Alternating current and direct current hybrid micro-grid control system based on transformer area direct current and flexible power supply

Abstract

The invention relates to the field of power grid control, in particular to an alternating current and direct current hybrid micro-grid control system based on transformer area direct current and flexible transformer area, which mainly comprises a high-precision synchronous acquisition sensor, a solid-state circuit breaker, converters (AC / DC and DC / DC), an edge computing node, a light storage integrated model, a flexible load plug and play model, an energy storage grouping rule and an island division model. According to the scheme, data are acquired by deploying hardware such as a sensor and a solid-state circuit breaker, a flexible topology is constructed, a power grid state is predicted by using algorithms such as a Gaussian-Newton method and an LSTM, a control strategy is optimized based on a power cost model and a heuristic algorithm, and alternating current and direct current cooperative stability is realized through a converter and droop control. Faults are quickly responded by means of traveling wave distance measurement and a solid-state circuit breaker, and finally, multi-scale optimization is driven by a digital twinning and MPC algorithm, so that a software and hardware collaborative microgrid control system is formed.

Description

Technical Field

[0001] This invention relates to the field of microgrid control, and more particularly to a hybrid AC / DC microgrid control system based on DC-flexible distribution transformers. Background Technology

[0002] With the advancement of the "dual carbon" goal, the AC / DC hybrid microgrid at the distribution substation level has become the core form of distribution network upgrade because it is compatible with DC energy sources such as photovoltaics and energy storage, as well as DC loads such as electric vehicle charging piles and data centers.

[0003] However, current technology faces multiple bottlenecks:

[0004] First, there is a lack of AC / DC coordinated control mechanism. Traditional independent control strategies cannot jointly regulate AC / DC bus voltage fluctuations and converter power flow, resulting in DC bus ripple reaching ±10% of the rated value and AC side frequency deviation exceeding ±0.5Hz, requiring manual intervention and making it difficult to achieve dynamic balance of "source-grid-load-storage".

[0005] Second, fault response and islanding control are inefficient. Traditional static islanding division results in a DC critical load recovery time of several minutes. The phase angle difference error of the loop closing operation is greater than 10°, which can easily cause the inrush current to exceed twice the rated value.

[0006] Third, the renewable energy absorption capacity is limited. The experience-based scheduling of photovoltaic and energy storage leads to a curtailment rate of more than 10%. The single energy storage cluster strategy results in a charging and discharging efficiency difference of more than 15%, and the comprehensive network loss of the distribution area is more than 20% higher than the theoretical value. Fourth, there is insufficient plug-and-play compatibility. The access of new equipment requires manual configuration of parameters (time > 30 minutes), and the lack of real-time modeling capabilities leads to lag in dynamic adjustment.

[0007] The shortcomings of existing technologies in terms of flexibility, reliability, and economy urgently necessitate control methods that integrate the characteristics of DC and flexible power grids to achieve efficient absorption, rapid self-healing, and flexible collaboration of multi-source equipment.

[0008] To address the aforementioned issues, this invention provides a hybrid AC / DC microgrid control system based on DC-flexible distribution network. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a DC-DC hybrid microgrid control system based on transformer substations, which realizes flexible interconnection, efficient absorption, fault self-healing, and multi-timescale adaptive stable operation of the DC-DC hybrid microgrid.

[0010] To achieve the above objectives, one aspect of the present invention is to provide a hybrid AC / DC microgrid control system based on DC-flexible distribution networks, comprising the following modules:

[0011] Multi-source sensing and dynamic modeling: High-precision synchronous sensors are deployed at key nodes of the AC / DC hybrid microgrid in the distribution area to collect AC parameters, DC parameters, photovoltaic and energy storage status and flexible load characteristics; at the same time, a flexible dynamic topology architecture is constructed, a dediscretized photovoltaic and energy storage integrated model and a flexible load plug-and-play model are established, and a microgrid digital twin model is generated to adapt to different operating states.

[0012] Feasible domain characterization and prediction: Based on the photovoltaic-storage model, the Gauss-Newton method is used to iteratively predict the grid state and stability before and after loop closure, and the feasible domain of the transformer area is characterized by a hybrid coordinate system power flow algorithm; through a deep learning model, AC and DC power and load regulation potential are predicted based on historical data, and dynamic calibration is performed using real-time weather data.

[0013] Power optimization and hierarchical control: Construct a power cost model, train it using historical data and combine it with calibrated weather forecasts to optimize power generation planning; Based on the photovoltaic-storage plan, energy storage clustering rules and converter instructions, generate candidate schemes through heuristic algorithms, and after verification by direct and flexible constraints, use droop control or virtual synchronous machine algorithm to achieve frequency-voltage coordinated stability of AC and DC subgrids;

[0014] Fault Response and Islanding Control: The fault point is located by traveling wave ranging, the fault area is quickly isolated by solid-state circuit breakers, the islanding scheme is obtained by calling the solver, and the critical DC loads are prioritized; after the fault is repaired, the optimal loop closing path is generated by heuristic algorithm to reduce grid connection impact.

[0015] Digital twin closed-loop optimization: Real-time acquisition of operational data and comparison with the digital twin model, online adjustment of converter reference values ​​and energy storage power; automatic model updates and rapid re-optimization for new equipment access, achieving multi-timescale adjustment from seconds to hours, ensuring efficient and stable system operation.

[0016] The second aspect of this invention provides a control method for a hybrid AC / DC microgrid based on transformer substation DC-flexible operation, used to implement a hybrid AC / DC microgrid control system based on transformer substation DC-flexible operation, comprising the following steps:

[0017] S100: Deploy high-precision synchronous acquisition sensors at key nodes of the AC / DC hybrid microgrid in the distribution area, while constructing a flexible dynamic topology architecture, and establishing a dediscretized integrated photovoltaic and energy storage model and a flexible load plug-and-play model to generate a digital twin model of the microgrid to adapt to different operating states of the AC / DC hybrid microgrid.

[0018] S200: Based on the aforementioned photovoltaic-storage integrated model, the Gauss-Newton iterative method is used to predict the grid state and system stability before and after loop closure, thereby characterizing the feasible region of the area and determining the feasible operating range of the microgrid under different operating conditions. At the same time, a deep learning model is used to predict the AC / DC power and flexible load regulation potential in the future based on historical data, and the prediction results are dynamically calibrated through real-time weather data.

[0019] S300: Construct a power cost model for a smart microgrid collaborative control system, train the model using historical data, and use calibrated weather forecast data to predict future power generation and operating costs, thereby improving the accuracy and economy of power generation planning;

[0020] S400: Combining the photovoltaic and energy storage power generation plan, energy storage grouping rules and converter power commands, a heuristic algorithm is used to generate candidate control schemes; after verification by direct current and flexible constraints, the converter output is adjusted by droop control or virtual synchronous machine algorithm according to the active and reactive power coordinated control strategy to achieve frequency-voltage coordinated stability of AC and DC subgrids.

