Hybrid enhanced power distribution network business expansion power flow and reliability cooperative verification method and system

Abstract

The invention discloses a hybrid enhanced power distribution network business expansion power flow and reliability cooperative verification method and system, and the method comprises the steps: inputting the topological data and historical power flow data of a power distribution network into a power flow calculation engine for large model reasoning acceleration, and obtaining the data of the power distribution network; inputting the data of the power distribution network and the wind, light and storage output models into a reliability evaluation algorithm containing new energy randomness to obtain a reliability index; performing digital twinning and federal learning data correction processing on the reliability index and the equipment real-time measurement data to obtain a corrected reliability index and an equipment parameter update value; performing N-1 fault simulation and self-healing strategy generation processing on the corrected reliability index, the equipment parameter update value and the equipment importance degree to obtain a fault simulation index and a self-healing strategy; and performing feedback optimization closed-loop processing on the fault simulation index and the self-healing strategy to obtain a structured improvement suggestion. According to the invention, an innovative solution is provided for business expansion verification of the high-proportion new energy grid-connected power distribution network.

Description

TECHNICAL FIELD

[0001] The application relates to a mixed enhanced power distribution network industry expansion power flow and reliability collaborative verification method and system, and belongs to the technical field of intelligent control of power distribution networks. BACKGROUND

[0002] Power distribution network industry expansion (power supply business expansion) scheme verification is a key link to ensure the safe operation of the power grid, and the power supply capacity of the scheme needs to be verified through power flow calculation, and the fault response capacity needs to be verified through reliability evaluation. With large-scale grid connection of new energy, the traditional verification method faces the following core problems:

[0003] Low efficiency of power flow calculation: the traditional forward-backward substitution method and Newton-Raphson method have many iterations (15-20 times) and long solution time (>5 seconds) for large-scale power distribution networks with node number >1000, which cannot meet the rapid verification demand of industry expansion scheme.

[0004] Insufficient accuracy of reliability evaluation: the traditional fault mode effect analysis (FMEA) algorithm does not consider the randomness of new energy output (such as wind power output fluctuation and photovoltaic output intermittency), and relies on static historical data without correction combined with real-time measurement data, resulting in a deviation of >30% between the evaluation result and the actual value.

[0005] Lack of closed-loop optimization mechanism: the existing verification only outputs the "qualified / unqualified" conclusion, does not generate targeted improvement suggestions, and cannot be fed back to the scheme generation link to form a "verification-optimization" closed loop.

[0006] For example, although the existing technology optimizes the power flow calculation steps, it does not combine large model acceleration, and does not involve reliability evaluation, and does not consider the correction of new energy randomness.

[0007] Therefore, a collaborative verification method integrating mixed enhanced intelligent algorithm is needed to solve the above problems. SUMMARY

[0008] Objective: In order to overcome the deficiencies in the prior art, the application provides a mixed enhanced power distribution network industry expansion power flow and reliability collaborative verification method and system.

[0009] Technical scheme: In order to solve the above technical problems, the technical scheme adopted by the application is:

[0010] In the first aspect, a mixed enhanced power distribution network industry expansion power flow and reliability collaborative verification method specifically comprises:

[0011] Step S1: inputting power distribution network topology data and historical power flow data into a large model reasoning accelerated power flow calculation engine to obtain node voltage curves, line active power curves, line reactive power curves and short-circuit current curves of the power distribution network.

[0012] Step S2: input the node voltage curve, line active power curve, line reactive power curve, short circuit current curve of the power distribution network, and wind, light, and storage output model into the reliability evaluation algorithm containing new energy randomness, to obtain the reliability index.

[0013] Step S3: perform digital twinning and federated learning data correction processing on the reliability index and real-time measurement data of the equipment, to obtain the corrected reliability index and equipment parameter update value.

[0014] Step S4: perform N-1 fault simulation and self-healing strategy generation processing on the corrected reliability index, equipment parameter update value, and equipment importance, to obtain the fault simulation index and self-healing strategy.

[0015] Step S5: perform feedback optimization closed-loop processing on the fault simulation index and self-healing strategy, to obtain the structured improvement suggestion.

[0016] Optionally, the large model inference acceleration power flow calculation engine includes a physical mechanism layer, a large model acceleration layer, and an adaptive decision layer.

