Intelligent arrangement response method and system based on power network environment
By collecting cross-regional power transmission data in real time through edge nodes and generating global optimization targets using federated prediction models and differential privacy technology, the problem of data sharing in cross-regional power grids has been solved. This has enabled privacy and security, as well as optimization effects in cross-regional power dispatch, and has improved the renewable energy absorption rate and the stability of the power system.
Patent Information
- Application Number
- CN202511003711.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-21
- Publication Date
- 2025-11-14
AI Technical Summary
Cross-regional power grid operators are unable to share sensitive information such as user load and equipment status due to data sovereignty and security compliance requirements, making it difficult to build a global optimization model during centralized data processing and hindering cross-regional power dispatch.
By collecting cross-regional power transmission data in real time through edge nodes, updating the parameter set of the output parameter of the federated prediction model, and using differential privacy encryption combined with a dynamic noise adjustment mechanism, a global optimization objective is generated. Combined with the noise parameter mapping table stored in the blockchain, the encrypted gradient is integrated to update the global model, generate a dynamic power transmission capacity allocation strategy and Pareto final solution set, and perform multi-objective optimization.
It achieves synergistic optimization of privacy, security, economy, reliability and environmental protection in cross-regional power dispatch, improves the absorption rate of new energy sources, reduces carbon emissions, and ensures the stability of key power transmission channels.
Smart Images
Figure CN120951346A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power network technology, and in particular to an intelligent orchestration response method and system based on a power network environment. Background Technology
[0002] A power grid is a complete system consisting of power generation equipment, transmission lines, substations, distribution lines, and power consumption equipment. It covers the entire process of electricity production, transmission, distribution, and consumption. The power grid is the core of the power grid, and its main tasks are to transmit and distribute electrical energy and change voltage. The power grid undertakes the important tasks of power transmission, distribution, and stable power supply, and is one of the important infrastructures of modern society.
[0003] With the rapid changes in the power network environment, current power network scheduling includes power generation planning, load allocation, regional dispatching, data acquisition and monitoring, etc. Power generation planning is mainly based on historical load curves and manual forecasting, arranging the output of thermal power, hydropower and other units, and issuing dispatching orders according to fixed patterns. Load allocation mainly relies on the experience of dispatchers to manually balance supply and demand, such as adjusting unit output or cutting off non-critical loads. Regional dispatching is mainly based on the division of dispatching authority according to administrative regions, with each level of dispatching center operating independently and coordinating cross-regional power support. Data acquisition and monitoring mainly rely on Internet of Things-based power equipment to collect real-time data.
[0004] Regarding the aforementioned technologies, the inventors believe that current power network orchestration relies on centralized data for global data sharing during cross-regional dispatch. However, due to data sovereignty and security compliance requirements, cross-regional power grid operators cannot share sensitive information such as user load and equipment status, making it difficult to construct a global optimization model during centralized data processing, which is inconvenient for cross-regional power dispatch. Summary of the Invention
[0005] To address the current issue that power network orchestration relies on centralized data sharing for cross-regional dispatch, but cross-regional power grid operators are unable to share sensitive information such as user load and equipment status due to data sovereignty and security compliance requirements, making it difficult to construct a global optimization model during centralized data processing and hindering cross-regional power dispatch, this application provides an intelligent orchestration response method and system based on the power network environment.
[0006] Firstly, this application provides an intelligent orchestration and response method based on a power network environment, employing the following technical solution: including:
[0007] Real-time cross-regional power transmission data is collected through edge nodes. A pre-set federated prediction model is trained based on the cross-regional power transmission data, a set of parameter updates is output, and the set of parameter updates is sent to a global aggregator in the cloud to generate a global optimization target.
[0008] Based on the global optimization objective and historical data, output and execute the inter-regional power transmission strategy;
[0009] Based on the aforementioned inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability, and carbon emission constraints, a weighted Chebyshev subproblem is established, and the Pareto final solution set is output.
[0010] The system acquires real-time operational data and compares it with the Pareto final solution set, outputting a deviation value. If the deviation value is greater than a threshold, the local model corresponding to the deviation value greater than the threshold is updated, and the local model is uploaded to the global aggregator to regenerate the global optimization objective.
