Power grid dispatching strategy optimization method and system

Abstract

The invention provides a power grid dispatching strategy optimization method and system, and the method comprises the steps: deploying a device based on a quantum entanglement technology between transformer substations, and collecting the state information of a power grid, and encrypting and transmitting the state information to a control center through a quantum network; a pulse neural network processor is used in a control center to extract spatial-temporal characteristics, and a quantum game theory model is used to generate an optimized scheduling strategy. Multi-modal verification data is collected, including visual deformation, abnormal sound monitoring, and operational resistance data. A genetic algorithm and deep reinforcement learning are utilized to optimize a decision model, and a dynamic weight distribution mechanism is included. And through a photon-quantum hybrid computing architecture execution model, a scheduling strategy is collaboratively optimized, and a result is fed back to a physical power grid. According to the method, the quantum technology and the spiking neural network are combined, new energy fluctuation is captured in real time, the multi-target weight is dynamically adjusted through the quantum game theory model, and the scheduling strategy robustness is improved. And a photon-quantum hybrid computing architecture is adopted, so that the feature extraction speed and the optimization efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching, and in particular to a power grid dispatching strategy optimization method and system. Background Art

[0002] In the field of power grid dispatch strategy optimization, traditional technologies primarily rely on mathematical programming methods and heuristic intelligent algorithms to achieve dispatch decisions. Economic dispatch models based on linear programming and mixed integer programming construct optimization models with minimizing power generation costs as the objective function, combining conditions such as unit combination constraints and power flow constraints to generate static dispatch plans in deterministic scenarios. Dynamic programming and model predictive control (MPC) use rolling optimization mechanisms to respond to time-varying load demands and enhance short-term dispatch flexibility. In recent years, with the development of artificial intelligence technology, deep learning frameworks have been used to predict renewable energy output, and reinforcement learning algorithms learn dispatch strategies through interaction with the environment. Some research has also achieved coordinated control of distributed power sources through multi-agent systems.

[0003] However, existing technologies have three limitations: First, traditional optimization methods are highly dependent on precise system model parameters. In scenarios with high penetration of new energy, the strong randomness and volatility of wind power / photovoltaic output lead to model mismatch, resulting in significant deviations between the optimization results and actual operating conditions; Second, although heuristic algorithms can handle nonlinear constraints, they have the defects of slow convergence speed and easy to fall into local optimality, making it difficult to meet the computational efficiency requirements of real-time scheduling; Third, existing methods mostly adopt a single-objective optimization framework, which makes it difficult to take into account the coordinated optimization of multi-dimensional indicators such as economy, environmental protection, and reliability. In particular, when dealing with supply and demand imbalances caused by extreme weather, they lack the intelligent decision-making ability to dynamically adjust scheduling strategies. Summary of the Invention

[0004] The present invention aims to at least solve the technical problem in the prior art that there is a significant deviation between the optimization results and the actual operating conditions, and in particular innovatively proposes a power grid dispatching strategy optimization method and system.

[0005] In order to achieve the above-mentioned object of the present invention, the present invention provides a method for optimizing a power grid dispatching strategy, the method comprising: S1. Deploy entangled photon pair generators based on spontaneous parametric down-conversion technology between substations to collect grid status information between substations through quantum entangled state correlation; S2. Building a quantum key distribution network based on the entangled photon pair generating device, and transmitting the power grid status information collected in S1 to a central control center through quantum encryption via the quantum key distribution network; S3. Deploy a pulse neural network processor in the central control center, input the encrypted power grid status information transmitted by S2 into the pulse neural network processor, extract the spatiotemporal feature sequence; based on the spatiotemporal feature sequence, generate a multi-objective optimization scheduling strategy through a quantum game theory model, and complete the simulation operation of the multi-objective optimization scheduling strategy on the digital twin; S4. Collect multimodal verification data of the multi-objective optimization scheduling strategy in S3 after running on the digital twin, wherein the multimodal verification data includes substation visual deformation data, transformer abnormal sound monitoring data, and virtual circuit breaker operation resistance data; S5, a feature selection module based on a genetic algorithm, performs chromosome encoding and crossover mutation operations on the load forecast error and the number of switch actions in the historical dispatch data, and outputs an optimized feature subset; a policy gradient module based on deep reinforcement learning, uses the multimodal verification data collected in S4 as state input, and adopts an exploration strategy to balance the sampling probability of new and old strategies in the experience replay pool; a decision optimization model is generated by jointly optimizing the results output by the genetic algorithm and the policy gradient module. The decision optimization model includes a dynamic weight allocation mechanism, which adjusts the optimization weights of power generation cost and network loss rate according to the urgency of the power grid; S6. Execute the decision optimization model through a photon-quantum hybrid computing architecture, wherein the photon-quantum hybrid computing architecture includes a silicon photonic neural network accelerator and a quantum annealer. The photon-quantum hybrid computing architecture is used to perform collaborative optimization of the optimization scheduling strategy using classical computing and quantum computing, and ultimately generate a scheduling strategy set including a feeder automatic segmentation scheme, distributed energy aggregation instructions, and energy storage system charging and discharging curves. The scheduling strategy set is then encrypted with a quantum key and fed back to the physical power grid for execution.

[0006] As an optional embodiment of the present invention, optionally, extracting the spatiotemporal feature sequence in step S3 includes: S301, decrypting the encrypted grid state information, and converting the decrypted grid state information into a pulse interval time; S302, capturing pulse intervals of different time scales using delay taps of a time-delay neural network layer based on the pulse intervals; S303, extracting local spatial patterns using a convolutional pulse layer based on the pulse interval, and allocating association weights between power grid topology nodes using a graph attention mechanism based on the local spatial patterns; S304, performing weighted fusion of the pulse interval time and the local spatial pattern to generate a composite feature including short-term historical dependency and spatial correlation; S305, screening key features from the composite features through information gain screening; S306: Convert the key features into real-valued vectors to obtain a spatiotemporal feature sequence.

[0007] As an optional embodiment of the present invention, optionally, generating a multi-objective optimization scheduling strategy through a quantum game theory model in step S3 includes: S307. Deploy a quantum game theory model in the central control center, where the quantum game theory model includes a quantum strategy encoder, a game participant simulator, and a quantum equilibrium solver; S308. Mapping the spatiotemporal feature sequence into a quantum bit superposition state using the quantum strategy encoder, and initializing the behavior of the game participants; S309, defining a revenue function of power generation cost, network loss rate, and load balance of the power grid, and simulating strategy interaction through a quantum gate sequence based on the quantum bit superposition state; S310, running a quantum approximate optimization algorithm in the quantum equilibrium solver, simulating quantum state evolution in strategy interaction through the quantum gate sequence to approximate Nash equilibrium and generate candidate scheduling strategies; S311 , verifying the feasibility of the candidate scheduling strategy and optimizing the multi-objective trade-offs to generate a multi-objective optimized scheduling strategy.

