Edge side data security co-processing method of power system
By deploying quantum key distribution modules and federated learning at the edge nodes of the power grid, combined with dynamic trust assessment, the security and accuracy issues of power grid terminals under the threat of quantum computing are solved, achieving efficient data processing and real-time monitoring, and ensuring the security and accuracy of the power system.
Patent Information
- Application Number
- CN202511318305.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-16
- Publication Date
- 2026-01-23
AI Technical Summary
Existing power grid terminals lack adequate encryption schemes when facing quantum computing threats. They have long key update cycles, high risk of key leakage, idle terminal computing resources, low data transmission and processing efficiency, and cannot monitor abnormal node behavior in real time, affecting the security and accuracy of the power system.
A lightweight security module with quantum key distribution is deployed at the edge node of the power grid to perform data encryption and preprocessing. Data security collaboration is achieved through federated learning and dynamic trust assessment mechanisms. Quantum random number generators are used to generate true random numbers. Combined with lattice cryptography homomorphic encryption and unsupervised anomaly detection, real-time monitoring and isolation of edge computing and model parameters are achieved.
It achieves high security against quantum computing, accelerates key updates, improves data processing efficiency, reduces latency, ensures the real-time security and accuracy of the power system, and prevents the lateral spread of attacks.
Smart Images
Figure CN121396433A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for comprehensive data management at the edge of a power grid. Background Technology
[0002] Existing power grid terminals or edge gateways primarily rely on symmetric or asymmetric encryption chips or software algorithms for data encryption. With the development of quantum computing technology, classical public-key cryptosystems such as RSA and ECC face the threat of "quantum cracking." Simultaneously, lightweight terminals have limited computing resources, making it difficult to deploy complex post-quantum cryptography (PQC) algorithms, resulting in long key update cycles and a high risk of key leakage. Furthermore, existing solutions typically send raw measurement data directly to the cloud or main station after simple compression, with the cloud handling data cleaning, feature extraction, and advanced analysis. This approach presents the following problems in existing power systems with a high proportion of renewable energy and power electronic equipment connected to the grid:
[0003] a) The data volume has expanded rapidly, highlighting the bottlenecks in transmission link bandwidth and main station computing power, and the latency is difficult to meet the millisecond-level protection and control requirements;
[0004] b) The raw data contains a large amount of invalid or redundant information, and sending it directly will waste network resources;
[0005] c) The computing power on the terminal side is idle, and the distributed computing resources are not fully utilized.
[0006] Currently, power grid terminal access authentication mostly uses one-time identity certificates or whitelists, which cannot detect abnormal behavior of nodes during operation in real time (such as malicious code injection or model poisoning attacks). Once a certain edge node is compromised, attackers can continuously upload forged data or tamper with model parameters, thereby affecting the accuracy and security of network-wide collaborative decision-making. Summary of the Invention
[0007] The purpose of this invention is to address the problems of poor accuracy and low security in the existing edge-side data security collaborative decision-making of power systems, and to propose an edge-side data security collaborative processing method for power systems.
[0008] The edge-side data security collaborative processing method for a power system according to the present invention includes the following steps:
[0009] A lightweight security module with quantum key distribution capability is deployed at the distributed edge nodes of the power grid. This module acts as a root of trust, generating and distributing unique quantum key pairs for the edge nodes. The quantum key pairs are used to encrypt the locally collected power data for the first time.
[0010] The edge node uses its local computing resources to preprocess the first encrypted power data. The preprocessing includes data cleaning, feature extraction, and preliminary anomaly detection to generate preprocessed data.
[0011] A collaborative analysis model based on federated learning is established, in which multiple edge nodes act as participants and collaboratively train a global model for power grid status assessment using local data without sharing the original preprocessed data. The parameter updates of the global model are protected by homomorphic encryption technology.
[0012] A dynamic trust assessment mechanism is set up to continuously monitor the behavioral data of each edge node. When any node is assessed as untrustworthy, the key revocation and model parameter isolation mechanism is automatically triggered to ensure the data security and model integrity of the collaborative system.
[0013] Furthermore, the quantum key distribution function is implemented by a quantum random number generator integrated into the lightweight security module. The quantum random number generator generates truly random numbers based on the principle of vacuum state fluctuations to generate the quantum key pair.
