Power emergency network congestion control method based on low-orbit broadband satellite

Abstract

The invention discloses an electric power emergency network congestion control method based on a low-orbit broadband satellite, which belongs to the technical field of electric power, and comprises the following steps: judging a network state as a non-uniform flow distribution or malicious node attack state according to congestion inducements; executing a flow scheduling strategy based on trust evaluation aiming at a flow distribution non-uniform state: calculating a link trust value to identify a key link, screening a key flow, and calculating a rerouting path by optimizing an objective function to perform flow migration; and aiming at a malicious node attack state, executing a congestion relieving strategy based on spatial-temporal characteristics and reinforcement learning: evaluating a node trust value and isolating malicious nodes based on the spatial-temporal characteristics, and dynamically adjusting a packet loss strategy of satellite nodes by utilizing a deep reinforcement learning model which is executed in a centralized training and distribution manner. According to the method, the problems of satellite link congestion misjudgment, malicious attack defense deficiency, slow sudden service response and the like in an electric power emergency scene are solved, and reliable and low-delay transmission of key electric power services in an extreme environment is remarkably guaranteed.

Description

Technical Field

[0001] This invention relates to the field of power technology, specifically to a method for congestion control in power emergency networks based on low-orbit broadband satellites. Background Technology

[0002] With the deepening of smart grid construction and the rapid development of the energy internet, the demand for real-time performance, reliability, and bandwidth in power services is growing exponentially. Especially during extreme emergencies such as earthquakes, typhoons, and snowstorms, the power infrastructure in affected areas is often paralyzed along with ground communication networks. In such situations, quickly establishing a stable and efficient emergency communication link to ensure the real-time transmission of disaster information and the accurate issuance of dispatch instructions is the lifeline for ensuring rapid power grid repair and restoration.

[0003] Currently, the power emergency communication system mainly consists of terrestrial fiber optic private networks, public wireless networks (4G / 5G), and traditional high-orbit (GEO) satellite communication. However, these methods have significant shortcomings in responding to extreme disasters: terrestrial fiber optics and base stations have poor resilience and are easily interrupted by tower collapses or power outages, and their coverage is insufficient in remote areas such as deep mountains and deserts; although traditional high-orbit satellites have wide coverage, their long transmission distances result in high round-trip latency, and their narrow channel bandwidth makes it difficult to support the high-bandwidth requirements of high-definition video monitoring and drone inspection image transmission in power emergency scenarios, and even more so to meet the stringent requirements of millisecond-level low latency for control services such as relay protection.

[0004] In recent years, a new generation of space communication technologies, represented by low Earth orbit (LEO) broadband satellite constellations, has flourished. LEO satellite networks, with their low orbital altitude, low transmission latency, large link bandwidth, and seamless global coverage, can effectively overcome geographical limitations and quickly fill the gaps during the "darkest moments" of terrestrial network failures. They provide access services to power emergency sites with a similar experience to terrestrial broadband, significantly improving the communication support capabilities of power systems in response to sudden disasters.

[0005] However, applying low-Earth orbit (LEO) satellite networks to complex power emergency scenarios still faces significant technical challenges in congestion control. First, LEO satellites move at high speeds relative to the Earth, resulting in a highly dynamic network topology and frequent switching between satellite-to-ground and inter-satellite links. Traditional TCP congestion control protocols often incorrectly identify random packet loss caused by link switching or rain attenuation as network congestion, blindly shrinking the congestion window and leading to a significant drop in throughput and extremely low bandwidth utilization. Second, the service traffic at power emergency sites is highly sudden and heterogeneous, including both latency-sensitive control commands and bandwidth-intensive panoramic video feeds. When momentary congestion occurs at network edge nodes, without a differentiated congestion management mechanism based on service priorities, critical control data can easily be blocked or even dropped by the video stream, causing serious power grid safety hazards. Furthermore, the long-latency bandwidth product characteristics of LEO satellite networks differ from those of terrestrial networks. Existing congestion control algorithms struggle to maintain high link utilization while ensuring low queuing latency, easily leading to buffer bloat and further deteriorating the experience of interactive emergency services.

[0006] Currently, some research has been conducted on congestion control for satellite networks, but most of it focuses on optimizing general internet services in static satellite scenarios, or only on physical layer link adaptation technology. Few studies have deeply integrated the specific QoS requirements of power emergency services with the highly dynamic link characteristics of low-Earth orbit satellites, and there is a lack of a congestion control scheme capable of cross-layer network state perception and intelligent decision-making.

[0007] Based on this, the present invention designs a power emergency network congestion control method based on low-orbit broadband satellites to solve the above problems. Summary of the Invention

[0008] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a power emergency network congestion control method based on low-orbit broadband satellite.