[0008] As an optional embodiment of the present invention, optionally, in step S3, performing a simulation operation of the multi-objective optimization scheduling strategy on the digital twin includes: S312. Build a digital twin of the power grid, configure a simulation environment in the digital twin, and then initialize the simulation environment; S313, mapping the multi-objective optimization scheduling strategy into digital twin executable instructions; S314: Execute the executable instructions in a simulation environment to simulate the operation status of the power grid and collect simulation data in real time.

[0009] As an optional embodiment of the present invention, optionally, in step S5, using the exploration strategy to balance the sampling probabilities of the new and old strategies in the experience replay pool includes: S501: Construct a dual experience replay pool and initialize sampling weights. The dual experience replay pool includes an old strategy pool and a new strategy pool. S502: Extracting spatiotemporal features of the substation visual deformation data, collecting time series data of transformer abnormal sound monitoring data, converting virtual circuit breaker operating resistance data into frequency domain data, and aligning the timestamps of the spatiotemporal features, time series data, and frequency domain data using a Transformer encoder to generate a unified state vector containing power grid topology information. S503: Input the unified state vector into the policy gradient module to generate an action probability distribution based on the exploration strategy; extract experience samples from the dual experience replay pool using dynamically adjusted sampling weights, where the sampling probability of the new policy pool increases exponentially with the training rounds, and the sampling probability of the old policy pool maintains the diversity of historical experience through a soft update mechanism; S504: Calculate the policy gradient based on the extracted experience samples, combine the optimized feature subset output by the genetic algorithm feature selection module, and update the neural network weights of the policy gradient module through a parameter sharing mechanism.

[0010] As an optional embodiment of the present invention, optionally, generating the scheduling policy set in step S6 includes: S601. Initialize the photon matrix core parameters of the silicon photonic neural network accelerator and the quantum bit coupling strength of the quantum annealer to build a classical-quantum hybrid computing task pipeline. S602: Input the spatiotemporal feature sequence of the decision optimization model into the silicon photonic neural network accelerator, perform photon convolution operations and photon recurrent neural network operations through the Mach-Zehnder interferometer array, and generate quantum annealing machine initial state encoding data; S603, mapping the power grid scheduling problem into a quadratic unconstrained binary optimization model through quantum bit superposition based on the quantum annealer initial state encoding data, running a quantum approximate optimization algorithm in the quantum annealer, and generating a candidate scheduling strategy set through transverse field quantum fluctuations and longitudinal field strategy constraints; S604. Perform a hybrid evaluation on the candidate scheduling strategy set based on the classical-quantum hybrid computing task pipeline to obtain an evaluation result. If the evaluation result does not meet a preset multi-objective optimization threshold, inject the intermediate calculation result of the silicon photonic neural network accelerator to perform iterative optimization of the scheduling strategy to obtain the scheduling strategy set that meets the multi-objective optimization threshold.

[0011] On the other hand, the present invention also provides a power grid dispatching strategy optimization system, characterized in that it includes the power grid dispatching strategy optimization method described above, and the system also includes a data preprocessing module, a feature extraction module, a policy gradient module, a genetic algorithm feature selection module, a decision optimization module, a classical-quantum hybrid computing module and a result output module; Among them, the data preprocessing module is used to clean, normalize and fill missing values of the historical dispatching data of the power grid; the feature extraction module is used to extract multimodal features from the historical dispatching data of the power grid; the policy gradient module is used to generate the power grid dispatching strategy based on the reinforcement learning algorithm; the genetic algorithm feature selection module is used to select and optimize the extracted multimodal features to obtain the feature subset; the decision optimization module is used to optimize the power grid dispatching decision based on the optimized feature subset and the output of the policy gradient module; the classical-quantum hybrid computing module is used to further optimize the output of the decision optimization module to generate a dispatching strategy set that meets the multi-objective optimization threshold; the result output module is used to output the final generated dispatching strategy set to the power grid dispatching system to realize the automation and intelligence of power grid dispatching.

[0012] The present invention has the following beneficial effects: By collecting grid status information through quantum entanglement and combining it with spatiotemporal feature extraction using a pulse neural network processor, the present invention can capture the random fluctuations of renewable energy output in real time. The quantum game theory model encodes multidimensional objectives (generation cost, grid loss rate, load balance) through quantum state superposition. When verifying strategies in a digital twin, the weights of these objectives can be dynamically adjusted, significantly improving the robustness of the scheduling strategy to wind power / photovoltaic output forecast errors. To address the slow convergence of heuristic algorithms, this paper adopts a hybrid photonic-quantum computing architecture. A silicon photonic neural network accelerator implements photon convolution operations through a Mach-Zehnder interferometer array, resulting in faster feature extraction compared to traditional GPU acceleration. When running the QAOA algorithm, a quantum annealer uses transverse field quantum fluctuations to achieve global search and longitudinal field strategy constraints to accelerate convergence to the optimal solution, significantly improving the optimization efficiency of power grid dispatch strategies.

[0013] This invention uses a dynamic weight allocation mechanism to automatically adjust the weights of power generation costs and network loss rates when the grid emergency index exceeds a threshold. Combining the MFCC signatures of transformer noise and the FFT spectrum of virtual circuit breaker operating resistance from multimodal validation data, the decision-making optimization model can more accurately assess grid status in emergencies, quickly generate optimal dispatch strategies, and effectively respond to grid emergencies.

[0014] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which: Figure 1 It is a flow chart of a power grid dispatching strategy optimization method of the present invention. DETAILED DESCRIPTION

[0016] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0017] Example 1 like Figure 1 As shown, a method for optimizing a power grid dispatching strategy comprises: S1. Deploy entangled photon pair generators based on spontaneous parametric down-conversion technology between substations to collect grid status information between substations through quantum entangled state correlation; It should be noted that the use of entangled photon pairs to collect grid status information in step S1 offers the advantages of high precision and real-time performance. Entangled photon pairs are generated based on spontaneous parametric down-conversion technology, which produces highly correlated quantum states. By measuring the state of one photon, the state of the other can be indirectly determined, enabling accurate remote collection of grid status information. This information collection method based on quantum entanglement is not limited by the bandwidth and latency of traditional communication lines, significantly improving the efficiency and accuracy of grid status information collection.

