An island micro-grid adaptive self-healing control method and system based on real-time health perception

Abstract

The application discloses a kind of based on real-time health perception's island microgrid adaptive self-healing control method and system, belong to microgrid control technical field. Including: collection microgrid electrical quantity data and carry out physical characteristic extraction, construct physical enhancement characteristic vector input SE-LightFCN model and carry out fault diagnosis, output fault type and confidence degree;Based on fault confidence, frequency deviation and energy storage state of charge calculate microgrid real-time health index, and compare with preset threshold value and execute hierarchical response, when health index is lower than critical threshold, trigger self-healing reconfiguration control;Activation agent based on proximal policy optimization algorithm, construct reward function containing energy storage state of charge dynamic virtual cost, output the active power set value of each distributed power supply and carry out self-healing control.The application can quickly diagnose fault under edge computing power limited and strong noise environment, consider the rapidity of frequency recovery and operation economy, improve the power supply reliability of island microgrid.

Description

Technical Field

[0001] This invention belongs to the field of microgrid control technology, and particularly relates to an adaptive self-healing control method and system for island microgrids based on real-time health perception. Background Technology

[0002] Island microgrids typically employ a multi-energy complementary topology consisting of diesel generators, gas turbines, photovoltaics, wind power, and energy storage, aiming to provide independent and reliable power supply to islands far from the mainland. Compared to large power grids, island microgrids exhibit significant characteristics of "low inertia and weak damping," and the increasing penetration rate of distributed power sources based on power electronic interfaces further exacerbates the frequency stability problem of the system under disturbances. Simultaneously, the island environment often presents harsh conditions such as strong winds and salt spray, making electrical equipment prone to complex faults such as high-impedance grounding and arcing, posing a serious threat to the safe operation of the microgrid.

[0003] In fault diagnosis, traditional physical model methods rely on the accuracy of line parameters, which can easily lead to missed or false diagnoses of high-impedance faults. Signal processing-based methods, such as Hilbert-Huang transform and wavelet transform, can enhance fault feature resolution, but their computational complexity increases exponentially with the number of decomposition layers, making them unsuitable for the limited computing resources of island edge devices. Existing data-driven methods based on convolutional neural networks have improved diagnostic efficiency, but most directly map the original voltage and current waveforms, lacking guidance from physical mechanisms. Their feature extraction capabilities significantly decrease under strong noise interference, resulting in insufficient diagnostic robustness.

[0004] In fault recovery and control, traditional network reconstruction methods often employ static optimization strategies such as genetic algorithms, which are computationally time-consuming and ill-suited to handling dynamic transient processes after a fault. Among existing deep reinforcement learning methods, deep Q-networks are only applicable to discrete action spaces, while deep deterministic policy gradients, although supporting continuous control, suffer from poor training stability. More importantly, existing research largely focuses on the single objective of frequency stability, neglecting the differences in fuel economic costs among different distributed power sources during fault recovery. In island scenarios, fuel relies on maritime refueling, which is extremely costly. How to rapidly smooth out frequency fluctuations while prioritizing the utilization of zero-marginal-cost renewable energy and limiting the output of high-cost fuel-powered units is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an adaptive self-healing control method for island microgrids based on real-time health sensing, comprising: Collect instantaneous voltage and current data at the point of common coupling of the island microgrid; Physical features are extracted from the collected data to construct a physical enhancement feature vector; The physical enhancement feature vector is input into the pre-trained SE-LightFCN fault diagnosis model, which outputs the fault type and fault confidence. Based on the fault confidence, microgrid frequency deviation, and energy storage state of charge, calculate the real-time health index of the microgrid. The microgrid's real-time health index is compared with a preset threshold, and a tiered response strategy is executed. If the microgrid's real-time health index is lower than a first preset threshold but higher than a second preset threshold, an early warning signal is issued. If the microgrid's real-time health index is lower than the second preset threshold, self-healing reconfiguration control is triggered. After triggering the self-healing reconfiguration control, the agent based on the proximal policy optimization algorithm is activated; A reward function is constructed that includes the dynamic virtual cost of energy storage state of charge. The agent performs self-healing control by outputting the active power setpoint of each distributed power source according to the current microgrid state.