[0009] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for congestion control in power emergency networks based on low-Earth orbit broadband satellites, comprising the following steps: Step 1: Monitor the network status of the power emergency network in real time, and determine the network status as uneven traffic distribution or malicious node attack state based on the congestion causes. Step 2: When the traffic distribution is determined to be uneven, execute a traffic scheduling strategy based on trust assessment: calculate link trust values ​​to identify critical links, filter critical flows, and calculate rerouting paths for traffic migration by optimizing the objective function; Step 3: When the state is determined to be a malicious node attack, execute a congestion mitigation strategy based on spatiotemporal features and reinforcement learning: First, evaluate the node trust value based on spatiotemporal features and isolate the malicious node. Then, use a deep reinforcement learning model that is centrally trained and distributed to dynamically adjust the packet loss strategy of the satellite node.

[0010] Furthermore, this invention provides a power emergency service traffic scheduling strategy (TrustFlow) based on trust assessment. This strategy is mainly applied during the normal operation phase of the power emergency network. Through the collaborative work of four core modules, it solves the load balancing problem under the highly dynamic changes of the satellite-to-ground link, ensuring that critical power emergency services do not experience congestion. Specifically, it includes: Power Emergency Link Selection Module: Addressing the high cost and significant stability variations of low-Earth orbit satellite links in power emergency networks, a link trust assessment model is constructed. The controller in the command center periodically acquires the overall network status and, considering link transmission loss, propagation delay, and bandwidth usage, calculates the trust value for each satellite-to-ground / inter-satellite link. :

[0011] in, For link path loss, To delay the transmission time, The current power service load carried by the link. This represents the total bandwidth. Links with trust values ​​below a preset threshold are marked as "critical links," which are high-risk links that may cause power dispatch commands to be blocked.

[0012] Power Critical Flow Identification and Selection Module: To prevent high-bandwidth-consuming services from crowding out high-priority control services, this module performs characteristic analysis on traffic on critical links. It calculates the data flow growth rate by monitoring the changes in the size of the power service data flow per unit time. :

[0013] in, This indicates the number of bytes received by the switch. The system identifies high-bandwidth services with growth rates exceeding a preset threshold (such as high-definition video transmission from disaster sites and drone inspection images) as "critical flows." These critical flows are the primary scheduling targets; migrating them can reserve "zero-congestion" transmission channels for small-packet critical services such as relay protection and remote control.

[0014] Critical Flow Rerouting Module: This module aims to find the optimal alternative paths for critical high-volume flows, such as selected power emergency field videos, while avoiding secondary congestion caused by migration. A multi-objective optimization model is established, ensuring network-wide load balancing, through the following objective function. Calculate the rerouting path:

[0015] This formula aims to minimize the weighted sum of the maximum link utilization and the total network load, ensuring that critical flows are scheduled to high-trust idle satellite links, thereby improving the overall throughput and stability of the power emergency network.

[0016] The routing policy generation and distribution module transforms the calculated rerouting paths into executable flow table entries. The controller compares the new path with the original shortest path, generating new OpenFlow flow table entries or updating routing rules only for satellite nodes that have changed. This on-demand update method rapidly completes path switching for power services, enabling dynamic shaping of emergency network traffic and ensuring that all types of power emergency services can obtain matching communication resources as needed during disasters, maintaining communication continuity.

[0017] Furthermore, this invention provides a power emergency network congestion mitigation strategy based on spatiotemporal features and reinforcement learning (ETM-DCDRL). This strategy aims to maintain the survivability of critical power services by constructing an active defense system of "identification-isolation-mitigation" when the power emergency network suffers malicious attacks or sudden abnormal traffic due to disasters. The main contents include:

[0018] (1) Trust record: Using the software-defined network flow table extension function, the total number of Packet_in requests of each node within a unit of time is recorded. And the number of successfully forwarded data packets (Data_S) and the number of failed forwarding packets (Data_U).

[0019] (2) Multidimensional trust assessment: Calculate trust values ​​based on time frequency. To detect flood attacks: Calculate the data forwarding trust value using Bayesian methods : Combined with control message forwarding trust value And neighbor node recommendations, calculate spatial forwarding trust value A composite trust value is generated by combining the above indicators. (3) Trust isolation: When the aggregate trust value is lower than the preset security threshold, a flow table instruction is issued to discard traffic from that node.

[0020] Furthermore, the congestion mitigation strategy based on deep reinforcement learning includes the following steps: Problem Modeling: Establish a joint optimization objective for satellite network congestion mitigation. Under the constraints of buffer capacity M and link bandwidth C, maximize the node transmission rate and minimize queuing length and packet loss. in, For transmission rate, The buffer queue length, For packet loss probability, To receive traffic.

