal axis represents time in milliseconds, recording the complete transient process from 0 to 100 milliseconds; the vertical axis represents the current surge rate in amperes per second.

[0132] In the figure, the light gray curve represents the traditional transfer control response without applying virtual impedance, while the dark green curve represents the optimized transfer control response after applying adaptive virtual impedance. The waveform evolution shows that after the power flow transfer command is triggered at 20 milliseconds, the light gray curve instantly exhibits violent oscillations, with its peak current mutation rate approaching 8500 amperes per second, accompanied by multiple large-amplitude decay fluctuations. This high-intensity transient impact can easily cause secondary physical damage to the already deteriorated transformer winding insulation.

[0133] In contrast, the dark green curve, at the same transfer trigger moment, shows a significantly suppressed peak current mutation rate of around 2800 amperes per second, and the oscillation energy rapidly decays and subsides within 10 milliseconds. This data comparison intuitively confirms the technical effectiveness of the adaptive virtual impedance model in the specific implementation, demonstrating that the methods of using virtual resistance to dampen active power oscillations and using virtual inductance to suppress current change rate can indeed effectively mitigate the electrical stress impact of power flow transfer transient processes on the fragile insulating phase.

[0134] Meanwhile, observing the detailed features of the dark green curve reveals that within the initial 20 to 25 milliseconds of triggering, the waveform still exhibits a certain degree of rising edge spikes and weak high-frequency jitter. This indicates that the current virtual impedance parameter mapping mechanism, when responding to extremely high-frequency transient electromagnetic physical processes, has its damping matching dynamic tracking accuracy limited by the sampling control cycle of the underlying hardware. This suggests that there is still room for further technological expansion and upgrades in the future for nanosecond-level transient overvoltage protection and the underlying broadband converter high-frequency decoupling control.

[0135] Furthermore, due to and It is based on dynamic calculation of the insulation transient vulnerability index. For phases with higher vulnerability, the system assigns greater virtual damping and inertia, thus achieving precise matching of protection levels.

[0136] Specifically, within the aforementioned control framework, to fundamentally eliminate the excitation sources causing transient oscillations, the system uses the insulation transient vulnerability index as a constraint, sets an adjustment time window and a maximum allowable change slope for the reactive power droop coefficient of the insulation-deteriorated phase, and generates a nonlinear, smooth transition droop coefficient adjustment trajectory, allowing the reactive power droop coefficient of the insulation-deteriorated phase to be continuously adjusted to the target value along the droop coefficient adjustment trajectory. Traditional distributed control, upon receiving a power flow transfer command, typically uses a step-based method to directly switch the original droop coefficient to the target droop coefficient. This step signal contains extremely rich broadband harmonic components, which are the direct cause of system transient impacts. To eliminate this cause, this embodiment performs trajectory planning for the droop coefficient adjustment process.

[0137] For example, the system establishes an inverse proportional relationship based on the calculated insulation transient vulnerability index. The higher the vulnerability index, the smaller the maximum allowable change slope is set for it, and the longer the adjustment time window is allocated.

[0138] Under these constraints, the system employs a fifth-order polynomial interpolation algorithm or an S-curve generator to generate a droop coefficient adjustment trajectory that smoothly transitions nonlinearly between the initial droop coefficient and the target droop coefficient. The mathematical characteristic of this trajectory is that it is not only continuous in the time domain, but its first derivative (rate of change) and second derivative (acceleration of change) also remain continuous throughout the entire adjustment time window, and the derivatives at the start and end points are constrained to zero.

[0139] When the control system executes the transfer strategy, the reactive power droop coefficient no longer undergoes abrupt changes, but instead slides slowly and continuously along this rigorously planned smooth curve to the target value. This mechanism makes the power flow from the insulation-deteriorated phase extremely gradual, and the power redistribution process is effectively lengthened over time. Since the system's input excitation does not contain any discontinuous high-frequency step components, combined with the damping buffering effect of the aforementioned adaptive virtual impedance model, the current and voltage of the entire distribution network segment can complete the transfer of the steady-state operating point within a very small fluctuation range, ensuring that the electrical stress borne by the transformer windings of the vulnerable phase remains at an extremely low level during this period.

[0140] like Figure 6 The graph shows the smooth transition trajectory curve of the reactive power droop coefficient with varying slope based on insulation transient vulnerability constraints. The horizontal axis represents time in milliseconds, showing the control cycle from 0 to 150 milliseconds, and the vertical axis represents the reactive power droop coefficient in per-unit values.

[0141] The light gray dashed line in the figure represents the traditional step switching trajectory without vulnerability constraints, while the dark blue solid line represents the smooth transition trajectory after applying variable slope planning.