[0021] S500: Locates the fault point of the distribution network using traveling wave ranging technology, quickly isolates the fault area using solid-state circuit breakers, calls the solver to solve the fault recovery model of the distribution network, obtains the islanding scheme of the distribution network during the fault time, prioritizes the power supply of DC critical loads, and rationally allocates energy storage power stations in combination with clustering rules. Then, it generates the optimal loop path through heuristic algorithm to reduce grid connection impact.

[0022] S600: Real-time acquisition of microgrid operation data and comparison with digital twin model, online adjustment of converter reference values ​​and energy storage charging and discharging power, and automatic updating of digital twin model to trigger rapid re-optimization for new equipment access, achieving multi-timescale adjustment from second to hour, ensuring that AC / DC hybrid microgrids work efficiently and stably under various operating conditions.

[0023] The third invention provides a computer device and a readable storage medium, characterized in that the computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize a DC-DC hybrid microgrid control system based on a transformer substation.

[0024] Compared with the prior art, the present invention provides an AC / DC hybrid microgrid control system based on DC-flexible distribution transformers, which has the following advantages:

[0025] 1. This scheme achieves frequency-voltage coordinated stability of AC / DC subgrids through hybrid power flow calculation, flexible constraints and deep learning prediction, suppresses DC bus ripple to within ±3%, frequency deviation < ±0.1Hz, and dynamically balances "source-grid-load-storage".

[0026] 2. This scheme adopts traveling wave ranging + solid-state circuit breaker (breaking time <100μs), dynamically divides AC and DC islands based on feasible domain, prioritizes DC load protection, predicts phase angle difference and optimizes path before loop closing, fault recovery time <500ms, and inrush current ≤1.2 times rated value;

[0027] 3. This solution reduces the curtailment rate to below 3% by constructing a power cost model, combining heuristic algorithms and energy storage clustering strategies, improves the energy storage charging and discharging efficiency by 8%, and reduces the overall grid loss in the distribution area by 15%, achieving multi-objective optimization and economical operation.

[0028] 4. This solution updates equipment parameters in real time by relying on a digital twin model, triggering rapid re-optimization (calculation < 1s), and automatically adapting to power-response characteristics. The response time for new equipment access is reduced from 30 minutes to less than 1 second, meeting the requirements of flexible interconnection and flexible adjustment of DC-flexible power grids.

[0029] This invention addresses the core challenges of flexibility, reliability, and economy in hybrid AC / DC microgrids in distribution areas through a complete technology chain encompassing multi-source sensing, collaborative modeling, hierarchical control, and digital twin optimization. Field testing has verified that all technical indicators outperform traditional solutions, providing a feasible technical path for upgrading distribution grids in scenarios with high-penetration renewable energy integration. This approach demonstrates significant engineering application value and market potential. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a block diagram of the overall system architecture of the present invention;

[0032] Figure 2 This is a flowchart of the control method of the present invention;

[0033] Figure 3 This is a flowchart illustrating the fault response and islanding control of the present invention.

[0034] Figure 4 This is a flowchart of the digital twin closed-loop optimization process of the present invention;

[0035] Figure 5 This is a block diagram showing the correspondence between the hardware and the algorithm of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0037] This addresses the problems of insufficient AC / DC coordinated control, inefficient fault response, limited renewable energy absorption capacity, and poor compatibility with new equipment in traditional AC / DC hybrid microgrids in distribution areas.

[0038] This proposal suggests a hybrid AC / DC microgrid control system based on the DC-flexible distribution network.

[0039] To address the issues of lack of AC / DC coordinated control mechanisms, inefficient fault response and islanding control, and limited renewable energy absorption capacity;

[0040] Based on the dynamic partitioning of AC / DC islands within the feasible region, priority is given to DC loads, and the phase angle difference is predicted and the path is optimized before loop closure.

[0041] It can reduce the curtailment rate to below 3%, improve the energy storage charging and discharging efficiency by 8%, reduce the overall grid loss in the distribution area by 15%, and achieve multi-objective optimization and economic operation;

[0042] By relying on digital twin models to update equipment parameters in real time, the response time for new equipment access is reduced from 30 minutes to less than 1 second, meeting the needs of flexible interconnection and flexible adjustment of DC and flexible power grids.

[0043] This solution achieves multi-source sensing modeling, AC / DC coordinated control, dynamic islanding, and digital twin optimization, while also enabling flexible interconnection of microgrids, efficient absorption, rapid self-healing, and plug-and-play adaptive stable operation of multiple devices.

[0044] Example 1, as Figure 1 As shown, an illustrative example is provided for the visualization selection method of acupuncture bronze figure combined with 3D digital human body modeling provided in the embodiments of this application.

[0045] S100: High-precision synchronization sensors are deployed at key nodes of the microgrid in the distribution area to collect AC / DC parameters, photovoltaic and energy storage status, and load characteristics. A flexible dynamic topology architecture is constructed based on graph theory, and the node connection relationship is described by a directed graph. A dediscrete photovoltaic and energy storage model (including photovoltaic single diode formula and energy storage RC circuit) and a flexible load plug-and-play model are established. Digital twins are generated through MATLAB and Unity, with synchronization accuracy ≤1μs, modeling error ≤2%, and adaptability to multiple operating states.

[0046] S200: Iteratively predicts the grid state before and after loop closure using the Gauss-Newton method, and characterizes the feasible region of the transformer area using a hybrid power flow algorithm. Constraints include DC voltage ±5% and frequency ±0.2Hz. A bidirectional LSTM neural network is used to predict AC and DC power. Historical weather and load data are input, and the predicted value for the next hour is output. The validation set RMSE ≤3%, and it is dynamically calibrated through real-time meteorological data, with an error ≤3%.

[0047] S300: Construct a multi-objective model that includes photovoltaic-storage costs, energy storage losses, and converter losses, and train it using the Adam algorithm until the validation set MSE < 0.01. Combined with calibrated weather forecasts, optimize the power generation plan for the next 24 hours, group energy storage according to SOC, prioritize photovoltaic consumption, achieve a curtailment rate of ≤3%, reduce energy storage losses by 27%, and realize economic dispatch and multi-objective optimization;

[0048] S400: The upper layer generates multi-objective schemes through the NSGA-II algorithm to optimize photovoltaic power, energy storage grouping, and converter reactive power. The lower layer executes droop control and VSG algorithm. The photovoltaic system adopts MPPT+voltage droop, and the energy storage system simulates synchronous machine inertia. The AC / DC voltage fluctuation is ≤±2%, and the frequency deviation is ≤±0.1Hz, achieving coordinated and stable control.