[0011] Preferably, the parameter update set includes: model gradient and supply and demand features;
[0012] The process involves real-time collection of cross-regional power transmission data through edge nodes, training a pre-defined federated prediction model based on the cross-regional update data, outputting a set of parameter update data, and sending the set of parameter update data to a global aggregator in the cloud to generate a global optimization target, including:
[0013] The cross-regional power transmission data is collected in real time via edge nodes based on the Internet of Things, and the federated prediction model is trained using the cross-regional power transmission data, and the model gradient is output.
[0014] The model gradient is encrypted using differential privacy, and after being appended with local noise distribution parameters, it is uploaded to the cloud-based global aggregator.
[0015] The global aggregator integrates all the model gradients, updates the global model, optimizes the parameters of the federated prediction model based on the global model, and distributes the optimized model parameters to the federated prediction model of the edge nodes.
[0016] The federated forecasting model is optimized according to the model optimization parameters, and the supply and demand characteristics are predicted according to the optimized federated forecasting model and transmitted to the global aggregator in the cloud in real time.
[0017] A global optimization objective is generated based on the global model and the parameter update set.
[0018] Preferably, the step of outputting and executing the inter-regional power transmission strategy based on the global optimization objective and historical data includes:
[0019] By calling upon the historical data stored in the blockchain, and integrating the data characteristics of different regions through vertical federated transfer learning, global absorption data is generated.
[0020] Based on the preset absorption rate threshold and global absorption data, a dynamic transmission capacity allocation strategy is generated through a hierarchical Q network, and a priority transmission channel list is generated.
[0021] Preferably, the step of outputting and executing the inter-regional power transmission strategy based on the global optimization objective and historical data includes:
[0022] Based on the carbon emission constraint intensity in the global optimization objective and the historical data, the model weights of different regions in the global model are dynamically allocated through a federal reward and punishment mechanism, and a cross-regional market clearing strategy is generated according to the model weights.
[0023] Based on the aforementioned cross-regional market clearing strategy, dedicated network slices are allocated to the power trading channel, and a priority trading link configuration scheme is output.
[0024] Preferably, the step of establishing a weighted Chebyshev subproblem based on the inter-regional power transmission strategy, minimizing inter-regional power transmission losses, power supply reliability, and carbon emission constraints, and outputting a Pareto final solution set, includes:
[0025] Based on the global optimization objective, the weighted Chebyshev decomposition method is used to transform the minimization of inter-regional transmission loss, power supply reliability and carbon emission constraints into multiple sets of single-objective sub-problems, each set of single-objective sub-problems corresponding to different weight combinations.
[0026] The control edge nodes independently solve the single-objective sub-problem based on the inter-regional power transmission strategy, generating a preliminary optimized solution set;
[0027] The global aggregator summarizes the preliminary optimized solution sets of each region, filters out non-dominated solutions, and generates the Pareto final solution set.
[0028] By injecting pre-defined cross-regional scheduling experience rules into the global model through knowledge distillation, the Pareto final solution set that satisfies the multilateral protocol is generated.
[0029] Preferably, after aggregating the preliminary optimized solution sets of each region through the global aggregator, filtering non-dominated solutions, and generating the Pareto final solution set, the method further includes:
[0030] The Pareto final solution set was verified using random scenarios through Monte Carlo simulation.
[0031] If any solution in the Pareto final solution set exceeds the threshold for power supply reliability degradation or carbon emission limits under the random scenario, the corresponding sub-solution in the Pareto final solution set is removed and the optimization is iterated again.
[0032] Preferably, after encrypting the model gradient using differential privacy, attaching local noise distribution parameters, and uploading it to the cloud-based global aggregator, the method further includes:
[0033] Noise intensity is classified according to the data sensitivity level of different regions, and the corresponding noise variance is determined according to the noise intensity of different regions.
[0034] The noise variance is dynamically adjusted based on the real-time inter-regional power transmission capacity: when the tie line power reaches the rated capacity, the noise variance is reduced to a preset ratio to retain key features, which include at least the tie line power and the reserve capacity.
[0035] By using blockchain to store the key features, noise intensity, and noise variance in a regional noise parameter mapping table, the noise distribution parameters of the global aggregator and edge nodes are updated synchronously.
[0036] Secondly, this application discloses an intelligent orchestration and response device based on a power network environment, which adopts the following technical solution, including:
[0037] The global module is used to collect cross-regional power transmission data in real time through edge nodes, train a preset federated prediction model based on the cross-regional power transmission data, output a set of parameter updates, and send the set of parameter updates to the global aggregator in the cloud to generate a global optimization target.