[0017] The physical mechanism layer includes a three-phase system sequence component decomposition module, an improved forward-backward substitution method module, and a DG equivalent admittance processing module.

[0018] The three-phase system sequence component decomposition module is configured to decompose three-phase voltage / current into positive sequence, negative sequence, and zero sequence networks.

[0019] The improved forward-backward substitution method module is configured to process three-phase unbalanced networks using an improved forward-backward substitution method, introduce sequence component transformation to decompose the three-phase system into positive sequence, negative sequence, and zero sequence networks, and establish power balance equations respectively.

[0020] The DG equivalent admittance processing module is configured to increase the DG equivalent admittance in the node admittance matrix for distributed power access, and process the constant power / constant voltage control mode of the inverter type DG.

[0021] The large model acceleration layer adopts an LSTM model.

[0022] The adaptive decision layer is configured to output the power flow solution using the improved forward-backward substitution method in the physical mechanism layer when the number of nodes of the power distribution network is less than or equal to 500, and to realize scene adaptation through a decision tree algorithm when the number of nodes is greater than 500, predict the voltage initial value using an LSTM model, and perform iteration according to the Newton-Raphson method to obtain the converged power flow solution.

[0023] Optionally, the step S2 specifically includes:

[0024] Step S2.1: According to the wind power output model, the photovoltaic output model, the wind-solar-storage combined output scene is generated, and the wind-solar-storage combined output typical scene is obtained by clustering.

[0025] Step S2.2: Traverse the device list to define the typical fault mode.

[0026] Step S2.3: According to the node voltage curve, the line active power curve, the line reactive power curve and the short-circuit current curve of the distribution network, the voltage out-of-limit node set and the power overload line set after the fault are calculated.

[0027] Step S2.4: The wind-solar-storage combined output typical scene is weighted, and the weighted fault impact time of the voltage out-of-limit node set and the power overload line set after the fault is calculated.

[0028] Step S2.5: According to the weighted fault impact time, the reliability index is calculated.

[0029] Optionally, the step S3 specifically comprises:

[0030] Step S3.1: A three-dimensional model of the distribution network is constructed based on GIS data.

[0031] Step S3.2: A device-level dynamic model is established.

[0032] Step S3.3: The state quantity of the device-level dynamic model in the three-dimensional model of the distribution network is mapped in real time according to the real-time measurement data of the device.

[0033] Step S3.4: The state quantity is input into the hierarchical aggregation model of federated learning to obtain the output prediction value.

[0034] Step S3.5: The reliability index and the device parameter are corrected according to the output prediction value to obtain the corrected reliability index and the updated value of the device parameter.

[0035] Optionally, the step S4 specifically comprises:

[0036] Step S4.1: The device importance is obtained according to the device importance scoring model.

[0037] Step S4.2: The fault probability is sorted according to the device importance to obtain a fault scene sequence.

[0038] Step S4.3: A multi-objective function of the self-healing strategy is constructed.

[0039] Step S4.4: The fault scene in the fault scene sequence is simulated according to the corrected reliability index and the updated value of the device parameter, and the multi-objective function of the self-healing strategy is solved to obtain the fault simulation index and the self-healing strategy.

[0040] Optionally, the equipment importance scoring model The expression is as follows:

[0041]

[0042] in, This is the load influence factor. This is the historical failure frequency coefficient. For the new energy access coefficient, This refers to the geographical environment coefficient.

[0043] Optional, multi-objective function of self-healing strategy The expression is as follows:

[0044]

[0045] in, This is the load loss amount. This represents the maximum voltage drop. For recovery time, , , These are the first weight, the second weight, and the third weight, respectively.

[0046] Optionally, step S5 specifically includes:

[0047] Step 5.1: Set a three-level threshold. When the fault index in the self-healing strategy exceeds the three-level threshold, proceed to step 5.2.

[0048] Step 5.2: Input the fault simulation indicators into the NLP structured suggestion generation model to obtain structured improvement suggestions.

[0049] In a second aspect, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a hybrid enhanced distribution network power flow and reliability co-verification method as described in any of the first aspects.

[0050] Thirdly, a computer device is introduced, including:

[0051] Memory is used to store instructions.

[0052] A processor is configured to execute the instructions, causing the computer device to perform operations of a hybrid enhanced distribution network power flow and reliability co-verification method as described in any of the first aspects.

