Quantum-coherent Ising machine-based power grid topology candidate optimization method and system

Abstract

The present disclosure provides a power grid topology candidate optimization method and system based on quantum coherent Ising machine sampling, relating to the technical field of power system operation control and quantum computing, comprising: training a reward agent network to guide the construction of an efficient topology library; filtering the topology with N-1 safety margin as a hard constraint; constructing a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable, submitting the coherent Ising machine after Ising transformation and 8-bit equal scaling, and outputting multiple low-energy state samples in parallel through single annealing; after energy sorting, deduplication, truncation and base repair, the average pairwise Hamming distance in the pool is used as the diversity criterion to optimize the candidate pool output. The present disclosure repositions the coherent Ising machine from an "optimal solution" tool to a "diverse candidate set generator", which can output a candidate pool considering quality and diversity for dispatchers to choose at one time. The present disclosure realizes the rapid generation of a candidate pool considering quality and diversity.

Description

Technical Field

[0001] This disclosure relates to the interdisciplinary field of power system operation control and quantum computing, specifically to a method and system for optimizing candidate power grid topologies based on quantum coherent Ising machine sampling. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] The high proportion of renewable energy integration in the new power system puts provincial power grids under dual pressures: increased volatility in net load on the source side and frequent grid-side congestion. Given that the unit combination scheme has already been determined, dispatchers need to make transmission line switching (topology adjustment) decisions on a 15-minute timescale within the day to alleviate congestion; however, this decision faces a huge combination search space.

[0004] In existing technologies, transmission line switching problems are typically solved using mixed-integer programming or metaheuristic algorithms (such as simulated annealing and genetic algorithms). These methods have the advantage of mature theory and the ability to obtain globally optimal or near-optimal solutions, but the solution time increases rapidly with the number of candidate lines, making it difficult to meet the real-time requirements of 15-minute online decision windows on large-scale systems. Another type of dedicated quantum computing hardware, such as the Coherent Ising Machine (CIM), offers advantages such as single-annealing parallel exploration of the solution space, room-temperature operation, fully connected topologies, and millisecond-level solution latency, providing a new approach for large-scale combinatorial optimization. However, applying CIM to power grid topology pre-selection still faces the following technical challenges: (1) Existing methods mostly use CIM to solve a single minimum energy solution, and the output is an optimal topology rather than a set of diverse candidate solutions. Schedulers lack comparable alternative solutions for manual secondary judgment. (2) CIM single annealing actually produces a large number of low-energy state samples in parallel. Existing methods only take the lowest energy solution and discard the rest of the samples, which does not make full use of this sampling characteristic to build a candidate pool that takes into account both quality and diversity. (3) If the candidate topology library is generated in a uniform and random manner, the proportion of high-value topologies in the library will be sparse, which will waste the expensive DC optimal power flow (DC-OPF) safety and cost assessment budget on these topologies. (4) In the optimization model, the N-1 safety margin and economic cost are indicators of different natures. If the safety constraint is optimized in the form of a penalty term and mixed with the cost target, the safety margin, cost, and diversity will be compromised, making it difficult to guarantee that each topology in the candidate pool strictly meets the hard safety access requirements. Summary of the Invention

[0005] To address the aforementioned issues, this disclosure proposes a method and system for power grid topology candidate optimization based on quantum coherent Ising machine sampling. By transforming the large-scale topology selection problem into a quadratic unconstrained binary optimization (QUBO) model, multiple low-energy state samples are generated in parallel using the Boltzmann sampling characteristics of the coherent Ising machine. A candidate pool that balances quality and diversity is selected using the average pairwise Hamming distance (APD) criterion. Filtering is performed with an N-1 safety margin as a hard constraint, and an efficient topology library is constructed by guiding the construction through a reward agent network.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions: A power grid topology candidate optimization method based on quantum coherent Ising machine sampling includes: Candidate lines were selected based on power grid operation mode and comprehensive influencing factors; Randomly perturb the candidate lines, use a reward agent network to guide the generation of feasible topologies and build a topology library; A QUBO objective function is constructed using "whether the topology is selected into the candidate pool" as the decision variable; After transforming the QUBO objective function into the Ising form, it is scaled proportionally based on the coefficient accuracy of the coherent Ising machine and submitted to the coherent Ising machine in sampling mode. Multiple low-energy state samples are generated by parallel solving in a single annealing. After sorting, deduplicating, truncating and repairing the energy of the low-energy samples, the average pairwise Hamming distance within the pool for each candidate combination is calculated, and the combination with the greatest diversity is selected as the candidate pool output.