[0021] (4) Model Construction and Training: The Deep Deterministic Policy Gradient (DDPG) algorithm based on the Actor-Critic architecture is used to construct a model architecture that includes an online / target Actor network and an online / target Critic network. The training content is as follows:

[0022] Define reward function : Update the Critic network parameters by minimizing the loss function. : Update the Actor network parameters using a deterministic policy gradient. : The target network parameters are synchronized using a soft update mechanism, the formula of which is as follows: in, This is for updating coefficients.

[0023] (5) Distributed execution: The trained Actor network model parameters are distributed to each satellite node; during communication, each satellite node only relies on the local state observed locally. The optimal packet drop probability is calculated and output in real time through the local Actor network.

[0024] The optimal packet drop probability The calculation formula is as follows: .

[0025] Compared to existing technologies, the advantages of this invention are as follows: This invention, tailored to the characteristics of power emergency networks, can accurately distinguish between two congestion modes: uneven traffic distribution and malicious attacks. By using the TrustFlow strategy to migrate large flows such as video, it reserves channels for critical smaller flows such as relay protection, achieving zero-congestion transmission in conventional scenarios.

[0026] This invention introduces the ETM mechanism to profile the behavior of power terminals from time and space dimensions, which can quickly identify and isolate flooding and packet loss attacks, and prevent malicious nodes from paralyzing the emergency network.

[0027] This invention employs a DCDRL algorithm with centralized training and distributed execution, enabling satellite nodes to possess independent intelligence. Even under extreme conditions such as limited satellite-to-ground links or explosive network traffic, the satellite can still control the queuing latency of critical services to an extremely low level through a millisecond-level dynamic packet loss strategy, significantly outperforming traditional congestion control algorithms. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0029] Figure 1 For power emergency communication scenarios; Figure 2 This is a flowchart of traffic scheduling based on trust assessment. Figure 3 This is a diagram illustrating a flood attack. Figure 4 This is a diagram illustrating a forwarding attack. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] This embodiment constructs a specific power emergency communication scenario to illustrate in detail the specific execution flow, mathematical model, and parameter settings of the power emergency network congestion control method based on low-Earth orbit broadband satellites of the present invention. The core logic of the method of the present invention is as follows: based on the specific causes of network congestion (uneven traffic distribution or malicious node attacks), either a traffic scheduling strategy based on trust assessment (TrustFlow) or a congestion mitigation strategy based on spatiotemporal features and reinforcement learning (ETM-DCDRL) is triggered.

[0032] Example 1: Please refer to the accompanying drawings in the instruction manual. Figure 1This embodiment sets the scenario as a severe geological disaster in a mountainous area, resulting in the interruption of terrestrial power communication fiber optic cables. The power emergency command center restores communication using a space-ground integrated network composed of low-Earth orbit (LEO) broadband satellites. Within this network, based on service requirements and behavioral characteristics, the following three types of typical terminal nodes are deployed:

[0033] Terminal A (High-priority control service): Located in the damaged substation, it needs to send relay protection trip commands or telemetry data. This service flow is small, but it is sensitive to latency and cannot tolerate congestion or packet loss.

[0034] Terminal B (High-Bandwidth Monitoring Service): Drones located at the repair site transmit high-definition panoramic video in real time. This service generates a large amount of data, and its unpredictable nature, depending on the content being filmed, makes it highly susceptible to becoming a "major surge" in the network and consuming satellite link bandwidth.

[0035] Terminal C (Abnormal / Malicious Node): A terminal may be controlled by a network attack, sending high-frequency connection requests to the satellite network (flooding attack) or deliberately dropping data packets (forwarding attack) in an attempt to paralyze the network.

[0036] In this scenario, the power emergency network congestion control method based on low-orbit broadband satellite of the present invention is executed as follows to ensure the smooth flow of instructions from terminal A, while rationally scheduling the video stream of terminal B and isolating attacks from terminal C.

[0037] Scenario 1: For congestion caused by uneven traffic distribution, trigger a traffic scheduling strategy based on trust assessment (TrustFlow strategy).

[0038] When the system detects no malicious attack characteristics in the network, but some links are overloaded, causing control command transmission from terminal A to be blocked, the congestion is determined to be caused by "uneven traffic distribution." At this point, a traffic scheduling policy based on trust assessment (TrustFlow policy) is initiated. The TrustFlow policy is executed collaboratively by four modules: a link selection module, a critical flow selection module, a critical flow rerouting module, and a routing policy generation module. The relationships between these modules are as follows: Figure 2 As shown.

[0039] The TrustFlow strategy mainly includes the following steps: 1.1 During the network initialization startup phase, the TrustFlow strategy first uses the shortest path routing strategy to pre-build a set of paths, and allocates paths to newly arriving data streams based on the source address and destination address, thereby ensuring the normal forwarding of initial traffic.

