Method for collaborative construction of alternating-current and direct-current hybrid external transmission power grid

Abstract

The invention, which relates to the technical field of power transmission, discloses a method for collaborative construction of an AC / DC hybrid external transmission power grid, and the method comprises the following steps: S1, collecting historical data of a power grid; s2, based on the collected data, an uncertainty variable set is defined, and the uncertainty variable set comprises wind and light output fluctuation, load random change and fault probability distribution; and S3, adopting Latin hypercube sampling to generate a multi-scene set. According to the method, the Kantorovich distance of each scene in the initial multi-scene set is calculated, the Kantorovich distance is defined as the optimal transportation cost between probability distributions, the difference between the minimized scenes is solved through linear programming, the similar scenes are iteratively combined according to the preset reduction threshold value, the high-probabilistic weight scene is reserved, and the optimal transportation cost of the scenes in the initial multi-scene set is obtained. According to the power grid collaborative construction method, the calculation efficiency is improved, the scene generation time consumption is reduced to 1.5 hours from 6.2 hours of a traditional Monte Carlo method, and the coverage rate of a typical coverage scene to extreme weather is improved.

Description

Technical Field

[0001] This invention relates to the field of power transmission technology, specifically a method for the collaborative construction of an AC / DC hybrid power grid for external transmission. Background Technology

[0002] Driven by the "dual carbon" goals, my country is accelerating the construction of its new power system, with the penetration rate of new energy power generation continuously increasing. Centralized energy bases and distributed power sources coexist, and the demand for cross-regional power transmission is becoming increasingly strong. The AC / DC hybrid power grid is a core hub connecting energy bases and load centers, possessing the characteristics of large-capacity, long-distance power transmission, providing important support for the optimized allocation of new energy sources. AC / DC power grid design is a core component of power system planning. Grid structure optimization and multi-scenario adaptability analysis are crucial for improving grid transmission efficiency, security, stability, and economy. The randomness of new energy output and load demand brings significant uncertainty. Traditional scenario generation and reduction technologies are inefficient and prone to overlooking key scenarios, resulting in insufficient adaptability of grid planning schemes. Existing transmission network planning often adopts a single-stage design approach, failing to fully consider the multi-stage evolution characteristics of the power grid from isolated operation to regional interconnection and intelligent interconnection. Scenario generation is also excessively time-consuming due to grid expansion, dense DC feeds, and extreme weather conditions.

[0003] The existing methods for collaboratively constructing AC / DC hybrid power transmission grids have the following drawbacks: 1. Patent document CN120389440A discloses a method for coordinated control of AC / DC hybrid microgrids, "including the following steps: Step S1: Constructing a multi-level microgrid architecture, including a main microgrid, a DC sub-microgrid, and an AC sub-microgrid, wherein the main microgrid and the DC sub-microgrid are connected through a bidirectional interconnected converter; Step S2: Deploying a hierarchical Multi-Agent control architecture, including a main microgrid agent, a DC sub-microgrid agent, and an AC sub-microgrid agent; Step S3: Executing three-level coordinated control, whereby the main microgrid agent receives the main grid scheduling instructions and decomposes them into sub-microgrid control objectives, coordinates the power distribution among sub-microgrids based on a global optimization algorithm, and simultaneously maintains voltage and frequency stability through local control algorithms. This invention significantly improves the grid absorption efficiency and operational stability in scenarios with a high proportion of new energy access." However, existing methods for coordinated construction of AC / DC hybrid power grids for external transmission take too long to generate scenarios, reducing the coverage of typical scenarios for extreme weather conditions. Summary of the Invention

[0004] The purpose of this invention is to provide a method for the collaborative construction of AC / DC hybrid power transmission grids, in order to solve the technical problems mentioned in the background art, such as the excessive time required for scenario generation and the reduced coverage of typical scenarios for extreme weather.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for collaboratively constructing an AC / DC hybrid power grid for external transmission, the method comprising the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, which includes fluctuations in wind and solar power output, random changes in load, and the probability distribution of failures; Step S3: Generate a multi-scene set using Latin hypercube sampling; Step S4: Reduce the scene set using a scene reduction technique based on Kantorovich distance to obtain typical scenes; Step S5: Apply typical scenarios to perform grid collaborative optimization. The grid collaborative optimization includes power flow calculation, fault simulation and resource scheduling, and outputs a construction scheme for AC / DC hybrid power grid. The construction scheme covers the grid topology, protection device configuration and operation strategy. Step S3 specifically involves: initializing the sampling dimension, which corresponds to the key indicators of the set of uncertain variables, including output fluctuation rate, load deviation coefficient and failure frequency; applying the Latin hypercube sampling algorithm to generate uniformly distributed scene samples in the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range, and generating an initial multi-scene set containing more than 1000 scenes.