[0049] S500: Traveling wave ranging and positioning accuracy ≤50m, solid-state circuit breaker breaking time <100μs, islanding based on MST algorithm, prioritizing the protection of critical DC loads. Before loop closing, pre-synchronization control optimizes the phase angle difference to <3°, inrush current ≤1.1 times the rated value, fault recovery time <500ms, improving system resilience;

[0050] S600: Utilizes the MPC algorithm to correct control deviations in real time. When a new device is connected, the digital twin model is updated within 500ms, triggering second-level re-optimization. Multi-timescale adjustment: Second-level suppression of voltage sags, minute-level adjustment of energy storage clusters, and hour-level updates to the power generation plan ensure dynamic system optimization with an adjustment error ≤1.5%.

[0051] Step S100 is used for multi-source sensing modeling and dynamic topology construction, providing an accurate data foundation and physical model support for subsequent feasible domain analysis and prediction, including:

[0052] S110:

[0053] S1101: Based on the topology of the AC / DC hybrid microgrid in the distribution area, sensors are deployed at energy conversion nodes (AC / DC converters, DC / DC converters), energy production nodes (photovoltaic array combiner boxes, energy storage battery clusters), and load nodes (DC charging piles, AC residential distribution boxes) to form a "three-layer sensing network".

[0054] Sensor configuration:

[0055] Communication side:

[0056] A three-phase voltage and current sensor (model: WT3000, accuracy class 0.2S, sampling rate 10kHz) is installed on the AC bus to collect voltage amplitude (range 0-400V), frequency (45-55Hz), phase difference (0-360°) and harmonic content (2nd-50th harmonic).

[0057] Frequency disturbance monitor (accuracy ±0.01Hz) detects power grid frequency fluctuations in real time;

[0058] DC side:

[0059] Hall current sensor (model: CSNP500, bandwidth 100kHz, accuracy ±0.5%) is deployed on the DC side of photovoltaic inverter and the DC end of energy storage converter to collect real-time current (range 0-500A).

[0060] High-precision voltage sensor (voltage divider type, accuracy ±0.1%, sampling rate 1kHz), installed on DC bus, monitors voltage ripple (peak-to-peak value ≤5V);

[0061] Photovoltaic panel temperature sensor (PT100, accuracy ±0.5℃), one sensor is deployed for every 10 photovoltaic panels, used to calculate the temperature correction coefficient;

[0062] The energy storage battery cluster SOC monitoring module (based on the extended Kalman filter algorithm, with an error of ≤2%) collects the voltage, current and temperature of individual cells in real time and transmits the data through the CAN bus;

[0063] Bluetooth 5.0 sensor (model: nRF52832), embedded in adjustable loads (such as V2G electric vehicles, smart air conditioners), broadcast parameters include: device type (Class A / B), rated power (kW), adjustable range (±%), response time (ms / second level);

[0064] Data transmission adopts the IEEE1588 precision clock synchronization protocol to ensure that the synchronization accuracy of multi-source sensor data is ≤1μs. At the same time, edge computing nodes (based on ARM Cortex-A72 architecture) aggregate data through ModbusTCP and MQTT protocols, and upload it to the cloud digital twin platform after preprocessing.

[0065] S120:

[0066] S1201: Define a directed graph G(N,E), where:

[0067] Node set N:

[0068] N AC= {AC bus, distribution network interface, AC load}, node attributes include voltage level (220V / 380V) and short-circuit capacity (MVA);

[0069] N DC = {DC bus, photovoltaic array, energy storage battery}, node attributes include rated voltage (380VDC) and maximum allowable power (kW);

[0070] N CONV ={AC / DC converter, DC / DC converter}, node attributes include converter type (voltage source type / current source type), rated capacity (kVA), and efficiency curve (%);

[0071] Node set E:

[0072] AC line side: The weight is the line impedance Z = R + jX (Ω), calculated based on the data of the transformer area geographic information system (GIS), taking into account the influence of temperature on the conductor resistance (R = R0(1 + αΔT), α = 0.004 / ℃);

[0073] DC line edge: weighted by line resistance R (Ω), inductance is ignored, voltage drop is calculated as follows:

[0074] ΔV = IR;

[0075] Converter edge: Bidirectional directed edge, weighted by the active power transmission limit P max (kW) and reactive power adjustment range Q range (kVar);

[0076] Topology changes are indicated by the on / off state (0 / 1) of a solid-state circuit breaker (SSCB, breaking time < 100 μs), such as disconnecting N in islanded mode. AC Connect to the distribution network (while deleting), or add a new flexible load node (while adding);

[0077] Establish the topology state transition matrix T ij =1 indicates that nodes i and j are connected, which is used to quickly calculate changes in power flow distribution;

[0078] S130:

[0079] S1301: Employs a single diode model; output power expression:

[0080]

[0081] Where: N s / N p G is the number of series and parallel connections. STC =1000W / m 2 T STC=25℃, q=1.6×10 -19 C;

[0082] k = 1.38 × 10 -23 J / K, A = 1.5 is the diode factor;

[0083] Considering the non-uniformity of spatial irradiance, the maximum power point (MPP) of each photovoltaic string is calculated using the particle swarm optimization (PSO) algorithm to avoid power loss caused by local shading;

[0084] The battery cluster adopts a second-order RC equivalent circuit model, and the terminal voltage equation is as follows:

[0085]

[0086] State variables: polarization voltage (R p (where τ is the polarization resistance and τ is the time constant);

[0087] Lifetime Loss Model: The equivalent total lifetime loss (DOD) for each charge-discharge cycle is... The objective function includes the loss cost c. DOD = 0.01 yuan / Wh·DOD;

[0088] Establish a load characteristic database, including:

[0089] For models with a shifting start-up time (such as water heaters, the start-up time can be delayed by Δt ≤ 2 hours);

[0090] Reduced power (e.g., air conditioner, which can reduce power ΔP ≤ 30% for 30 minutes);

[0091] Injection type (such as V2G electric vehicles, which can reverse power supply P≤5kW);

[0092] Mechanical loads (such as air conditioning compressors): Start-up delay t delay =3-5 minutes, adjust step size ΔP = 0.5kW;

[0093] 1%;

[0095] Upon connection, the load nameplate data is read using an automatic identification protocol (such as IEEE 2423) to generate an equivalent current source model I. LD (t)=P(t) / V dc (DC load) or (AC load);

[0096] S140:

[0097] S1401: It adopts a joint simulation architecture of MATLAB / Simulink and Unity3D. Simulink is responsible for electrical system modeling (photovoltaics, energy storage, converter), and Unity is responsible for 3D visualization and interaction.

[0098] Key component library:

[0099] Converter library: Includes AC / DC converter models based on PWM modulation (switching frequency 20kHz) and bidirectional DC / DC converter models (efficiency ≥96%).