[0007] As one embodiment, the step of randomly perturbing candidate lines, using a reward agent network to guide the generation of feasible topologies, and constructing a topology library includes: Training a reward agent network to approximate the immediate reward of a single-step flip action; exist N s The sum of the absolute values ​​of the reward agent outputs of each candidate line flipping action is aggregated under each random system state to obtain the line sampling weight containing the value bias component and the uniform coverage component. The flipped lines are selected according to the sampling weights of the lines and their feasibility is verified to generate feasible topologies, which are then constructed into a topology library.

[0008] As one embodiment, a QUBO objective function is constructed using "whether the topology is selected into the candidate pool" as the decision variable, including: With "the first k Whether a topology is selected into the candidate pool is a binary decision variable. Construct the QUBO objective function:

[0009] in, For topology Normalized cost; For topology With topology The normalized Hamming distance between them; Number of candidate lines; For diversity weight; Cardinality-constrained weights; The size of the candidate pool; The topology library size is represented by the first term, the topology quality term, the diversity reward term, and the cardinality constraint term. The decision variables are: The connection / disconnection status coding of the topology itself They belong to different levels; Representing topology respectively i Topology j The candidate pool decision variable takes the value {0,1}, where 1 indicates that the topology is selected into the candidate pool and 0 indicates that it is not selected. Representing topology respectively i Topology j Its own line connectivity status encoding vector; i , j For topology index, i < j This represents the summation of the diversity term over all topological pairs in the pool.

[0010] As one embodiment, the process of transforming the QUBO objective function into the Ising form and then scaling it proportionally based on the coefficient accuracy of the coherent Ising machine, followed by parallel solving via single annealing to produce multiple low-energy state samples, includes: First, the QUBO objective function is processed... Transform into the Ising Hamiltonian:

[0011] in, For Ising spin variable, The local field coefficients acting on a single spin. The coupling coefficient between spin pairs; They represent the first i The, the j One Ising spin variable; i , j Spin index, i < j This indicates that the coupling term is summed over all spin pairs, with each pair counted only once.

[0012] Secondly, based on the scaling factor α Perform global proportional scaling to [ [127,127] Integer range adapted to the 8-bit coefficient precision of the coherent Ising machine hardware; submitted to the coherent Ising machine in sampling mode, producing N low-energy spin samples in parallel with a single annealing.

[0013] As one embodiment, the step of sorting, deduplicating, truncating, and restoring the low-energy state samples by energy, calculating the average pairwise Hamming distance within each candidate combination, and selecting the combination with the highest diversity as the candidate pool output includes: For low-energy state samples, sort by energy, remove duplicates, truncate and retain the top M samples. For each candidate combination, repair the base number to K. If there are fewer than K, add them to the unselected topology in ascending order of energy. If there are more than K, remove them in descending order of energy. After energy sorting, deduplication, energy truncation and cardinality repair, the pool APD of each candidate combination is calculated according to the APD selection criterion.

[0014] As one embodiment, a safety filtering operation is performed before constructing the QUBO objective function or after sampling and optimization. The safety filtering operation uses an N-1 safety margin as a hard constraint to remove topologies with a safety margin lower than a threshold, so that the candidate pool contains only schemes that satisfy the N-1 safety requirement.

[0015] According to some embodiments, the present disclosure adopts the following technical solutions: A power grid topology candidate optimization system based on quantum coherent Ising machine sampling includes: The screening module is used to screen candidate lines based on the power grid operation mode and comprehensive impact factors. The topology library construction module is used to randomly perturb candidate lines, guide the generation of feasible topologies using a reward agent network, and build a topology library. The objective function construction module is used to construct a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable; The objective function uses an optimization module to transform the QUBO objective function into the Ising form and then scale it proportionally according to the coefficient accuracy of the coherent Ising machine. The sampled function is then submitted to the coherent Ising machine and solved in parallel by a single annealing to produce multiple low-energy state samples. The output selection module is used to sort, deduplicate, truncate and repair the low-energy state samples by energy, calculate the average pairwise Hamming distance of each candidate combination in the pool, and select the one with the greatest diversity as the candidate pool output.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0017] According to some embodiments, the present disclosure adopts the following technical solutions: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0019] Compared with the prior art, the beneficial effects of this disclosure are as follows: This disclosure presents a power grid topology candidate optimization method based on quantum coherent Ising machine sampling. The method consists of three mutually cooperating functional steps: security filtering, QUBO modeling, and sampling optimization, supplemented by a reward-guided topology library construction. Security filtering can be performed before sampling optimization (pre-filtering to reduce the solution space) or after sampling optimization (post-filtering to eliminate unsafe solutions obtained from sampling). Furthermore, at each decision step, the method adaptively determines whether to re-execute the sampling optimization operation based on three types of signals: system state changes, historical solution quality decay, and timeliness, thereby achieving rapid generation of a candidate pool that balances quality and diversity.