[0018] S2. Building a quantum key distribution network based on the entangled photon pair generating device, and transmitting the power grid status information collected in S1 to a central control center through quantum encryption via the quantum key distribution network; It should be noted that the specific process of building a quantum key distribution network in step S2 involves leveraging the correlation properties of quantum entangled states to ensure the security of information transmission. First, a quantum entangled channel is established between substations using a device that generates entangled photon pairs. Subsequently, a quantum key distribution protocol, such as the BB84 protocol or its improved version, is used to achieve quantum encrypted transmission of power grid status information. During this process, the measurement results of the entangled photon pairs are closely related to the generation of the key. Any eavesdropping or interference in the transmission process will increase the key error rate, which can be detected by the central control center. This encrypted transmission method based on quantum key distribution can effectively resist external attacks and ensure the secure transmission of power grid status information.

[0019] The construction of a quantum key distribution network specifically includes the establishment of quantum channels, the generation and distribution of quantum keys, and the application of quantum encryption algorithms. The establishment of quantum channels relies on optical fibers or other quantum communication media to ensure that entangled photon pairs can be efficiently transmitted between substations and central control centers. The generation of quantum keys is based on the measurement results of entangled photon pairs, and a specific quantum algorithm is used to generate highly random and unpredictable keys. These keys are then distributed to each substation through quantum channels for the encryption of grid status information. During the encryption process, quantum encryption algorithms, such as quantum one-time pad, are used to combine grid status information with keys to generate encrypted information. S3. Deploy a pulse neural network processor in the central control center, input the encrypted power grid status information transmitted by S2 into the pulse neural network processor, extract the spatiotemporal feature sequence; based on the spatiotemporal feature sequence, generate a multi-objective optimization scheduling strategy through a quantum game theory model, and complete the simulation operation of the multi-objective optimization scheduling strategy on the digital twin; S4. Collect multimodal verification data of the multi-objective optimization scheduling strategy in S3 after running on the digital twin, wherein the multimodal verification data includes substation visual deformation data, transformer abnormal sound monitoring data, and virtual circuit breaker operation resistance data; S5, a feature selection module based on a genetic algorithm, performs chromosome encoding and crossover mutation operations on the load forecast error and the number of switch actions in the historical dispatch data, and outputs an optimized feature subset; a policy gradient module based on deep reinforcement learning, uses the multimodal verification data collected in S4 as state input, and adopts an exploration strategy to balance the sampling probability of new and old strategies in the experience replay pool; a decision optimization model is generated by jointly optimizing the results output by the genetic algorithm and the policy gradient module. The decision optimization model includes a dynamic weight allocation mechanism, which adjusts the optimization weights of power generation cost and network loss rate according to the urgency of the power grid; S6. Execute the decision optimization model through a photon-quantum hybrid computing architecture, wherein the photon-quantum hybrid computing architecture includes a silicon photonic neural network accelerator and a quantum annealer. The photon-quantum hybrid computing architecture is used to perform collaborative optimization of the optimization scheduling strategy using classical computing and quantum computing, and ultimately generate a scheduling strategy set including a feeder automatic segmentation scheme, distributed energy aggregation instructions, and energy storage system charging and discharging curves. The scheduling strategy set is then encrypted with a quantum key and fed back to the physical power grid for execution.

[0020] The principle of a power grid dispatching strategy optimization method in this embodiment is: First, by acquiring high-precision information from quantum entangled states, combined with the efficient feature extraction capabilities of a spiking neural network processor, real-time monitoring and accurate description of the power grid state were achieved. Subsequently, a multi-objective optimization scheduling strategy was generated using a quantum game theory model and simulated on a digital twin to verify the strategy's effectiveness and feasibility. During the strategy verification process, multimodal verification data was collected, providing a rich information foundation for subsequent decision-making optimization.

[0021] Furthermore, the present invention constructs a decision-making optimization model through the combined optimization of genetic algorithms and deep reinforcement learning. This model not only considers key characteristics such as load forecast error and the number of switching operations, but also introduces a dynamic weight allocation mechanism that can flexibly adjust the optimization weights of power generation cost and network loss rate based on the grid's urgency, thereby achieving refined control of grid dispatch strategies.

[0022] Ultimately, the present invention utilizes a hybrid photonic-quantum computing architecture to execute the decision-making optimization model. This architecture leverages the photon convolution speed of a silicon photonic neural network accelerator and the global search capabilities of a quantum annealer, achieving coordinated optimization of classical and quantum computing for scheduling strategies. This architecture then generates a scheduling strategy set, including an automatic feeder segmentation scheme, distributed energy aggregation instructions, and energy storage system charge and discharge profiles. This strategy set is then encrypted with a quantum key and fed back to the physical power grid for execution, achieving automated, intelligent, and efficient grid scheduling.

[0023] As an optional embodiment of the present invention, optionally, extracting the spatiotemporal feature sequence in step S3 includes: S301, decrypting the encrypted grid state information, and converting the decrypted grid state information into a pulse interval time; It should be noted that the purpose of converting the grid state information into pulse intervals in step S301 is to match the input format of the spiking neural network processor. A spiking neural network is a neural network model based on the spiking mechanism of biological neurons, and its input is a pulse sequence. Therefore, it is necessary to convert the grid state information into pulse intervals to simulate the pulse transmission process between neurons.

[0024] S302, capturing pulse intervals of different time scales using delay taps of a time-delay neural network layer based on the pulse intervals; It should be noted that by adjusting the delay tap parameters in step S302, the time scale of interest can be flexibly selected, thereby more accurately describing the dynamic behavior of the power grid. The time-delay neural network layer can also perform nonlinear transformations on the pulse intervals, further enhancing the ability to express features and providing rich and accurate information input to the quantum game theory model. After capturing the pulse interval characteristics at different time scales, these features are further used to construct a spatiotemporal feature sequence to comprehensively reflect the changing trends and patterns of the power grid state.

[0025] S303, extracting local spatial patterns using a convolutional pulse layer based on the pulse interval, and allocating association weights between power grid topology nodes using a graph attention mechanism based on the local spatial patterns; It should be noted in step S303 that by extracting the local spatial pattern of the pulse interval time through the convolution pulse layer, the spatial correlation between different areas in the power grid can be captured. The convolution pulse layer can effectively extract the local spatial features in the power grid status information, such as the power flow relationship between substations, line load conditions, etc. Furthermore, based on the extracted local spatial pattern, the association weights between the power grid topology nodes are allocated through the graph attention mechanism. The graph attention mechanism can dynamically adjust the connection strength between different nodes, reflecting the degree of mutual influence between different parts of the power grid. This mechanism not only takes into account the physical connection structure of the power grid, but also incorporates the dynamic changes of the power grid status information, thereby more accurately describing the topological characteristics of the power grid.