[0038] The power transmission module is used to output and execute inter-regional power transmission strategies based on the global optimization objective and historical data.
[0039] The optimization module is used to establish a weighted Chebyshev subproblem based on the inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability and carbon emission constraints, and output the Pareto final solution set;
[0040] The iteration module is used to acquire actual running data in real time, compare the actual running data with the Pareto final solution set, output the deviation value, and if the deviation value is greater than a threshold, update the local model corresponding to the deviation value greater than the threshold, upload the local model to the global aggregator, and regenerate the global optimization objective.
[0041] Thirdly, this application also provides a control device, the device comprising:
[0042] It includes a memory and a processor, wherein the memory stores a computer program that can be loaded and executed by the processor, such as the intelligent orchestration response method based on the power network environment described above.
[0043] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above regarding the intelligent orchestration response method based on a power network environment.
[0044] In summary, the system in this application collects cross-regional power transmission data in real time through edge nodes, updates the output parameter set using a federated prediction model, and employs a differential privacy encryption combined with a dynamic noise adjustment mechanism: noise intensity is classified according to the data sensitivity level of each region, and the noise variance is dynamically adjusted based on the real-time cross-regional power transmission capacity, while retaining key features to ensure the accuracy of the strategy. A global aggregator integrates encrypted gradients to update the global model, generates a global optimization objective, and synchronizes multi-regional noise configuration using a noise parameter mapping table stored on the blockchain. A dynamic power transmission capacity allocation strategy and Pareto final solution set are generated based on a hierarchical Q-network and a weighted Chebyshev decomposition method. The robustness of the solution set is verified through Monte Carlo simulation, eliminating schemes that cause power supply reliability degradation exceeding the threshold or carbon emission exceeding the standard, and retaining the solution set that satisfies the multilateral protocol after iterative optimization. The closed-loop verification process compares the deviation between the actual operating data and the solution set, triggers incremental training to update the local model and feeds it back to the global optimization, forming a technical link of "dynamic adjustment - privacy protection - multi-objective optimization - closed-loop verification". This enables the coordinated optimization of privacy security, economy, reliability and environmental protection in cross-regional power dispatch, improves the renewable energy consumption rate, reduces carbon emissions and ensures the stability of key power transmission channels. Attached Figure Description
[0045] Figure 1 This is a flowchart illustrating an intelligent orchestration and response method based on a power network environment.
[0046] Figure 2 This is a structural block diagram of an intelligent orchestration and response device based on a power network environment. Detailed Implementation
[0047] The following combination Figure 1 - Figure 2 This application will be described in further detail.
[0048] The current power network orchestration system has the following technical characteristics and limitations:
[0049] Traditional dispatching models rely on human experience: power generation planning is based on historical load curves and manual prediction algorithms, and the output of conventional units such as thermal power and hydropower is dispatched in a fixed mode. Load allocation depends on the dispatcher's experience to balance supply and demand, and the decision-making process is subject to response delays and subjective errors.
[0050] Regional dispatching faces administrative barriers: Cross-regional power dispatching adopts a provincial-municipal hierarchical management model, with each administrative region's dispatching center operating independently. Cross-regional power support requires multi-level coordination, resulting in problems such as low information transmission efficiency and high coordination costs.
[0051] Data silos hinder global optimization: Due to data sovereignty and security compliance requirements, cross-regional power grid operators cannot share sensitive information such as user load curves and equipment operating status, and centralized data processing models are difficult to build global optimization models;
[0052] Disconnect between real-time perception and decision-making: Although IoT devices have achieved full-network data collection, real-time data is only used for local monitoring and has not been deeply involved in cross-regional scheduling decisions, resulting in an application bottleneck of rich data but lack of knowledge.
[0053] Based on this, this application discloses an intelligent orchestration and response method based on a power network environment. The execution entity is the control system. This application collects cross-regional power transmission data in real time through edge nodes, updates the parameter set using the output parameter set of the federated prediction model, and employs differential privacy encryption combined with a dynamic noise adjustment mechanism. A global aggregator integrates encrypted gradients to update the global model, generating a global optimization objective. Compared to traditional centralized dispatching, this solution replaces the sharing of original data with model gradients, eliminating the need to upload original data. This reduces the risk of leakage of sensitive information such as user load and equipment status, while simultaneously improving the cross-regional renewable energy consumption rate and shortening the response delay of dispatching instructions, providing a safe and efficient collaborative optimization framework for new power systems.