[0020] This disclosure presents a power grid topology candidate optimization method based on quantum coherent Ising machine sampling. The candidate pool paradigm outputs K diverse and safe candidate topologies instead of a single optimal solution. These topologies are interpretable and can be selected by the dispatcher. The method repositions CIM from an "optimization tool" to a "diversity candidate set generator," solving problems such as how to select a candidate pool that balances quality and diversity from a large number of low-energy state samples, and how to pre- / post-filter the N-1 safety margin in the form of hard constraints.

[0021] This disclosed method for power grid topology candidate optimization based on quantum coherent Ising machine sampling features coordinated library construction and sampling. The topology library constructed by rewarding agents enriches high-value topologies within the library, providing a high-quality search space for CIM sampling and APD optimization. Safety is used as a hard constraint filter rather than a soft penalty, ensuring that the overall candidate pool meets N-1 safety requirements. Both the N-1 fault set and the monitored set are limited to those selected by the line outage distribution factor (LODF). There are candidate lines, and the complexity is... Down to .

[0022] This disclosed method for power grid topology candidate optimization based on quantum coherent Ising machine sampling applies quantum computing to the field of power system topology optimization control. It utilizes the characteristic of coherent Ising machines to generate multiple low-energy state samples in parallel during a single annealing process, and adaptive triggering saves quantum resources. The trigger value is constructed by taking the maximum value of three signals: system state change, historical solution quality decay, and timeliness. Quantum computing resources are only re-called when the trigger value exceeds a threshold; otherwise, the historical candidate pool from the previous decision step is reused. Experimental results show that the number of CIM calls is reduced by approximately 58.6% while the quality of the candidate pool remains largely unchanged. Attached Figure Description

[0023] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0024] Figure 1 This is a schematic diagram of the overall process of the CIM topology candidate pool pre-selection method according to an embodiment of the present disclosure; Figure 2 This is a multi-dimensional comparison chart of candidate pool quality and solution speed in embodiments of this disclosure; Figure 3 This diagram illustrates the coupling relationship between diversity weights and 8-bit hardware coefficient precision in embodiments of this disclosure. Figure 4 The candidate pool size N in this embodiment of the disclosure t The impact curve of APD on solution delay; Figure 5 This is a comparison chart of the number of coherent Ising machine calls between the adaptive triggering and the triggering baseline of each step in this embodiment of the present disclosure. Detailed Implementation

[0025] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] Example 1 One embodiment of this disclosure provides a power grid topology candidate optimization method based on quantum coherent Ising machine sampling, the method steps of which include: Step 1: Screen candidate lines based on power grid operation mode and comprehensive influencing factors; Step 2: Randomly perturb the candidate lines, use the reward agent network to guide the generation of feasible topologies and build a topology library; Step 3: Construct a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable; Step 4: After transforming the QUBO objective function into the Ising form, the overall scale is proportionally adjusted according to the coefficient accuracy of the coherent Ising machine. The sampled function is then submitted to the coherent Ising machine in sampling mode, and multiple low-energy state samples are generated by parallel solving in a single annealing process. Step 5: After sorting, deduplicating, truncating and repairing the energy of the low-energy state samples, calculate the average pairwise Hamming distance of each candidate combination in the pool, and select the one with the greatest diversity as the candidate pool output.