[0026] S304, performing weighted fusion of the pulse interval time and the local spatial pattern to generate a composite feature including short-term historical dependency and spatial correlation; S305, screening key features from the composite features through information gain screening; It should be noted in step S305 that information gain is an indicator that measures the ability of a feature to predict the target variable. By calculating the reduction in information entropy of each feature before and after the data set is divided, the contribution of the feature to the classification or regression task can be quantified. In the present invention, the information gain screening algorithm is used to screen out key features from the composite features. These key features not only contain the short-term historical dependency information of the power grid state, but also reflect the spatial correlation characteristics between the nodes of the power grid. These key features will be used as input to the quantum game theory model to generate a multi-objective optimization scheduling strategy, thereby realizing effective simulation and verification of the strategy on the digital twin.

[0027] S306: Convert the key features into real-valued vectors to obtain a spatiotemporal feature sequence.

[0028] As an optional embodiment of the present invention, optionally, generating a multi-objective optimization scheduling strategy through a quantum game theory model in step S3 includes: S307. Deploy a quantum game theory model in the central control center, where the quantum game theory model includes a quantum strategy encoder, a game participant simulator, and a quantum equilibrium solver; It should be noted in step S307 that the quantum strategy encoder is used to encode grid state information into quantum states for strategy generation within the quantum computing framework. This encoder leverages the superposition and entanglement properties of quantum bits to map the complex state of the grid into a high-dimensional quantum state space, enabling a more comprehensive description of the grid's operating state. The game participant simulator simulates the various participants in the grid, such as power plants, substations, and users, who formulate scheduling strategies based on their own interests and objectives. These strategies are simulated and evolved within the quantum computing framework to find the optimal scheduling solution. The quantum equilibrium solver is used to solve the equilibrium solution of the quantum game—that is, the optimal strategy chosen by each participant given the strategies of the other participants. This solver leverages the efficiency and parallelism of quantum algorithms to quickly find a set of scheduling strategies that meets multi-objective optimization criteria.

[0029] S308. Mapping the spatiotemporal feature sequence into a quantum bit superposition state using the quantum strategy encoder, and initializing the behavior of the game participants; It's important to note that step S308 not only preserves key information about the grid's state but also imbues this information with the properties and potential of a quantum state. Subsequently, the behavior of the game participants is initialized, defining their initial strategies and states within the game. These initial behaviors reflect the actual operating conditions and scheduling requirements of each grid participant, ensuring the authenticity and effectiveness of the game process.

[0030] S309, defining a revenue function of power generation cost, network loss rate, and load balance of the power grid, and simulating strategy interaction through a quantum gate sequence based on the quantum bit superposition state; It should be noted in step S309 that the purpose of defining the profit function is to quantify the effect of the grid dispatching strategy, where the power generation cost reflects the economic efficiency of the grid operation, the network loss rate reflects the energy loss during the grid transmission process, and the load balance ensures the stability and reliability of the grid. These profit functions serve as optimization objectives in the quantum game theory model, guiding the formulation and selection of dispatching strategies. By simulating strategy interactions through quantum gate sequences, it is possible to efficiently simulate the strategy game process between the various participants in the grid and capture the mutual influence and dynamic evolution between strategies. Under the quantum computing framework, this simulation not only has a high degree of parallelism and computational efficiency, but can also use characteristics such as quantum superposition and entanglement to explore more possible strategy combinations, thereby finding a better dispatching solution.

[0031] S310, running a quantum approximate optimization algorithm in the quantum equilibrium solver, simulating quantum state evolution in strategy interaction through the quantum gate sequence to approximate Nash equilibrium and generate candidate scheduling strategies; It should be noted in step S310 that the quantum approximate optimization algorithm is a heuristic optimization algorithm based on quantum computing, which can efficiently search for approximate optimal solutions under the quantum computing framework. In the present invention, the algorithm uses a sequence of quantum gates to simulate the evolution of quantum states in strategy interactions, and by continuously adjusting the strategies of the participants, the benefit function of the entire power grid system is optimized or approximately optimized. In this process, the evolution of the quantum state approaches the Nash equilibrium, that is, it reaches a stable state in which the strategy of each participant is the optimal choice given the strategies of other participants. Ultimately, the quantum approximate optimization algorithm outputs a series of candidate scheduling strategies that, while meeting the multi-objective optimization conditions of the power grid, also reflect the balance of interests and scheduling needs of each participant in the power grid.

[0032] S311 , verifying the feasibility of the candidate scheduling strategy and optimizing the multi-objective trade-offs to generate a multi-objective optimized scheduling strategy.

[0033] As an optional embodiment of the present invention, optionally, in step S3, performing a simulation operation of the multi-objective optimization scheduling strategy on the digital twin includes: S312. Build a digital twin of the power grid, configure a simulation environment in the digital twin, and then initialize the simulation environment; It should be noted in step S312 that the digital twin, as a virtual mirror of the power grid, can accurately reflect the actual operating status and characteristics of the power grid. After initializing the simulation environment, the multi-objective optimization scheduling strategy is input into the digital twin to simulate the operation of the power grid under different scheduling strategies. In this process, the digital twin can capture the dynamic changes of the power grid in real time, including load fluctuations, equipment failures, line congestion, etc., and evaluate and optimize the scheduling strategy based on this information. Through simulation operation, the effectiveness, stability and economy of the scheduling strategy can be verified, providing a scientific basis for subsequent scheduling decisions. At the same time, the digital twin can also simulate the operating status of the power grid under abnormal conditions such as extreme weather and emergencies, evaluate the adaptability and robustness of the scheduling strategy under extreme conditions, and ensure that the power grid maintains safe and stable operation in a complex and changing environment.

[0034] S313, mapping the multi-objective optimization scheduling strategy into digital twin executable instructions; S314: Execute the executable instructions in a simulation environment to simulate the operation status of the power grid and collect simulation data in real time.

[0035] It should be noted in step S14 that the digital twin simulation allows for real-time collection of grid operating status data under various dispatch strategies, such as load distribution, line flow, and voltage levels. This data not only reflects the actual effectiveness of the dispatch strategy, but also allows for further evaluation of dispatch strategy performance indicators, such as power generation costs, network loss rates, and load balance, enabling refined control and optimization of the dispatch strategy. Simulation data can also be used to train and optimize machine learning models, enhancing the intelligence and automation of grid dispatch decision-making.