[0142] At the 20-millisecond power flow transfer trigger moment, the light gray dashed line instantly changes from 0.02 to 0.05. This extremely high slope of change will directly induce high-frequency harmonics and transient electrical shocks in the system at the physical control level.

[0143] In contrast, the dark blue solid line did not experience a sudden abrupt change at this moment. Instead, the system automatically set an adjustment time window of 80 milliseconds based on the pre-calculated insulation transient vulnerability index. Its reactive power droop coefficient started from 0.02, and the first and second rates of change of the curve remained continuous at both the beginning and end, exhibiting a nonlinear smooth transition characteristic of slow acceleration followed by gradual deceleration, finally reaching the target value of 0.05 smoothly at 100 milliseconds. This trajectory distribution characteristic directly confirms that by lengthening the time axis of power redistribution, the transfer rate of transient power flow can be effectively limited, thereby avoiding the application of secondary electrical stress impact to the vulnerable insulation deteriorated phase winding due to abrupt changes in control parameters.

[0144] Meanwhile, careful observation of the rising and stable segments of the dark blue solid line reveals that the curve exhibits slight stepped sawtooth characteristics and minor numerical jitter after reaching steady state. This objectively reflects that when the underlying converter digital processor performs discretization interpolation calculations, it is limited by the control step size resolution of the actual main frequency and floating-point truncation error, resulting in a certain discretization deviation in the output trajectory. This indicates that in the future, this distributed control system still has certain technical upgrade potential for reconstructing high-precision continuous-time control algorithms for underlying hardware platforms with higher sampling frequencies.

[0145] It is important to note that traditional technologies, when dealing with heavily loaded transformers or transformers experiencing insulation alarms, typically rely on abrupt load shedding or reconfiguration. This not only interrupts the normal power supply to users but also generates strong operational overvoltages during the mechanical breaking process of the circuit breaker, causing secondary damage to the insulation of surrounding power equipment. Even the few studies that address flexible power flow control largely overlook the severe secondary electrical damage caused to already deteriorated insulation media by transient electrical shocks during control parameter switching. The technical solution detailed in this embodiment, however, deeply integrates underlying high-frequency power electronic control technology with the physical insulation state of the transformer, constructing a highly targeted transient buffer protection layer while performing power flow scheduling.

[0146] In summary, through the accurate calculation of the insulation transient vulnerability index, the dynamic superposition of the adaptive virtual impedance model, and the smooth planning of the variable slope droop trajectory in this embodiment, this technical solution can provide a targeted and flexible power flow transfer mechanism for transformer windings with severe insulation degradation without affecting the grid's continuous power supply and the ultimate goal of global power flow scheduling.

[0147] This approach significantly improves upon the passive impact situation caused by parameter mutations in traditional control, enabling the system to proactively reduce the load on vulnerable phases to delay aging while effectively avoiding secondary risks of insulation breakdown induced by the control behavior itself. This method demonstrates extremely high engineering safety value and application prospects in complex distribution network scenarios, including the renovation of aging transformer substations, extreme load dispatching during peak summer periods, and scenarios involving a large number of unbalanced nonlinear loads. It significantly improves the operational reliability of power equipment and the safety of system dispatching at the end of its entire lifecycle.

[0148] Example 3:

[0149] In actual power distribution network operation, especially under extreme conditions such as continuous high temperatures and peak summer demand, the entire distribution network is often under global heavy load. In such scenarios, all healthy phases of transformers within the topology network bear extremely high loads, and their remaining carrying capacity may be nearly exhausted. If severe insulation degradation of a phase is detected at this time, relying solely on the electrical side topology transfer mechanism in Example 1, the graph convolutional neural network will struggle to find a receiving network segment with sufficient margin and meeting safety constraints, thus facing the dilemma of the electrical transfer path being locked by physical capacity. If reactive power droop transfer is forcibly executed, it will inevitably lead to cascading overload of adjacent transformer areas and trigger large-area low-voltage over-limit; if no intervention is taken, the transformer in the degraded phase will quickly reach the thermal breakdown critical point.

[0150] To address the capacity bottleneck caused by relying solely on the electrical-side topology transfer mechanism, the method described in this embodiment is further extended to electrothermal synergistic optimization, specifically including a dynamic intervention step for active heat dissipation based on electrothermal digital twins and marginal cost game theory.

[0151] Specifically, based on the transformer electrothermal coupling digital twin model described in step S2, feedforward simulation is used to calculate the expected cooling range of the transformer active cooling system on the hot spot temperature of the winding under different operating levels.

[0152] In the physical structure of a transformer, the active cooling system typically includes multiple cooling fans and a forced oil circulation guide pump. These devices form an independent cooling circuit and have multiple discrete or continuous operating speeds according to control commands.