[0014] Furthermore, the feature extraction in the preprocessing step specifically includes:
[0015] Extracting time-series features of voltage, current, and frequency from power data;
[0016] Wavelet transform is used to extract transient signal features to identify early signs of power grid faults.
[0017] Furthermore, the collaborative training process of the federated learning includes:
[0018] Each edge node trains its local model using its preprocessed data and generates local gradients.
[0019] The local gradient is encrypted using a lattice-based homomorphic encryption scheme and then uploaded to the aggregation server.
[0020] The aggregation server aggregates all encrypted local gradients, generates a global gradient, and distributes it to each edge node to update its local model.
[0021] Furthermore, the dynamic trust assessment mechanism specifically includes:
[0022] Collect behavioral data from each edge node, including data upload frequency, model parameter update magnitude, and consistency of anomaly detection results;
[0023] An anomaly detection algorithm based on unsupervised learning is used to identify suspicious nodes that deviate from group behavior;
[0024] When the trust score of a suspicious node falls below a preset threshold, the key revocation and model parameter isolation mechanism is triggered.
[0025] Furthermore, the key revocation and model parameter isolation mechanism includes:
[0026] Broadcast the revocation command to all edge nodes and mark the quantum keys of suspicious nodes as invalid;
[0027] Before the next round of federated learning training, the suspicious nodes are removed from the list of participants and their local model parameters are sandboxed to prevent contamination of the global model.
[0028] Furthermore, the edge-side data security collaborative processing method also includes:
[0029] Deploy smart terminals with edge computing capabilities at key nodes of the power grid. The smart terminals integrate the lightweight security module and the local computing resources.
[0030] The smart terminal communicates with the aggregation server via a 5G or fiber optic network.
[0031] Furthermore, the intelligent terminal also has an adaptive power consumption management function, which can dynamically adjust its computing and communication resources according to the power grid load.
[0032] Compared with the prior art, the present invention has the following advantages:
[0033] This invention employs a lightweight security module integrating a vacuum fluctuation quantum random number generator (QRNG) to dynamically generate unique quantum key pairs for each edge node, achieving "one-time pad". Compared to existing RSA / ECC schemes, it possesses resistance to quantum computing cracking capabilities, a security lifespan of ≥20 years, and meets the long-term evolution requirements of power systems. Data cleaning, feature extraction, and preliminary anomaly detection are moved to the edge nodes, with local CPU / GPU / NPU collaborative computation. The original data compression rate is ≥80%, and the end-to-end latency is reduced from the traditional second level to within 5 ms, meeting the millisecond-level operational requirements of relay protection, frequency emergency control, and other applications. Based on unsupervised anomaly detection, real-time monitoring of node behavior (upload frequency, gradient update amplitude, and detection result consistency) is performed. Once the trust score falls below the threshold, quantum key revocation, node removal, and model parameter sandbox isolation can be completed within <1 second, preventing the lateral spread of poisoning attacks and maintaining system availability >99.9%. Therefore, the secure collaborative processing method described in this invention offers high accuracy and security in processing edge data.
[0034] This invention is applicable to distributed renewable energy power stations without mains power. Attached Figure Description
[0035] Figure 1This is a flowchart of a power system edge-side data security collaborative processing method as described in Specific Implementation 1. Detailed Implementation
[0036] Specific Implementation Method 1: Combination Figure 1 This embodiment describes a method for edge-side data security collaborative processing in a power system, which includes the following steps:
[0037] A lightweight security module with quantum key distribution capability is deployed at the distributed edge nodes of the power grid. This module acts as a root of trust, generating and distributing unique quantum key pairs for the edge nodes. The quantum key pairs are used to encrypt the locally collected power data for the first time.
[0038] The edge node uses its local computing resources to preprocess the first encrypted power data. The preprocessing includes data cleaning, feature extraction, and preliminary anomaly detection to generate preprocessed data.
[0039] A collaborative analysis model based on federated learning is established, in which multiple edge nodes act as participants and collaboratively train a global model for power grid status assessment using local data without sharing the original preprocessed data. The parameter updates of the global model are protected by homomorphic encryption technology.
[0040] A dynamic trust assessment mechanism is set up to continuously monitor the behavioral data of each edge node. When any node is assessed as untrustworthy, the key revocation and model parameter isolation mechanism is automatically triggered to ensure the data security and model integrity of the collaborative system.