[0040] 1.2 During operation, the link selection module will periodically collect traffic data and calculate the link trust value. If the trust value of a link is found to be lower than the set threshold, it means that the link is at risk of congestion. The system will mark it as a critical link and extract the critical flow from it and send it to the rerouting module.

[0041] 1.3 Subsequently, the rerouting module uses the preset path set to plan new transmission paths for critical flows, guiding nodes to switch traffic from the original path to the new path, thereby reducing the occupancy rate of low-trust links.

[0042] 1.4 Finally, the routing policy generation module compares the new paths of critical flows with the shortest path set. For overlapping parts, the existing routing rules are directly reused for forwarding. Only for non-overlapping parts, the routing table is updated as needed. In this way, the system overhead of traffic scheduling is effectively reduced.

[0043] The following section will provide a detailed introduction to the four modules of the TrustFlow strategy: I. Link Selection Module: To alleviate link congestion and achieve load balancing in the power emergency network, the primary task is to identify critical links requiring traffic scheduling. This module assesses the trust of links and assigns trust values ​​based on two indicators: link transmission cost and link utilization. Links below the trust threshold are identified as critical links. On the one hand, given the high construction cost of satellite links in the power emergency communication architecture, their security and stability must be prioritized over terrestrial links. On the other hand, link load rate is an important reference indicator for assessing whether power emergency service links urgently require scheduling. These indicators can be obtained in real time through network topology discovery and port scanning technologies. Furthermore, for satellite and terrestrial links, if their load ratios are the same, the priority for traffic scheduling on satellite links should be higher than that on terrestrial links. To accurately capture link characteristics and avoid scheduling traffic to links with high transmission costs or instability, the trust value e of each link is calculated using the following formula. :

[0044] The first term of the formula captures the characteristics and differences between satellite links and terrestrial links, where, For the path loss on link e, The first term describes the propagation delay; the second term describes how link load affects the trust value, where, This refers to the bandwidth currently used by the power services carried by the link. The larger the ratio in the second item, the heavier the load, and the more likely link e is to become congested.

[0045] Finally, links in the current network with trust values ​​below the trust threshold are selected and marked as "critical links," and added to the critical link set E.crucial middle.

[0046] II. Critical Flow Selection Module: This module determines critical links based on the evaluation results of link trust values, combined with differences in link physical characteristics and load rate indicators, and further filters critical flows from these. If the set of critical links is not empty, it indicates that the power emergency network may face congestion risks, and path replanning of service flows is urgently needed. In order to ensure communication effectiveness while reducing scheduling overhead, this method focuses on detecting and scheduling high-bandwidth service flows in critical links. The network controller can obtain service traffic data recorded by forwarding nodes within a time window. Given the positive correlation between data flow size and its duration, the classification of large and small flows can be completed by periodically monitoring the change in the total number of bytes transmitted by power service flows. Calculate the data flow f within the time window [ The rate of change in size within the data stream, i.e., the data stream growth rate. :

[0047] in, They represent The number of bytes received by the switch at a given time point.

[0048] When making traffic scheduling decisions for critical links in the power emergency network, the system focuses on high-bandwidth service flows. To effectively reduce the computational burden on the network control center and avoid system overhead caused by scheduling massive amounts of small service flows, this method does not migrate small flows. Instead, it migrates flows whose rate of change exceeds a set value (set in this embodiment). Electricity business flows exceeding 10% are identified as critical flows and added to the critical flow set. Further processing will be carried out in the following stages.

[0049] III. Critical Flow Rerouting Module: To fully exploit the potential of redundant links in the power emergency network, this method reroutes high-bandwidth power services on critical links to nearby idle links, thereby improving overall network throughput and optimizing emergency communication performance. To achieve load balancing, this module calculates the optimal path for power service flows by solving an optimization problem, designing variables, an optimization objective function, and constraints adapted to the emergency scenario. Typically, network-wide link load balancing is a crucial indicator of emergency network performance, but focusing solely on this single indicator can have negative effects; for example, in pursuit of ultimate load balancing, critical power service flows might be rerouted to excessively long and circuitous paths. This long-distance detour not only increases the overall traffic burden of the power emergency network but may also affect the transmission of other emergency service flows. Therefore, this traffic scheduling objective considers two indicators simultaneously: link load balancing performance and the total link load in the network.

[0050] The objective function is optimized as follows. Calculate the rerouting path: This formula aims to minimize the weighted sum of the maximum link utilization and the total network load, ensuring that critical flows are scheduled to high-trust idle satellite links, thereby improving the overall throughput and stability of the power emergency network.