[0153] In normal operation, the start and stop of the cooling system are often triggered by hard logic based solely on a simple top-level oil temperature threshold, lacking foresight. In this embodiment, however, the system introduces the concept of electrothermal synergy into the low-level control. When an insulation degradation trend is detected, the system does not immediately issue a power flow transfer command, but instead calls a high-precision electrothermal coupled digital twin model to enter feedforward simulation mode.

[0154] This digital twin model integrates fluid dynamics and thermodynamic heat transfer equations. The system sequentially inputs different operating parameters of the cooling system into the model, which is equivalent to increasing the convective heat transfer coefficient and the mass flow rate of the cooling medium. Through transient thermodynamic simulations on the time axis, the model can accurately calculate the expected temperature reduction of the winding hotspot after one thermal time constant cycle, assuming the current ambient temperature and load remain constant, if the active cooling system is upgraded to the corresponding operating level. This simulation process precisely quantifies the heat dissipation potential of the physical device.

[0155] Optionally, the expected temperature drop is substituted into the calculation constraints of the real-time insulation aging rate of each phase to reverse-calculate the maximum safe apparent power threshold that can be equivalently increased at each operating level. After obtaining the expected temperature drop at different operating levels, the system begins to perform a cross-dimensional mapping from physical potential to electrical capacity. Based on the thermodynamic properties of the insulation material, the decrease in the winding hot spot temperature will directly lead to a slowdown in the rate of decrease in the polymerization degree of the internal insulation paper.

[0156] The system resubmits the simulated reduced temperature value into the function matrix of the real-time insulation aging rate. If the expected temperature drop causes the winding hot spot temperature to fall back, the corresponding insulation aging rate will decrease accordingly. Under the physical limit of ambient temperature, the system evaluates whether the reduced aging rate has fallen below the safe limit.

[0157] At this point, the system maintains the insulation aging safety limit as a rigid constraint, and in the digital twin model, the input current parameter is gradually increased in reverse until the insulation aging rate touches the safety limit again. The corresponding input apparent power at this point is the additional capacity space gained by physical cooling.

[0158] To accurately express this equivalent capacity expansion process, the system uses the following formula to calculate the maximum safe apparent power threshold that can be equivalently expanded at each operating level:

[0159] ;

[0160] In the formula, For the first The maximum safe apparent power threshold for equivalent amplification under operating conditions. This represents the baseline maximum safe apparent power threshold without active cooling intervention. The first result obtained from feedforward simulation of the digital twin model The expected cooling rate corresponding to the operating speed. This is the initial value of the currently measured winding hot spot temperature. The ambient temperature.

[0161] In practical applications, the engineering and physical significance of this formula lies in its direct conversion of the transformer's mechanical air-cooling / water-cooling capabilities into its electrical load-bearing capacity. This translates to the equivalent amplified maximum safe apparent power threshold. Essentially, this is a form of "virtual capacity." Without altering any electrical connection topology of the system, simply by accelerating internal heat dissipation, windings that were previously in an overload or deterioration-prone range can regain the internal thermal capacity to legally and compliantly handle larger loads. This mechanism provides heavy-load distribution networks with a local self-healing method that does not rely on external grid support.

[0162] It's also important to note that while increasing the operating speed of the active cooling system can gain virtual capacity, this comes at a cost. Forcing all fans and oil pumps to operate will significantly increase the internal power consumption of transformer auxiliaries, and prolonged high-load operation will accelerate the wear and tear on mechanical moving parts such as bearings. Therefore, the system must find the most economical and reasonable balance between on-site physical cooling and the transfer of electrical power across the entire network.

[0163] Specifically, we construct an active heat dissipation marginal cost function based on the auxiliary power consumption and mechanical wear of the transformer active heat dissipation system, and a power flow transfer marginal cost function based on the transmission network loss and the insulation aging of the line's health.

[0164] For example, in order to achieve comprehensive optimization of economic and security factors across multiple dimensions, the system quantifies the costs of these two different technical paths into the economic dimension and constructs the following marginal cost game difference formula for comparative decision-making:

[0165] ;

[0166] In the formula, For the first The marginal cost difference function between active cooling and power transfer at different operating speeds. This represents the unit electricity price for the current period. For the first The auxiliary active power consumption of the active cooling system under operating conditions. To optimize the time step for the set control, The economic loss penalty coefficient for mechanical wear. For the first Mechanical wear and tear depreciation index at different operating gears. The increase in active power due to line transmission losses caused by implementing the corresponding capacity power flow transfer. This is the economic penalty coefficient for insulation life degradation. This refers to the additional insulation aging depreciation index generated by the healthy phase if the corresponding capacity power flow transfer is performed.