[0053] Beneficial effects: The hybrid enhanced power flow and reliability co-verification method and system for distribution networks provided by this invention have the following advantages compared with the prior art:

[0054] 1. Efficiency Improvement: The large model acceleration reduces power flow calculation time by more than 85%, supporting second-level verification of large-scale distribution networks.

[0055] 2. Improved accuracy: Stochastic modeling of new energy sources and real-time data correction reduce the reliability assessment error from >30% to <10%.

[0056] 3. Closed-loop optimization: Generate structured improvement suggestions to achieve a closed loop of "solution generation-verification-optimization", reducing the number of times the business expansion solution is repeatedly modified (from 3-4 times to 1-2 times). Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating a hybrid enhanced distribution network power flow and reliability co-verification method according to the present invention.

[0058] Figure 2 This is a schematic diagram of the power flow calculation engine for accelerating large-scale model inference according to the present invention.

[0059] Figure 3 This is a schematic diagram of the reliability assessment algorithm for new energy randomness in this invention.

[0060] Figure 4 This is a schematic diagram of the fault simulation and self-healing strategy generation algorithm of the present invention.

[0061] Figure 5 This is a schematic diagram of the feedback optimization closed-loop processing algorithm of the present invention. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0063] The present invention will be further described below with reference to specific embodiments.

[0064] Example 1:

[0065] This embodiment introduces a hybrid enhanced method for co-verification of power flow expansion and reliability in distribution networks, such as... Figure 1 As shown, it specifically includes:

[0066] Step S1: Input the distribution network topology data and historical power flow data into the power flow calculation engine that accelerates large model inference to obtain the node voltage curve, line active power curve, line reactive power curve and short-circuit current curve of the distribution network.

[0067] Among them, the power flow calculation engine for accelerating large model inference, such as Figure 2 As shown, it includes: a physical mechanism layer, a large model acceleration layer, and an adaptive decision layer.

[0068] The physical mechanism layer includes: a three-phase system sequence component decomposition module, an improved forward-backward substitution method module, and a DG (distributed generation) equivalent admittance processing module.

[0069] The three-phase system sequence component decomposition module is used to decompose the three-phase voltage / current into positive sequence, negative sequence, and zero sequence networks.

[0070] The improved forward-backward substitution module is used to process three-phase unbalanced networks using the improved forward-backward substitution method. It introduces sequence component transformation (symmetric component method) to decompose the three-phase system into positive-sequence, negative-sequence, and zero-sequence networks, and establishes power balance equations for each. The expressions for the power balance equations are as follows:

[0071]

[0072] , , Let be the positive-sequence, negative-sequence, and zero-sequence active power of node i, respectively. This represents the positive-sequence voltage at node i. Let represent the positive sequence voltage at node j. This represents the negative sequence voltage at node i. Let represent the negative sequence voltage at node j. This represents the positive-order admittance between node i and node j. This represents the negative-order admittance between node i and node j. This represents the positive-sequence phase angle of node i. This represents the positive-order phase angle of node j. This represents the negative-order phase angle of node j. This represents the negative-order phase angle of node j. The phase angle represents the positive-sequence admittance between node i and node j. The phase angle represents the negative-order admittance between node i and node j. This represents the zero-sequence voltage at node i. This represents the zero-sequence voltage at node j. This represents the zero-order admittance between node i and node j. This represents the zero-sequence phase angle of node i. This represents the zero-sequence phase angle of node j. The phase angle represents the zero-sequence admittance between node i and node j.

[0073] The DG equivalent admittance processing module is used to add DG equivalent admittance to the node admittance matrix for distributed generation (DG) access, and to handle the constant power / constant voltage control mode of inverter-type DG.

[0074] The large model acceleration layer uses the LSTM model.

[0075] The LSTM model structure includes: an input layer with 128-dimensional features (node ​​historical voltage / power, DG output, timestamp, meteorological data), a hidden layer with 3 layers (64 neurons per layer), and an output layer with 2 dimensions (voltage amplitude, initial phase angle). The Adam optimizer (learning rate 0.001) and Dropout (0.2) are used to prevent overfitting.

[0076] Training dataset augmentation: Historical data was augmented with noise (injecting ±5% Gaussian noise to simulate measurement error) to generate 100,000 samples, reducing the model's generalization error to 3.2%.