[0036] As an optional embodiment of the present invention, optionally, in step S5, using the exploration strategy to balance the sampling probabilities of the new and old strategies in the experience replay pool includes: S501: Construct a dual experience replay pool and initialize sampling weights. The dual experience replay pool includes an old strategy pool and a new strategy pool. It's important to note in step S501 that the dual experience replay pools are designed to effectively manage and utilize historical data to improve the training efficiency and policy performance of deep reinforcement learning algorithms. The old policy pool stores historical scheduling data, reflecting the performance and effectiveness of past scheduling policies; while the new policy pool stores data for new policies currently being tested and optimized. Initializing sampling weights ensures that both old and new policy data are sampled and utilized appropriately during training.

[0037] S502: Extracting spatiotemporal features of the substation visual deformation data, collecting time series data of transformer abnormal sound monitoring data, converting virtual circuit breaker operating resistance data into frequency domain data, and aligning the timestamps of the spatiotemporal features, time series data, and frequency domain data using a Transformer encoder to generate a unified state vector containing power grid topology information. It should be noted in step S502 that this step aims to integrate information from different monitoring methods to construct a comprehensive description of the power grid status. By extracting the spatiotemporal features of the substation visual deformation data, the changing trends of the physical equipment in the power grid can be captured. At the same time, collecting time series data of transformer abnormal sound monitoring data can reflect the dynamic changes in the transformer's operating status. Converting the virtual circuit breaker operating resistance data into frequency domain data helps analyze the mechanical vibration characteristics during operation. Using the Transformer encoder to align the timestamps of this data can ensure that information from different sources remains consistent in time, thereby generating a unified state vector containing power grid topology information. This vector not only integrates multiple monitoring data but also retains information in the time dimension, providing rich input features for subsequent deep reinforcement learning algorithms, which helps improve the accuracy and robustness of power grid dispatch strategies.

[0038] S503: Input the unified state vector into the policy gradient module to generate an action probability distribution based on the exploration strategy; extract experience samples from the dual experience replay pool using dynamically adjusted sampling weights, where the sampling probability of the new policy pool increases exponentially with the training rounds, and the sampling probability of the old policy pool maintains the diversity of historical experience through a soft update mechanism; It should be noted in step S503 that the policy gradient module generates an action probability distribution based on the unified state vector, which reflects the possibility of different scheduling actions under the current power grid state. In order to balance exploration and utilization, the present invention adopts a dynamically adjusted sampling weight mechanism. In the new strategy pool, the sampling probability increases exponentially with the increase in training rounds, which helps the algorithm to quickly converge to the optimal solution of the new strategy. At the same time, the sampling probability in the old strategy pool is adjusted through a soft update mechanism to maintain the diversity of historical experience and avoid the algorithm from falling into local optimality. This mechanism ensures that the algorithm can effectively utilize historical experience during the training process and continuously explore new scheduling strategies, thereby improving the intelligence level and adaptability of power grid scheduling.

[0039] S504: Calculate the policy gradient based on the extracted experience samples, combine the optimized feature subset output by the genetic algorithm feature selection module, and update the neural network weights of the policy gradient module through a parameter sharing mechanism.

[0040] It should be noted in step S504 that the calculation of the policy gradient depends on the experience samples extracted from the dual experience replay pool. These samples contain the state transition and reward information of the power grid under different scheduling strategies. By calculating the policy gradient, the performance of the current strategy under a given state can be evaluated and the update direction of the policy parameters can be guided. At the same time, combined with the optimized feature subset output by the genetic algorithm feature selection module, the input features can be further refined to improve the efficiency and accuracy of the policy gradient calculation. The parameter sharing mechanism is used to update the neural network weights of the policy gradient module. By sharing weights, the redundancy of network parameters can be reduced, the training process can be accelerated, and the generalization ability of the model can be improved. This step improves the intelligence level of the power grid scheduling strategy.

[0041] As an optional embodiment of the present invention, optionally, in step S504, the expression for updating the neural network weights of the policy gradient module through the parameter sharing mechanism is: ; in, Indicates that the policy gradient module is The neural network weight parameters of the iteration, Indicates that the policy gradient module is The neural network weight parameters of the iteration, represents the learning rate of the neural network weight update, represents the gradient of the objective function of the policy gradient module with respect to the weight, represents the gradient weighting coefficient of genetic algorithm feature selection, Represented by the optimized feature subset The resulting binary mask matrix, represents the Hadamard product, represents the fitness function of the genetic algorithm feature selection module, represents the multimodal validation dataset.

[0042] As an optional embodiment of the present invention, optionally, in step S5, the decision optimization model includes: ; ; ; ; ; ; in, represents the overall optimization objective function, Indicates the total time length of the cycle, and represents the dynamic weight coefficient, represents the power generation cost function, express The grid dispatch decision vector at time , represents the network loss rate calculation function, represents the policy consistency constraint coefficient, represents the deep reinforcement learning policy network, Represents the state vector after multimodal data fusion, represents the feature mask matrix output by the genetic algorithm, represents the optimized feature subset selected by the genetic algorithm, represents the feature selection regularization coefficient, represents the total number of candidate features, represents the genetic algorithm feature scoring function, Indicates the first The original features, represents the genetic algorithm selector, represents a collection of historical scheduling data, represents the set of trainable parameters of the genetic algorithm feature selection module, represents the multimodal feature encoder, express Visual deformation data of substation at each moment, express Real-time transformer abnormal noise monitoring data, express Virtual circuit breaker operation resistance data at the moment, represents the weight adjustment sensitivity parameter, Indicates the grid emergency threshold, represents the grid emergency index, represents the Sigmoid activation function, represents the trainable weight matrix of the urgency assessment network, represents the temporal difference of visual deformation data, represents the Mel frequency cepstral coefficient extraction function, represents the fast Fourier transform function.

[0043] As an optional embodiment of the present invention, optionally, generating the scheduling policy set in step S6 includes: S601. Initialize the photon matrix core parameters of the silicon photonic neural network accelerator and the quantum bit coupling strength of the quantum annealer to build a classical-quantum hybrid computing task pipeline. It should be noted in step S601 that this step aims to take advantage of the advantages of silicon photonic neural network accelerators and quantum annealers to build an efficient classical-quantum hybrid computing framework to accelerate the generation of scheduling strategy sets. Silicon photonic neural network accelerators achieve high-speed parallel computing through photon matrix cores, which can significantly improve the reasoning speed and energy efficiency of neural networks. The quantum annealer, on the other hand, uses the principles of quantum mechanics to demonstrate excellent performance in solving combinatorial optimization problems. By initializing the photon matrix core parameters and the quantum bit coupling strength, it is possible to ensure that computing tasks are reasonably distributed between classical and quantum resources to maximize computing efficiency. By building a classical-quantum hybrid computing task pipeline, the entire process from data input to result output can be automated, thereby improving the automation and flexibility of scheduling strategy generation.