[0167] In practical applications, the active cooling marginal cost function (the first half of the right-hand side of the formula) primarily measures the direct energy consumption bill of cooling equipment and the implicit costs of its lifespan reduction. The power flow transfer marginal cost function (the second half of the right-hand side of the formula), on the other hand, comprehensively calculates the conductor transmission losses caused by transmitting loads over long distances to other distribution areas, as well as the insulation lifespan reduction of previously healthy transformers at distant locations due to prolonged exposure to additional loads. The purpose of constructing this cost difference function is to enable distributed control systems to make rational economic decisions based on the entire lifecycle cost, avoiding the resource waste caused by recklessly prioritizing power supply.

[0168] Specifically, before generating the phase power transfer ratio in step S3, the marginal cost of active heat dissipation under each operating level is compared with the corresponding marginal cost of power transfer. When it is determined that the total remaining carrying capacity of the healthy phase in the entire network is lower than the capacity to be transferred, or the marginal cost of active heat dissipation is lower than the marginal cost of power transfer, an instruction to increase the operating power is preferentially issued to the transformer active heat dissipation system to physically reduce the actual winding hot spot temperature. This execution logic establishes the priority mechanism for intervention.

[0169] When the first condition is triggered, namely when the total remaining carrying capacity of the healthy phase of the entire network is lower than the capacity to be transferred, due to the objective constraints of the physical network capacity, even if the economic cost of active heat dissipation is higher than the cost of power flow transfer, the system will still prioritize active heat dissipation to intervene in order to avoid equipment overload and power outage.

[0170] When the second condition is triggered, although the entire network has capacity, in the economic game, the system, after calculating the cost difference function, finds that the overall cost of solving part of the heat problem locally by increasing the fan speed is far lower than the network loss and depreciation caused by scheduling remote resources. At this time, the control system sends an upgrade operation command to the transformer temperature control box through the communication bus, using the frequency converter driver to increase the speed of the fan or oil pump, thereby enhancing the physical heat exchange rate and truly and physically reducing the winding hot spot temperature.

[0171] like Figure 7 This figure presents a Pareto front distribution diagram of the marginal cost game between active cooling intervention and electrical power flow transfer under extreme heavy load conditions. The horizontal axis represents the comprehensive cost of active cooling in yuan, ranging from 20 to 220, and the vertical axis represents the comprehensive cost of power flow transfer in yuan, ranging from 50 to 450.

[0172] The light gray dots in the figure represent a large set of policy scheduling solutions generated by the system under different active cooling operation levels and different power flow transfer ratios. The dark red solid line and the dots form the Pareto optimal front.

[0173] The macroscopic distribution of the scattered points shows that as the overall cost of active heat dissipation increases, the overall cost of power flow transfer shows a significant downward trend. This objectively confirms the economic game logic that physical heat dissipation and cooling described in the specific implementation can be equivalently converted into electrical capacity.

[0174] Within the low active cooling cost range of 50 to 100 yuan on the horizontal axis, the power flow transfer cost corresponding to the vertical axis is generally higher than 300 yuan, indicating that the system mainly relies on long-distance power dispatch at this time, resulting in high costs for line transmission network loss and additional insulation aging of the receiving phase.

[0175] When the horizontal axis increases to around 150 yuan, the power flow transfer cost corresponding to the vertical axis decreases to around 150 yuan. The dark red solid line shows a clear inflection point in this area, which is the optimal marginal cost balance point that takes into account both auxiliary power consumption depreciation and remote network loss.

[0176] Simultaneously, observing the deep red Pareto front line reveals that it is not a smooth, continuous inverse proportional curve; rather, it exhibits stepped flatness and slight numerical fluctuations in local intervals. This indicates that the current computational model is limited by the mechanical characteristics of the discrete operating speeds of the transformer fan and oil pump, and the resulting non-dominated solution set still carries a certain degree of discretization error. This suggests that the control system still has room for technological upgrades in the future, particularly in terms of introducing continuously variable frequency cooling control hardware and more refined continuous state-space optimization algorithms.

[0177] It should also be noted that once the active cooling system is activated and takes effect, the original electrical dispatching requirements will be fundamentally changed.

[0178] Specifically, using the equivalently amplified maximum safe apparent power threshold, part of the overflow load of the insulation-degraded phase is offset in situ, the reduced power flow transfer demand base is recalculated, and the phase power flow transfer ratio in step S3 is updated based on the reduced power flow transfer demand base.

[0179] The overflow load of the insulation-deteriorated phase refers to the portion of the current actual active load that exceeds the base maximum safe apparent power threshold. Due to active cooling intervention, the maximum safe apparent power threshold is effectively increased, and this increased safe capacity directly absorbs the corresponding overflow load within the transformer itself. The system subtracts the in-situ offset portion from the original total load to be transferred, resulting in a significantly reduced power flow transfer demand base.