[0006] Preferably, the historical power grid data includes AC power grid parameters, DC power grid parameters, renewable energy output data, and load demand data. The AC power grid parameters include node voltage, line impedance, and transformer capacity. The DC power grid parameters include converter station characteristics, DC line resistance, and power modulation capability. Step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set. The Kantorovich distance is defined as the optimal transportation cost between probability distributions. The differences between scenarios are minimized by solving linear programming. Based on a preset reduction threshold, similar scenarios are iteratively merged while retaining scenarios with high probability weights, ultimately reducing the number of typical scenarios to 50-100.

[0007] Preferably, in step S3, dynamically adjusting the parameters of Latin hypercube sampling further includes optimizing the scene generation accuracy. Specifically, this involves: real-time monitoring of power grid operation data streams; identifying the non-steady-state characteristics of uncertain variables through online learning algorithms; dynamically updating the sampling dimension and distribution function; introducing a sensitivity analysis module; and automatically adjusting the number of sampling layers and sample size of the LHS based on a variance reduction rate feedback mechanism; and triggering sampling re-iteration when a sudden change in new energy output or a load peak is detected, ensuring that the multi-scenario set captures short-term fluctuation characteristics and improves the fit of typical scenarios to the real-time power grid.

[0008] Preferably, the parameters for dynamically adjusting the Latin hypercube sampling are based on a prediction model trained on a long short-term memory network using historical power grid data. Specifically, the long short-term memory network is trained using historical power grid data, with the input being the output time series of new energy sources, load change trends, and historical fault records. The output is a priority list of sampling dimensions and a distribution function correction coefficient, achieving parameter pre-adjustment 1-6 hours in advance. When the prediction confidence level is ≥90%, the pre-adjusted parameters are directly loaded to reduce online calculation latency.

[0009] Preferably, in step S4, a hybrid distance metric model is constructed by integrating Kantorovich distance and entropy weighting. Specifically, when calculating Kantorovich distance, information entropy theory is integrated, and each scene is assigned a weight factor through entropy weighting to quantify its uncertainty contribution. The normalized hybrid distance formula is designed as: D_hybrid=α*D_kantorovich+β*H_entropy, where α and β are adjustable coefficients, D_kantorovich is the Kantorovich distance, and H_entropy is the scene entropy value. A greedy algorithm is applied to optimize the reduction process to reduce the number of iterations, while ensuring the consistency of the probability distribution of typical scene sets.

[0010] Preferably, the hybrid distance metric model further includes a hierarchical parallel computing architecture to accelerate the hybrid distance, specifically: the initial scene set is divided into K subsets, where K≥4, and each subset is assigned to an independent computing node; each node performs hybrid distance calculation and local scene reduction in parallel to generate intermediate typical scene subsets; the intermediate results are aggregated through the master node, and quadratic Kantorovich distance fusion is used to ensure the consistency of the global probability distribution.

[0011] Preferably, step S5 further includes an intelligent decision-making module, specifically: in the power grid collaborative optimization stage, a deep reinforcement learning agent is deployed. The agent is based on the Q-learning framework, and its inputs are a typical scenario set and the real-time power grid status. The agent outputs adaptive control instructions, including power flow adjustment thresholds, fault recovery priorities, and energy storage charging and discharging strategies. The agent is trained through a reward function to achieve scenario-driven dynamic closed-loop optimization.

[0012] Preferably, the reward function is to minimize grid loss or maximize renewable energy consumption. The intelligent decision-making module embeds a safety boundary verification module, which is used to constrain the decision space of the reinforcement learning agent. Specifically, it defines the grid safety operation domain, including voltage stability margin, DC commutation failure threshold, and N-1 fault passage. Before the agent outputs the control command, it verifies whether it satisfies the safety boundary equation set: V min ≤V i ≤V max Node voltage constraints; P dc_loss ≤P dc_limitDC loss constraint; Δƒ≤0.5H, frequency deviation constraint; secondary correction of over-limit commands to ensure the transient stability of the power grid during typical application scenarios.