[0054] Reference Figure 1 The embodiments of this application include at least steps S10 to S40.
[0055] S10 collects cross-regional power transmission data in real time through edge nodes, trains a pre-set federated prediction model based on the cross-regional power transmission data, outputs a set of parameter updates, and sends the set of parameter updates to the global aggregator in the cloud to generate a global optimization target.
[0056] S20, based on global optimization objectives and historical data, outputs and executes inter-regional power transmission strategies.
[0057] S30 establishes a weighted Chebyshev subproblem based on inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability, and carbon emission constraints, and outputs the Pareto final solution set.
[0058] S40 acquires the actual running data in real time, compares the actual running data with the Pareto final solution set, outputs the deviation value, and if the deviation value is greater than the threshold, updates the local model corresponding to the deviation value greater than the threshold, uploads the local model to the global aggregator, and regenerates the global optimization objective.
[0059] Specifically, cross-regional power transmission data is collected in real time through edge nodes. A federated prediction model is used to generate a set of encrypted gradient parameters locally. Differential privacy is used to dynamically adjust noise intensity before uploading the data to a global aggregator in the cloud. The aggregator integrates gradients from multiple regions to generate a global optimization objective (e.g., renewable energy absorption rate ≥ 85%, carbon emissions ≤ 300g / kWh). Based on historical data and the global objective, a hierarchical Q-network is used to generate cross-regional power transmission strategies. A weighted Chebyshev decomposition method is used to transform transmission losses, power supply reliability, and carbon emission constraints into a multi-objective optimization problem, and a Pareto final solution set is selected. Monte Carlo simulations are used to verify the robustness of the solution set under extreme scenarios, and feasible solutions are output after eliminating out-of-target schemes. When the deviation between the actual operating data and the solution set exceeds the limit, the incremental training module is triggered to update the local model parameters and feed them back to the global aggregator, dynamically correcting the optimization objective and forming a closed-loop feedback. This achieves coordinated optimization of economy, reliability, and environmental protection in cross-regional power dispatch, improving renewable energy absorption rate and reducing dispatch command response latency to milliseconds.
[0060] In some embodiments, the parameter update set includes: model gradient and supply and demand characteristics; step S10 specifically includes the following steps: collecting cross-regional power transmission data in real time through edge nodes based on the Internet of Things, training a federated prediction model through the cross-regional power transmission data, and outputting the model gradient; encrypting the model gradient using differential privacy, attaching local noise distribution parameters, and uploading it to the global aggregator in the cloud; the global aggregator integrates all model gradients, updates the global model, and optimizes the parameters of the federated prediction model according to the global model, and distributes the optimized model parameters to the federated prediction model of the edge nodes; optimizing the federated prediction model according to the optimized model parameters, predicting supply and demand characteristics according to the optimized federated prediction model, and transmitting them to the global aggregator in the cloud in real time; generating a global optimization target based on the global model and the parameter update set.
[0061] Specifically, cross-regional power transmission data is collected in real time through edge nodes, a federated prediction model is trained locally to generate model gradients, and the gradients are encrypted using differential privacy technology, with the addition of local noise distribution parameters (dynamically adjusting the variance σ). 2 After preserving key features, the data is uploaded to the global aggregator. The global aggregator integrates gradients from multiple regions, updates the global model parameters, and distributes them to edge nodes to optimize local federated prediction models. Based on the optimized models, it predicts supply and demand characteristics and synchronizes them in real-time to the cloud to generate global optimization objectives. A dynamic noise adjustment mechanism strikes a balance between privacy protection and feature fidelity, and a blockchain-based noise parameter mapping table enables cross-regional synchronization. Finally, knowledge distillation injects scheduling rules into the global model, forming a Pareto solution set for multi-objective optimization, supporting the coordinated optimization of economy, environmental protection, and reliability in cross-regional power dispatching.
[0062] In some embodiments, step S20 specifically includes the following steps: calling historical data stored in the blockchain, integrating data features of different regions through vertical federated transfer learning to generate global absorption data; and generating a dynamic transmission capacity allocation strategy through a hierarchical Q network based on a preset absorption rate threshold and global absorption data to generate a priority transmission channel list.