[0044] S602: Input the spatiotemporal feature sequence of the decision optimization model into the silicon photonic neural network accelerator, perform photon convolution operations and photon recurrent neural network operations through the Mach-Zehnder interferometer array, and generate quantum annealing machine initial state encoding data; It should be noted that step S602 processes the spatiotemporal feature sequence of the decision optimization model using a silicon photonic neural network accelerator. As a core component, the Mach-Zehnder interferometer array is capable of performing high-speed and precise photon convolution and photon recurrent neural network operations. These operations can capture the complex spatiotemporal dependencies of power grid status. This step enables efficient preprocessing of power grid scheduling strategies.

[0045] S603, mapping the power grid scheduling problem into a quadratic unconstrained binary optimization model through quantum bit superposition based on the quantum annealer initial state encoding data, running a quantum approximate optimization algorithm in the quantum annealer, and generating a candidate scheduling strategy set through transverse field quantum fluctuations and longitudinal field strategy constraints; It should be noted in step S603 that this step utilizes the unique advantages of the quantum annealer to map the power grid scheduling problem into a quadratic unconstrained binary optimization model. Through the superposition state of quantum bits, multiple possible scheduling strategies can be explored simultaneously, greatly expanding the search space. Running the quantum approximate optimization algorithm in the quantum annealer can efficiently find the approximate optimal solution. At the same time, by introducing transverse field quantum fluctuations and longitudinal field strategy constraints, the search process can be further guided to ensure that the generated candidate scheduling strategy set not only meets the actual needs of power grid operation, but also has high quality and diversity.

[0046] The expression of the above quantum approximate optimization algorithm is: in, The Problem Hamiltonian for the grid scheduling problem encodes scheduling objectives (such as power generation cost and network loss rate) as qubit interaction terms; represents the number of qubits; Indicates the The local field coefficient of each quantum bit corresponds to the linear weight of the power generation cost or network loss rate; represents the Pauli-Z operator, the spin state of the qubits (encoding |0> or |1>); Representing quantum bits and The coupling coefficient between them reflects the interaction strength of the grid topology constraints or dispatch strategies.

[0047] represents the Pauli-Z operator, The spin state of the qubits; The variational quantum circuit representing the QAOA is composed of alternating angle-parameterized quantum gates; Represents the phase angle of the Hamiltonian of the problem, controlling the speed at which the quantum state approaches the optimal solution; Represents the rotation angle of the hybrid operator, which controls the exploration range of the quantum state between the calculation base states; The circuit depth parameter of QAOA determines the number of repetitions of the quantum gate sequence (affecting the solution quality and the balance between computing resources); It represents the key parameter that controls the evolution of the hybrid Hamiltonian in the quantum approximate optimization algorithm. By driving the qubits to oscillate between superposition states, it enhances the algorithm's global exploration capability and ultimately generates a set of candidate strategies that meet the grid scheduling constraints. Indicates the The phase angle of the problem Hamiltonian in layer quantum circuits; represents the mixed Hamiltonian along .

[0048] S604. Perform a hybrid evaluation on the candidate scheduling strategy set based on the classical-quantum hybrid computing task pipeline to obtain an evaluation result. If the evaluation result does not meet a preset multi-objective optimization threshold, inject the intermediate calculation result of the silicon photonic neural network accelerator to perform iterative optimization of the scheduling strategy to obtain the scheduling strategy set that meets the multi-objective optimization threshold.

[0049] It should be noted in step S604 that this step emphasizes the ability of the classical-quantum hybrid computing framework to evaluate and optimize the scheduling policy set. Through the constructed classical-quantum hybrid computing task pipeline, the candidate scheduling policy set can be evaluated efficiently and comprehensively. The evaluation results will directly reflect the performance of the policy set in multi-objective optimization, including key indicators such as the economy, safety and stability of power grid operation. If the evaluation results do not meet the pre-set multi-objective optimization threshold, it means that the current policy set still has room for optimization. At this time, the system will inject the intermediate calculation results of the silicon photonic neural network accelerator and use these intermediate results to iteratively optimize the scheduling policy. Through continuous iteration and adjustment, the optimal solution can be gradually approached, and finally a scheduling policy set that meets the multi-objective optimization threshold is obtained.

[0050] As an optional embodiment of the present invention, optionally, the expression of the approximate optimization algorithm in step S603 is: ; ; in, The Ising model Hamiltonian representing the target problem, represents the number of qubits, Indicates the The local field coefficients of the qubits, Pauli- The operator acts on the qubits, Representing quantum bits and The coupling coefficient between Pauli- The operator acts on the qubits, represents the variational quantum circuit of QAOA, represents the number of layers of the mixing operator, Indicates the The rotation angle of the layer blending operator, represents the Hamiltonian of the mixing operator, Indicates the The phase angle of the Hamiltonian for the layer problem; In step S604, the expression of the classical-quantum hybrid computing task pipeline is: in, Represents the post-processing flow of mixed calculation results, represents the optimal quantum bit measurement result, Represents a traversal function, represents the optimal final state of the variational quantum circuit, represents the parameters of the variational quantum circuit, Represents the power grid dispatch instruction set, represents the result decoding function, represents the constraint matrix, represents the load tensor, represents the variational parameter of the next iteration, represents the optimization function, represents the parameter gradient of the loss function, Indicates the scheduling effect feedback value, Represents the verification function, Indicates the load distribution after actual execution.

[0051] Example 2 A power grid dispatching strategy optimization system, characterized in that it includes the power grid dispatching strategy optimization method described above, and the system also includes a data preprocessing module, a feature extraction module, a policy gradient module, a genetic algorithm feature selection module, a decision optimization module, a classical-quantum hybrid computing module, and a result output module; Among them, the data preprocessing module is used to clean, normalize and fill missing values of the historical dispatching data of the power grid; the feature extraction module is used to extract multimodal features from the historical dispatching data of the power grid; the policy gradient module is used to generate the power grid dispatching strategy based on the reinforcement learning algorithm; the genetic algorithm feature selection module is used to select and optimize the extracted multimodal features to obtain the feature subset; the decision optimization module is used to optimize the power grid dispatching decision based on the optimized feature subset and the output of the policy gradient module; the classical-quantum hybrid computing module is used to further optimize the output of the decision optimization module to generate a dispatching strategy set that meets the multi-objective optimization threshold; the result output module is used to output the final generated dispatching strategy set to the power grid dispatching system to realize the automation and intelligence of power grid dispatching.