[0029] As one embodiment, this disclosure presents a power grid topology candidate optimization method based on quantum coherent Ising machine sampling. This method consists of three mutually cooperating functional steps: security filtering, QUBO objective function modeling, and sampling optimization, supplemented by a topology library construction guided by a reward agent. Security filtering can be performed before sampling optimization (pre-filtering, reducing the solution space) or after sampling optimization (post-filtering, eliminating unsafe solutions obtained from sampling), either order is acceptable. Furthermore, at each decision step, the method adaptively determines whether to re-execute the sampling optimization operation based on three types of signals: system state changes, historical solution quality decay, and timeliness. The specific implementation process is as follows: Step 1: Screen candidate lines based on power grid operation mode and comprehensive influencing factors; Specifically, the selection is based on the LODF (Line Outage Distribution Factor). LODF measures the proportion of power flow shifting to other lines after a line outage, and defines a normalized sensitivity:

[0030] in, For the line The distribution factor vector of the power flow on other lines is interrupted. mean(·) represents the arithmetic mean of the absolute values ​​of the vector components; std(·) represents the standard deviation of the absolute values ​​of the vector components; the ratio of the two measures the power flow on the line. i The degree of concentration of the impact of the flow shift after the shutdown (the higher the ratio, the more concentrated the impact is on a few lines, and the higher the screening value).

[0031] Combining ground state current The comprehensive impact factor was obtained as follows:

[0032] in, Indicates the line i The LODF comprehensive influence factor (taking into account both the transfer ratio and the ground state power level). Indicates the line i The ground-state power flow active power (per unit value); Indicates the line i The normalized outage distribution factor (mean / std ratio); Indicates taking The smaller value in 1 (top truncation normalization to prevent single-line weight overflow); according to The top 30 candidate lines are selected in descending order. LODF's cumulative contribution is approximately 99.94%.

[0033] Step 2: Randomly perturb the candidate lines, use the reward agent network to guide the generation of feasible topologies and build a topology library; Specifically, a reward agent network (hereinafter referred to as Critic) is trained to approximate the instantaneous reward change of a single-step flip action. The Critic input consists of the system state (load rate, net load, predicted characteristics, and current topology coding of the first few heavily loaded lines) concatenated with the one-hot coding of the flip action to be evaluated. The network is a multilayer perceptron (hidden layer dimensions are 64, 32, and 16 respectively, and the output is a scalar). Using the actual single-step cost or reward change calculated by DC-OPF as the regression label and the mean squared error as the loss function, after approximation... The Spearman correlation coefficient of each (state, action, reward) sample trained offline to the validation set is no less than 0.99.

[0034] Furthermore, the specific process of topology generation and topology library construction is to provide an immediate reward for training the reward agent network that approximates a single-step flipping action, in N s The sum of the absolute values ​​of the reward agent outputs for each candidate line flipping action is aggregated under each random system state to obtain the line sampling weights containing the value bias component and the uniform coverage component. Flipping lines are selected according to these weights and their feasibility is verified to generate an offline feasible topology library, as detailed below: exist Predict the flipping action of each candidate line in a random system state. For each candidate line Aggregate the magnitude of its reward change across all states:

[0035]

[0036] in, Candidate lines exist The sum of the magnitudes of the Critic reward changes in each state; This represents the sampling weight of the line; Configurable trade-off ratio (preferred embodiment) ); Configurable number of random state samples (preferred embodiment) , Time-weighted estimation tends to stabilize). Number of candidate lines; This represents the line traversal index when summing all candidate lines; This represents the sum of the magnitudes of the reward changes across all candidate lines, used to normalize the value bias component; It indicates the system status, including the load rate, net load, predicted characteristics, and current topology code of the top few heavily loaded lines; This indicates the line flipping action to be evaluated. =l indicates the on / off state of candidate line l; the first term (value bias component) biases sampling towards high-value lines, and the second term (uniform coverage component) ensures that low-value lines are not missed. Selecting inverted paths without replacement and verifying connectivity and feasibility, a feasible topology library is generated offline. ,in This represents the size of the topology library (i.e., the total number of feasible topologies in the library). Each topology is encoded. Includes offline pre-calculated DC-OPF normalized cost With an N-1 safety margin.

[0037] Furthermore, candidate lines sampling weights ,in , , .