[0100] Load library: integrates the power-time characteristic curve of adjustable loads (e.g., it takes 2 minutes for the charging power of an electric vehicle to linearly increase from 0 to 7kW);

[0101] Edge computing nodes send real-time data (voltage, current, SOC, etc.) to the cloud every 100ms. Unity receives the data via WebSocket and drives the virtual device to update dynamically.

[0102] Establish an error calibration module: compare the output power deviation between the physical system and the digital twin (e.g., when the error between the photovoltaic predicted power and the measured power is >5%), and automatically adjust the model parameters (e.g., photovoltaic panel temperature coefficient, energy storage internal resistance);

[0103] Based on the integrated photovoltaic-storage model generated by S100, the aforementioned feasible region boundary and prediction data provide a safe operating boundary and decision input for the generation of intelligent collaborative control strategies. S200 is used for feasible region characterization and multi-step state prediction, including:

[0104] S210:

[0105] S2101: AC side initialization: Assume AC bus voltage V AC_ref =1.0∠0° (per unit value), iteratively calculate the voltage phasors of each node using the Newton-Raphson method and Jacobian matrix elements;

[0106]

[0107] Convergence condition: Nodal power imbalance < 10 -5 per-unit value;

[0108] S2102: Based on the AC-side converter output power PAC / DC and the DC load P LD_dc Calculate the DC bus voltage:

[0109]

[0110] Among them, G dc_eq For the equivalent conductance of the DC network, consider the converter loss P. loss =0.02Pconv_in ;

[0111] S2103: Converter power limit: |P conv |≤P rated ,

[0112] DC bus voltage: V dc ∈[361V,399V](rated value 380V±5%);

[0113] Frequency deviation: f∈[49.8Hz,50.2Hz];

[0114] S220:

[0115] S2201: The systems on both sides of the loop closure point are a microgrid (parameter subscript m) and a distribution network (subscript g), respectively, and the state variable is the phase angle difference Δθ = θ. m -θ g Frequency difference Δf = f m -f g Voltage difference ΔV = V m -V g ;

[0116] Collect the mean data 100ms before loop closure. Calculate the initial deviation:

[0117]

[0118] S2202: Inrush current after loop closure (Ignoring the transient effects of frequency difference), where V avg =(V m +V g ) / 2, loop impedance Z loop =Z m +Z g (Z m Z is the equivalent impedance of the microgrid. g (Equivalent impedance of the distribution network);

[0119] Jacobi matrix elements:

[0120] Iterative updates:

[0121] Among them, I surge_limit =1.5I rated As the threshold, iterate to |I surge -I surge_limit |<0.1I rated stop;

[0122] S230:

[0123] S2301: Input layer: 12-dimensional feature vector, containing:

[0124] Weather data: Current irradiance (W / m²) 2 ), temperature (°C), wind speed (m / s), humidity (%), cloud cover (oktas);

[0125] Historical data: Photovoltaic power, energy storage SOC, and load power for the previous hour (one point every 15 minutes, for a total of 4 points);

[0126] Hidden layers: Bidirectional LSTM layers (2 layers, 128 neurons per layer), capturing the sequential dependencies of time series, with tanh as the activation function;

[0127] Output layer: 3D vector, predicting photovoltaic power (kW), energy storage adjustable power (kW), and flexible load adjustable amount (kW) for the next 15 minutes, 30 minutes, and 1 hour;

[0128] Training process:

[0129] Dataset: Data from a certain TV station area in 2024, totaling 8760 hours, divided into training set, test set, and validation set in a 7:2:1 ratio;

[0130] Preprocessing: Standardizing characteristics such as irradiance and temperature. The load data is normalized (to the 0-1 range);

[0131] Optimizer: Adam algorithm, learning rate 0.001, batch size 32, training period 50 epochs;

[0132] Performance metrics: The validation set RMSE is 1.8kW (photovoltaic power), 0.5kW (energy storage power), and 1.2kW (load adjustability), which meets engineering requirements (error ≤3%).

[0133] Based on the prediction results output by S200, an economical and efficient energy dispatch benchmark is formed by optimizing the photovoltaic-storage power generation plan and energy storage charging and discharging strategy. This provides the core objective function and constraints for the multi-objective optimization of subsequent control strategies. Step S300 is used for power cost modeling and power generation planning optimization, including:

[0134] S310:

[0135] S3101: Objective function definition: minC = C PV +C BESS +C conv +C load_shift ;

[0136] The costs for each item are as follows:

[0137] Cost of photovoltaic power generation: (Including equipment depreciation and maintenance costs), Δt = 15 minutes (0.25 hours);

[0138] Energy storage lifespan depletion costs: Cap BESS =20kWh (energy storage capacity), with the change in SOC equivalent to DOD;

[0139] Converter loss cost:

[0140] Converter efficiency η conv =0.96, η dc =0.97;

[0142] Load shifting incentive (negative cost): (Peak-valley electricity price difference) encourages the shifting of peak loads to off-peak periods;

[0143] Photovoltaic power generation constraints:

[0144] 0≤P PV (t)≤P PV_pred (t), -P BESS_max_chg ≤P BESS (t)≤P BESS_max_dis ;

[0145] SOC constraint: 20% ≤ SOC(t) ≤ 80%;

[0146] Converter power constraint: P AC / DC (t)≤S AC / DC_rated • cosφ (power factor cosφ ≥ 0.9);

[0147] S3102: Remove abnormal data: When the photovoltaic power is greater than 1.2 times the predicted value or the energy storage SOC is less than 10%, mark it as an invalid data point;

[0148] Feature engineering: Add "seasonal factors" (spring / summer / autumn / winter, one-hot encoding) and "weekday type" (weekday / weekend) to improve the model's ability to predict periodic loads;

[0149] Initialize model parameters: Use the Kaiming initialization method to avoid gradient vanishing / exploding.

[0150] Forward propagation: Input historical feature data for 7 days, output the cost-optimal power generation plan for the next 24 hours;

[0151] Backpropagation: Calculate the loss function (mean squared error MSE) and update the parameters using the AdamW optimizer (with weight decay);

[0152] Early stopping mechanism: If the validation set MSE does not decrease for 5 consecutive epochs, training is stopped to prevent overfitting;

[0153] Typical day (sunny): The error between the predicted power generation plan and the actual operating cost is 3.8%, and the curtailment rate decreases from 12% to 2.5%;

[0154] Extreme weather (heavy rain): The model automatically adjusts the energy storage discharge strategy to ensure power supply to critical loads, reducing load shedding by 40%;

[0155] S3103 scheduled updates: daily at 00:00, 06:00, 12:00, and 18:00, based on the latest weather forecast (resolution improved to 500m×500m) to re-optimize the 24-hour plan;

[0156] Event trigger: When the measured photovoltaic power deviates from the predicted value by more than 10% or the energy storage SOC is less than 25%, rolling optimization is initiated (optimization time domain of 4 hours, step size of 1 hour);

[0157] Photovoltaic-storage power generation plan: Photovoltaic active power command and energy storage charging and discharging power command every 15 minutes (e.g., 10:00-10:15, photovoltaic tracking MPPT, energy storage charging 5kW);

[0158] Converter control modes: When the AC / DC converter is connected to the grid, it operates in PQ mode (independent active and reactive power control), and switches to V / f mode (voltage and frequency support) when islanded.