[0063] Specifically, blockchain is used to store encrypted historical data across regions, ensuring data immutability and compliant sharing. Vertical federated transfer learning is employed to align data characteristics across different regions (e.g., the photovoltaic output time series of region A and the load demand curve of region B), generating global absorption data without sharing the original data. Based on preset absorption rate thresholds and global absorption data, a hierarchical Q-network dynamically optimizes transmission capacity allocation: the upper-level Meta-Controller generates a priority transmission region list based on real-time tie-line power and the N-1 criterion, while the lower-level Sub-Controller allocates specific channel capacity according to unit combination constraints. The blockchain records the strategy execution effect, improving the cross-regional renewable energy absorption rate and reducing transmission losses.
[0064] Furthermore, step S20 also includes the following steps: based on the carbon emission constraint intensity in the global optimization target and historical data, dynamically allocate the model weights of different regions in the global model through a federal reward and punishment mechanism, generate a cross-regional market clearing strategy based on the model weights; allocate dedicated network slices for the power trading channel based on the cross-regional market clearing strategy, and output a priority trading link configuration scheme.
[0065] Specifically, the system dynamically assesses regional contributions based on the carbon emission constraint intensity in the global optimization objective through a federal reward and punishment mechanism: regions with high absorption rates and low emissions receive increased weight, while those with lower absorption rates receive decreased weight, driving the cross-regional market clearing strategy towards low-carbon directions. Combining historical transaction data stored on the blockchain, a clearing strategy is generated that considers real-time electricity price signals and economic objectives. Based on this clearing strategy, 5G network slicing technology allocates uRLLC slices to high-priority transaction channels and eMBB slices to ordinary transactions, forming a priority transaction link configuration scheme. This can improve the cross-regional absorption rate of new energy and reduce carbon emission intensity.
[0066] In some embodiments, step S30 specifically includes the following steps: based on the global optimization objective, the weighted Chebyshev decomposition method is used to transform the minimization of inter-regional transmission loss, power supply reliability and carbon emission constraints into multiple sets of single-objective sub-problems, each set of single-objective sub-problems corresponding to different weight combinations; the control edge nodes independently solve the single-objective sub-problems based on the inter-regional transmission strategy to generate a preliminary optimized solution set; the preliminary optimized solution sets of each region are aggregated through a global aggregator, non-dominated solutions are screened, and a Pareto final solution set is generated; the preset inter-regional scheduling experience rules are injected into the global model through knowledge distillation to generate a Pareto final solution set that satisfies the multilateral protocol.
[0067] Specifically, the multi-objective constraints are decomposed into multiple sets of single-objective sub-problems using the weighted Chebyshev decomposition method, such as an economic priority group, an environmental priority group, and a balance group. Edge nodes independently solve these sub-problems based on their local inter-regional power transmission strategies to generate an initial solution set. The global aggregator uses a non-dominated sorting algorithm to filter the solution set; for example, a solution is removed if its carbon emission intensity or transmission loss exceeds the limit by 5%. Finally, a Pareto solution set containing non-dominated solutions is generated. Simultaneously, knowledge distillation technology transforms scheduling rules such as the N-1 criterion into a model constraint layer. This effectively improves the energy absorption rate of inter-regional power transmission and increases its feasibility.
[0068] In some embodiments, step S30 further includes the following steps: performing random scenario verification on the Pareto final solution set through Monte Carlo simulation; if any scheme in the Pareto final solution set has a power supply reliability degradation value exceeding a threshold or a carbon emission exceeding the standard under random scenarios, then the corresponding sub-solution in the Pareto final solution set is removed and iterated optimization is performed again.
[0069] Specifically, by using Monte Carlo simulations to randomly generate extreme scenarios such as fluctuations in renewable energy output, sudden load changes, and tie-line failures, the final Pareto solution set is stress-tested. After removing invalid solutions, the weighted Chebyshev decomposition method is iteratively applied again to generate new solutions. This process ultimately ensures that the solution set remains usable in complex environments, improving the new power system's ability to cope with uncertainties and its applicability to complex environments.