[0051] in, To maximize link utilization; Total link load; y is a weighting coefficient, taking a value of (0,1), used to balance load balancing and transmission costs. Constraint 1 sets the transmission path for power emergency service flows, stipulating that any independent power service flow can only be forwarded through a single path during transmission, where y f p The flag indicates whether traffic f is rerouted on path p; Constraint 2 is set for the link load of the power emergency network, where C represents the maximum transmission capacity of the satellite link and E represents the set of all links in the network. This constraint requires that at any time, the total power service load carried by the link must not exceed its maximum bandwidth limit.

[0052] IV. Routing Policy Generation Module: After the critical flow rerouting module receives the power emergency service flow to be migrated and its newly planned path, the routing policy generation module is responsible for converting the path into a routing deployment policy executable by network nodes. This policy consists of flow table forwarding entries and basic shortest path routing rules. Its core logic is to generate specific flow table control instructions for each rerouting power emergency service path, and use the existing shortest path routing mechanism in conjunction with the newly issued flow table entries to construct a new service transmission channel.

[0053] Specifically, the calculated rerouting paths are transformed into executable flow table entry policies. The controller compares the new path with the original shortest path, generating new OpenFlow flow table entries or updating routing rules only for satellite nodes that have changed. Through this on-demand update method, path switching for power services can be quickly completed, enabling dynamic shaping of emergency network traffic and ensuring that all types of power emergency services can obtain matching communication resources as needed during disasters, maintaining communication continuity.

[0054] In this embodiment, the controller periodically monitors the network status and performs the following operations: a. Link Identification: A continuous increase in the link load between terminal B and satellite Sat-1 is detected, and the trust value of this link is calculated. If the link drops below a preset threshold, the link selection module marks it as a critical link.

[0055] b. Traffic Filtering: The controller analyzes the traffic on critical links and finds that the traffic of terminal A is extremely small and stable, while the video stream of terminal B is affected by sudden changes in the image. The growth rate is calculated. If the percentage is greater than 10%, the critical stream selection module will lock the video stream of terminal B as the critical stream.

[0056] c. Path Calculation: The critical flow rerouting module calculates a new path via satellite Sat-2 based on the optimization objective function. Although this path has slightly more hops, Sat-2 currently has low load and can meet the bandwidth requirements.

[0057] d. Policy Execution: The routing policy generation module generates and issues flow table update instructions. The video stream of terminal B is seamlessly migrated to Sat-2 for transmission, freeing up bandwidth resources on Sat-1 and ensuring that the relay protection instructions of terminal A can be transmitted with low latency on Sat-1, successfully resolving the congestion caused by uneven traffic distribution.

[0058] Scenario 2: For congestion caused by malicious node attacks, trigger a congestion mitigation strategy based on spatiotemporal features and reinforcement learning (ETM-DCDRL strategy).

[0059] During the operation of the power emergency communication network, if the system detects an abnormal surge in packet-in requests (suspected flooding attack, such as...) Figure 3 (As shown) or a sudden drop in packet forwarding success rate (suspected forwarding attack / packet loss, such as...) Figure 4 When the congestion occurs (as shown), the cause is determined to be "malicious node attack". At this time, the congestion mitigation strategy based on spatiotemporal features and reinforcement learning (ETM-DCDRL strategy) is activated. The ETM-DCDRL strategy uses the two key processes of attack detection and congestion mitigation to reduce the impact of the attack, so as to achieve more effective network management and protection.

[0060] The ETM-DCDRL strategy mainly includes the following: 2.1 Anomaly Node Attack Detection Mechanism (ETM) Based on Spatiotemporal Features In order to accurately identify malicious nodes that attempt to paralyze the power emergency network, this invention first constructs a complete trust management closed loop, covering three stages: trust recording, multi-dimensional trust assessment, and trust isolation.

[0061] (1) Collection and recording of trust data The system utilizes the flow table extension function of the software-defined network architecture to record the behavioral characteristics of power terminal nodes in real time. To combat flooding attacks, the controller maintains a Packet_in counter table and sets an interval counter for each source node to count the total number of Packet_in message requests it receives per unit of time. To counter forwarding attacks, the data flow table is expanded with "Number of successful forwardings (Data_S)" and "Number of failed forwardings (Data_U)" fields. When node i sends a data packet to the next-hop node j, a listening mechanism is started. If node j successfully forwards the packet, Data_S is incremented by 1; otherwise, Data_U is incremented by 1.

[0062] (2) Multidimensional Trust Assessment Based on Spatiotemporal Characteristics Based on the collected behavioral characteristics, the credibility of nodes is quantified from two dimensions: time (packet sending frequency) and space (forwarding success rate).