[0180] Subsequently, the system re-inputs this significantly reduced base number into the graph convolutional neural network. Due to the reduction in the total transfer amount, the network can distribute the remaining load more easily across the entire network. Transfer paths that were originally rejected due to exceeding capacity limits may meet the conditions again, ultimately generating updated phase power flow transfer ratio instructions with higher feasibility and safety margin.

[0181] Based on existing technology analysis, traditional power distribution network dispatching and protection schemes suffer from insufficient cross-disciplinary collaboration, as electrical control and thermal management are typically two independent subsystems. When faced with equipment overload alarms, traditional electrical-side measures are often limited to adjusting transformer tap changers, reconfiguring the status of distribution network feeder switches, or even directly implementing load shedding and power outage operations when capacity is limited. These methods neglect the inherent cooling potential of transformers as physical entities.

[0182] The technical solution described in this embodiment successfully breaks down the professional barriers between electrical network power flow and physical thermal field. By constructing a deep electrothermal twin model and introducing marginal cost game theory, this solution can effectively expand electrical capacity by actively improving physical heat dissipation efficiency under extreme conditions where the electrical topology capacity is nearing its limit, thus achieving in-situ suppression of insulation aging. This closed-loop mechanism, which offsets electrical overload through physical cooling and thereby reduces external power flow transfer pressure, effectively alleviates the grid bottleneck of the distribution network during peak summer electricity consumption periods and avoids local grid paralysis caused by the depletion of available transfer paths.

[0183] This solution not only significantly reduces the overall insulation loss of all network equipment, but also optimizes the overall operating cost of the system while ensuring power supply continuity, providing a highly reliable engineering solution for building a modern smart distribution network with high resilience and strong self-healing capabilities.

[0184] Example 4:

[0185] like Figure 8 As shown in the figure, this embodiment provides a power equipment insulation collaborative monitoring and transformer network segment distributed control system.

[0186] In light of the background technology, AC distribution network transformers have long faced the problem of three-phase load imbalance. Traditional monitoring and control technologies can only rely on the overall parameters of the three phases for passive early warning, lacking the ability to assess the insulation status of individual phases and intervene in dynamic power flow. This system embodiment provides a hardware architecture solution with engineering feasibility by constructing a closed-loop system composed of edge sensing devices, digital twin hosts, regional autonomous computing networks, and underlying power electronic actuators.

[0187] A collaborative monitoring system for insulation of power equipment and a distributed control system for transformer network segments, applied to AC distribution networks, includes: a multi-modal acquisition and purification calculation module, a twin-driven and power constraint solving module, a phase-by-phase topology construction and path optimization module, and a phase-by-phase distributed droop and imbalance mitigation module. Among them:

[0188] The multi-modal acquisition and purification calculation module is used to deploy sensing units in each phase of the transformer to synchronously acquire the real-time leakage current of each phase. The winding hot spot temperature, bus reference phase voltage, and neutral point displacement voltage of the AC distribution network affect the real-time leakage current of each phase. Perform filtering and purification to obtain the actual leakage current of each phase. And calculate the dielectric loss tangent and the three-phase imbalance.

[0189] In actual power equipment hardware deployments, this module primarily relies on an edge data acquisition gateway and a multi-type high-precision sensor array. Specifically, by installing high-frequency, broadband Rogowski coils on the grounding circuit of each phase bushing of the transformer, real-time leakage current at the microampere level is acquired in a non-destructive through-core manner. High-temperature resistant fluorescent fiber optic temperature probes are pre-embedded at the estimated hot spot locations of each phase winding inside the transformer to obtain the winding hot spot temperature; the reference phase voltage of the bus and the neutral point displacement voltage are obtained by using the three-phase voltage transformer (PT) on the low-voltage side of the distribution transformer area.

[0190] To ensure strict synchronization and alignment of multi-source data across time, this module is equipped with a high-resolution analog-to-digital converter (ADC) and a field-programmable gate array (FPGA). The FPGA uses the zero-crossing point of the A-phase bus voltage at the power frequency as the global hard synchronization trigger signal to drive synchronous sampling across all acquisition channels. Subsequently, the built-in digital signal processor (DSP) runs high-speed calculation algorithms to process the noisy real-time leakage current. By performing cascaded morphological and wavelet packet filtering, high-frequency switching noise and power frequency background interference are effectively filtered out, resulting in a high signal-to-noise ratio output of the actual leakage current. The phasor calculation of the dielectric loss tangent and the three-phase imbalance is performed in real time within the DSP.

[0191] The twin-driven and power constraint solving module is used to construct a digital twin model of the transformer's electrothermal coupling, inputting the actual leakage current of each phase. The winding hot spot temperature, combined with the additional temperature rise correction caused by the three-phase imbalance, is used to calculate the real-time insulation aging rate of each phase; with the real-time insulation aging rate not exceeding the limit as a constraint, the maximum safe apparent power threshold, active power upper limit and reactive power adjustment margin range of each phase are solved, and each phase is divided into insulation deterioration phase and bearing healthy phase based on the real-time insulation aging rate.