[0070] In some embodiments, step S40 further includes the following steps: classifying noise intensity according to the data sensitivity level of different regions, and determining the corresponding noise variance according to the noise intensity of different regions; dynamically adjusting the noise variance according to the real-time cross-regional transmission capacity: when the tie line power reaches the rated capacity, reducing the noise variance to a preset ratio to retain key features, the key features including at least: tie line power and reserve capacity; storing the key features, noise intensity and noise variance in the regional noise parameter mapping table through the blockchain to ensure that the noise distribution parameters of the global aggregator and the edge nodes are updated synchronously.
[0071] Specifically, differentiated noise intensities are set through regional data sensitivity classification, and the noise variance is dynamically adjusted based on real-time transmission capacity, prioritizing the retention of key parameters such as tie-line power fluctuation characteristics and reserve capacity requirements. A noise parameter mapping table is stored on a blockchain, enabling real-time synchronization between the global aggregator and edge nodes, ensuring a dynamic balance between privacy protection and model accuracy. When transmission pressure eases, the initial noise variance is automatically restored, reducing errors in renewable energy absorption rates and carbon emission intensity in cross-regional collaborative optimization, and improving the robustness of cross-regional dispatch strategies in complex scenarios.
[0072] The implementation principle of the intelligent orchestration response method based on the power network environment in this application embodiment is as follows: The system collects cross-regional power transmission data in real time through edge nodes, updates the output parameter set using a federated prediction model, and adopts a differential privacy encryption combined with a dynamic noise adjustment mechanism: noise intensity is divided according to the regional data sensitivity level, and the noise variance is dynamically adjusted according to the real-time cross-regional power transmission capacity, retaining key features to ensure the accuracy of the strategy. A global aggregator integrates encrypted gradients to update the global model, generates a global optimization objective, and synchronizes multi-region noise configuration with a noise parameter mapping table stored in the blockchain. A dynamic power transmission capacity allocation strategy and Pareto final solution set are generated based on a hierarchical Q-network and a weighted Chebyshev decomposition method. The robustness of the solution set is verified through Monte Carlo simulation, and schemes that cause power supply reliability degradation exceeding the threshold or carbon emission exceed the standard are eliminated. After iterative optimization, the solution set that meets the multilateral protocol is retained. The closed-loop verification process compares the deviation between the actual operating data and the solution set, triggers incremental training to update the local model and feeds it back to the global optimization, forming a technical link of "dynamic adjustment - privacy protection - multi-objective optimization - closed-loop verification". This enables the coordinated optimization of privacy security, economy, reliability and environmental protection in cross-regional power dispatch, improves the renewable energy consumption rate, reduces carbon emissions and ensures the stability of key power transmission channels.
[0073] Figure 1 This is a flowchart illustrating an intelligent orchestration and response method based on a power network environment in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated otherwise, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0074] Based on the same technical concept, referring to Figure 2 This application also provides an intelligent orchestration and response device based on a power network environment, which adopts the following technical solution: The device includes:
[0075] The global module is used to collect cross-regional power transmission data in real time through edge nodes, train a preset federated prediction model based on the cross-regional power transmission data, output a set of parameter updates, and send the set of parameter updates to the global aggregator in the cloud to generate a global optimization target.
[0076] The power transmission module is used to output and execute inter-regional power transmission strategies based on global optimization objectives and historical data;
[0077] The optimization module is used to establish a weighted Chebyshev subproblem based on inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability and carbon emission constraints, and output the Pareto final solution set;
[0078] The iteration module is used to acquire actual running data in real time, compare the actual running data with the final Pareto solution set, output the deviation value, and if the deviation value is greater than the threshold, update the local model corresponding to the deviation value greater than the threshold, upload the local model to the global aggregator, and regenerate the global optimization objective.
[0079] In some embodiments, the parameter update set includes: model gradient and supply and demand features;
[0080] The global module is specifically used to collect cross-regional power transmission data in real time through edge nodes based on the Internet of Things, train a federated prediction model using the cross-regional power transmission data, and output the model gradient.
[0081] The gradient of the differential privacy encryption model is uploaded to the cloud global aggregator after being supplemented with local noise distribution parameters;
[0082] The global aggregator collects all model gradients, updates the global model, optimizes the parameters of the federated prediction model based on the global model, and distributes the optimized model parameters to the federated prediction models of the edge nodes.
[0083] The federated forecasting model is optimized based on the model optimization parameters. The supply and demand characteristics are predicted based on the optimized federated forecasting model and transmitted to the global aggregator in the cloud in real time.
[0084] A global optimization objective is generated based on the global model and the set of parameter updates.