[0159] Flexible load dispatch: During off-peak hours (23:00-7:00), priority is given to dispatching electric vehicle charging (7kW power), and during peak hours (12:00-18:00), the power of air conditioning is reduced by 15%;

[0160] Step S400 is used for the generation and real-time execution of hierarchical control strategies, including:

[0161] S410:

[0162] S4101: Algorithm Selection: The Non-Dominated Sorting Genetic Algorithm (NSGA-II) is adopted, with advantages including:

[0163] Supports parallel optimization of multiple objectives (minimum grid loss, maximum renewable energy utilization, and minimum energy storage loss);

[0164] Crowding degree calculation maintains the diversity of the solution set and avoids getting trapped in local optima;

[0165] Photovoltaic storage active power P PV_ref ∈[0,P PV_pred ], P BESS_ref ∈[-P chg_max,P dis_max ];

[0166] Converter reactive power Q conv_ref ∈[-Q max Q max ](Q max =S rated ·sinφ max , φ max =30°);

[0167] Number of energy storage clusters N group ∈{2,3,4,5};

[0168] Inter-group allocation strategy S alloc ∈{SOC priority, efficiency priority, capacity priority};

[0169] S420:

[0170] S4201: SOC Level Grouping: Energy storage power stations are divided into three groups according to SOC: high (>60%), medium (30%-60%), and low (<30%). The standard deviation of SOC of power stations within each group is ≤5%.

[0171] Efficiency grouping: Based on the energy storage charge and discharge efficiency curves (e.g., lithium battery efficiency > 95%, lead-acid battery efficiency 85%-90%), the batteries are divided into high-efficiency group and ordinary group.

[0172] Peak shaving task allocation: When network loss optimization is prioritized, high SOC groups undertake discharge tasks, and low SOC groups undertake charging tasks, allocated according to group capacity ratio. Among them, C group This represents the total energy storage capacity within the group.

[0173] Frequency modulation task allocation: When efficiency is prioritized, the high-efficiency group responds first to high-frequency regulation (cycle < 1 minute), and the ordinary group responds to low-frequency regulation (cycle > 5 minutes);

[0174] Power allocation is based on the proportion of regulating capacity: Power commands for each power station within the group are as follows: Let i be the adjustable capacity of power station i (SOC × capacity during discharge, (1-SOC) × capacity during charging);

[0175] S430:

[0176] S4301: Employs voltage droop control + MPPT correction: P PV_ctrl =P MPPT -k dc_droop (V dc -V dc_ref )k dc_droop =0.1kW / V, when the DC bus voltage is higher than the reference value (380V), the photovoltaic active power output is reduced to suppress voltage rise;

[0177] In islanded mode, a virtual synchronous generator (VSG) is used for control to simulate the inertia of a synchronous generator.

[0178]

[0179] Moment of inertia J = 2 kg·m 2 (Equivalent to a 100kVA synchronous machine), damping coefficient D = 100N·m·s / rad, maintaining frequency stability;

[0180] AC / DC converter: Based on voltage-reactive droop characteristic: Q conv =Q ref +k q (V AC_ref -V AC )k q =-10kVar / V, when the AC bus voltage is lower than the rated value, it automatically generates reactive power to raise the voltage to above 95%;

[0181] The harmonic current components are calculated using instantaneous reactive power theory, and the converter injects reverse harmonic current to make THD < 3%.

[0182] This hierarchical control architecture transforms global optimization results into executable real-time control commands, ensuring that the microgrid maintains dynamic equilibrium under multi-objective constraints.

[0183] Step 500 is used for fault response and islanding adaptive control, generating the optimal loop-closing path through a heuristic algorithm to reduce grid connection impact, including:

[0184] S510:

[0185] S5101: Install transient current sensors at both ends of the power distribution line (bus side M, load side N) to collect the arrival time t of the first traveling wave pulse after the fault. M ,t N ;

[0186] Fault distance calculation:

[0187] The total line length is L, the traveling wave velocity is v = 0.3 km / μs (overhead line) or v = 0.15 km / μs (cable), and the positioning accuracy is ≤ 50 m.

[0188] When a fault current greater than 5 times the rated value is detected, the SSCB is triggered to trip. The tripping process is as follows:

[0189] The pre-charged capacitor discharges, driving the mechanical contacts to separate rapidly (within 50μs);

[0190] An electric arc is generated between the contacts, and the arc is extinguished within 50μs by a magnetic blowout arc extinguishing device;

[0191] Once the disconnection is complete, a disconnection signal is sent to the control system (total delay < 100 μs);

[0192] By analyzing transient current waveforms using wavelet transform, we can distinguish between short-circuit faults (rich in high-frequency components) and inrush currents (dominantly low-frequency components) to avoid malfunctions.

[0193] S520:

[0194] S5201: Critical load: Hospital DC emergency power supply, data center UPS input terminal, priority 1 (absolutely no power interruption);

[0195] Adjustable load: Electric vehicle charging station (priority 2, charging can be paused);

[0196] Photovoltaic-storage unit: Photovoltaic array and energy storage converter (priority 3, ensuring continuous power supply);

[0197] Residential lighting load (priority 1);

[0198] Commercial air conditioning load (priority 2, power can be reduced by 30%);

[0199] Non-critical industrial loads (priority 3, can be completely cut off);

[0200] The Minimum Spanning Tree (MST) algorithm is used to construct the topology of isolated islands. The objective is:

[0201] DC islanding: includes at least one energy storage converter and photovoltaic array, with minimal path impedance from load node to power node;

[0202] AC islanding: The energy storage converter acts as the root node, prioritizing the connection of high-priority loads and disconnecting low-priority loads until power balance is achieved; 530:

[0204] 5301: The microgrid enters the grid connection preparation state, and the AC / DC converter switches to PQ-V / f hybrid mode to track the voltage phase and frequency of the distribution network in real time;

[0205] Calculate the phase difference Δθ = θ micro -θ grid Frequency difference Δf = f micro -f grid Voltage difference ΔV = V micro -V grid ;

[0206] The converter reference value is adjusted using a proportional-integral (PI) controller.

[0207] θ ref =θ grid +k pΔθ+k i ∫Δθdt;

[0208] f ref =f grid +k p Δf+k i ∫ΔfdtV ref =V grid +k p ΔV+k i ∫ΔVdt;

[0209] Control parameter: k p =0.5,k i =0.1, the target will reduce Δθ to <3°, Δf <0.1Hz, and ΔV <5% within 5 seconds;

[0210] When multiple loop paths exist, Dijkstra's algorithm is used to select the path with the minimum loop impedance to avoid excessive inrush current.