[0013] Preferably, the construction method further includes multi-timescale collaborative construction of scenarios: dividing typical scenario sets into short-time, medium-time, and long-time scales, where short-time is at the minute level, medium-time is at the hour level, and long-time is at the day level; designing differentiated optimization objectives for each scale, with the short-time scale focusing on rapid fault isolation, the medium-time scale optimizing power balance, and the long-time scale planning grid expansion; and achieving cross-scale data synchronization through a scenario coupling module to generate a comprehensive construction scheme.

[0014] Preferably, the scenario coupling module further includes cross-scale scenario dynamic verification, which includes the following: establishing a digital twin of the AC / DC hybrid power grid and mapping the physical power grid topology and equipment status in real time; inputting a typical scenario set into the twin to simulate the execution effect of short-term, medium-term, and long-term scale optimization schemes; and correcting the scenario weights based on the simulation results to generate a Pareto optimal solution set.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention calculates the Kantorovich distance of each scenario in an initial multi-scenario set, where the Kantorovich distance is defined as the optimal transportation cost between probability distributions. It minimizes the differences between scenarios through linear programming, iteratively merges similar scenarios and retains scenarios with high probability weights based on a preset reduction threshold, and finally reduces the number of typical scenarios to 50-100. This improves the computational efficiency of the grid collaborative construction method, reduces the scenario generation time from 6.2 hours in the traditional Monte Carlo method to 1.5 hours, and enhances the coverage of typical scenarios to improve coverage of extreme weather conditions. 2. This invention trains a Long Short-Term Memory (LSTM) network using historical power grid data. The input includes the power output time series of new energy sources, load change trends, and historical fault records. The output includes a sampling dimension priority list and distribution function correction coefficients, enabling parameter pre-adjustment 1-6 hours in advance. When the prediction confidence level is ≥90%, the pre-adjusted parameters are directly loaded, reducing online computation latency. An LSTM model is trained to predict the wind and solar power fluctuations of the next day, outputting the sampling dimension priority as wind / solar power > load > fault. When the predicted power output fluctuation is >25%, the number of sampling layers is automatically increased to 20, and the sample size is expanded to 2000. The Kantorovich distance fusion entropy weight method (α=0.7, β=0.3) generates 60 typical scenarios. Resampling is triggered during midday load peaks, and the scenario set is updated within 30 seconds. This reduces the parameter adjustment latency from 15 minutes to 22 seconds, improving the robustness of scenarios to prediction errors and increasing accuracy. 3. This invention divides the initial scene set into K subsets, where K≥4, and each subset is assigned to an independent computing node. Each node performs hybrid distance calculation and local scene reduction in parallel to generate intermediate typical scene subsets. The intermediate results are aggregated by the master node, and a second Kantorovich distance fusion is used to ensure the consistency of the global probability distribution. The 10,000 initial scenes are divided into 8 subsets and assigned to 8 GPU nodes. Each node solves D_hybrid=0.6D_kantorovich+0.4H_entropy in parallel. The master node performs a second fusion of the intermediate results, and finally retains 100 typical scenes. The tidal calculation task is dynamically allocated to idle nodes, which reduces the total time of scene reduction, meets the real-time scheduling requirements, and reduces server resource consumption. 4. This invention ensures the transient stability of the power grid during typical application scenarios by performing secondary corrections on over-limit commands. It sets a voltage safety range of 0.95pu-1.05pu and a commutation failure threshold of <5 times per year. The DQN network inputs typical scenarios and real-time status, outputs energy storage charging and discharging commands, automatically corrects over-limit commands, and prioritizes the disconnection of non-critical loads. This achieves the functions of reducing the decision-making safety violation rate to make risks controllable, shortening fault recovery time, and improving energy storage utilization. Attached Figure Description

[0016] Figure 1 This is a comparison data table of the generation and reduction of the basic scene in this invention; Figure 2 This is a comparison data table of dynamic parameter optimization scenarios for the present invention; Figure 3 A comparison data table of accelerated computation using the parallel architecture of this invention; Figure 4 This is a comparison data table of safety constraint reinforcement learning for the present invention; Figure 5 This is a multi-scale digital twin verification comparison data table for the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments 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 scope of protection of the present invention.