[0085] In some embodiments, the power transmission module is specifically used to call historical data stored in the blockchain, integrate the data features of different regions through vertical federated transfer learning, and generate global absorption data;
[0086] Based on the preset absorption rate threshold and global absorption data, a dynamic transmission capacity allocation strategy is generated through a hierarchical Q network, and a priority transmission channel list is generated.
[0087] In some embodiments, the power transmission module is specifically used to dynamically allocate model weights of different regions in the global model based on the carbon emission constraint intensity in the global optimization objective and historical data, and to generate cross-regional market clearing strategies based on the model weights.
[0088] Based on the cross-regional market clearing strategy, dedicated network slices are allocated to the power trading channel, and a priority trading link configuration scheme is output.
[0089] In some embodiments, the optimization module is specifically used to transform the minimization of inter-regional transmission loss, power supply reliability and carbon emission constraints into multiple sets of single-objective sub-problems based on the global optimization objective and using the weighted Chebyshev decomposition method. Each set of single-objective sub-problems corresponds to a different weight combination.
[0090] The control edge nodes independently solve single-objective sub-problems based on the cross-regional power transmission strategy, generating a preliminary optimized solution set;
[0091] The preliminary optimization solution sets of each region are aggregated by a global aggregator, non-dominated solutions are filtered, and the final Pareto solution set is generated.
[0092] By injecting pre-defined cross-regional scheduling experience rules into the global model through knowledge distillation, a Pareto final solution set that satisfies the multilateral protocol is generated.
[0093] In some embodiments, the optimization module is also used to perform random scenario verification of the Pareto final solution set through Monte Carlo simulation;
[0094] If any solution in the Pareto final solution set experiences a power supply reliability degradation exceeding a threshold or carbon emission exceeding the limit under random scenarios, the corresponding sub-solution in the Pareto final solution set is removed and the optimization is iterated again.
[0095] In some embodiments, the optimization module is further configured to classify noise intensity according to the data sensitivity level of different regions, and determine the corresponding noise variance according to the noise intensity of different regions;
[0096] The noise variance is dynamically adjusted according to the real-time inter-regional transmission capacity: when the tie line power reaches the rated capacity, the noise variance is reduced to a preset ratio to retain key characteristics, which include at least the tie line power and reserve capacity.
[0097] By using blockchain to store key features, noise intensity, and noise variance in a regional noise parameter mapping table, the noise distribution parameters of the global aggregator and edge nodes are updated synchronously.
[0098] This application also discloses a computer-readable storage medium.
[0099] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed, such as the intelligent orchestration response method based on the power grid environment described above. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0100] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A smart orchestration response method based on a power network environment, characterized in that, include: Real-time cross-regional power transmission data is collected through edge nodes. A pre-set federated prediction model is trained based on the cross-regional power transmission data, a set of parameter updates is output, and the set of parameter updates is sent to a global aggregator in the cloud to generate a global optimization target. Based on the global optimization objective and historical data, output and execute the inter-regional power transmission strategy; Based on the aforementioned inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability, and carbon emission constraints, a weighted Chebyshev subproblem is established, and the Pareto final solution set is output. The system acquires real-time operational data and compares it with the Pareto final solution set, outputting a deviation value. If the deviation value is greater than a threshold, the local model corresponding to the deviation value greater than the threshold is updated, and the local model is uploaded to the global aggregator to regenerate the global optimization objective.
2. The intelligent orchestration and response method based on a power network environment according to claim 1, characterized in that, The parameter update set includes: model gradient and supply and demand characteristics; The process involves real-time collection of cross-regional power transmission data through edge nodes, training a pre-defined federated prediction model based on the cross-regional update data, outputting a set of parameter update data, and sending the set of parameter update data to a global aggregator in the cloud to generate a global optimization target, including: The cross-regional power transmission data is collected in real time via edge nodes based on the Internet of Things, and the federated prediction model is trained using the cross-regional power transmission data, and the model gradient is output. The model gradient is encrypted using differential privacy, and after being appended with local noise distribution parameters, it is uploaded to the cloud-based global aggregator. The global aggregator integrates all the model gradients, updates the global model, optimizes the parameters of the federated prediction model based on the global model, and distributes the optimized model parameters to the federated prediction model of the edge nodes. The federated forecasting model is optimized according to the model optimization parameters, and the supply and demand characteristics are predicted according to the optimized federated forecasting model and transmitted to the global aggregator in the cloud in real time. A global optimization objective is generated based on the global model and the parameter update set.