[0077] The adaptive decision-making layer:

[0078] Dynamic switching of calculation modes: When the number of distribution network nodes is ≤500, the pure forward-backward substitution method in the physical mechanism layer is more efficient and outputs the power flow solution (node ​​voltage magnitude and phase angle); when the number of nodes is >500, the hybrid mode of LSTM model initial value + Newton-Raphson method in the large model acceleration layer is automatically enabled. The scenario is adapted through the decision tree algorithm. First, LSTM is used to quickly predict the initial voltage value, and then the Newton-Raphson method is used for accurate iteration. Finally, the converged power flow solution (node ​​voltage magnitude and phase angle) can be obtained.

[0079] Furthermore, in one embodiment, the performance of the power flow calculation engine for accelerating large model inference is verified:

[0080] Test scenario: 1200-node distribution network (including 300 DG access points, of which 150 are photovoltaic, 100 are wind power, and 50 are energy storage), three-phase imbalance 8%. Comparison results:

[0081] Computational method Number of iterations Solving time (seconds) Voltage calculation error (%) Conventional Newton method 18 5.2 2.1 Inventive hybrid method 6 0.7 1.8

[0082] Step S2: Input the node voltage curves, line active power curves, line reactive power curves, short-circuit current curves, and wind, solar, and energy storage output models of the distribution network into a reliability assessment algorithm that incorporates the randomness of new energy sources, to obtain reliability indices, including: SAIDI (Simplified Average Outage Duration), SAIFI (System Average Outage Frequency), and CAIDI (User Average Outage Duration). Figure 3 As shown, it specifically includes:

[0083] Step S2.1: Based on the wind power output model and the photovoltaic power output model, generate wind-solar-storage combined power output scenarios, and obtain typical wind-solar-storage combined power output scenarios through clustering.

[0084] Among them, wind power output model Using the Weibull distribution and introducing a "wind speed-output mapping function" (where v is the wind speed), the following results were obtained based on measured wind farm data:

[0085]

[0086] in, To cut into wind speed, Rated wind speed, To cut off the wind speed, This is the rated power.

[0087] Photovoltaic power output model :

[0088] The randomness of light intensity is simulated using a Beta distribution, with the following probability density function:

[0089]

[0090] Where G is the light intensity, , The shape parameters are obtained by fitting based on historical data. .

[0091] Scene generation: 500 combined wind, solar and energy storage power generation scenarios were generated using Latin hypercube sampling (LHS), and then reduced to 20 typical scenarios through K-means clustering to reduce computational load.

[0092] Step S2.2: Traverse the equipment list (lines, transformers, switches) and define 32 typical fault modes (such as single-phase grounding of lines and gas protection operation of transformers).

[0093] Step S2.3: Based on the node voltage curve, line active power curve, line reactive power curve and short-circuit current curve of the distribution network, calculate the set of nodes with voltage exceeding the limit and the set of lines with power overload after the fault (e.g., voltage < 0.9 pu or > 1.1 pu).

[0094] Step S2.4: Assign weights to typical scenarios of combined wind, solar and energy storage power output (according to scenario probability distribution), and calculate the weighted fault impact time of voltage over-limit node set and power overload line set.

[0095]

[0096] in, The weights of scene s, The duration of the fault in scenario s.

[0097] Step S2.5: Calculate the reliability index based on the weighted failure duration.

[0098] In addition to the traditional SAIDI / SAIFI, a new "Renewable Energy Power Supply Reliability Index (RESR)" has been added:

[0099]

[0100] in, The duration of power supply to DG during the k-th fault. This represents the number of users affected in the k-th failure.

[0101] Step S3: Perform digital twin and federated learning data correction processing on the reliability indicators and real-time measurement data of the equipment to obtain the corrected reliability indicators and updated equipment parameter values.

[0102] Step S3.1: Construct a three-dimensional model of the power distribution network (accuracy 0.5m) based on GIS data, including the spatial coordinates of towers, conductors, and transformers.

[0103] Step S3.2: Establish a device-level dynamic model (e.g., a transformer temperature rise model). (where I is the load current and t is the duration).

[0104] Step S3.3: Based on real-time measurement data from the equipment, map state variables to the equipment-level dynamic model in the 3D model of the distribution network in real time. This involves mapping over 2000 state variables, including node voltage, line power, and equipment temperature, with a data refresh delay of ≤50ms.