[0018] Comparative examples 1-5 were constructed using existing data, employing Monte Carlo sampling, Euclidean distance reduction, and manual decision-making.

[0019] The test conditions for Examples 1-5 and Comparative Examples 1-5 are as follows: AC / DC hybrid power grid: three ±800kV DC lines, 220kV AC ring network, renewable energy penetration rate of 35%, peak load of 12GW; Hardware environment: CPU: 2 x Intel Xeon Gold 6348 @ 2.6GHz, GPU: NVIDIA A100 x 4, RAM: 512GB DDR4; Software platform: CloudPSSv4.0 including renewable energy fluctuation simulation library, PSCADv5.0.3; Dataset: Wind and solar power output: 10 years of historical data from the National Meteorological Administration, resolution 1min; Load: 5 years of dispatch records from provincial power grids; Evaluation period: continuous operation for 30 days, simulating 4 extreme weather events, namely typhoons and sandstorms.

[0020] Example 1: Please refer to Figure 1 A method for collaboratively constructing an AC / DC hybrid power transmission grid, the method comprising the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, which includes fluctuations in wind and solar power output, random changes in load, and the probability distribution of failures; Step S3: Generate a multi-scene set using Latin hypercube sampling; Step S4: Reduce the scene set using a scene reduction technique based on Kantorovich distance to obtain typical scenes; Step S5: Apply typical scenarios to perform grid collaborative optimization. The grid collaborative optimization includes power flow calculation, fault simulation and resource scheduling, and outputs a construction scheme for AC / DC hybrid power grid. The construction scheme covers the grid topology, protection device configuration and operation strategy. Step S3 specifically involves: initializing the sampling dimensions, which correspond to key indicators of the uncertainty variable set, including output fluctuation rate, load deviation coefficient, and fault occurrence frequency. The Latin hypercube sampling algorithm is applied to generate uniformly distributed scenario samples within the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range. An initial multi-scenario set is generated, containing more than 1000 scenarios. The historical power grid data includes AC grid parameters, DC grid parameters, renewable energy output data, and load demand data. The AC grid parameters include node voltage, line impedance, and transformer capacity. The DC grid parameters include converter station characteristics, DC... The flow line resistance and power modulation capability, specifically step S4, involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set, where the Kantorovich distance is defined as the optimal transportation cost between probability distributions; minimizing the differences between scenarios through linear programming; iteratively merging similar scenarios and retaining scenarios with high probability weights according to a preset reduction threshold; and finally reducing the number of typical scenarios to 50-100. This improves the computational efficiency of the grid collaborative construction method, reducing the scenario generation time from 6.2 hours in the traditional Monte Carlo method to 1.5 hours, and enhancing the coverage of typical scenarios to improve coverage of extreme weather conditions.