3. The intelligent orchestration and response method based on a power network environment according to claim 2, characterized in that, The step of outputting and executing inter-regional power transmission strategies based on the global optimization objective and historical data includes: By calling upon the historical data stored in the blockchain, and integrating the data characteristics of different regions through vertical federated transfer learning, global absorption data is generated. Based on the preset absorption rate threshold and global absorption data, a dynamic transmission capacity allocation strategy is generated through a hierarchical Q network, and a priority transmission channel list is generated.
4. The intelligent orchestration and response method based on a power network environment according to claim 2, characterized in that, The step of outputting and executing inter-regional power transmission strategies based on the global optimization objective and historical data includes: Based on the carbon emission constraint intensity in the global optimization objective and the historical data, the model weights of different regions in the global model are dynamically allocated through a federal reward and punishment mechanism, and a cross-regional market clearing strategy is generated according to the model weights. Based on the aforementioned cross-regional market clearing strategy, dedicated network slices are allocated to the power trading channel, and a priority trading link configuration scheme is output.
5. The intelligent orchestration and response method based on a power network environment according to claim 2, characterized in that, The weighted Chebyshev subproblem, established based on the inter-regional power transmission strategy, minimization of inter-regional power transmission losses, power supply reliability, and carbon emission constraints, outputs a Pareto final solution set, including: Based on the global optimization objective, the weighted Chebyshev decomposition method is used to transform the minimization of inter-regional transmission loss, power supply reliability and carbon emission constraints into multiple sets of single-objective sub-problems, each set of single-objective sub-problems corresponding to different weight combinations. The control edge nodes independently solve the single-objective sub-problem based on the inter-regional power transmission strategy, generating a preliminary optimized solution set; The global aggregator summarizes the preliminary optimized solution sets of each region, filters out non-dominated solutions, and generates the Pareto final solution set. By injecting pre-defined cross-regional scheduling experience rules into the global model through knowledge distillation, the Pareto final solution set that satisfies the multilateral protocol is generated.
6. The intelligent orchestration and response method based on a power network environment according to claim 5, characterized in that, After aggregating the preliminary optimized solution sets of each region through the global aggregator, filtering non-dominated solutions, and generating the Pareto final solution set, the method further includes: The Pareto final solution set was verified using random scenarios through Monte Carlo simulation. If any solution in the Pareto final solution set exceeds the threshold for power supply reliability degradation or carbon emission limits under the random scenario, the corresponding sub-solution in the Pareto final solution set is removed and the optimization is iterated again.
7. The intelligent orchestration response method based on a power network environment according to claim 2, characterized in that, After the model gradient is encrypted using differential privacy, and local noise distribution parameters are appended before being uploaded to the cloud-based global aggregator, the process further includes: Noise intensity is classified according to the data sensitivity level of different regions, and the corresponding noise variance is determined according to the noise intensity of different regions. The noise variance is dynamically adjusted based on the real-time inter-regional power transmission capacity: when the tie line power reaches the rated capacity, the noise variance is reduced to a preset ratio to retain key features, which include at least the tie line power and the reserve capacity. By using blockchain to store the key features, noise intensity, and noise variance in a regional noise parameter mapping table, the noise distribution parameters of the global aggregator and edge nodes are updated synchronously.
8. An intelligent orchestration and response device based on a power network environment, characterized in that, The device includes: The global module is used to collect cross-regional power transmission data in real time through edge nodes, train a preset federated prediction model based on the cross-regional power transmission data, output a set of parameter updates, and send the set of parameter updates to the global aggregator in the cloud to generate a global optimization target. The power transmission module is used to output and execute inter-regional power transmission strategies based on the global optimization objective and historical data; The optimization module is used to establish a weighted Chebyshev subproblem based on the inter-regional power transmission strategy, minimization of inter-regional power transmission loss, power supply reliability and carbon emission constraints, and output the Pareto final solution set. The iteration module is used to acquire actual running data in real time, compare the actual running data with the Pareto final solution set, output the deviation value, and if the deviation value is greater than a threshold, update the local model corresponding to the deviation value greater than the threshold, upload the local model to the global aggregator, and regenerate the global optimization objective.
9. A control device, characterized in that, The device includes: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.