[0052] The working principle of the power grid dispatching strategy optimization system in this embodiment is: The system first preprocesses historical grid dispatch data using the data preprocessing module to ensure data quality and consistency. The feature extraction module then extracts multimodal features from the preprocessed data. These features comprehensively reflect the grid's operating status and historical dispatch history. The policy gradient module, based on a reinforcement learning algorithm, uses these features to generate preliminary grid dispatch strategies.

[0053] The genetic algorithm feature selection module optimizes the extracted multimodal features and, through the powerful search capabilities of the genetic algorithm, selects the feature subset that has the greatest impact on the generation of the scheduling strategy. This step effectively reduces the dimensionality of the problem and improves the efficiency and accuracy of subsequent calculations.

[0054] The decision optimization module further optimizes grid dispatch decisions based on the optimized feature subset and the output of the policy gradient module. It comprehensively considers multiple objectives such as the grid's economy, security, and stability to generate a more refined and reliable dispatch strategy.

[0055] The classical-quantum hybrid computing module leverages the respective strengths of classical and quantum computing to further optimize the output of the decision optimization module. By building a classical-quantum hybrid computing task pipeline, it enables efficient evaluation and optimization of candidate scheduling policy sets. This step significantly improves the quality and diversity of the scheduling policy set, ensuring that the resulting set meets multi-objective optimization thresholds.

[0056] Finally, the output module outputs the resulting dispatch strategy set to the power grid dispatch system, achieving automated and intelligent grid dispatch. Through the collaborative work of various modules, the entire system achieves comprehensive optimization of the grid dispatch strategy, effectively improving the grid's operational efficiency and stability.

[0057] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and alterations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A method for optimizing power grid dispatching strategy, characterized in that: The method comprises: S1. Deploy entangled photon pair generators based on spontaneous parametric down-conversion technology between substations to collect grid status information between substations through quantum entangled state correlation; S2. Building a quantum key distribution network based on the entangled photon pair generating device, and transmitting the power grid status information collected in S1 to a central control center through quantum encryption via the quantum key distribution network; S3. Deploy a pulse neural network processor in the central control center, input the encrypted power grid status information transmitted by S2 into the pulse neural network processor, extract the spatiotemporal feature sequence; based on the spatiotemporal feature sequence, generate a multi-objective optimization scheduling strategy through a quantum game theory model, and complete the simulation operation of the multi-objective optimization scheduling strategy on the digital twin; S4. Collect multimodal verification data of the multi-objective optimization scheduling strategy in S3 after running on the digital twin, wherein the multimodal verification data includes substation visual deformation data, transformer abnormal sound monitoring data, and virtual circuit breaker operation resistance data; S5, a feature selection module based on a genetic algorithm, performs chromosome encoding and crossover mutation operations on the load forecast error and the number of switch actions in the historical dispatch data, and outputs an optimized feature subset; a policy gradient module based on deep reinforcement learning, uses the multimodal verification data collected in S4 as state input, and adopts an exploration strategy to balance the sampling probability of new and old strategies in the experience replay pool; a decision optimization model is generated by jointly optimizing the results output by the genetic algorithm and the policy gradient module. The decision optimization model includes a dynamic weight allocation mechanism, which adjusts the optimization weights of power generation cost and network loss rate according to the urgency of the power grid; S6. Execute the decision optimization model through a photon-quantum hybrid computing architecture, wherein the photon-quantum hybrid computing architecture includes a silicon photonic neural network accelerator and a quantum annealer. The photon-quantum hybrid computing architecture is used to perform collaborative optimization of the optimization scheduling strategy using classical computing and quantum computing, and ultimately generate a scheduling strategy set including a feeder automatic segmentation scheme, distributed energy aggregation instructions, and energy storage system charging and discharging curves. The scheduling strategy set is then encrypted with a quantum key and fed back to the physical power grid for execution.

2. A method for optimizing power grid dispatching strategy according to claim 1, characterized in that: Extracting the spatiotemporal feature sequence in step S3 includes: S301, decrypting the encrypted grid state information, and converting the decrypted grid state information into a pulse interval time; S302, capturing pulse intervals of different time scales using delay taps of a time-delay neural network layer based on the pulse intervals; S303, extracting local spatial patterns using a convolutional pulse layer based on the pulse interval, and allocating association weights between power grid topology nodes using a graph attention mechanism based on the local spatial patterns; S304, performing weighted fusion of the pulse interval time and the local spatial pattern to generate a composite feature including short-term historical dependency and spatial correlation; S305, screening key features from the composite features through information gain screening; S306: Convert the key features into real-valued vectors to obtain a spatiotemporal feature sequence.

3. A method for optimizing power grid dispatching strategy according to claim 1, characterized in that: In step S3, the multi-objective optimization scheduling strategy is generated by the quantum game theory model, including: S307. Deploy a quantum game theory model in the central control center, where the quantum game theory model includes a quantum strategy encoder, a game participant simulator, and a quantum equilibrium solver; S308. Mapping the spatiotemporal feature sequence into a quantum bit superposition state using the quantum strategy encoder, and initializing the behavior of the game participants; S309, defining a revenue function of power generation cost, network loss rate, and load balance of the power grid, and simulating strategy interaction through a quantum gate sequence based on the quantum bit superposition state; S310, running a quantum approximate optimization algorithm in the quantum equilibrium solver, simulating quantum state evolution in strategy interaction through the quantum gate sequence to approximate Nash equilibrium and generate candidate scheduling strategies; S311 , verifying the feasibility of the candidate scheduling strategy and optimizing the multi-objective trade-offs to generate a multi-objective optimized scheduling strategy.

4. A method for optimizing power grid dispatching strategy according to claim 1, characterized in that: In step S3, the multi-objective optimization scheduling strategy is implemented on the digital twin to simulate the multi-objective optimization scheduling strategy, including: S312. Build a digital twin of the power grid, configure a simulation environment in the digital twin, and then initialize the simulation environment; S313, mapping the multi-objective optimization scheduling strategy into digital twin executable instructions; S314: Execute the executable instructions in a simulation environment to simulate the operation status of the power grid and collect simulation data in real time.