[0021] Example 2: Please refer to Figure 2A method for collaboratively constructing an AC / DC hybrid power grid for power transmission includes the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, including wind and solar power output fluctuations, random load changes, and fault probability distribution; Step S3: Generate a multi-scenario set using Latin hypercube sampling; Step S4: Reduce the scenario set using a scenario reduction technique based on Kantorovich distance to obtain typical scenarios; Step S5: Apply typical scenarios to perform power grid collaborative optimization, which includes power flow calculation, fault simulation, and resource scheduling, and output a construction scheme for the AC / DC hybrid power grid, the construction scheme covering the power grid... The topology, protection device configuration, and operation strategy are defined. Step S3 specifically involves: initializing sampling dimensions, where each dimension corresponds to key indicators of the uncertainty variable set, including output fluctuation rate, load deviation coefficient, and fault occurrence frequency. A Latin hypercube sampling algorithm is applied to generate uniformly distributed scenario samples within the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range. This generates an initial multi-scenario set containing more than 1000 scenarios. The historical power grid data includes AC grid parameters, DC grid parameters, renewable energy output data, and load demand data. The AC grid parameters include node voltage, line impedance, and transformer capacity. The DC grid parameters include converter... The characteristics of the station, DC line resistance, and power modulation capability are considered. Step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set. The Kantorovich distance is defined as the optimal transportation cost between probability distributions. Linear programming is used to minimize the differences between scenarios. Based on a preset reduction threshold, similar scenarios are iteratively merged while retaining scenarios with high probability weights, ultimately reducing the number of typical scenarios to 50-100. In step S3, dynamically adjusting the parameters of the Latin hypercube sampling also includes optimizing the scenario generation accuracy. Specifically, this involves: real-time monitoring of the power grid operation data stream, identifying the non-steady-state characteristics of uncertain variables through an online learning algorithm, and dynamically adjusting the... New sampling dimensions and distribution functions; the introduction of a sensitivity analysis module, based on a variance reduction rate feedback mechanism, automatically adjusts the number of sampling layers and sample size of the LHS; when a sudden change in new energy output or a load peak is detected, parameter adjustment triggers sampling re-iteration to ensure that the multi-scenario set captures short-term fluctuation characteristics and improves the fit of typical scenarios to the real-time power grid. The dynamic adjustment of the parameters of the Latin hypercube sampling is based on a prediction model trained with a long short-term memory network using historical power grid data. Specifically, the long short-term memory network is trained using historical power grid data, with the input being the time series of new energy output, load change trends, and fault history records; the output is a sampling dimension priority list and distribution function correction coefficients, achieving parameter pre-adjustment 1-6 hours in advance.When the prediction confidence level is ≥90%, pre-tuned parameters are directly loaded to reduce online computation latency. An LSTM model is trained to predict the next day's wind and solar load fluctuations. The output sampling dimension priority is wind and solar load > load > fault. When the predicted output fluctuation is >25%, the number of sampling layers is automatically increased to 20, and the sample size is expanded to 2000. The Kantorovich distance fusion entropy weight method (α=0.7, β=0.3) generates 60 typical scenarios. Resampling is triggered during midday load peaks, and the scenario set is updated within 30 seconds. This reduces the parameter adjustment latency from 15 minutes to 22 seconds, improving the robustness of scenarios to prediction errors and increasing accuracy.

[0022] Example 3: Please refer to Figure 3A method for collaboratively constructing an AC / DC hybrid power grid for power transmission includes the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, including wind and solar power output fluctuations, random load changes, and fault probability distribution; Step S3: Generate a multi-scenario set using Latin hypercube sampling; Step S4: Reduce the scenario set using a scenario reduction technique based on Kantorovich distance to obtain typical scenarios; Step S5: Apply typical scenarios to perform power grid collaborative optimization, which includes power flow calculation, fault simulation, and resource scheduling, outputting a construction scheme for the AC / DC hybrid power grid, the construction scheme covering the power grid topology, Protection device configuration and operation strategy; Step S3 specifically involves: initializing sampling dimensions, where each dimension corresponds to key indicators of the uncertainty variable set, including output fluctuation rate, load deviation coefficient, and fault occurrence frequency; applying the Latin hypercube sampling algorithm to generate uniformly distributed scenario samples within the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range; generating an initial multi-scenario set containing more than 1000 scenarios; the historical power grid data including AC power grid parameters, DC power grid parameters, renewable energy output data, and load demand data; the AC power grid parameters including node voltage, line impedance, and transformer capacity; and the DC power grid parameters including converter station characteristics and DC line resistance. Regarding power modulation capability, step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set. The Kantorovich distance is defined as the optimal transportation cost between probability distributions. The difference between scenarios is minimized by solving linear programming. Based on a preset reduction threshold, similar scenarios are iteratively merged while retaining scenarios with high probability weights, ultimately reducing the number of typical scenarios to 50-100. In step S4, a hybrid distance metric model is constructed by integrating Kantorovich distance and entropy weighting. Specifically, when calculating the Kantorovich distance, information entropy theory is integrated, and each scenario is assigned a weight factor through entropy weighting to quantify its uncertainty contribution. The normalized hybrid distance formula is designed as: D_hybrid=α*D_kantorovich+β*H_entropy, where α and β are adjustable coefficients, D_kantorovich is the Kantorovich distance, and H_entropy is the scene entropy value. A greedy algorithm is applied to optimize the reduction process to reduce the number of iterations, while ensuring the consistency of the probability distribution of typical scene sets. The hybrid distance metric model also includes a hierarchical parallel computing architecture to accelerate the hybrid distance. Specifically, the initial scene set is divided into K subsets, where K≥4, and each subset is assigned to an independent computing node. Each node performs hybrid distance calculation and local scene reduction in parallel to generate intermediate typical scene subsets.By aggregating intermediate results through the master node and employing quadratic Kantorovich distance fusion to ensure global probability distribution consistency, 10,000 initial scenes are divided into 8 subsets and distributed across 8 GPU nodes. Each node solves D_hybrid=0.6D_kantorovich+0.4H_entropy in parallel. The master node then performs a second fusion of intermediate results, ultimately retaining 100 typical scenes. Tidal calculation tasks are dynamically allocated to idle nodes, achieving reduced overall time for scene reduction, meeting real-time scheduling requirements, and minimizing server resource consumption.