[0038] Step 3: Using "the first" Whether a pre-computed topology is selected into the candidate pool is a binary decision variable. Construct the QUBO objective function:

[0039] in, For topology Normalized cost (containing only cost items, with safety constraints pre-filtered); For topology With topology The normalized Hamming distance between them; Diversity weight (preferred) ); Radix-constrained weights (preferred) ); Candidate pool size (preferred) The first term is the topology quality term (making high-value topologies more likely to be selected), the second term is the diversity reward term (making the greater the difference between topologies, the lower the energy), and the third term is the cardinality constraint term (penalizing the selection of exactly the cardinality). It should be specifically pointed out that: decision variables (Whether the topology is selected) and the on / off status coding of the topology itself. (Which lines are connected / disconnected) belongs to different levels—QUBO is constructed at the "topology selection" level rather than the "line connectivity / disconnection" level, thereby avoiding quadratic terms in line-level coding under the single-step flip constraint (only one line is switched at a time). Two obstacles exist: the constant zero value leads to complete failure, and the cost of DC-OPF cannot be linearly superimposed by independent weights at the line level. Representing topology respectively i Topology j The candidate pool decision variable takes the value {0,1}, where 1 indicates that the topology is selected into the candidate pool and 0 indicates that it is not selected. Representing topology respectively i Topology j Its own line connectivity status encoding vector ( Each component corresponds to the voting / retreat status of a candidate line. i , j For the index of the topology (candidate pool decision variables), i <j This represents the summation of the diversity term over all topological pairs in the pool.

[0040] Furthermore, the diversity reward items are .

[0041] Step 4: After transforming the QUBO objective function into the Ising form, the overall scale is proportionally adjusted according to the coefficient accuracy of the coherent Ising machine. The sampled function is then submitted to the coherent Ising machine in sampling mode, and multiple low-energy state samples are generated by parallel solution through single annealing. Specifically, firstly, the QUBO objective function is processed... Transform into the Ising Hamiltonian: ; in, For Ising spin variable, The local field coefficients acting on a single spin. The coupling coefficient between spin pairs; They represent the first i The, the j Each Ising spin variable (corresponding one-to-one with the topological decision variable): = 1 corresponds to =1, meaning topology k is selected); i , j Indexing for spin (i.e., topological decision variables), i <j This indicates that the coupling term is summed over all spin pairs, with each pair counted only once.

[0042] Secondly, based on the scaling factor α Perform global proportional scaling to [ [127,127] Integer range adapted to the 8-bit coefficient precision of the coherent Ising machine hardware; submitted to the coherent Ising machine in sampling mode, producing N low-energy spin samples in parallel with a single annealing.

[0043] Furthermore, scaling factor α .

[0044] in, α This represents the global proportional scaling factor, which applies to all coefficients of the Hamiltonian; This represents the maximum absolute value of all spin-pair coupling coefficients; This represents the maximum absolute value among all local field coefficients.

[0045] Step 5: After sorting, deduplicating, truncating and repairing the energy of the low-energy samples, calculate the average pairwise Hamming distance of each candidate combination in the pool, and select the one with the greatest diversity as the candidate pool output.

[0046] Specifically, the candidate pool is optimized using the average pairwise Hamming distance within the pool as a measure of diversity. Samples are sorted by energy, deduplicated, and truncated before retention. (preferred) For each candidate combination, the base number is repaired to exactly... Less than Then, fill in the unselected topology in ascending order of energy, if there are more than [a certain number of elements]. Then, remove items in descending order of energy level and calculate the APD within the pool:

[0047] in, For candidate pool; For topology With topology The line connectivity code uses different numbers of bits. The one with the largest APD is selected as the final candidate pool. Output the data and attach an N-1 safety margin label to each topology for the scheduler to make decisions. APD (average pairwise Hamming distance): the average of the different bits in the on / off codes of all topologies in the candidate pool, which is normalized by Lc and ranges from [0,1]. The larger the APD, the greater the difference between the candidate schemes in the pool (the better the diversity), and it is used to select the best candidate pool.

[0048] Furthermore, at each decision step, an adaptive decision is made as to whether to re-execute the sampling optimization operation: Calculate the state change signals respectively Quality attenuation signal and time-sensitive signals All three signals were normalized and truncated to ; The trigger value is constructed by taking its maximum value and then using a linear mapping:

[0049] in, This represents the change in system state between the current decision step and the previous decision step (taking the maximum absolute value of the difference between each state component). The normalization benchmark for representing state changes (preferred embodiment is 0.1); This indicates the initial quality of the candidate pool (the pool score at the time of initial generation). This represents the historical best pool score; age represents the number of decision steps (pool age) that the current candidate pool has undergone since the last refresh. Indicates the upper limit of the candidate pool time limit (10 steps in the preferred embodiment); , These are the lower and upper limits of the trigger amount, respectively (in a preferred embodiment, they are 0.05 and 0.50); ​​when the trigger amount... If the preset threshold is exceeded, the sampling optimization operation is re-executed and the candidate pool is updated; otherwise, the historical candidate pool from the previous decision step is reused.