[0105] Among them, the “MQTT + edge computing” architecture is adopted. The edge gateway (deployed in the substation) preprocesses the real-time measurement data of the equipment (filtering, outlier removal) and then uploads it to the device-level dynamic model in the digital twin platform through an encrypted channel (AES-256), reducing bandwidth usage by 60%.

[0106] Step S3.4: Input the state variables into the hierarchical aggregation model of federated learning to obtain the output prediction value.

[0107] The layered aggregation model of federated learning includes: edge layer and cloud layer.

[0108] Edge layer (client): 10 regional edge nodes (each node manages 10-20 smart meters / PMUs), local training of sub-models (such as LSTM for photovoltaic output prediction), training period of 1 hour;

[0109] Cloud layer (server): The "federated average + dynamic weighting" aggregation algorithm is used to aggregate the model parameters of edge nodes according to data quality (the weight of nodes with measurement accuracy > 98% is increased by 20%), and the global model update cycle is 2 hours.

[0110] Privacy protection mechanisms: Differential privacy technology is used (Laplace noise is injected, with an intensity of 1.5); model parameters are homomorphically encrypted before transmission to prevent data leakage.

[0111] Step S3.5: Correct the reliability index and equipment parameters based on the output prediction value to obtain the corrected updated values ​​of the reliability index and equipment parameters.

[0112] In a pilot scenario, the photovoltaic output prediction error decreased from 15% to 5.8%, and the accuracy of line fault rate updates increased to 92%.

[0113] Step S4: Perform N-1 fault simulation and self-healing strategy generation processing on the corrected reliability indicators, updated equipment parameter values, and equipment importance to obtain fault simulation indicators and self-healing strategies. For example... Figure 4 As shown, it specifically includes:

[0114] Step S4.1: Obtain the equipment importance based on the equipment importance scoring model.

[0115] Among them, the equipment importance scoring model The expression is as follows:

[0116]

[0117] in, This is the load influence factor. This is the historical failure frequency coefficient. For the new energy access coefficient, This refers to the geographical environment coefficient.

[0118] Step S4.2: Sort the failure probabilities according to the importance of the equipment to obtain a failure scenario sequence.

[0119] Of these, line faults accounted for 70% (e.g., L13: S=0.85, high risk), and transformer faults accounted for 30%.

[0120] Step S4.3: Construct the multi-objective function of the self-healing strategy The expression is as follows:

[0121]

[0122] in, This is the load loss amount. This represents the maximum voltage drop. For recovery time, For weights.

[0123] Step S4.4: Simulate the fault scenarios in the fault scenario sequence based on the corrected reliability index and updated equipment parameter values, solve the multi-objective function of the self-healing strategy, and obtain the fault simulation index and the self-healing strategy.

[0124] The self-healing strategy generation process includes:

[0125] Initial policy library: 5000 historical policies are pre-stored (based on 10 years of fault handling records).

[0126] Large model matching: The BERT model is used to match the current fault features (voltage drop, deficiency) with the policy library and output the Top 5 candidate policies;

[0127] Simulation verification: Evaluate candidate strategies through digital twin simulation and select the strategy with the smallest F value (e.g., in a certain scenario, the F value of strategy 3 is 28% lower than that of strategy 1).

[0128] Step S5: Perform feedback optimization closed-loop processing on the fault simulation indicators and self-healing strategies to obtain structured improvement suggestions. For example... Figure 5 As shown, it specifically includes:

[0129] Step 5.1: Set three threshold levels (early warning value, limit value, emergency value). When the fault indicators in the self-healing strategy exceed the three threshold levels, proceed to step 5.2.

[0130] Step 5.2: Input the fault simulation indicators into the NLP structured suggestion generation model to obtain structured improvement suggestions.

[0131] Among them, the generative model of structured suggestions in NLP (Natural Language Processing) is as follows:

[0132] Technical Indicators → Natural Language Mapping:

[0133] A fine-tuned BERT model for the power sector is adopted (the training corpus consists of 100,000 historical business expansion suggestions). The input is a fault index vector (such as [22% voltage drop, 1200kW loss]), and the output is the suggestion template filling content. The suggestion generation rule base contains 120 equipment improvement rules (such as "conductor cross-section upgrade rule: when the load factor is >85% and the N-1 fault voltage drop is >15%, it is recommended to upgrade the cross-section").