[0063] Time-based trust assessment: Calculate a time-frequency-based trust value based on the high-frequency request behavior (flooding characteristics) of power terminals. The more frequently a node sends requests, the lower its trust value. The calculation formula is as follows:

[0064] Where i is the evaluation node, j is the node being evaluated, and t represents the time slot. For frequency factors, This represents the total number of message requests.

[0065] Spatial Dimension Trust Assessment: For forwarding attacks, Bayesian methods are used to calculate data forwarding trust values. : Where i is the evaluation node, and j is the node being evaluated. It is the number of successful forwards about node j recorded by node i. It's the number of failures. This represents the total number of data forwards.

[0066] Similarly, a trust value for controlling message forwarding can be calculated. .

[0067] in, It is a node Records about nodes The number of times control messages were successfully forwarded. It controls the total number of forwards. It represents the number of failures; This leads to the direct forwarding trust value. : in, α It is the trust weight for data forwarding. β It controls the forwarding trust weight. α+β =1; Furthermore, by combining neighbor node recommendations, the spatial forwarding trust value is calculated. : in, It is with nodes Neighbors who can communicate normally The node being evaluated The neighbor node index, It is the direct forwarding trust value that node i evaluates for node j in time slot t; Aggregated Trust and Isolation: Combining the temporal and spatial trust values ​​mentioned above, an aggregated trust value is generated for the node. : in, It is the weight of time-forwarding trust in aggregated trust. It is the weight of spatial forwarding trust in aggregated trust; (3) Trust Reporting and Isolation: Satellite nodes periodically report aggregated trust values ​​to the ground controller. When the aggregated trust value of a node is lower than the preset security threshold, the controller determines that it is a malicious node, immediately issues a flow table instruction, discards all traffic from that node, cuts off the injection of attack traffic at the source, and prevents junk data from crowding out the valuable bandwidth of the power emergency network.

[0068] 2.2 Dynamic Congestion Relief Based on Deep Reinforcement Learning (DCDRL) For attack traffic that has not been fully isolated, network congestion caused by attack remnants, and congestion caused by legitimate burst traffic, this invention constructs an intelligent congestion mitigation model based on the Deep Deterministic Policy Gradient (DDPG) algorithm. This includes the following:

[0069] (1) Modeling of satellite network congestion mitigation problem To quantify the transmission performance of the power emergency network, key indicators are first defined: 1) Receive traffic Received traffic metrics This indicator represents the total power emergency service traffic that satellite node i receives from all nodes in the previous hop within a unit of time. This metric is not only used to monitor in real time whether there is a risk of receiver overload on satellite nodes, thus determining whether traffic optimization or congestion adjustment mechanisms need to be activated, but also to assess the traffic distribution and load balancing level among different nodes in the power emergency network. This provides crucial situational awareness information and decision-making basis for the network control center to formulate global resource allocation strategies.

[0070] 2) Delay In this invention, the end-to-end delay of data packet transmission is defined as consisting of two parts: propagation delay and queuing delay, i.e. .

[0071] Among them, the delay in transmission The calculation formula is: in, Let be the distance between nodes, and c be the speed of light.

[0072] The formula for calculating queuing delay is: in, Where C is the number of bytes in the queue for time slot t, and C is the link bandwidth. This represents the number of queued data packets.

[0073] 3) Sending rate When node i receives a data packet from the previous hop node and continues transmission, the data packet transmission rate of node i is expressed as: .

[0074] Based on this, the power industry congestion mitigation problem is modeled as a joint optimization problem. The congestion mitigation strategy employs a proactive intervention mechanism, which is not triggered only after the buffer is completely saturated, but rather before the buffer occupancy reaches a preset threshold, i.e., based on probability. Packet dropping is implemented. Given that the sender's buffer often reaches its capacity limit before the bottleneck link, this strategy aims to accelerate the transmission and clearing of queued data to improve the overall performance of the power emergency network. Therefore, the joint optimization objective of this strategy is set as: maximizing the node's sending rate while simultaneously minimizing the average queue length and packet dropping rate within the buffer.

[0075] The objective function and constraints are as follows: in, For transmission rate, The buffer queue length, For packet loss probability, To receive traffic, the constraints are as follows: first, the number of data packets queued in the buffer should not exceed the buffer size; second, the maximum sending rate should not exceed the link's maximum capacity; third, the amount of data received by a node should not be negative; and fourth, the probability of data packet drop should be between 0 and 1. M is the buffer size, and C is the link's maximum transmission capacity.

[0076] (2) Model construction based on the DDPG algorithm: The above optimization problem is transformed into a Markov Decision Process (MDP), and solved using a Centralized Training Distributed Execution (CTDE) architecture. The Markov process is defined as a quadruple: Where S represents the state space, A represents the action space, p represents the state transition probability, and r represents the reward value.