[0023] Example 4: Please refer to Figure 4A method for collaboratively constructing an AC / DC hybrid power grid for power transmission includes the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, including wind and solar power output fluctuations, random load changes, and fault probability distribution; Step S3: Generate a multi-scenario set using Latin hypercube sampling; Step S4: Reduce the scenario set using a scenario reduction technique based on Kantorovich distance to obtain typical scenarios; Step S5: Apply typical scenarios to perform power grid collaborative optimization, which includes power flow calculation, fault simulation, and resource scheduling, and outputs a construction scheme for the AC / DC hybrid power grid. The proposed construction scheme encompasses the power grid topology, protection device configuration, and operation strategy. Step S3 specifically involves: initializing sampling dimensions, where each dimension corresponds to key indicators of the uncertainty variable set, including output fluctuation rate, load deviation coefficient, and fault occurrence frequency. The Latin hypercube sampling algorithm is applied to generate uniformly distributed scenario samples within the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range. This generates an initial multi-scenario set containing more than 1000 scenarios. The historical power grid data includes AC grid parameters, DC grid parameters, renewable energy output data, and load demand data. The AC grid parameters include node voltage and line resistance. The DC grid parameters include converter station characteristics, DC line resistance, and power modulation capability. Step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set, where the Kantorovich distance is defined as the optimal transportation cost between probability distributions; minimizing the differences between scenarios through linear programming; iteratively merging similar scenarios and retaining scenarios with high probability weights based on a preset reduction threshold; and finally reducing the number of typical scenarios to 50-100. Step S5 also includes an intelligent decision-making module, specifically: deploying a deep reinforcement learning agent during the grid collaborative optimization phase. Based on the Q-learning framework, the input consists of a set of typical scenarios and real-time grid status. The agent outputs adaptive control commands, including power flow adjustment thresholds, fault recovery priorities, and energy storage charging and discharging strategies. The agent is trained through a reward function to achieve scenario-driven dynamic closed-loop optimization. The reward function minimizes grid losses or maximizes renewable energy consumption. An intelligent decision-making module embeds a safety boundary verification module, which constrains the decision space of the reinforcement learning agent. Specifically, it defines the grid safety operating domain, including voltage stability margin, DC commutation failure threshold, and N-1 fault passage. Before the agent outputs control commands, it verifies whether they satisfy the safety boundary equations: V min ≤V i ≤V max Node voltage constraints; P dc_loss ≤P dc_limitDC loss constraint; Δƒ≤0.5H, frequency deviation constraint; secondary correction of over-limit commands to ensure grid transient stability during typical application scenarios; voltage safety range set at 0.95pu-1.05pu; commutation failure threshold <5 times per year; DQN network input of typical scenario and real-time status, output of energy storage charging and discharging commands; verification before command execution: ; Automatic correction of over-limit commands and priority cut-off of non-critical loads have enabled the decision-making safety violation rate to be reduced, making risks controllable, shortening fault recovery time, and improving energy storage utilization.