[0050] As one embodiment, a simulated annealing solver will be used as a backup solver.

[0051] As one embodiment, a safety filtering operation is performed before constructing the QUBO objective function or after sampling and optimization. This safety filtering operation uses an N-1 safety margin as a hard constraint, filtering out operations with a safety margin below a threshold. (Preferred embodiment) The topology is eliminated if the power flow of the line does not exceed 95% of the thermal stability limit, so that the candidate pool contains only schemes that meet the N-1 safety requirement. The verification of the N-1 safety constraint limits both the fault set and the monitored set to a subset of candidate lines selected by the line outage distribution factor.

[0052] Example 2 One embodiment of this disclosure provides a power grid topology candidate optimization method based on quantum coherent Ising machine sampling. The verification platform is an IEEE RTS-96 standard single-area test system (75 buses / 145 branches / 32 generators). The data source is the 2023-2025 forecast data of unified dispatch load and wind and solar power output (15-minute resolution) disclosed by a provincial power grid. The quantum hardware used is a coherent Ising machine (fully connected, room temperature operation, single-annealing millisecond-level accuracy, coefficient accuracy approximately 8-bit) invoked via a cloud platform in sampling mode. The key QUBO parameter values ​​involved are shown in Table 1.

[0053] Table 1. Key Parameter Values ​​for QUBO

[0054] (1) Candidate line screening: number of candidate lines The selection was based on the LODF (Line Outage Distribution Factor). LODF measures the proportion of power flow shifting to other lines after a line outage; a normalized sensitivity is defined. ,in For the line The distribution factor vector of the power flow on the remaining lines is interrupted. Combined with the ground-state power flow... Obtain the comprehensive impact factor ,according to The top 30 entries in descending order are selected, and LODF's cumulative contribution is approximately 99.94%.

[0055] (2) Topology library construction: 2000 feasible topologies were generated from 30 candidate lines by random perturbation (1 to 5 bits flipped, followed by breadth-first search connectivity verification). DC-OPF costs and N-1 safety margins were pre-calculated under four load levels (0.70, 0.85, 1.00, 1.15, per unit). DC-OPF is a linear programming method that solves for minimum cost generation allocation and verifies over-limits given a topology under DC approximation. Topologies with a safety margin lower than the threshold τ=0.05 were filtered online.

[0056] (3) The QUBO model was input into the coherent Ising machine and the traditional classical method for solution comparison. The traditional method includes three types: GreedyQ (greedy cost method, which only calculates the normalized cost). Before ascending order selection (Topology), GreedyD (Greedy Diversity Approach, which explicitly adds a diversity reward within a greedy framework). SA (Simulated Annealing), a metaheuristic algorithm that gradually decreases the temperature parameter, randomly samples the search space, and accepts inferior solutions to escape local optima. The solution results are, for example... Figure 2 As shown in Table 2.

[0057] Table 2 Comparison of candidate pool quality and solution speed ( N t =150, average of 5 scenarios)

[0058] Note: GreedyQ (quality cost method): In each step, the topology that optimizes the quality term of the objective function is greedily selected and added to the candidate pool until K topologies are filled; it is the fastest solution but the resulting solutions are highly similar and the candidate pool has poor diversity. GreedyD (Diversity Greedy Algorithm): Greedy construction compromises between quality and diversity. In each step, it selects a topology that balances quality and has a large Hamming distance from the already selected set. Diversity within the pool is preferred over quality greedy algorithm. SA (Simulated Annealing): A classic metaheuristic solver that iteratively optimizes the solution space using a temperature-controlled random acceptance criterion and runs independently multiple times to produce multiple samples. It serves as a classic reference and backup solver for this invention. CIM (Coherent Ising Machine): The dedicated optical quantum hardware used in this invention maps the problem to the Ising Hamiltonian and generates N low-energy state samples in parallel with a single annealing (N is the number of samples in a single annealing, and the preferred embodiment is 2000).

[0059] The comparison shows that while GreedyQ solves the problem the fastest (0.001 ms), its APD is only 0.057, and it only reaches 7.4 candidate lines. The candidate pool is also very similar, making it unsuitable for schedulers who require truly different alternatives. CIM outperforms SA (0.189 vs 0.176, a 7.4% improvement) and GreedyD (0.189 vs 0.161, a 17.4% improvement) in APD, while its solution latency (2.90 ms) is significantly lower than SA (15.78 ms). It is not inferior to simulated annealing in both quality and speed dimensions, and is strictly superior in at least one dimension.