[0077] State space: Defines the state space for time slot t. Including satellite node locations ,load Link load Buffer queue length Number of bags in queue Receive traffic Sending capability and the amount of data sent out .

[0078] Action space : Represents the discard decision vector That is, the probability of active packet loss for each satellite node in the current time slot.

[0079] reward function To guide the agent in learning the optimal policy, the reward function is defined as follows: in, For the reward function, For transmission rate, The length of the buffer queue. For packet loss probability, To receive traffic. , , These are the weighting coefficients.

[0080] (3) Algorithm Training and Parameter Update: In the congestion mitigation scenario of the power emergency network, satellite nodes need to dynamically adjust the packet drop probability according to the real-time changing buffer state. Since this drop probability is a continuously changing value between 0 and 1, rather than a discrete switching quantity, traditional reinforcement learning algorithms based on value functions are difficult to directly apply to optimization in such a continuous action space. In view of this, this invention adopts a Deep Deterministic Policy Gradient (DDPG) algorithm model based on the Actor-Critic architecture to achieve efficient learning and decision-making for continuous congestion control strategies.

[0081] The DDPG model constructed in this invention mainly consists of two core logical parts: an Actor network and a Critic network. The Actor network, as the executor of actions, is responsible for mapping the observed network state to specific continuous actions (i.e., drop probability); while the Critic network acts as a "referee," evaluating the merits of the Actor's chosen actions by calculating the Q-value, guiding the Actor to optimize its strategy towards higher rewards.

[0082] To address the slow convergence and instability issues of traditional Actor-Critic algorithms during training, this embodiment constructs a composite structure comprising four neural networks during centralized training on the controller side: firstly, an online Actor network. With the online Critic network One is used for real-time policy generation and value evaluation; the other is the corresponding target Actor network. With the target Critic network This dual-network structure, combined with an experience replay mechanism, effectively breaks the temporal correlation of training data and smooths fluctuations in target values, thereby significantly improving the training stability and convergence efficiency of the model in complex power emergency environments.

[0083] The training process is as follows: Step 1, Exploration and Experience Collection: To balance "exploration" and "exploitation" in the early stages of training and prevent the policy from getting trapped in local optima, Gaussian noise is added to the deterministic actions output by the online Actor network. ,Right now The controller will generate a state transition quadruple from its interaction with the environment. The data is stored in the experience replay pool. During training, the system randomly selects batches of samples from the pool for gradient calculation, thus eliminating temporal correlations between data.

[0084] Step 2, Critic Network Update (Value Evaluation): The training objective of the Critic network is to make the predicted Q-value as close as possible to the target Q-value. The system achieves this by minimizing the loss function. To update the parameters of the online Critic network The loss function is defined as:

[0085] in, It is the training loss function of the Critic network. These are the parameters of the online Critic network. It is the current state. It is the action performed in state s. It is the reward obtained at the current moment. It is a discount factor. , It is the next state reached after performing action 'a'. It is in state The following actions are generated by the target Actor network. It is the set of parameters for the target Critic network.

[0086] The target value is calculated by the target Critic network and the target Actor network, i.e. .

[0087] Step 3, Actor Network Update (Policy Optimization): The optimization objective of the Actor network is to maximize the value score given by the Critic network. Its performance objective function is... It is expressed as expected cumulative reward, parameter The policy parameters are updated using a deterministic policy gradient algorithm, that is, the policy parameters are adjusted along the direction of the Q-value gradient.

[0088] Step 4, Target Network Soft Update: To further ensure the stability of training, the parameters of the target network are not directly copied from the online network, but are iteratively updated using a soft update mechanism. That is, each time, only a small portion of the parameter weights of the online network are synchronized to the target network. in, To update the coefficients, ensure that the target network parameters change smoothly and avoid training divergence caused by drastic parameter fluctuations.

[0089] (4) Distributed execution: After training, the Actor network model is distributed. The data is then distributed to each satellite node. In power emergency communications, satellite node i only needs to rely on its locally observed state. The optimal packet drop probability is calculated in real time. :