[0211] Within 10 cycles after loop closure, an adaptive filtering algorithm is used to suppress the non-periodic component in the inrush current to ensure that the protection device does not operate.

[0212] Step S600 is used for closed-loop adaptive optimization driven by digital twins. It forms a closed loop of "sensing-analysis-control-optimization" through digital twins, ensuring that the microgrid continuously optimizes its operating state when equipment changes or the environment changes. This is the core guarantee for the dynamic adaptability of the preceding steps. Specifically:

[0213] S610:

[0214] S6101: Prediction Time Domain N p =10 (1 second, sampling period 100ms), control time domain N c =5 (control quantity applied in the first 5 cycles);

[0215] State variable: x(t) = [V dc ,f,SOC,P PV ,P LD ], Control variables:

[0216] u(t)=[P BESS_ref Q conv_ref V dc_ref ];

[0217] Objective function:

[0218]

[0219] Weighting coefficients: λ1 = 100, λ2 = 50, λ3 = 1, balancing voltage frequency stability and control smoothness;

[0220] The quadratic programming (QP) algorithm is used for online solution, with a computation time of <50ms, which meets the real-time requirements;

[0221] S620:

[0222] S6201: When a new device (such as a DC charging pile) is connected, the device type, rated parameters, and communication interface information are broadcast via the IEEE 2030.5 standard protocol.

[0223] S6202: Edge computing nodes parse data and generate equivalent models of devices;

[0224] Charging pile: Constant power model P EV =7kW (charging mode) or P EV = -5kW (V2G mode), response time <100ms;

[0225] S6203: Update the digital twin model, triggering a rapid re-optimization:

[0226] Add new nodes to the topology graph and recalculate the AC / DC power flow.

[0227] The MPC model adds a state variable P. EV Optimize the regeneration of control sequences in the time domain;

[0228] If the actual operating parameters of the equipment deviate from the nameplate data by more than 10% (e.g., the actual power of the charging pile is 6.3kW), the model parameters are identified online using the recursive least squares method, and the database is updated.

[0229] S630:

[0230] S6301: Voltage dip (e.g., voltage drop to 85% due to motor startup);

[0231] The converter rapidly injects reactive power (response time < 50ms), formula:

[0232] Q comp =k vr (V ref -V meas )k vr =20kVar / V, the target is to restore the voltage to over 95% within 200ms;

[0233] Variations in photovoltaic irradiance cause DC bus voltage fluctuations > ±3%;

[0234] Adjusting the boost control value of the DC / DC converter, the calculation formula is as follows:

[0235] V boost_ref =V dc_ref +ΔV ripple ·kpi (1+T is 1 );

[0236] Proportional-integral parameter: k pi =0.5,T i =2 seconds, suppressing ripple to within ±2%;

[0237] When the weather forecast is updated to heavy rain, the predicted photovoltaic power output drops by 50%.

[0238] The S300 power cost model was re-run, the energy storage charging and discharging plan was adjusted, and the charging amount during off-peak hours was increased (e.g., charging 10kWh from 23:00 to 7:00) to ensure the discharge demand during high-load periods during the day.

[0239] System integration and testing:

[0240] The Advantech UNO-3083G features an Intel i7 processor and an integrated FPGA acceleration module (Xilinx Zynq-7000) to achieve synchronous sampling data processing and fast control algorithms (latency <100μs).

[0241] Supports Ethernet, RS-485, and CAN bus, and is compatible with OPCUA, MQTT, and Modbus protocols;

[0242] AC / DC converter: ABBPCS100, rated capacity 100kVA, switching frequency 20kHz, supports seamless switching of PQ / Vf modes;

[0243] DC / DC converter: TIC2000 series controller, efficiency ≥97%, supports maximum power point tracking (MPPT) and droop control;

[0244] Lithium battery cluster: CATL 51.2V / 200Ah, SOC monitoring accuracy ±2%, charge / discharge efficiency ≥95%;

[0245] Real-time operating system (RTOS): QNXNeutrino, task scheduling accuracy ≤10μs, running real-time tasks such as MPC and power flow calculation;

[0246] Human-Machine Interface (HMI): Developed based on Qt, it provides functions such as topology visualization, parameter setting, and fault alarm, and supports mobile access;

[0247] Signal processing: FFT harmonic analysis, wavelet transform (for fault detection);

[0248] Optimization algorithms: NSGA-II, PSO, Dijkstra (used for policy generation and path planning);

[0249] Control algorithms: VSG, droop control, MPC (for real-time adjustment);

[0250] Scenario testing:

[0251] 1. Conditions: Irradiance from 800 W / m 2 The sudden drop to 200W / m 2 (Simulated cloud cover), lasting 10 minutes;

[0252] Result: The DC bus voltage dropped from 382V to 375V (fluctuation -1.8%), and recovered to 380V±1% within 200ms;

[0253] The energy storage converter starts discharging within 50ms, increasing the output power from 0 to 30kW while maintaining the frequency at 50.0±0.05Hz;

[0254] 2. Conditions: The power distribution network connection is manually disconnected, entering island mode, with a total load of 60kW (40kW AC and 20kW DC).

[0255] Results: Solid-state circuit breaker breaking time 80μs, islanding completion time 300ms;

[0256] The AC islanding frequency stabilizes to 50.0Hz within 1 second, the DC bus voltage is 380V±2%, and there is no power interruption for critical loads;

[0257] 3. Conditions: Before loop closure, phase angle difference is 10°, frequency difference is 0.5Hz, and voltage difference is 8%.

[0258] Results: Pre-synchronization control reduced the phase angle difference to 2.5°, frequency difference to 0.08Hz, and voltage difference to 3% within 4.2 seconds;

[0259] The loop-closing inrush current was 1.1 times the rated value; the system did not activate any protection devices and returned to normal operation.

[0260] Experimental Example 1:

[0261] Experimental scenario: A hybrid AC / DC microgrid in a certain distribution area, including:

[0262] AC side: Distribution network interface (220V / 50Hz), residential load (total 20kW, power factor 0.9);

[0263] DC side: Photovoltaic array (peak power 50kW), lithium battery energy storage (capacity 20kWh, initial SOC 50%), DC charging pile (7kW×2);

[0264] Converters: AC / DC converter (rated capacity 30kVA), DC / DC converter (rated power 50kW);

[0265] Experimental objective: To verify the effectiveness of each step in the method and materials, from S100 to S600, with a focus on testing:

[0266] The modeling accuracy of digital twin models;

[0267] The effect of AC / DC coordinated control on voltage and frequency stability;

[0268] Fault response speed and the rationality of islanding;

[0269] Renewable energy utilization rate and grid loss optimization effect;

[0270] Experimental steps:

[0271] Hall current sensors (accuracy ±0.5%) are deployed on the DC side of the photovoltaic system, three-phase voltage sensors (accuracy 0.2S class) are installed on the AC bus, and SOC monitoring modules (error ≤2%) are connected to the energy storage cluster.