[0041] In this embodiment, for the first time, four technology chains—quantum key distribution, edge preprocessing, federated learning, and dynamic trust—are integrated into the same process framework to achieve an end-to-end secure collaborative closed loop; the overall security of the system is high.
[0042] Specific Implementation Method Two: This implementation method further defines the edge-side data security collaborative processing method for a power system described in Specific Implementation Method One. In this implementation method, the quantum key distribution function is implemented by a quantum random number generator integrated into the lightweight security module. The quantum random number generator generates truly random numbers based on the principle of vacuum state fluctuations to generate the quantum key pair.
[0043] In this embodiment, by adopting QRNG based on the principle of vacuum fluctuations, the quality of the true random number entropy source is ≥0.997, the generation rate of quantum key pairs is increased from 10 kbps in the soft algorithm mode to ≥2 Mbps, and the key rotation cycle is shortened from hours to minutes, greatly reducing the long-term key exposure window.
[0044] Specific Implementation Method Three: This implementation method further defines the edge-side data security collaborative processing method for power systems described in Specific Implementation Method One. In this implementation method, the feature extraction in the preprocessing step specifically includes:
[0045] Extracting time-series features of voltage, current, and frequency from power data;
[0046] Wavelet transform is used to extract transient signal features to identify early signs of power grid faults.
[0047] In this embodiment, after simultaneously extracting time-series features (RMS, harmonics, frequency drift) and wavelet transient features locally at the edge node, the detection accuracy in the early stage of fault (≤3 ms) is improved from 78% to 93% by the traditional FFT method, and the false alarm rate is reduced by 65%, providing a high-confidence trigger signal for millisecond-level protection.
[0048] Specific Implementation Method Four: This implementation method further defines the edge-side data security collaborative processing method for a power system described in Specific Implementation Method One. In this implementation method, the collaborative training process of federated learning includes:
[0049] Each edge node trains its local model using its preprocessed data and generates local gradients.
[0050] The local gradient is encrypted using a lattice-based homomorphic encryption scheme and then uploaded to the aggregation server.
[0051] The aggregation server aggregates all encrypted local gradients, generates a global gradient, and distributes it to each edge node to update its local model.
[0052] In this embodiment, homomorphic encryption using lattice cryptography (such as BGV-512) is used to encrypt the gradient, which increases the aggregation calculation latency by less than 15 ms, but greatly increases the cracking complexity. Compared with existing plaintext gradient transmission, a round of federated training can be completed under completely zero leakage conditions, meeting the compliance requirements for cross-border data flow in the energy industry.
[0053] Specific Implementation Method Five: This implementation method further defines the edge-side data security collaborative processing method for power systems described in Specific Implementation Method One. In this implementation method, the dynamic trust assessment mechanism specifically includes:
[0054] Collect behavioral data from each edge node, including data upload frequency, model parameter update magnitude, and consistency of anomaly detection results;
[0055] An anomaly detection algorithm based on unsupervised learning is used to identify suspicious nodes that deviate from group behavior;
[0056] When the trust score of a suspicious node falls below a preset threshold, the key revocation and model parameter isolation mechanism is triggered.
[0057] In this implementation, by introducing unsupervised anomaly detection (LOF+AutoEncoder hybrid model), the identification time for poisoning attacks and node spoofing is shortened from the existing manual audit of T+3 days to T+30 seconds, the abnormal node recovery rate is ≥99%, and the attacker's incubation window is significantly reduced.
[0058] Specific Implementation Method Six: This implementation method further defines the edge-side data security collaborative processing method for a power system described in Specific Implementation Method One. In this implementation method, the key revocation and model parameter isolation mechanism includes:
[0059] Broadcast the revocation command to all edge nodes and mark the quantum keys of suspicious nodes as invalid;
[0060] Before the next round of federated learning training, the suspicious nodes are removed from the list of participants and their local model parameters are sandboxed to prevent contamination of the global model.
[0061] In this implementation, the dual-action mechanism of key revocation broadcasting + model parameter sandbox isolation achieves "one-click chain disconnection", completing node removal and pollution parameter isolation before the start of the next training round (as fast as 1 minute), avoiding continuous pollution of the global model, and the system's average recovery time MTTR < 2 minutes.
[0062] Specific Implementation Method Seven: This implementation method further defines the edge-side data security collaborative processing method for a power system described in Specific Implementation Method One. In this implementation method, the edge-side data security collaborative processing method further includes:
[0063] Deploy smart terminals with edge computing capabilities at key nodes of the power grid. The smart terminals integrate the lightweight security module and the local computing resources.