[0060] Diversity weight The coupling relationship with 8-bit precision is as follows: Figure 3 As shown. The CIM hardware limits the Ising coefficient to 0. Integer range (approximately 8 bits), linear terms generated by cardinality constraints after QUBO → Ising transform Dominant global scaling factor, when Increase or When the value is too large, the diversity coupling term It disappears during quantization rounding. Experiments show... The candidate pool APD was significantly higher than that of the time pool. (Approximately 2-fold difference), verifying low make The rationale for a design that survives in 8-bit environments.

[0061] Candidate pool size The impact on APD, such as Figure 4 As shown. CIM's diversity is superior to GreedyD's; back Mass disappearance, GreedyD gains the upper hand; to The two are equal. This boundary arises from the imbalance in the dynamic range of the linear and quadratic terms in the QUBO→Ising transform with 8-bit coefficient precision. Higher precision hardware can shift this boundary.

[0062] In a specific hardware configuration (a coherent Ising quantum machine cloud platform and a general-purpose single-core CPU), the online CIM solution (including 2000 samples) took 2.90 ms, which is about 5.4 times faster than the CPU's single-core simulated annealing (approximately 15.78 ms). It should be noted that this is a comparison between dedicated quantum hardware and a general-purpose CPU, not a fair comparison under equivalent computing power; those skilled in the art can obtain different absolute times under different hardware environments, and this disclosure does not rely on the absolute performance of a specific hardware configuration.

[0063] Furthermore, at each decision step, the system adaptively determines whether to re-execute the sampling optimization operation based on the following three normalized trigger signals: Calculate the three normalized signals for state change, mass decay, and aging respectively, and construct the trigger quantity by taking the maximum value through linear mapping. (This embodiment takes) , , ).when If the threshold is exceeded, the sampling optimization operation is re-executed and the candidate pool is updated; otherwise, the historical candidate pool from the previous decision step is reused, avoiding the need to call the coherent Ising machine. Simultaneously, lifecycle management is applied to the historical candidate pool: each candidate pool is accompanied by… and count, Exceeding limits results in elimination; when the reuse rate falls below a threshold for several consecutive steps, a forced recalculation and an increase in the number of samples are triggered. In a typical day's test with 96 rolling decision steps and 5 episodes, compared to the baseline of calling CIM at every step, adaptive triggering reduced the number of CIM calls from 45.4 to 18.8 (a reduction of approximately 58.6%), while the average cost of the candidate pool showed no significant difference. Figure 5 As shown.

[0064] Example 3 One embodiment of this disclosure provides a power grid topology candidate optimization system based on quantum coherent Ising machine sampling, comprising: The screening module is used to screen candidate lines based on the power grid operation mode and comprehensive impact factors. The topology library construction module is used to randomly perturb candidate lines, guide the generation of feasible topologies using a reward agent network, and build a topology library. The objective function construction module is used to construct a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable; The objective function uses an optimization module to transform the QUBO objective function into the Ising form and then scale it proportionally according to the coefficient accuracy of the coherent Ising machine. The sampled function is then submitted to the coherent Ising machine and solved in parallel by a single annealing to produce multiple low-energy state samples. The output selection module is used to sort, deduplicate, truncate and repair the low-energy state samples by energy, calculate the average pairwise Hamming distance of each candidate combination in the pool, and select the one with the greatest diversity as the candidate pool output.

[0065] Example 4 One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0066] Example 5 One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When executed by a processor, the computer instructions implement the power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0067] Example 6 One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the power grid topology candidate optimization method based on quantum coherent Ising machine sampling.

[0068] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create a machine 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.

[0069] 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.

[0070] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A power grid topology candidate optimization method based on quantum coherent Ising machine sampling, characterized in that, include: Candidate lines were selected based on power grid operation mode and comprehensive influencing factors; Randomly perturb the candidate lines, use a reward agent network to guide the generation of feasible topologies and build a topology library; A QUBO objective function is constructed using "whether the topology is selected into the candidate pool" as the decision variable; After transforming the QUBO objective function into the Ising form, it is scaled proportionally based on the coefficient accuracy of the coherent Ising machine and submitted to the coherent Ising machine in sampling mode. Multiple low-energy state samples are generated by parallel solving in a single annealing. After sorting, deduplicating, truncating and repairing the energy of the low-energy samples, the average pairwise Hamming distance within the pool for each candidate combination is calculated, and the combination with the greatest diversity is selected as the candidate pool output.