[0272] A hybrid AC / DC topology model was constructed, with photovoltaics using a single-diode model and energy storage using a second-order RC circuit model. The charging pile was set as an adjustable load (adjustable range 0-7kW, response time <100ms). A digital twin model was generated, synchronizing real-time data (sampling frequency 1kHz).

[0273] Recorded data:

[0274] Sensor type Measurement parameters Measured value Model predictions error Photovoltaic current sensor DC current 85A 83.5A +1.76％ AC voltage sensor bus voltage 221V 220.5V +0.23％ Energy storage SOC monitoring SOC 50.1％ 50.0％ +0.20％

[0275] Run AC / DC coordinated power flow calculation, set DC bus voltage constraint of 380V±5%, and converter response speed ≤10ms;

[0276] Predict photovoltaic power for the next hour using an LSTM model (inputting 3 days of historical data), and measure real-time irradiance (800W / m²). 2 )calibration:

[0277] Recorded data:

[0278]

[0279]

[0280] A power cost model is constructed, with the objective function including the cost of photovoltaic storage (0.3 yuan / kWh), energy storage loss (50 yuan / cycle), and converter loss (0.1 yuan / kWh).

[0281] Input the calibrated weather forecast, optimize the power generation plan for the next 24 hours, and set the energy storage SOC to be maintained at 30%-70%;

[0282] Recorded data:

[0283]

[0284] The upper layer generates a control scheme through the PSO algorithm: photovoltaic power is fully fed into the grid, energy storage discharges 10kW from 10:00 to 12:00, and charging piles charge during off-peak hours (23:00-7:00);

[0285] The lower layer implements active and reactive power coordinated control, and the DC bus voltage is maintained at 380±2V through the DC / DC converter, with AC frequency fluctuation ≤±0.1Hz;

[0286] Recorded data:

[0287]

[0288]

[0289] Simulating a short-circuit fault in an AC line, the traveling wave ranging and positioning time is <1ms, and the breaking time of the solid-state circuit breaker is 80μs.

[0290] Divide the system into DC islands (photovoltaic + energy storage + charging piles) and AC islands (residential loads + VSG energy storage), and prioritize power supply to charging piles.

[0291] After the fault was repaired, the phase angle difference of the closed loop was predicted by the Gauss-Newton method (optimized to 2.5°), and the closed loop inrush current was 1.1 times the rated value.

[0292] When a new 7kW DC charging pile is connected, the digital twin model automatically updates the topology, triggering a rapid re-optimization (computation time 800ms).

[0293] The energy storage discharge power is adjusted by the MPC algorithm to ensure that the DC bus voltage fluctuation is less than ±1.5%.

[0294] Recorded data:

[0295] Optimization metrics Fluctuations before access Fluctuations after access Adjustment cycle DC voltage ripple ±1.2％ ±1.4％ ＜1s Energy storage SOC deviation ±2％ ±2.1％ dynamic equilibrium

[0296] Experimental conclusion:

[0297] Modeling accuracy: The prediction error of the digital twin model for key parameters (voltage, current, SOC) is less than 2%, which meets the requirements of real-time control.

[0298] Control effect: AC / DC coordinated control makes the bus voltage fluctuation ≤ ±2% and the frequency deviation < ±0.1Hz, which is better than the traditional independent control scheme (voltage fluctuation ±5%, frequency deviation ±0.3Hz).

[0299] Fault resilience: Fault isolation and islanding time <500ms, no interruption of critical loads, 38.9% reduction in loop inrush current, and significantly improved system recovery efficiency.

[0300] Economic optimization: The curtailment rate of solar power was reduced from 12% to 2.8%, energy storage losses and converter losses were reduced by 27.1% and 26.7% respectively, and the overall network loss in the distribution area was reduced by more than 15%.

[0301] Flexibility: When a new device is connected, the system can complete the model update and control strategy re-optimization within 1 second, achieving "plug and play" adaptive adjustment;

[0302] This experiment verified the effectiveness of the control method in the material, and all indicators met the design goals, providing reliable technical support for the engineering application of DC-DC hybrid microgrids in the distribution area.

[0303] Example 2: A hybrid AC / DC microgrid control system based on DC-flexible distribution network, comprising the following modules;

[0304] Multi-source sensing and dynamic modeling: High-precision synchronous sensors are deployed at key nodes of the AC / DC hybrid microgrid in the distribution area to collect AC parameters, DC parameters, photovoltaic and energy storage status and flexible load characteristics; at the same time, a flexible dynamic topology architecture is constructed, a dediscretized photovoltaic and energy storage integrated model and a flexible load plug-and-play model are established, and a microgrid digital twin model is generated to adapt to different operating states.

[0305] Feasible domain characterization and prediction: Based on the photovoltaic-storage model, the Gauss-Newton method is used to iteratively predict the grid state and stability before and after loop closure, and the feasible domain of the transformer area is characterized by a hybrid coordinate system power flow algorithm; through a deep learning model, AC and DC power and load regulation potential are predicted based on historical data, and dynamic calibration is performed using real-time weather data.

[0306] Power optimization and hierarchical control: Construct a power cost model, train it using historical data and combine it with calibrated weather forecasts to optimize power generation planning; Based on the photovoltaic-storage plan, energy storage clustering rules and converter instructions, generate candidate schemes through heuristic algorithms, and after verification by direct and flexible constraints, use droop control or virtual synchronous machine algorithm to achieve frequency-voltage coordinated stability of AC and DC subgrids;

[0307] Fault Response and Islanding Control: The fault point is located by traveling wave ranging, the fault area is quickly isolated by solid-state circuit breakers, the islanding scheme is obtained by calling the solver, and the critical DC loads are prioritized; after the fault is repaired, the optimal loop closing path is generated by heuristic algorithm to reduce grid connection impact.

[0308] Digital twin closed-loop optimization: Real-time acquisition of operational data and comparison with the digital twin model, online adjustment of converter reference values ​​and energy storage power; automatic model updates and rapid re-optimization for new equipment access, achieving multi-timescale adjustment from seconds to hours, ensuring efficient and stable system operation.

[0309] The above module is implemented using the AC / DC hybrid microgrid control method based on transformer substations in Example 1.

[0310] Example 3: A computer device and a readable storage medium, characterized in that the computer device includes: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize a DC-DC hybrid microgrid control system based on a transformer substation.