[0064] The smart terminal communicates with the aggregation server via 5G or fiber optic networks to ensure the real-time performance and stability of data transmission.
[0065] In this embodiment, intelligent terminals integrating "quantum modules + edge computing power" are deployed at key nodes. Through 5GuRLLC (1 ms air interface) or direct fiber optic connection, the end-to-end communication latency is stabilized within 3 ms. Compared with the existing three-level architecture of "terminal + aggregation + core", the number of network hops is reduced by 2, and the link reliability is improved by an order of magnitude.
[0066] Specific Implementation Method Eight: This implementation method further defines the edge-side data security collaborative processing method for a power system described in Specific Implementation Method Seven. In this implementation method, the intelligent terminal also has an adaptive power consumption management function, which can dynamically adjust its computing and communication resources according to the power grid load to optimize energy consumption.
[0067] In this embodiment, by introducing adaptive power management, the terminal automatically switches between three modes: "power saving / balanced / performance" under different load scenarios. The power consumption is less than 2 W under low load, and the computing power can be instantly boosted to 15 TOPS under peak load. The average annual power consumption is reduced by 30%, and the battery backup time is extended by 40%, which is suitable for distributed new energy power stations without mains power.
[0068] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for secure collaborative processing of edge-side data in a power system, characterized in that, Includes the following steps: A lightweight security module with quantum key distribution capability is deployed at the distributed edge nodes of the power grid. This module acts as a root of trust, generating and distributing unique quantum key pairs for the edge nodes. The quantum key pairs are used to encrypt the locally collected power data for the first time. The edge node uses its local computing resources to preprocess the first encrypted power data. The preprocessing includes data cleaning, feature extraction, and preliminary anomaly detection to generate preprocessed data. A collaborative analysis model based on federated learning is established, in which multiple edge nodes act as participants and collaboratively train a global model for power grid status assessment using local data without sharing the original preprocessed data. The parameter updates of the global model are protected by homomorphic encryption technology. A dynamic trust assessment mechanism is set up to continuously monitor the behavioral data of each edge node. When any node is assessed as untrustworthy, the key revocation and model parameter isolation mechanism is automatically triggered to ensure the data security and model integrity of the collaborative system.
2. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The quantum key distribution function is implemented by a quantum random number generator integrated into the lightweight security module. The quantum random number generator generates truly random numbers based on the principle of vacuum state fluctuations to generate the quantum key pair.
3. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The feature extraction in the preprocessing step specifically includes: Extracting time-series features of voltage, current, and frequency from power data; Wavelet transform is used to extract transient signal features to identify early signs of power grid faults.
4. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The collaborative training process of the federated learning includes: Each edge node trains its local model using its preprocessed data and generates local gradients. The local gradient is encrypted using a lattice-based homomorphic encryption scheme and then uploaded to the aggregation server. The aggregation server aggregates all encrypted local gradients, generates a global gradient, and distributes it to each edge node to update its local model.
5. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The dynamic trust assessment mechanism specifically includes: Collect behavioral data from each edge node, including data upload frequency, model parameter update magnitude, and consistency of anomaly detection results; An anomaly detection algorithm based on unsupervised learning is used to identify suspicious nodes that deviate from group behavior; When the trust score of a suspicious node falls below a preset threshold, the key revocation and model parameter isolation mechanism is triggered.
6. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The key revocation and model parameter isolation mechanism includes: Broadcast the revocation command to all edge nodes and mark the quantum keys of suspicious nodes as invalid; Before the next round of federated learning training, the suspicious nodes are removed from the list of participants and their local model parameters are sandboxed to prevent contamination of the global model.
7. The edge-side data security collaborative processing method for a power system according to claim 1, characterized in that, The edge-side data security collaborative processing method also includes: Deploy smart terminals with edge computing capabilities at key nodes of the power grid. The smart terminals integrate the lightweight security module and the local computing resources. The smart terminal communicates with the aggregation server via 5G or fiber optic networks to ensure the real-time performance and stability of data transmission.
8. The edge-side data security collaborative processing method for a power system according to claim 7, characterized in that, The intelligent terminal also has an adaptive power consumption management function, which can dynamically adjust its computing and communication resources according to the power grid load to optimize energy consumption.