[0006] Optionally, the construction of the SE-LightFCN fault diagnosis model includes: extracting local features of time series data through a one-dimensional convolutional layer; introducing a compression-activation module to compress feature channels and generate channel weight coefficients through global average pooling, and recalibrating the features of different channels; performing dimensionality reduction through a global average pooling layer, and outputting classification results through a Softmax layer.

[0007] Optionally, the calculation of the microgrid real-time health index comprehensively considers the system frequency stability index, voltage stability index, and energy adequacy index; wherein, the energy adequacy index includes an energy storage support force function, which is a function that decays nonlinearly when the energy storage state of charge deviates from a preset midpoint.

[0008] Optionally, the dynamic virtual cost of the energy storage state of charge is as follows: when the output of the renewable energy inverter is provided only by renewable energy, the virtual cost is set to zero; when the energy storage unit needs to discharge to support it, the virtual cost is proportional to the energy storage state of charge penalty factor, which increases nonlinearly as the energy storage state of charge decreases; when the energy storage state of charge is below the minimum safety threshold, a linear penalty is applied.

[0009] Optionally, the state space of the near-end policy optimization agent includes frequency deviation, frequency change rate, output power of each distributed power source and energy storage state of charge; the action space is the active power adjustment of the diesel generator, gas turbine and renewable energy inverter.

[0010] Optionally, the physical feature extraction includes: calculating the positive-sequence, negative-sequence, and zero-sequence components of voltage and current, calculating the power flow direction, and calculating differential dynamic features.

[0011] This invention also proposes an adaptive self-healing control system for island microgrids based on real-time health sensing, comprising: The data acquisition module is configured to collect electrical quantity data of the microgrid in real time. The edge computing module has a built-in processor and memory, which stores the SE-LightFCN model. It is used to extract physical features from the collected data and construct a physical enhancement feature vector. The physical enhancement feature vector is then input into the SE-LightFCN model for fault diagnosis, and the fault type and fault confidence are output. The decision control module is used to calculate the real-time health index of the microgrid based on the fault confidence, microgrid frequency deviation and energy storage state of charge, compare the real-time health index of the microgrid with a preset threshold to execute a graded response strategy, and activate an agent based on a near-end strategy optimization algorithm after triggering self-healing reconfiguration control, construct a reward function that includes the dynamic virtual cost of energy storage state of charge, and output the active power setpoint of each distributed power source. The actuators, including the underlying controllers of each distributed power source, are used to receive and respond to power commands issued by the decision control module.

[0012] Optionally, the decision control module interacts with the deep learning environment and simulation environment through interface files.

[0013] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0014] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention integrates physical features with a lightweight SE-LightFCN model, significantly compressing model parameters while enhancing fault feature capture capabilities under high-noise environments, achieving millisecond-level robust fault diagnosis at the edge. By constructing a real-time health index for the microgrid and implementing a hierarchical response mechanism, it can accurately distinguish between sub-healthy and critical states, effectively avoiding unnecessary tripping under non-serious faults. By introducing a near-end policy optimization agent that includes the dynamic virtual cost of energy storage charge state for self-healing reconfiguration control, the system autonomously learns to prioritize scheduling zero-marginal-cost renewable energy and limit the output of fuel-fired units. This significantly reduces fossil fuel consumption while rapidly restoring frequency stability, achieving synergistic optimization of operational economy and frequency stability, and possessing strong robustness to system parameter perturbations. Attached Figure Description