[0311] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A hybrid AC / DC microgrid control system based on DC-flexible distribution network, characterized in that, Includes the following modules: Multi-source sensing and dynamic modeling: High-precision synchronous sensors are deployed at key nodes of the AC / DC hybrid microgrid in the distribution area to collect AC parameters, DC parameters, photovoltaic and energy storage status and flexible load characteristics; at the same time, a flexible dynamic topology architecture is constructed, a dediscretized photovoltaic and energy storage integrated model and a flexible load plug-and-play model are established, and a microgrid digital twin model is generated to adapt to different operating states. Feasible domain characterization and prediction: Based on the photovoltaic-storage model, the Gauss-Newton method is used to iteratively predict the grid state and stability before and after loop closure, and the feasible domain of the transformer area is characterized by a hybrid coordinate system power flow algorithm; through a deep learning model, AC and DC power and load regulation potential are predicted based on historical data, and dynamic calibration is performed using real-time weather data. Power optimization and hierarchical control: Construct a power cost model, train it using historical data and combine it with calibrated weather forecasts to optimize power generation planning; Based on the photovoltaic-storage plan, energy storage clustering rules and converter instructions, generate candidate schemes through heuristic algorithms, and after verification by direct and flexible constraints, use droop control or virtual synchronous machine algorithm to achieve frequency-voltage coordinated stability of AC and DC subgrids; Fault Response and Islanding Control: The fault point is located by traveling wave ranging, the fault area is quickly isolated by solid-state circuit breakers, the islanding scheme is obtained by calling the solver, and the critical DC loads are prioritized; after the fault is repaired, the optimal loop closing path is generated by heuristic algorithm to reduce grid connection impact. Digital twin closed-loop optimization: Real-time acquisition of operational data and comparison with the digital twin model, online adjustment of converter reference values ​​and energy storage power; automatic model updates and rapid re-optimization for new equipment access, achieving multi-timescale adjustment from seconds to hours, ensuring efficient and stable system operation.

2. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The high-precision synchronous sensors include: a three-phase voltage and current sensor deployed on the AC side (accuracy 0.2S level, sampling rate ≥10kHz), a Hall current sensor deployed on the DC side (bandwidth 100kHz, accuracy ±0.5%), and a SOC monitoring module for the photovoltaic storage unit (error ≤2%), used to collect voltage, current, frequency, phase, temperature and state of charge data in real time.

3. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The flexible dynamic topology architecture is based on graph theory to construct a directed graph G(N,E), where node N includes AC bus, DC bus, photovoltaic storage unit and flexible load, and edge E represents line or converter connection, with weight being line impedance or converter power limit; the topology is dynamically reconstructed through the on / off state of solid-state circuit breakers, supporting rapid switching between islanded / grid-connected modes.

4. The AC / DC hybrid microgrid control system based on transformer substation DC-flexible distribution as described in claim 1, characterized in that: In the dediscretized photovoltaic-storage integrated model, the photovoltaic output power adopts a single diode model combined with temperature correction: Among them, P STC The standard test condition power is given by G, where G is the real-time irradiance and α is the irradiance. T Here is the temperature coefficient (value 0.004 / ℃), and T is the temperature of the photovoltaic panel.

5. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The hybrid coordinate system power flow algorithm solves the AC side fundamental phasor equation and the DC side algebraic equation simultaneously, and introduces flexible constraints: DC bus voltage fluctuation ≤ ±5% of rated value, converter switching rate ≤ 100kW / ms, and frequency deviation ≤ ±0.2Hz.

6. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The deep learning model is a bidirectional LSTM neural network. The input layer contains historical weather data, load curves, and solar power output data. The hidden layer is a 2-layer, 128-neuron structure. The output is a predicted value of AC and DC power for the next hour. The verification root mean square error (RMSE) is ≤3%.

7. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The heuristic algorithm includes particle swarm optimization (PSO) or non-dominated sorting genetic algorithm (NSGA-II) to generate multi-objective optimization schemes. The optimization objectives include minimizing network loss, maximizing renewable energy utilization, and minimizing energy storage lifetime loss. The constraints include energy storage SOC ∈ [20%, 80%] and converter power limit.

8. The AC / DC hybrid microgrid control system based on transformer substation DC-Flexible as described in claim 1, characterized in that: The solid-state circuit breaker has a breaking time of ≤100μs and a traveling wave ranging and positioning accuracy of ≤50m. Before the loop closing operation, the phase angle difference is optimized to ≤5° using the Gauss-Newton method, and the inrush current is controlled within 1.2 times the rated value.

9. A control method for a hybrid AC / DC microgrid based on transformer substation DC-flexibility, used to implement the hybrid AC / DC microgrid control system based on transformer substation DC-flexibility as described in any one of claims 1-8, characterized in that, Includes the following steps: S100: Deploy high-precision synchronous acquisition sensors at key nodes of the AC / DC hybrid microgrid in the distribution area, while constructing a flexible dynamic topology architecture, and establishing a dediscretized integrated photovoltaic and energy storage model and a flexible load plug-and-play model to generate a digital twin model of the microgrid to adapt to different operating states of the AC / DC hybrid microgrid. S200: Based on the aforementioned photovoltaic-storage integrated model, the Gauss-Newton iterative method is used to predict the grid state and system stability before and after loop closure, thereby characterizing the feasible region of the area and determining the feasible operating range of the microgrid under different operating conditions. At the same time, a deep learning model is used to predict the AC / DC power and flexible load regulation potential in the future based on historical data, and the prediction results are dynamically calibrated through real-time weather data. S300: Construct a power cost model for a smart microgrid collaborative control system, train the model using historical data, and use calibrated weather forecast data to predict future power generation and operating costs, thereby improving the accuracy and economy of power generation planning; S400: Combining the photovoltaic and energy storage power generation plan, energy storage grouping rules and converter power commands, a heuristic algorithm is used to generate candidate control schemes; after verification by direct current and flexible constraints, the converter output is adjusted by droop control or virtual synchronous machine algorithm according to the active and reactive power coordinated control strategy to achieve frequency-voltage coordinated stability of AC and DC subgrids. S500: Locates the fault point of the distribution network using traveling wave ranging technology, quickly isolates the fault area using solid-state circuit breakers, calls the solver to solve the fault recovery model of the distribution network, obtains the islanding scheme of the distribution network during the fault time, prioritizes the power supply of DC critical loads, and rationally allocates energy storage power stations in combination with clustering rules. Then, it generates the optimal loop path through heuristic algorithm to reduce grid connection impact. S600: Real-time acquisition of microgrid operation data and comparison with digital twin model, online adjustment of converter reference values ​​and energy storage charging and discharging power, and automatic updating of digital twin model to trigger rapid re-optimization for new equipment access, achieving multi-timescale adjustment from second to hour, ensuring that AC / DC hybrid microgrids work efficiently and stably under various operating conditions.

10. A computer device and a readable storage medium, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the AC / DC hybrid microgrid control system based on the DC / flexible distribution network as described in any one of claims 1 to 8.

Images