[0192] In practical engineering applications, this module can be implemented by an industrial control computer (IPC) deployed within the intelligent transformer distribution cabinet or a smart fusion terminal for the distribution area. This physical terminal contains a pre-built electrothermal coupled digital twin software model characterizing the transformer's physical structure and material properties. During operation, the industrial control computer receives structured data output from the multi-modal acquisition and purification / calculation module via its internal high-speed data bus. This data is then used to analyze actual leakage current. The insulation dielectric loss characteristics reflected, combined with the measured winding hot spot temperature, are used by the industrial control computer to perform iterative calculations of the thermodynamic differential equations in real time. Considering the long-standing load asymmetry problem in actual power distribution networks, the physical terminal introduces a correction for the additional temperature rise caused by leakage magnetic eddy currents in transformer structural components due to three-phase imbalance into the calculation engine, thereby outputting a quantified real-time insulation aging rate for each phase.

[0193] Subsequently, the system executes a reverse optimization algorithm based on physical boundary constraints to obtain the power operating boundary of each phase under the premise of ensuring insulation safety, and generates a status identification register for each phase (marked as insulation deterioration phase or taking over healthy phase) to provide data support for subsequent power flow control.

[0194] The phase-by-phase topology construction and path optimization module is used to construct the phase-by-phase distribution network segment topology map. It uses a graph convolutional neural network to calculate the power flow acceptance score of the healthy phase based on the insulation status and remaining carrying capacity of each phase, and generates the phase-by-phase power flow transfer ratio.

[0195] This module is primarily deployed in the regional autonomous controller host of the distribution network, or distributed within a peer-to-peer communication network formed by adjacent distribution area controllers. In terms of hardware connectivity, each distribution area controller interacts with topology information and node status in real time via fiber optic industrial Ethernet or 5G low-latency slicing networks. Within the regional controller's memory unit, the traditional single-line distribution network diagram is expanded into a phase-separated distribution network segment topology diagram containing independent operating parameters for phases A, B, and C. The neural network acceleration chip (such as an NPU or tensor computing unit) on the controller is responsible for running the graph convolutional neural network model.

[0196] The hardware acceleration unit uses the insulation health tags, remaining carrying capacity, and measured impedance parameters of each network node as input tensors to perform deep aggregation of graph structure features. The system calculates a quantified power flow carrying capacity score for each healthy node with carrying capacity in the topology network and automatically converts it into digital control commands, namely the phase power flow transfer ratio, to ensure that the power flow transfer path has sufficient safety margin on the physical line.

[0197] The phase-distributed droop and imbalance management module is used to implement active power-frequency droop control and independent reactive voltage droop control for each phase in the distribution network. Based on the phase power flow transfer ratio and various power constraint ranges, it dynamically adjusts the reactive power droop coefficients of the insulation-deteriorated phase and the receiving healthy phase to achieve power flow transfer; simultaneously, it suppresses the neutral point displacement voltage through zero-sequence reactive power injection to manage three-phase imbalance.

[0198] This module is the physical execution end of the entire control system. Its main entity is a three-phase four-wire intelligent power electronic converter connected to the distribution network, such as a distributed energy storage converter (PCS) or a flexible power distribution device with a power electronic interface. These converters typically employ a four-arm topology and have the physical capability to independently control the three-phase currents and the neutral current. The main control microprocessor inside the converter updates the reactive power droop coefficient setpoint in the converter control algorithm in real time based on the phase power flow transfer ratio received from the communication interface.

[0199] For the phase with deteriorated insulation, the microprocessor increases its corresponding droop coefficient parameter and uses the active power upper limit setting control command limiting module; for the healthy phase, its droop coefficient is reduced accordingly. The adjustment of the command parameters causes the converter's underlying IGBT drive module to output a specific port voltage, thereby guiding the load current to spontaneously transfer from the deteriorated phase to the healthy phase on the physical conductor. In this transfer condition, the converter's fourth bridge arm (neutral bridge arm) outputs a zero-sequence reactive current with a specific amplitude and phase through pulse width modulation (PWM) technology based on the collected neutral point displacement voltage signal, clamping the offset neutral point potential back to the safe reference value, thus realizing closed-loop power quality management at the physical level.

[0200] In actual power distribution network operation, the control system of this embodiment achieves independent quantification of the insulation aging rate of each single-phase winding through the collaboration of multimodal sensing hardware and digital twin host, making up for the physical defect of insufficient perception of local degradation in traditional three-phase overall monitoring. When the system identifies local insulation risk, the regional computing network and the underlying intelligent converter work closely together to dynamically adjust the independent droop control external characteristics of each phase. Under the premise of ensuring the continuity of power supply on the user side, the power flow of the distribution network is actively and smoothly transferred from the single-phase vulnerable winding to the healthy node with capacity. Furthermore, the neutral point displacement voltage is physically suppressed by the zero-sequence current injection of the four-arm converter.