5. A method for optimizing power grid dispatching strategy according to claim 1, characterized in that: In step S5, the exploration strategy is used to balance the sampling probabilities of the new and old strategies in the experience replay pool, including: S501: Construct a dual experience replay pool and initialize sampling weights. The dual experience replay pool includes an old strategy pool and a new strategy pool. S502: Extracting spatiotemporal features of the substation visual deformation data, collecting time series data of transformer abnormal sound monitoring data, converting virtual circuit breaker operating resistance data into frequency domain data, and aligning the timestamps of the spatiotemporal features, time series data, and frequency domain data using a Transformer encoder to generate a unified state vector containing power grid topology information. S503: Input the unified state vector into the policy gradient module to generate an action probability distribution based on the exploration strategy; extract experience samples from the dual experience replay pool using dynamically adjusted sampling weights, where the sampling probability of the new policy pool increases exponentially with the training rounds, and the sampling probability of the old policy pool maintains the diversity of historical experience through a soft update mechanism; S504: Calculate the policy gradient based on the extracted experience samples, combine the optimized feature subset output by the genetic algorithm feature selection module, and update the neural network weights of the policy gradient module through a parameter sharing mechanism.

6. A method for optimizing power grid dispatching strategy according to claim 5, characterized in that: In step S504, the expression for updating the neural network weights of the policy gradient module through the parameter sharing mechanism is: ； in, Indicates that the policy gradient module is The neural network weight parameters of the iteration, Indicates that the policy gradient module is The neural network weight parameters of the iteration, represents the learning rate of the neural network weight update, represents the gradient of the objective function of the policy gradient module with respect to the weight, represents the gradient weighting coefficient of genetic algorithm feature selection, Represented by the optimized feature subset The resulting binary mask matrix, represents the Hadamard product, represents the fitness function of the genetic algorithm feature selection module, represents the multimodal validation dataset.

7. A method for optimizing power grid dispatching strategy according to claim 1 or 5, characterized in that: In step S5, the decision optimization model includes: ； ； ； ； ； ； in, represents the overall optimization objective function, Indicates the total time length of the cycle, and represents the dynamic weight coefficient, represents the power generation cost function, express The grid dispatch decision vector at time , represents the network loss rate calculation function, represents the policy consistency constraint coefficient, represents the deep reinforcement learning policy network, Represents the state vector after multimodal data fusion, represents the feature mask matrix output by the genetic algorithm, represents the optimized feature subset selected by the genetic algorithm, represents the feature selection regularization coefficient, represents the total number of candidate features, represents the genetic algorithm feature scoring function, Indicates the first The original features, represents the genetic algorithm selector, represents a collection of historical scheduling data, represents the set of trainable parameters of the genetic algorithm feature selection module, represents the multimodal feature encoder, express Visual deformation data of substation at each moment, express Real-time transformer abnormal noise monitoring data, express Virtual circuit breaker operation resistance data at the moment, represents the weight adjustment sensitivity parameter, Indicates the grid emergency threshold, represents the grid emergency index, represents the Sigmoid activation function, represents the trainable weight matrix of the urgency assessment network, represents the temporal difference of visual deformation data, represents the Mel frequency cepstral coefficient extraction function, represents the fast Fourier transform function.

8. A method for optimizing power grid dispatching strategy according to claim 1, characterized in that: The scheduling policy set generated in step S6 includes: S601. Initialize the photon matrix core parameters of the silicon photonic neural network accelerator and the quantum bit coupling strength of the quantum annealer to build a classical-quantum hybrid computing task pipeline. S602: Input the spatiotemporal feature sequence of the decision optimization model into the silicon photonic neural network accelerator, perform photon convolution operations and photon recurrent neural network operations through the Mach-Zehnder interferometer array, and generate quantum annealing machine initial state encoding data; S603, mapping the power grid scheduling problem into a quadratic unconstrained binary optimization model through quantum bit superposition based on the quantum annealer initial state encoding data, running a quantum approximate optimization algorithm in the quantum annealer, and generating a candidate scheduling strategy set through transverse field quantum fluctuations and longitudinal field strategy constraints; S604. Perform a hybrid evaluation on the candidate scheduling strategy set based on the classical-quantum hybrid computing task pipeline to obtain an evaluation result. If the evaluation result does not meet a preset multi-objective optimization threshold, inject the intermediate calculation result of the silicon photonic neural network accelerator to perform iterative optimization of the scheduling strategy to obtain the scheduling strategy set that meets the multi-objective optimization threshold.

9. A method for optimizing power grid dispatching strategy according to claim 8, characterized in that: The expression of the approximate optimization algorithm in step S603 is: ； ； in, The Ising model Hamiltonian representing the target problem, represents the number of qubits, Indicates the The local field coefficients of the qubits, Pauli- The operator acts on the qubits, Representing quantum bits and The coupling coefficient between Pauli- The operator acts on the qubits, represents the variational quantum circuit of QAOA, represents the number of layers of the mixing operator, Indicates the The rotation angle of the layer blending operator, represents the Hamiltonian of the mixing operator, Indicates the The phase angle of the Hamiltonian for the layer problem; In step S604, the expression of the classical-quantum hybrid computing task pipeline is: ； in, Represents the post-processing flow of mixed calculation results, represents the optimal quantum bit measurement result, Represents a traversal function, represents the optimal final state of the variational quantum circuit, represents the parameters of the variational quantum circuit, Represents the power grid dispatch instruction set, represents the result decoding function, represents the constraint matrix, represents the load tensor, represents the variational parameter of the next iteration, represents the optimization function, represents the parameter gradient of the loss function, Indicates the scheduling effect feedback value, Represents the verification function, Indicates the load distribution after actual execution.

10. A power grid dispatching strategy optimization system, characterized in that: The method for optimizing power grid dispatching strategy according to any one of claims 1 to 9, wherein the system further comprises a data preprocessing module, a feature extraction module, a policy gradient module, a genetic algorithm feature selection module, a decision optimization module, a classical-quantum hybrid computing module, and a result output module; Among them, the data preprocessing module is used to clean, normalize and fill missing values of the historical dispatching data of the power grid; the feature extraction module is used to extract multimodal features from the historical dispatching data of the power grid; the policy gradient module is used to generate the power grid dispatching strategy based on the reinforcement learning algorithm; the genetic algorithm feature selection module is used to select and optimize the extracted multimodal features to obtain the feature subset; the decision optimization module is used to optimize the power grid dispatching decision based on the optimized feature subset and the output of the policy gradient module; the classical-quantum hybrid computing module is used to further optimize the output of the decision optimization module to generate a dispatching strategy set that meets the multi-objective optimization threshold; the result output module is used to output the final generated dispatching strategy set to the power grid dispatching system to realize the automation and intelligence of power grid dispatching.

Images