[0024] Example 5: Please refer to Figure 5A method for collaboratively constructing an AC / DC hybrid power grid for power transmission includes the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, including wind and solar power output fluctuations, random load changes, and fault probability distribution; Step S3: Generate a multi-scenario set using Latin hypercube sampling; Step S4: Reduce the scenario set using a scenario reduction technique based on Kantorovich distance to obtain typical scenarios; Step S5: Apply typical scenarios to perform power grid collaborative optimization, which includes power flow calculation, fault simulation, and resource scheduling, outputting an AC / DC hybrid power grid. The network construction scheme covers the power grid topology, protection device configuration, and operation strategy. Step S3 specifically involves: initializing sampling dimensions, where each dimension corresponds to key indicators of the uncertainty variable set, including output fluctuation rate, load deviation coefficient, and fault occurrence frequency. The Latin hypercube sampling algorithm is applied to generate uniformly distributed scenario samples within the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range, generating an initial multi-scenario set containing more than 1000 scenarios. The historical power grid data includes AC power grid parameters, DC power grid parameters, renewable energy output data, and load demand data. The network parameters include node voltage, line impedance, and transformer capacity. The DC grid parameters include converter station characteristics, DC line resistance, and power modulation capability. Step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set. The Kantorovich distance is defined as the optimal transportation cost between probability distributions. The differences between scenarios are minimized through linear programming. Based on a preset reduction threshold, similar scenarios are iteratively merged while retaining scenarios with high probability weights, ultimately reducing the number of typical scenarios to 50-100. The construction method also includes multi-timescale collaborative scenario construction: typical scenarios... The scenario set is divided into short-term, medium-term, and long-term scales, with short-term at the minute level, medium-term at the hour level, and long-term at the day level. Differentiated optimization objectives are designed for each scale: the short-term scale focuses on rapid fault isolation, the medium-term scale optimizes power balance, and the long-term scale plans grid expansion. Cross-scale data synchronization is achieved through a scenario coupling module to generate a comprehensive construction scheme. The scenario coupling module also includes cross-scale scenario dynamic verification, which includes: establishing a digital twin of the AC / DC hybrid power grid and mapping the physical grid topology and equipment status in real time; inputting typical scenario sets into the twin to simulate the execution effect of the short-term, medium-term, and long-term scale optimization schemes.Based on simulation results feedback, scenario weights are adjusted to generate Pareto optimal solution sets. Short-term scenarios include wind turbine disconnection, medium-term scenarios include load transfer, and long-term scenarios include grid expansion planning. A virtual grid with 120,000 nodes is constructed, mapping the physical equipment aging model. Short-term scenarios test protection action timing, and long-term scenarios evaluate the lifetime benefits of expansion schemes. Pareto optimization outputs a solution that balances economy and reliability, achieving a reduced number of scheme adjustments and a lower physical grid failure rate with closed-loop verification functionality.

[0025] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for the coordinated construction of an AC / DC hybrid power grid for external transmission, characterized in that, The construction method includes the following steps: Step S1: Collect historical power grid data; Step S2: Based on the collected data, define a set of uncertain variables, which includes fluctuations in wind and solar power output, random changes in load, and fault probability distribution; Step S3: Generate a multi-scene set using Latin hypercube sampling; Step S4: Reduce the scene set using a scene reduction technique based on Kantorovich distance to obtain typical scenes; Step S5: Apply typical scenarios to perform grid collaborative optimization. The grid collaborative optimization includes power flow calculation, fault simulation and resource scheduling, and outputs a construction scheme for AC / DC hybrid power grid. The construction scheme covers the grid topology, protection device configuration and operation strategy. Step S3 specifically involves: initializing the sampling dimension, which corresponds to the key indicators of the set of uncertain variables, including output fluctuation rate, load deviation coefficient and failure frequency; applying the Latin hypercube sampling algorithm to generate uniformly distributed scene samples in the probability distribution space, ensuring that the variable values ​​of each dimension are independent and fully cover their distribution range, and generating an initial multi-scene set containing more than 1000 scenes.

2. The method for collaborative construction of an AC / DC hybrid power grid according to claim 1, characterized in that: The historical power grid data includes AC power grid parameters, DC power grid parameters, renewable energy output data, and load demand data. The AC power grid parameters include node voltage, line impedance, and transformer capacity. The DC power grid parameters include converter station characteristics, DC line resistance, and power modulation capability. Step S4 specifically involves: calculating the Kantorovich distance of each scenario in the initial multi-scenario set. The Kantorovich distance is defined as the optimal transportation cost between probability distributions. The differences between scenarios are minimized by solving linear programming. Based on a preset reduction threshold, similar scenarios are iteratively merged while retaining scenarios with high probability weights, ultimately reducing the number of typical scenarios to 50-100.

3. The method for collaborative construction of an AC / DC hybrid power grid according to claim 2, characterized in that: In step S3, dynamically adjusting the parameters of Latin hypercube sampling also includes optimizing the scene generation accuracy. Specifically, this involves: real-time monitoring of power grid operation data streams, identifying the non-steady-state characteristics of uncertain variables through online learning algorithms, and dynamically updating the sampling dimension and distribution function; introducing a sensitivity analysis module, and automatically adjusting the number of sampling layers and sample size of LHS based on a variance reduction rate feedback mechanism; and triggering sampling re-iteration when a sudden change in new energy output or a load peak is detected, to ensure that the multi-scenario set captures short-term fluctuation characteristics and improves the fit of typical scenarios to the real-time power grid.