[0201] The deployment of this system effectively reduced the operating load of the transformer's deteriorated phase windings, alleviated the risk of thermal breakdown, slowed down the overall insulation aging process, and improved the operational safety and stability of AC distribution network equipment.

[0202] Example 5:

[0203] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0204] like Figure 9 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0205] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0206] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0207] The memory 103 stores a computer program corresponding to the power equipment insulation collaborative monitoring and transformer network segment decentralized control method of the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0208] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 9 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0209] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for collaborative monitoring of insulation in power equipment and decentralized control of transformer network segments, characterized in that, include: S1. Sensor units are deployed in each phase of the transformer to synchronously collect the real-time leakage current, winding hot spot temperature, bus reference phase voltage and neutral point displacement voltage of each phase. The real-time leakage current of each phase is filtered and purified to obtain the actual leakage current of each phase, and the dielectric loss tangent and three-phase unbalance are calculated. S2. Construct a transformer electrothermal coupling digital twin model. Calculate the total phase loss including dielectric loss based on the dielectric loss tangent. Input the total phase loss, the actual leakage current of each phase, and the winding hot spot temperature into the transformer electrothermal coupling digital twin model. Combine this with the additional temperature rise correction caused by three-phase imbalance to calculate the real-time insulation aging rate of each phase. With the real-time insulation aging rate not exceeding the limit as a constraint, solve for the maximum safe apparent power threshold, active power upper limit, and reactive power adjustment margin range of each phase. Based on the real-time insulation aging rate, divide each phase into insulation deterioration phases and healthy phases. S3. Construct a topology diagram of the phase distribution network segment, and calculate the power flow acceptance score of the healthy phase based on the insulation status and remaining carrying capacity of each phase using a graph convolutional neural network, and generate the phase power flow transfer ratio. S4. Implement active power-frequency droop control and independent reactive power voltage droop control for each phase in the distribution network. Based on the phase power flow transfer ratio and various power constraint ranges, dynamically adjust the reactive power droop coefficients of the insulation-deteriorated phase and the receiving healthy phase to achieve power flow transfer. Wherein: for the insulation-deteriorated phase, increase the reactive power droop coefficient of the corresponding phase, and at the same time use the upper limit of active power and the reactive power adjustment margin range of the corresponding phase obtained by solving as upper and lower limit constraints. For the healthy phase, the reactive power droop coefficient of the corresponding phase is reduced according to the phase power flow transfer ratio; simultaneously, the neutral point displacement voltage is suppressed by zero-sequence reactive power injection to manage the three-phase imbalance.

2. The method according to claim 1, characterized in that, Step S1, which involves filtering and purifying the real-time leakage current of each phase and calculating the dielectric loss tangent, specifically includes: Based on the adaptive matching structure element of the real-time leakage current signal variance within the sliding time window, morphological opening-closing cascade filtering and closing-opening cascade filtering are independently performed on the real-time leakage current of each phase, and the mean of the two is taken to obtain the time-domain filtered signal. Wavelet packet decomposition is performed on the time-domain filtered signal, and the high-frequency noise subband coefficients are reconstructed after threshold shrinkage processing to obtain the actual leakage current of each phase. Based on the bus reference phase voltage, the actual leakage current is decomposed into phasor values ​​to obtain the actual resistive and capacitive current amplitudes, which are then used to calculate the dielectric loss tangent. The insulation degradation of each phase is then calculated by integrating the temperature rise characteristics.

3. The method according to claim 1, characterized in that, In step S2, the calculation of the real-time insulation aging rate of each phase specifically includes: The total loss of each phase specifically includes the dielectric loss and the winding copper loss; Based on a third-order thermal network, the total loss of each phase and the ambient temperature are input, and an additional temperature rise correction term caused by the three-phase imbalance is introduced to solve for the updated hot spot temperature of each phase winding. Based on the Arrhenius thermal aging theory, a phase voltage offset correction term is introduced to calculate the real-time insulation aging rate of each phase. The specific formula is as follows: ； In the formula, The real-time insulation aging rate, This is the pre-inflation factor for insulation aging. As the activation energy for insulation aging, Let be the ideal gas constant. This refers to the hot spot temperature of the phase winding. The reference phase voltage of the bus is This is the rated phase voltage amplitude. This is the voltage aging index.