[0134] In one case, after two rounds of iteration, SAIDI decreased from 215 min / household to 168 min / household, meeting the target value (≤180 min / household).

[0135] The core innovations of this invention are as follows: focusing on solving the key technical bottlenecks in the business expansion verification of distribution networks under the scenario of new energy grid connection, highlighting the engineering breakthroughs and collaborative closed-loop logic of hybrid enhanced intelligent algorithms:

[0136] Hybrid augmented intelligence power flow calculation acceleration mechanism: Breaking through the traditional power flow calculation mode of pure physical mechanism or pure data, it pioneers a hybrid framework of "physical mechanism model + large model inference". It constructs the power balance physical equation of distribution network by forward backscalar method to ensure calculation accuracy; it introduces LSTM large model to predict the initial value of power flow (node ​​voltage, phase angle) and shortens the number of iterations of Newton-Raphson method from 15-20 times to 5-8 times, realizing the second-level solution (≤1 second) of distribution network with 1000+ nodes, and solving the problem of low power flow calculation efficiency in large-scale new energy grid-connected scenarios.

[0137] Optimization of Reliability Assessment Algorithm Incorporating New Energy Stochasticity: Addressing the deficiency of traditional reliability assessments that ignore fluctuations in new energy output, an improved scheme integrating stochastic modeling is proposed. This involves introducing a wind power output simulation model based on the Weibull distribution (quantifying randomness through shape parameter k and scale parameter λ), combined with photovoltaic output scenario generation technology, to construct a full-scenario reliability assessment system encompassing wind, solar, and energy storage. The traditional FMEA algorithm is improved by weighting the calculation of SAIDI, SAIFI, and CAIDI indicators using "new energy output scenario weights," thereby increasing the assessment accuracy for high-proportion new energy scenarios by over 25% and resolving the issue of large deviations between reliability assessments and actual operation.

[0138] Dynamic data correction technology using digital twins and federated learning: This innovative dual-dimensional correction mechanism, combining real-time digital twin interaction with privacy-preserving federated learning, establishes a second-level data synchronization channel between the physical power grid and the digital twin model, enabling real-time acquisition of measurement data from smart meters and PMUs. Based on federated learning, it achieves "local training-parameter upload-global aggregation" of edge terminal data, dynamically correcting the prediction error of new energy output (reducing it from 15% to 6%) and equipment failure rate parameters (based on weighted updates of historical and real-time data) while protecting data privacy, thus solving the problem of lag in traditional static data evaluation.

[0139] Intelligent collaborative generation of fault simulation and self-healing strategies: Construct an integrated mechanism of "N-1 fault scenario library + digital twin simulation + large model decision-making", automatically generate the Top 20 high-risk fault scenarios (line tripping, transformer shutdown) according to equipment importance, accurately simulate voltage drop, load deficit and other states after the fault through digital twin, and output multi-dimensional self-healing strategies (load shedding priority, energy storage charging and discharging control, microgrid island switching) in combination with large model inference, realize the quantitative assessment and proactive response of fault impact, and break through the limitation of traditional fault simulation only outputting "qualified / unqualified".

[0140] The structured feedback mechanism of "verification-optimization" closed loop: It is the first to create a full-process transformation logic from fault indicators to solution suggestions. Through natural language processing (NLP), the key indicators of fault simulation (maximum voltage drop, load loss) are transformed into structured improvement suggestions (such as conductor cross-section upgrade, energy storage configuration optimization, etc.). Feedback is fed back to the business expansion solution generation stage to form a closed loop of "solution design-verification evaluation-optimization iteration", which reduces the number of times the solution is repeatedly modified (from 3-4 times to 1-2 times) and improves the efficiency of the business expansion process.

[0141] In summary, the core breakthrough lies in the organic integration of physical mechanism models, large-scale models, federated learning, digital twins, and other technologies to achieve the triple goals of "efficiency improvement, accuracy enhancement, and closed-loop optimization," providing an innovative solution for business expansion verification of distribution networks with a high proportion of new energy grid connection.

[0142] Example 2:

[0143] This embodiment describes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a hybrid enhanced distribution network power flow and reliability co-verification method as described in any of Embodiment 1.

[0144] Example 3:

[0145] This embodiment describes a computer device, including:

[0146] Memory is used to store instructions.