4. The method for collaborative construction of an AC / DC hybrid power grid according to claim 3, characterized in that: The dynamically adjusted parameters of the Latin hypercube sampling are based on a prediction model trained with a long short-term memory network using historical power grid data. Specifically, the long short-term memory network is trained using historical power grid data, with the input being the output time series of new energy sources, load change trends, and historical fault records. The output is a priority list of sampling dimensions and a distribution function correction coefficient, achieving parameter pre-adjustment 1-6 hours in advance. When the prediction confidence level is ≥90%, the pre-adjusted parameters are directly loaded to reduce online calculation latency.

5. The method for collaborative construction of an AC / DC hybrid power grid according to claim 1, characterized in that: In step S4, a hybrid distance metric model is constructed by integrating Kantorovich distance and entropy weighting. Specifically, when calculating Kantorovich distance, information entropy theory is integrated, and weight factors are assigned to each scene using entropy weighting to quantify their uncertainty contribution. The normalized hybrid distance formula is designed as: D_hybrid=α*D_kantorovich+β*H_entropy, where α and β are adjustable coefficients, D_kantorovich is the Kantorovich distance, and H_entropy is the scene entropy value. A greedy algorithm is applied to optimize the reduction process to reduce the number of iterations while ensuring the consistency of the probability distribution of typical scene sets.

6. The method for collaborative construction of an AC / DC hybrid power grid according to claim 5, characterized in that: The hybrid distance metric model also includes a hierarchical parallel computing architecture to accelerate hybrid distance, specifically: the initial scene set is divided into K subsets, where K≥4, and each subset is assigned to an independent computing node; each node performs hybrid distance calculation and local scene reduction in parallel to generate intermediate typical scene subsets; the intermediate results are aggregated through the master node, and quadratic Kantorovich distance fusion is used to ensure the consistency of the global probability distribution.

7. The method for collaborative construction of an AC / DC hybrid power grid according to claim 1, characterized in that: Step S5 also includes an intelligent decision-making module, specifically: in the grid collaborative optimization stage, a deep reinforcement learning agent is deployed. The agent is based on the Q-learning framework, and the input is a typical scenario set and the real-time grid status. The agent outputs adaptive control instructions, including power flow adjustment threshold, fault recovery priority and energy storage charging and discharging strategy. The agent is trained through a reward function to achieve scenario-driven dynamic closed-loop optimization.

8. The method for collaborative construction of an AC / DC hybrid power grid according to claim 7, characterized in that: The reward function is to minimize grid loss or maximize renewable energy consumption. The intelligent decision-making module embeds a safety boundary verification module, which constrains the decision space of the reinforcement learning agent. Specifically, it defines the grid safety operation domain, including voltage stability margin, DC commutation failure threshold, and N-1 fault passage. Before the agent outputs control commands, it verifies whether they satisfy the safety boundary equations: V min ≤V i ≤V max Node voltage constraints; P dc_loss ≤P dc_limit DC loss constraint; Δƒ≤0.5H, frequency deviation constraint; secondary correction of over-limit commands to ensure the transient stability of the power grid during typical application scenarios.

9. The method for collaborative construction of an AC / DC hybrid power grid according to claim 1, characterized in that: The construction method also includes multi-timescale collaborative construction of scenarios: the typical scenario set is divided into short-time, medium-time and long-time scales, where short-time is at the minute level, medium-time is at the hour level and long-time is at the day level; differentiated optimization objectives are designed for each scale, with the short-time scale focusing on rapid fault isolation, the medium-time scale optimizing power balance, and the long-time scale planning grid expansion; Cross-scale data synchronization is achieved through the scene coupling module, generating a comprehensive construction solution.

10. A method for collaborative construction of an AC / DC hybrid power grid for external transmission according to claim 9, characterized in that: The scenario coupling module also includes cross-scale scenario dynamic verification, which includes the following: establishing a digital twin of the AC / DC hybrid power grid and mapping the physical power grid topology and equipment status in real time; inputting typical scenario sets into the twin to simulate the execution effect of short-term, medium-term, and long-term scale optimization schemes; and correcting scenario weights based on simulation results to generate a Pareto optimal solution set.

Images