[0016] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the island microgrid topology according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the network structure of the SE-LightFCN fault diagnosis model according to an embodiment of the present invention; Figure 4 This is a spatial distribution comparison diagram of the original features and physically enhanced features under noise interference in an embodiment of the present invention. Figure 5 This is a comparison chart of the frequency recovery effect of the method of the present invention with that of PID and DDPG methods in an embodiment of the present invention; Figure 6 This is a schematic diagram illustrating the dynamic power allocation method among different power sources according to an embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0019] Example 1 like Figure 1 As shown, this embodiment provides an adaptive self-healing control method for island microgrids based on real-time health sensing, including: Collect instantaneous voltage and current data at the point of common coupling of the island microgrid; Physical features are extracted from the collected data to construct a physical enhancement feature vector; The physical enhancement feature vector is input into the pre-trained SE-LightFCN fault diagnosis model, which outputs the fault type and fault confidence. Based on the fault confidence, microgrid frequency deviation, and energy storage state of charge, calculate the real-time health index of the microgrid. The microgrid's real-time health index is compared with a preset threshold, and a tiered response strategy is executed. If the microgrid's real-time health index is lower than a first preset threshold but higher than a second preset threshold, an early warning signal is issued. If the microgrid's real-time health index is lower than the second preset threshold, self-healing reconfiguration control is triggered. After triggering the self-healing reconfiguration control, the agent based on the proximal policy optimization algorithm is activated; A reward function is constructed that includes the dynamic virtual cost of energy storage state of charge. The agent performs self-healing control by outputting the active power setpoint of each distributed power source according to the current microgrid state.

[0020] Furthermore, the construction process of the SE-LightFCN model includes: One-dimensional convolutional layers (1D-Conv) are used to extract local features from time-series data; A compression-excitation (SE) module is introduced to compress feature channels through global average pooling (GAP), generate channel weight coefficients, and recalibrate the features of different channels to enhance the weight of key fault features. A global average pooling layer is used to replace the fully connected layer for dimensionality reduction, and the classification result is output through a Softmax layer. The model's inference time is less than 2ms, and the number of parameters is 1 / 10 of that of a general 2D-CNN model.

[0021] Furthermore, the calculation formula for the microgrid real-time health index (MRHI) comprehensively considers the system frequency stability index, voltage stability index, and energy adequacy index. Among them, the energy adequacy index includes the energy storage support force function f(SOC), which is a Gaussian-like function centered at SOC=0.5. When SOC deviates from 0.5, the health index shows a non-linear decay.

[0022] Furthermore, the construction logic for the dynamic virtual cost of energy storage SOC is as follows: When the output of the renewable energy inverter (REI) is provided solely by wind and solar resources, the virtual cost is set to zero; When energy storage unit discharge support is required, the virtual cost is proportional to the SOC penalty factor; the penalty factor increases rapidly and nonlinearly as the SOC decreases. When the SOC falls below the minimum safety threshold, a linear penalty is applied to force a stop to discharge.

[0023] Furthermore, the state space of the PPO agent includes frequency deviation, frequency change rate, output power of each power source, and energy storage SOC; the action space includes the active power adjustment of the diesel generator, gas turbine, and renewable energy inverter. By interacting with the environment, the intelligent agent autonomously learns a strategy to prioritize rapid power ramp-up via inverters to fill the power deficit in the early stages of a frequency drop, while limiting the generation of additional gas turbines.

[0024] Furthermore, physical feature extraction includes: calculating the positive-sequence, negative-sequence, and zero-sequence components of voltage and current, calculating power flow direction, and calculating differential dynamic features.

[0025] This embodiment also discloses an adaptive self-healing control system for island microgrids based on real-time health sensing, including: The data acquisition module is configured to collect electrical quantity data of the microgrid in real time. The edge computing module has a built-in processor and memory, which stores the SE-LightFCN model. It is used to extract physical features from the collected data and construct a physical enhancement feature vector. The physical enhancement feature vector is then input into the SE-LightFCN model for fault diagnosis, and the fault type and fault confidence are output. The decision control module is used to calculate the real-time health index of the microgrid based on the fault confidence, microgrid frequency deviation and energy storage state of charge, compare the real-time health index of the microgrid with a preset threshold to execute a graded response strategy, and activate an agent based on a near-end strategy optimization algorithm after triggering self-healing reconfiguration control, construct a reward function that includes the dynamic virtual cost of energy storage state of charge, and output the active power setpoint of each distributed power source. The actuators, including the underlying controllers of each distributed power source, are used to receive and respond to power commands issued by the decision control module.

[0026] Furthermore, the decision control module interacts with the deep learning environment and simulation environment through interface files.

[0027] Example 2 This embodiment establishes a co-simulation environment using MATLAB / Simulink and Python / PyTorch, enabling millisecond-level data exchange between the two through interface files. The island microgrid topology is as follows: Figure 2 As shown, the island microgrid is set with a rated voltage of 380V and a rated frequency of 50Hz. The initial active power values ​​of each power source are set as follows: gas turbine 600kW, diesel generator 200kW, and renewable energy inverter 50kW.