[0090] In this embodiment, terminal C (a malicious node) launches a flooding attack on the satellite network. The system first detects the Packet-in request count of terminal C through an anomaly node attack detection mechanism (ETM) based on spatiotemporal characteristics. The surge led to its time trust value The value decreased exponentially, eventually causing the aggregate trust value to fall below the security threshold. The controller immediately identified it as a malicious node and issued an isolation command. Subsequently, to address link congestion caused by attack remnants or concurrent services, the low-Earth orbit satellite used its locally deployed Actor network model to calculate the optimal packet drop probability in real time. Based on this, the satellite proactively discarded some low-priority data, successfully controlling the queuing delay of critical control commands within a certain time, thus ensuring the survivability of power emergency communications.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for congestion control in a power emergency network based on low-Earth orbit broadband satellites, characterized in that, Includes the following steps: Step 1: Monitor the network status of the power emergency network in real time, and determine the network status as uneven traffic distribution or malicious node attack state based on the congestion causes. Step 2: When the traffic distribution is determined to be uneven, execute a traffic scheduling strategy based on trust assessment: calculate link trust values ​​to identify critical links, filter critical flows, and calculate rerouting paths for traffic migration by optimizing the objective function; Step 3: When the state is determined to be a malicious node attack, execute a congestion mitigation strategy based on spatiotemporal features and reinforcement learning: First, evaluate the node trust value based on spatiotemporal features and isolate the malicious node. Then, use a deep reinforcement learning model that is centrally trained and distributed to dynamically adjust the packet loss strategy of the satellite node.

2. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 1, characterized in that, In step 1, the logic for determining the congestion cause is as follows: if no abnormal surge in packet-in requests or a sharp drop in packet forwarding success rate is detected in the network, but the load rate of some links is continuously exceeding the preset threshold and there is obstruction of power service transmission, it is determined to be an uneven traffic distribution state; if the number of packet-in requests per unit time exceeds the safety threshold or the packet forwarding success rate is lower than the normal level, it is determined to be a malicious node attack state.

3. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 2, characterized in that, In step 2, the traffic scheduling strategy based on trust assessment specifically includes: Link selection: Taking into account both link transmission cost and load rate, the trust value of link e is calculated using the following formula. : in, For link path loss, To delay the transmission time, The current power service load carried by the link. Total bandwidth; links with trust values ​​below a preset threshold are marked as critical links; Critical Flow Selection: Calculate the data flow f in the critical link within the time window [ Internal data flow growth rate : in, They represent The number of bytes received by the point-in-time switch; Traffic exceeding a preset percentage is identified as critical traffic; Critical Flow Rerouting: Constructing a Joint Optimization Objective Function Under the premise of satisfying link capacity constraints, calculate the optimal rerouting path for the critical flow: in, To maximize link utilization; Total link load; is the weighting coefficient, with a value of (0,1); Routing policy generation: The controller converts the calculated rerouting paths into OpenFlow flow table entries and sends them to the relevant satellite nodes.

4. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 3, characterized in that, Step 3, the node trust value evaluation based on spatiotemporal features, specifically includes: Trust Records: Utilizing the software-defined network flow table extension feature, the total number of Packet_in requests for each node within a unit of time is recorded. And the number of successfully forwarded packets (Data_S) and the number of failed packets (Data_U); Time-based assessment: Calculating trust scores based on time frequency To detect flood attacks: in, For frequency factors; Spatial Dimension Assessment: Calculating Data Forwarding Trust Values ​​Using Bayesian Methods To detect forwarding attacks: Combined with the trust value for controlling message forwarding And neighbor node recommendations, calculate spatial forwarding trust value ; Aggregate Trust Calculation: Generates an aggregate trust value by combining the above indicators. : 。 5. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 4, characterized in that, In step 3, isolating malicious nodes specifically means that when the aggregate trust value of a node is lower than a preset security threshold, the controller determines that the node is a malicious node and sends a discard command to the associated satellite nodes to block traffic from the malicious node.

6. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 5, characterized in that, In step 3, the method for dynamically adjusting the packet loss strategy of satellite nodes using a deep reinforcement learning model with centralized training and distributed execution is as follows: Establish a joint optimization objective for satellite network congestion mitigation, maximizing node transmission rate and minimizing queuing length and packet loss while satisfying the constraints of buffer capacity M and link bandwidth C: ; in, For transmission rate, The buffer queue length, For packet loss probability, To receive traffic.

7. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 6, characterized in that, The deep reinforcement learning model employs a deep deterministic policy gradient algorithm based on the Actor-Critic architecture, and the training process includes: Define reward function : Centralized training is performed on the SDN controller side, and the Critic network parameters are updated by minimizing the loss function. : Update the Actor network parameters using a deterministic policy gradient. : The target network parameters are synchronized using a soft update mechanism.

8. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 7, characterized in that, The formula for the soft update mechanism is as follows: in, This is for updating coefficients.

9. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 8, characterized in that, The dynamically adjusted packet loss strategy for satellite nodes is specifically implemented through a distributed execution process: the trained Actor network model parameters are distributed to each satellite node; during communication, each satellite node only relies on its locally observed state. The optimal packet drop probability is calculated and output in real time through the local Actor network.

10. The power emergency network congestion control method based on low-orbit broadband satellites according to claim 9, characterized in that, The optimal packet drop probability The calculation formula is as follows: 。

Images