[0147] A processor is configured to execute the instructions, causing the computer device to perform operations of a hybrid enhanced distribution network power flow and reliability co-verification method as described in any of Embodiment 1.

[0148] Example 4:

[0149] This embodiment introduces the simulation process of a hybrid-enhanced distribution network business expansion power flow and reliability collaborative verification method. Taking the business expansion scheme verification of a 10kV distribution network (including 1200 nodes, 200MW photovoltaic, 50MW wind power, and 100MWh energy storage) in a prefecture-level city as an example, the implementation process is described in detail:

[0150] 1. Data preparation:

[0151] Distribution network topology data: Node numbers 1-1200, 30 10kV lines (conductor cross-section 120-240mm²), 10 35kV transformers (capacity 50-100MVA).

[0152] Historical data: node voltage and power data from January to December 2023 (sampling frequency 15 minutes / time), wind farm / photovoltaic power plant output data, and historical equipment failure data (2021-2023).

[0153] Real-time measurement data: 100 smart meters (user load), 20 PMUs (node ​​voltage / phase angle), data refresh frequency 1 second / time.

[0154] 2. Step S1 implementation:

[0155] LSTM model training: Using data from January to September 2023 as the training set (70%), validation set (20%), and test set (10%), the initial values ​​were output after training. The number of iterations was reduced from 18 to 6.

[0156] Power flow calculation results: node voltage deviation < 5% (compliant with GB / T12325-2020 requirements), maximum line power load rate 85%, solution time 0.7 seconds.

[0157] 3. Step S2 implementation:

[0158] Weibull distribution parameters: Based on wind farm data from 2023, fit \(k=2.2\) \(\lambda=170MW\) to generate 100 wind power output scenarios.

[0159] Reliability index calculation: Traditional algorithm SAIDI = 180 min / household, improved algorithm SAIDI = 215 min / household (considering the fluctuation of new energy sources).

[0160] 4. Step S3 implementation:

[0161] Federated learning iteration: 10 iterations with 100 smart meter clients reduced the model loss function from 0.12 to 0.04.

[0162] Error correction: The photovoltaic output prediction error has been reduced from 15% to 6%, and the line failure rate has been updated to 0.23 times / year.

[0163] Corrected index: SAIDI = 208 min / household (more realistic).

[0164] 5. Step S4 implementation:

[0165] N-1 fault scenario: Select the tripping scenario of line L13 (conductor cross-section 120mm², load rate 85%).

[0166] Simulation results: Voltage drop of 22% (nodes 500-600), load deficit of 1200kW.

[0167] Self-healing strategy: Disconnect 300kW of interruptible industrial loads, discharge 800kW of energy storage, supply 200kW of microgrid isolated power, restore voltage to 92%, and reduce load deficit by 100kW.

[0168] 6. Step S5 implementation:

[0169] Improvement suggestion: "It is recommended to increase the cross-sectional area of ​​conductor L13 of 10kV line from 120mm² to 185mm², which is expected to reduce the maximum voltage drop by 12% and reduce the load loss by 800kW."

[0170] Solution optimization: After optimization, the load rate of line L13 was reduced to 72%, and the SAIDI was re-verified to be 172 min / household, which meets the business expansion requirement of "SAIDI≤180 min / household".

[0171] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0173] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0175] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A hybrid enhanced distribution network power flow and reliability co-verification method, characterized in that: Specifically comprising: Step S1: input the power distribution network topology data and historical power flow data into the large model inference accelerated power flow calculation engine to obtain the node voltage curve, line active power curve, line reactive power curve and short-circuit current curve of the power distribution network; Step S2: input the node voltage curve, line active power curve, line reactive power curve, short-circuit current curve of the power distribution network and wind, light and storage output model into the reliability evaluation algorithm containing new energy randomness to obtain the reliability index; Step S3: perform digital twinning and federated learning data correction processing on the reliability index and device real-time measurement data to obtain the corrected reliability index and device parameter update value; Step S4: perform N-1 fault simulation and self-healing strategy generation processing on the corrected reliability index, device parameter update value and device importance to obtain the fault simulation index and self-healing strategy; Step S5: perform feedback optimization closed-loop processing on the fault simulation index and self-healing strategy to obtain the structured improvement suggestion.