[0028] The method described in Example 1 can be divided into three core layers: the perception and diagnosis layer, the evaluation and decision-making layer, and the execution and control layer.

[0029] 1. Perception and Diagnosis Layer: SE-LightFCN model construction, the network structure of the SE-LightFCN model is as follows Figure 3 As shown; To address the challenges of limited computing power and strong noise interference at the edge of islands, this invention constructs an SE-LightFCN model that integrates physical features.

[0030] (1) Physical feature fusion and preprocessing; Traditional deep learning directly uses raw voltage and current waveforms, which are easily affected by noise. This invention introduces prior physical knowledge into the input layer.

[0031] Specifically, the positive-sequence, negative-sequence, and zero-sequence components of voltage and current, power flow direction, and differential dynamic characteristics are extracted using the symmetrical component method.

[0032] Let the collected three-phase voltage be , , The ordinal components are calculated using the following formula: ; in .

[0033] Power Flow characteristics calculate the active and reactive power of each branch; Differential dynamic features ( ) Calculate the first-order difference between the current time and the previous time of the wave ( ).

[0034] Experiments show that introducing physical features significantly improves the inter-class separation of fault samples in the feature space. For example... Figure 4 As shown, even under strong noise, the physical enhancement characteristics keep the clusters of different fault types compactly separated. The fault types include combinations of four AC positions (Cnv: converter; Dsl: diesel generator; Gas: gas turbine; Com: common power generation bus) and four AC fault types (AG: single-phase grounding; AB: phase-to-phase short circuit; ABG: two-phase grounding short circuit; ABC: three-phase short circuit), as well as DC side faults and normal operating conditions.

[0035] (2) SE-LightFCN network structure; Convolutional layers: Two 1D-Conv layers are used, with a kernel size of 3, a stride of 1, and the activation function is ReLU.

[0036] SE module: for feature maps First, channel descriptors are obtained through global average pooling (GAP): ; Then, weights s are generated through two fully connected layers: .

[0037] Finally, recalibrate the features: This allows the model to automatically focus on channels sensitive to asymmetric faults, such as negative-order components. Classification output: After dimensionality reduction via the GAP layer, the fault category is output through Softmax.

[0038] (3) Performance verification; Random noise was introduced into the test set and t-SNE visualization analysis was performed. Traditional WPT-SVM feature extraction takes more than 2ms, which is difficult to meet real-time requirements; general 2D-CNN has high accuracy but sluggish inference. In contrast, the LightFCN of this invention takes less than 2ms for inference and has a diagnostic accuracy of more than 97%, perfectly adapting to the computing power constraints of island edge terminals.

[0039] 2. Assessment of the decision-making level: MRHI and graded response; To prevent the system from overreacting to minor disturbances, this invention establishes the Microgrid Real-Time Health Index (MRHI).

[0040] (1) Definition of MRHI model; MRHI is a comprehensive indicator that is calculated across three dimensions: Electrical stability index ( ): Reflects the transient maintenance capability of a microgrid system under disturbances.

[0041] ( (for weights) Among them, voltage offset Frequency offset .

[0042] Fault severity index ( ): Fault type probability vector output by LightFCN Perform weighted calculations. Set the damage coefficients for different failure types. In this embodiment, a three-phase short circuit... Two-phase short circuit Single-phase grounding .

[0043] ; Energy adequacy index ( ): Reflects the system's survivability as an isolated system after fault isolation.

[0044] ; Among them, the energy storage support force is a nonlinear function based on the battery's state of charge (SOC). The state of health is highest when the state of charge (SOC) is 50%; the reserve capacity rate .

[0045] In conclusion, MRHI is defined as the system comprehensive health index at time t : ; wherein, is a weight coefficient of each sub-index, and . This function can dynamically reflect whether the system is still capable of maintaining the current topology operation after a fault occurs.

[0046] (2) Hierarchical response mechanism; Sub-healthy state (50 < MRHI < 85): the system issues an early warning but does not take any action.

[0047] Critical state (MRHI < 50): the system triggers reconstruction control immediately.