4. The method according to claim 1, characterized in that, In step S3, the calculation of the power flow acceptance score of the healthy phase based on the insulation state and remaining carrying capacity of each phase using a graph convolutional neural network specifically includes: Each transformer substation is abstracted as a node, and the connecting lines are abstracted as edges. The insulation degradation degree and remaining carrying capacity of each phase corresponding to the node are extracted as node attributes, and the rated current carrying capacity and impedance of each phase corresponding to the line are extracted as edge attributes. Define a phase-to-phase power flow carrying capacity index to quantify the potential of a single phase to take over the transferred power flow. Its calculation logic is the ratio of the remaining carrying capacity of the node to the rated capacity of the single phase, and the weighted sum of the ratio of the line current carrying capacity margin to the line rated current carrying capacity. The topology of the phase distribution network segment is processed by a two-layer graph convolutional neural network. The first layer aggregates the phase attribute information of neighboring nodes to generate hidden layer features, and the second layer combines global topology information to output the power flow carrying capacity score of each phase of each node.

5. The method according to claim 1, characterized in that, In step S4, the calculation method for suppressing the neutral point displacement voltage through zero-sequence reactive power injection is as follows: The phase compensation reactive power is equal to the product of the set unbalance compensation gain coefficient and the difference between the measured neutral point displacement voltage and the neutral point reference voltage.

6. The method according to claim 1, characterized in that, The step S4 is preceded by a phase insulation trend feedforward pre-coordination step: A long short-term memory network phase prediction model is constructed. The input data are the insulation degradation degree of each phase, phase load, ambient temperature and imbalance degree sequence data within the historical time window. The output data are the phase insulation degradation trend and phase power threshold change trend for a future set time period. When it is predicted that the maximum safe apparent power threshold of a certain phase will drop below a preset percentage threshold of the single-phase rated capacity in the future, a pre-coordination instruction is sent to the corresponding healthy phase of the selected optimal receiving network segment before the time threshold, reserving power receiving margin.

7. The method according to claim 1, characterized in that, The method also includes a phase-specific four-level flexible protection mechanism, which independently determines the protection level according to the calculated insulation degradation degree of each phase. The mechanism includes a first threshold, a second threshold, and a third threshold, wherein the first threshold is less than the second threshold, and the second threshold is less than the third threshold. Level 1 Normal Operation: The insulation degradation does not exceed the first threshold, the corresponding phase operates at rated capacity, and the reactive power droop coefficient is maintained at the benchmark value. Level 2 warning load reduction: When the insulation degradation is between the first threshold and the second threshold, the various power constraint intervals solved in step S2 are activated and the reactive power droop coefficient is adjusted according to the preset first step length to limit the power inflow under heavy load conditions. Level 3 Cooperative Transfer Stage: When the insulation degradation is between the second threshold and the third threshold, the phase power flow transfer ratio generation mechanism described in step S3 is activated to reduce the corresponding phase load to below the safety threshold. Level 4 Non-Impact Cut-off Stage: When the insulation degradation is higher than the third threshold, the corresponding phase load is reduced to no load by the generated phase power flow transfer ratio. After confirming that there is no current, the single-phase disconnection device is triggered to cut off the current after the preset delay time is reached.

8. The method according to claim 1, characterized in that, Step S3, after generating the phase-separated power flow transfer ratio, also includes a pre-strategy verification step: The transformer electrothermal coupling digital twin model was used to simulate and preview the phase load distribution and unbalance distribution of the entire network after applying the phase power flow transfer ratio; Verify whether all loads connected to the healthy phase exceed the active power limit and the maximum safe apparent power threshold of the corresponding phase, and simultaneously verify whether the voltage imbalance and current imbalance of the entire network after the transfer are within the preset voltage imbalance limit and the preset current imbalance limit, respectively. If any check fails, the phase power flow transfer ratio of each phase is iteratively adjusted by a preset step size until all constraints are met.

9. The method according to claim 1, characterized in that, The method also includes an online closed-loop self-optimization step for the control strategy: Construct a phase-dimension reinforcement learning environment, using the current insulation degradation, load rate, voltage deviation and imbalance of each phase to form multi-dimensional state space features; A multi-objective weighted reward function is constructed using insulation protection reward items, voltage quality reward items, load balancing reward items, and imbalance mitigation reward items; The control strategy network is trained using a near-end strategy optimization algorithm. The output reactive power droop coefficient correction ratio, phase power flow transfer rate, and imbalance compensation gain coefficient are used as the action space for optimization, forming a closed loop of control execution, data acquisition, and parameter update.

10. The method according to claim 1, characterized in that, When the AC distribution network segment in which this method is applied is equipped with a distributed energy storage system with phase-by-phase access: The phase control module of the energy storage converter of the distributed energy storage system is incorporated into a unified network segment distributed control system. During the transient process of power flow transfer according to the phase-by-phase power flow transfer ratio, the distributed energy storage system is scheduled to synchronously participate in phase-by-phase active power buffering and absorption and reactive power voltage support.

Images