2. The method of claim 1, wherein the method further comprises: The large model inference accelerated power flow calculation engine comprises a physical mechanism layer, a large model acceleration layer and an adaptive decision layer; The physical mechanism layer comprises a three-phase system sequence component decomposition module, an improved forward-backward substitution method module and a DG equivalent admittance processing module; The three-phase system sequence component decomposition module is configured to decompose three-phase voltage / current into positive sequence, negative sequence and zero sequence networks; The improved forward-backward substitution method module is configured to process three-phase unbalanced networks by using an improved forward-backward substitution method, introducing sequence component transformation to decompose the three-phase system into positive sequence, negative sequence and zero sequence networks, and establishing power balance equations respectively; The DG equivalent admittance processing module is configured to increase DG equivalent admittance in the node admittance matrix for distributed power supply access, and process the constant power / constant voltage control mode of the inverter type DG; The large model acceleration layer adopts an LSTM model; The adaptive decision layer is configured to output the power flow solution by using the improved forward-backward substitution method in the physical mechanism layer when the number of nodes of the power distribution network is less than or equal to 500, and to realize scene adaptation by using a decision tree algorithm, predict the voltage initial value by using the LSTM model, and obtain the converged power flow solution by using the Newton-Raphson method when the number of nodes is greater than 500.

3. The method of claim 1, wherein the method further comprises: The step S2 specifically comprises: Step S2.1: generating a wind-light-storage combined output scene according to a wind power output model and a photovoltaic output model, and obtaining a wind-light-storage combined output typical scene by clustering; Step S2.2: traversing a device list to define a typical fault mode; Step S2.3: calculating the voltage out-of-limit node set and the power overload line set after the fault according to the node voltage curve, the line active power curve, the line reactive power curve and the short-circuit current curve of the power distribution network; Step S2.4: assigning weights to the wind-light-storage combined output typical scene to calculate the weighted fault impact time of the voltage out-of-limit node set and the power overload line set after the fault; Step S2.5: calculating the reliability index according to the weighted fault impact time.

4. The method of claim 1, wherein the method further comprises: determining a reliability index for each of the plurality of candidate expansion plans; and selecting the candidate expansion plan having the highest reliability index. The step S3 specifically comprises: Step S3.1: constructing a three-dimensional model of the power distribution network based on GIS data; Step S3.2: establishing a device-level dynamic model; Step S3.3: Real-time mapping state variables of the equipment-level dynamic model in the three-dimensional model of the power distribution network according to real-time equipment measurement data; Step S3.4: Inputting the state variables into the hierarchical aggregation model of federated learning to obtain output prediction values; Step S3.5: Correcting the reliability index and the equipment parameter according to the output prediction values to obtain updated values of the corrected reliability index and the equipment parameter.

5. The method of claim 1, wherein the method further comprises: The step S4 specifically comprises: Step S4.1: Obtaining the equipment importance according to the equipment importance scoring model; Step S4.2: Sequencing the fault probability according to the equipment importance to obtain a fault scenario sequence; Step S4.3: Constructing a multi-objective function of the self-healing strategy; Step S4.4: Simulating the fault scenarios in the fault scenario sequence according to the corrected reliability index and the updated values of the equipment parameter, solving the multi-objective function of the self-healing strategy, and obtaining the fault simulation index and the self-healing strategy.

6. The method of claim 5, wherein the method further comprises: The device importance scoring model The expression is as follows: ； wherein, is a load influence coefficient, is a historical failure frequency coefficient, is a new energy access coefficient, is a geographical environment coefficient.

7. The method of claim 5, wherein the method further comprises: The multi-objective function of the self-healing strategy The expression is as follows: ； wherein, is the load loss amount, is the maximum voltage drop value, is the recovery time, , , are the first weight, the second weight, and the third weight, respectively.

8. The method of claim 1, wherein the method further comprises: determining a reliability index for each of the plurality of candidate expansion plans; and selecting the candidate expansion plan having the highest reliability index. The step S5 specifically comprises: Step 5.1: Setting three-level thresholds, and when the fault index in the self-healing strategy exceeds the three-level thresholds, entering step 5.2; Step 5.2: Inputting the fault simulation index into a generation model of an NLP structured suggestion to obtain a structured improvement suggestion.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is executed by a processor to implement the method of claim 1-8.

10. A computer device, comprising: Comprise: A memory for storing instructions; A processor for executing the instructions, so that the computer device executes the operations of the method of claim 1-8.

Images