[0048] Simulation verification: Scenario 1: A high-resistance grounding fault occurs in phase A of the diesel engine at t=5.01s (minor fault). The confidence of LightFCN rises to 0.98, MRHI drops to 74.86 (sub-healthy region), the system only issues an early warning and maintains operation, which avoids unnecessary power outages.

[0049] Scenario 2: A three-phase short-circuit fault occurs at the outlet of the diesel engine at t=5.01s (serious fault). The confidence of LightFCN rises to 0.99, MRHI plummets to 19.81 (critical region), and a Level 2 response is triggered immediately. At t=5.05s, the protection acts to remove the faulty diesel generator, and then the PPO agent intervenes.

[0050] 3. Execution control layer: economic frequency recovery based on PPO; After the fault is removed, the system faces a huge power deficit. This embodiment adopts the proximal policy optimization (PPO) algorithm for secondary frequency regulation.

[0051] (1) Reinforcement learning environment construction; State space: includes frequency deviation , frequency change rate df / dt, output power of each unit, and energy storage SOC.

[0052] Action space: continuous action space, corresponding to power adjustment instructions of each controllable power source (gas turbine, energy storage, wind and photovoltaic inverter).

[0053] (2) Reward function design (Reward Function); In order to guide the agent to achieve the dynamic balance between reliability and economy during the millisecond-level frequency recovery process, the present invention designs a reward function including frequency stability, dynamic economy and safety constraints.

[0054] ; Frequency stability reward ( The purpose is to enable the system frequency to return to its rated value without error, while suppressing oscillations during the recovery process.

[0055] ; Dynamic economic incentives ( ): By constructing a dynamic cost model based on SOC state awareness, the optimal output of the renewable energy inverter (REI) can be determined.

[0056] ]; in, This refers to the unit power generation cost of a diesel engine. It is the unit power generation cost of a gas turbine. It is the REI dynamic virtual cost, the REI's output. Real-time power of wind and solar power and energy storage charging and discharging power Composition. Since wind and solar power are cost-free, the cost of REI (Resource Isolation) depends primarily on the state of the energy storage units. SOC (State of Charge) penalty factor. The definition is as follows: ; when Virtual cost when provided solely by wind and light The intelligent agent will prioritize the use of renewable energy. When energy storage discharge support is required ( Virtual costs and Proportional. When SOC is high, The cost is very small, and the overall cost of REI is still lower than that of fuel, so intelligent agents tend to generate more. When SOC is low, The cost of fuel rapidly increased, even exceeding that of fuel, so the intelligent agent limited the output of the REI (Refrigerant Ion) and instead increased the output of the diesel / gas turbine to prevent over-discharge of the battery.

[0057] Safety boundary penalty ( To prevent microgrid collapse, hard constraints are imposed to punish over-limit behaviors.

[0058] ; Specifically, when the battery level falls below a minimum threshold, a significant linear penalty is imposed, forcing the agent to stop discharging.

[0059] (3) Analysis of control effect; like Figure 5 As shown, after the hair was removed at t=5.05s: PPO strategy (this invention): The high-response inverter is rapidly activated in 5.1s to maintain the lowest frequency point (Nadir) at 49.76 Hz, and then smoothly recovers to 50 Hz without overshoot in 5.4s.

[0060] DDPG strategy: Due to the lack of new energy incentives, it relies on a high-inertia synchronous machine, which causes Nadir to drop to 49.6 Hz and oscillate near the steady state.

[0061] PID strategy: Due to hysteresis, Nadir drops to 49.45 Hz, which is close to the 49.5 Hz load reduction threshold, posing a risk of secondary crash.

[0062] (4) Economic analysis; like Figure 6 As shown, in terms of power allocation, the PPO agent rapidly instructs the inverter output to climb and stabilize at 213kW within 5.1 seconds to fill the gap (an increase of over 160kW); while the high-cost gas turbine (GT) only maintains at 637kW (an increase of no more than 40kW). The increase in renewable energy power generation is more than four times that of the gas turbine. This indicates that the agent has successfully learned a strategy of prioritizing the scheduling of zero-marginal-cost wind, solar, and energy storage resources while limiting expensive fossil fuels, achieving rapid frequency stabilization while minimizing fuel costs.