2. The power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in claim 1, characterized in that, The step of randomly perturbing the candidate lines, using a reward agent network to guide the generation of feasible topologies, and constructing a topology library includes: Training a reward agent network to approximate the immediate reward of a single-step flip action; exist N s The sum of the absolute values ​​of the reward agent outputs of each candidate line flipping action is aggregated under each random system state to obtain the line sampling weight containing the value bias component and the uniform coverage component. The flipped lines are selected according to the sampling weights of the lines and their feasibility is verified to generate feasible topologies, which are then constructed into a topology library.

3. The power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in claim 1, characterized in that, Construct a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable, including: With "the k Whether a topology is selected into the candidate pool is a binary decision variable. Construct the QUBO objective function: in, For topology Normalized cost; For topology With topology The normalized Hamming distance between them; Number of candidate lines; For diversity weight; Cardinality-constrained weights; The size of the candidate pool; The topology library size is represented by the first term, the topology quality term, the diversity reward term, and the cardinality constraint term. The decision variables are: The connection / disconnection status coding of the topology itself They belong to different levels; Representing topology respectively i Topology j The candidate pool decision variable takes the value {0,1}, where 1 indicates that the topology is selected into the candidate pool and 0 indicates that it is not selected. Representing topology respectively i Topology j Its own line connectivity status encoding vector; i , j For topology index, i < j This represents the summation of the diversity term over all topological pairs in the pool.

4. The power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in claim 1, characterized in that, The process involves transforming the QUBO objective function into the Ising form and then scaling it proportionally based on the coefficient accuracy of the coherent Ising machine. Multiple low-energy state samples are generated through parallel solving using a single annealing process, including: First, the QUBO objective function is processed... Transform into the Ising Hamiltonian: in, For Ising spin variable, The local field coefficients acting on a single spin. The coupling coefficient between spin pairs; They represent the first i The, the j One Ising spin variable; i , j Spin index, i < j This indicates that the coupling term is summed over all spin pairs, with each pair counted only once. Secondly, based on the scaling factor α Perform global proportional scaling to [ [127,127] Integer range adapted to the 8-bit coefficient precision of the coherent Ising machine hardware; submitted to the coherent Ising machine in sampling mode, producing N low-energy spin samples in parallel with a single annealing.

5. The power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in claim 1, characterized in that, The process of sorting, deduplicating, truncating, and restoring the low-energy samples by energy, calculating the average pairwise Hamming distance within each candidate pool, and selecting the pool with the highest diversity as the candidate pool output includes: For low-energy state samples, sort by energy, remove duplicates, truncate and retain the top M samples. For each candidate combination, repair the base number to K. If there are fewer than K, add them to the unselected topology in ascending order of energy. If there are more than K, remove them in descending order of energy. After energy sorting, deduplication, energy truncation and baseline repair, the pool APD of each candidate combination is calculated according to the APD selection criterion.

6. The power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in claim 1, characterized in that, Before constructing the QUBO objective function or after sampling and optimization, a safety filtering operation is performed. The safety filtering operation uses an N-1 safety margin as a hard constraint to remove topologies with a safety margin lower than the threshold, so that the candidate pool contains only schemes that satisfy N-1 safety.

7. A power grid topology candidate optimization system based on quantum coherent Ising machine sampling, characterized in that, include: The screening module is used to screen candidate lines based on the power grid operation mode and comprehensive impact factors. The topology library construction module is used to randomly perturb candidate lines, guide the generation of feasible topologies using a reward agent network, and build a topology library. The objective function construction module is used to construct a QUBO objective function with "whether the topology is selected into the candidate pool" as the decision variable; The objective function uses an optimization module to transform the QUBO objective function into the Ising form and then scale it proportionally according to the coefficient accuracy of the coherent Ising machine. The sampled function is then submitted to the coherent Ising machine and solved in parallel by a single annealing to produce multiple low-energy state samples. The output selection module is used to sort, deduplicate, truncate and repair the low-energy state samples by energy, calculate the average pairwise Hamming distance of each candidate combination in the pool, and select the one with the greatest diversity as the candidate pool output.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in any one of claims 1-6.

10. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the power grid topology candidate optimization method based on quantum coherent Ising machine sampling as described in any one of claims 1-6.

Images