[0063] (5) Robustness verification; Considering the uncertainty of system inertia caused by parameter drift in actual operation, the working condition of ±40% perturbation of the underlying VSG virtual inertia was verified. The results show that the strategy of the present invention can adaptively adjust the active power target value without retuning the parameters, keeping the frequency deviation within a safe range.

[0064] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0065] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0066] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An adaptive self-healing control method for island microgrids based on real-time health sensing, characterized in that, include: Collect instantaneous voltage and current data at the point of common coupling of the island microgrid; Physical features are extracted from the collected data to construct a physical enhancement feature vector; The physical enhancement feature vector is input into the pre-trained SE-LightFCN fault diagnosis model, which outputs the fault type and fault confidence. Based on the fault confidence, microgrid frequency deviation, and energy storage state of charge, calculate the real-time health index of the microgrid. The microgrid's real-time health index is compared with a preset threshold, and a tiered response strategy is executed. If the microgrid's real-time health index is lower than a first preset threshold but higher than a second preset threshold, an early warning signal is issued. If the microgrid's real-time health index is lower than the second preset threshold, self-healing reconfiguration control is triggered. After triggering the self-healing reconfiguration control, the agent based on the proximal policy optimization algorithm is activated; A reward function is constructed that includes the dynamic virtual cost of energy storage state of charge. The agent performs self-healing control by outputting the active power setpoint of each distributed power source according to the current microgrid state.

2. The method according to claim 1, characterized in that, The construction of the SE-LightFCN fault diagnosis model includes: extracting local features of time series data through a one-dimensional convolutional layer; introducing a compression-activation module to compress feature channels and generate channel weight coefficients through global average pooling, and recalibrating the features of different channels; performing dimensionality reduction through a global average pooling layer, and outputting classification results through a Softmax layer.

3. The method according to claim 1, characterized in that, The calculation of the microgrid real-time health index comprehensively considers the system frequency stability index, voltage stability index, and energy adequacy index; wherein, the energy adequacy index includes an energy storage support force function, which is a function that decays nonlinearly when the energy storage state of charge deviates from a preset midpoint.

4. The method according to claim 1, characterized in that, The dynamic virtual cost of the energy storage state of charge is as follows: when the output of the renewable energy inverter is provided only by renewable energy, the virtual cost is set to zero; when the energy storage unit needs to discharge to support it, the virtual cost is proportional to the energy storage state of charge penalty factor, which increases nonlinearly as the energy storage state of charge decreases; when the energy storage state of charge is below the minimum safety threshold, a linear penalty is applied.

5. The method according to claim 4, characterized in that, The state space of the near-end strategy optimization agent includes frequency deviation, frequency change rate, output power of each distributed power source, and energy storage state of charge; the action space is the active power adjustment of the diesel generator, gas turbine, and renewable energy inverter.

6. The method according to claim 1, characterized in that, The physical feature extraction includes: calculating the positive-sequence, negative-sequence, and zero-sequence components of voltage and current, calculating the power flow direction, and calculating differential dynamic features.

7. An adaptive self-healing control system for island microgrids based on real-time health sensing, characterized in that, include: The data acquisition module is configured to collect electrical quantity data of the microgrid in real time. The edge computing module has a built-in processor and memory, which stores the SE-LightFCN model. It is used to extract physical features from the collected data and construct a physical enhancement feature vector. The physical enhancement feature vector is then input into the SE-LightFCN model for fault diagnosis, and the fault type and fault confidence are output. The decision control module is used to calculate the real-time health index of the microgrid based on the fault confidence, microgrid frequency deviation and energy storage state of charge, compare the real-time health index of the microgrid with a preset threshold to execute a graded response strategy, and activate an agent based on a near-end strategy optimization algorithm after triggering self-healing reconfiguration control, construct a reward function that includes the dynamic virtual cost of energy storage state of charge, and output the active power setpoint of each distributed power source. The actuators, including the underlying controllers of each distributed power source, are used to receive and respond to power commands issued by the decision control module.

8. The system according to claim 7, characterized in that, The decision control module interacts with the deep learning environment and simulation environment through interface files.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-6.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-6.

Images