CatBoost-based low-orbit internet communication system transient fault detection and reinforcement method

Abstract

The application discloses a low-orbit Internet communication system instantaneous fault detection and reinforcement method based on CatBoost, which comprises the following steps: preliminarily establishing a low-orbit Internet communication system instantaneous fault model; performing fault injection experiments on different stages of the communication system to obtain target low-orbit Internet vulnerability nodes and components prone to instantaneous faults, performing feature extraction on the vulnerability nodes prone to instantaneous faults, tracking the fault propagation track when the instantaneous fault occurs, and reconstructing the Internet communication system instantaneous fault model; constructing and training a fault type prediction model based on CatBoost; and predicting the communication program segment prone to instantaneous faults in the target low-orbit Internet communication process by using the model and performing reinforcement processing. Compared with other methods, the application has a higher instantaneous fault detection rate and can effectively reduce the time and space costs generated during detection. The application can improve the reliability of the low-orbit Internet communication system.

Description

Technical Field

[0001] This invention belongs to the field of soft hardening and trusted software technology, and in particular, it is a method for instantaneous fault detection in a low-orbit Internet communication system based on CatBoost. Background Technology

[0002] In recent years, with the rapid development of satellite technology, electronic technology, and new materials technology, numerous domestic and international companies have proposed plans to build internet constellations comprising thousands to tens of thousands of low-Earth orbit (LEO) satellites, such as the US's Starlink and OneWeb, and my country's "Hongyun" and "Hongyan" satellite constellation systems. Most of these satellite internet constellation systems utilize polar orbits with altitudes of 1000–1200 km and orbital angles of 85°–90°. However, LEO satellites inevitably experience the harsh and complex space radiation environment during their operation, leading to varying degrees of degradation or even damage to satellite components. Single-event events, in particular, have a significant impact on LEO satellites in orbit.

[0003] Soft errors are transient fault phenomena in semiconductor circuits, typically caused by high-energy particles acting on semiconductor devices, triggering abnormal transient changes in the logic state of data storage or instruction-related devices. Single-event upsets (SEUs) are a significant factor leading to satellite outages and affecting satellite availability. In recent years, with advancements in semiconductor technology, FPGA gate counts have surged, core voltages have decreased, and devices have become more resistant to total dose effects, but they are also more prone to SEUs, causing a rapid increase in the overall soft error rate (SER).

[0004] The earliest fault prediction methods simulated transient hardware failures and SDC (Susceptible Distress Capability) issues through fault injection. Simple fault injection requires injecting faults into a large number of instructions in the program, potentially hundreds or thousands of times per instruction, necessitating numerous scripts and configuration files, and consuming significant time. Therefore, optimized fault injection methods emerged to reduce the time required. Hari et al. proposed a fault point analysis method called Relyzer, which aims to compress the fault space and verify fault results or equivalent faults, allowing for selective fault injection of a small subset of faults, thus reducing the overhead of the fault injection system. Li et al. proposed a smart fault injection framework, SmartInjector, which can prune faults or find equivalent faults without prior fault injection, effectively compressing the fault space and improving the efficiency of fault injection. Xu et al. proposed the CriticalFault fault injection framework, which removes benign faults from the fault pool through instruction-level vulnerability analysis, compressing the fault space and reducing overhead. While fault injection can yield very accurate SDC vulnerability values, it comes with significant time overhead and tedious injection work.

[0005] Machine learning-based instruction SDC vulnerability prediction techniques analyze and extract features related to instruction SDC vulnerabilities, inject faults into a subset of instructions to obtain their true SDC vulnerability values, and then train a prediction model using appropriate machine learning algorithms to predict the SDC vulnerabilities of the instructions. Yang et al. proposed the ProPVInsiden method, which extracts features from instruction attributes and error propagation characteristics in a program to form a dataset, and uses Classification and Regression Trees (CART) to predict the SDC vulnerabilities of linker instructions in the program. Lu et al. proposed an empirical model, SDCTune, to predict SDC vulnerabilities. This model uses decision regression trees to predict the vulnerabilities of store and compare instructions in a program. Zhang Qianwen et al. extracted dependency and intrinsic features from LLVM intermediate code layer instructions to form a dataset, and used Support Vector Regression (SVR) for prediction, demonstrating good universality.

[0006] In summary, although a considerable amount of research has been conducted both domestically and internationally on software-based error detection technologies for satellite components, and some good results have been achieved, simple fault injection methods are accompanied by huge time and space overhead and tedious injection work. Existing machine learning-based error detection technologies only extract instruction features at the level of inherent features and do not consider the propagation characteristics of fault-causing instructions in the program, so they cannot be well applied to communication processes in low-Earth orbit Internet. Summary of the Invention

[0007] The purpose of this invention is to address the problems existing in the prior art by extracting the inherent characteristics of instructions and the propagation characteristics related to fault propagation in the execution process of a target program in a low-Earth orbit (LEO) internet communication system through fault injection. It proposes a CatBoost-based method for detecting transient faults in LEO internet communication systems. This method eliminates the need for exhaustive fault injection and can automatically identify communication program segments in the communication system prone to transient faults. This information can then guide efficient soft hardening strategies, particularly addressing transient faults caused by radiation environments in LEO internet communication. Applying this invention can improve the reliability of LEO internet communication.

[0008] The technical solution for achieving the objective of this invention is: a method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost, the method comprising the following steps:

[0009] Step 1: Analyze the different execution stages and existing failure cases of the low-Earth orbit (LEO) Internet communication system, and initially establish a transient failure model for the LEO Internet communication system.

[0010] Step 2: Conduct fault injection experiments at different stages of the communication system to identify vulnerable nodes and components of the target low-Earth orbit Internet that are prone to transient faults, and collect and analyze the corresponding fault information.

[0011] Step 3: Extract features from vulnerable nodes that are prone to transient failures at different stages, track the failure propagation trajectory when a transient failure occurs, and analyze the correlation between different working stages of the target LEO Internet after the transient failure occurs.

[0012] Step 4: Based on Steps 2 and 3, reconstruct the instantaneous fault model of the Internet communication system;

[0013] Step 5: Based on the instantaneous fault model of the Internet communication system, construct and train a fault type prediction model based on CatBoost according to the collected fault data and feature information.

[0014] Step 6: Use the CatBoost-based fault prediction model to predict the communication program segments that are prone to transient faults during the target low-orbit Internet communication process, and strengthen the predicted communication program segments that are prone to transient faults to enable them to have the function of autonomous detection of transient faults.

[0015] Furthermore, the construction process of the transient fault model of the low-orbit internet communication system described in step 1 is as follows:

[0016] Step 1.1: Based on the execution flow of the low-Earth orbit (LEO) Internet communication system, the LEO Internet communication system is divided into two stages: target network modeling and communication mechanism modeling. The first stage is network topology analysis, and the second stage is further divided into three parts: information reception, transmission, and data transmission.

[0017] Step 1.2, in the target network modeling stage, construct the fault model of the target network according to the six-tuple shown in equation (1):

[0018] W = {V,E,B,T,d,f,S,E} s} (1)

[0019] In the formula, V = (v1, v2, ..., v n ) represents the set of nodes in the network, v n It is the nth node, where n is the number of nodes in the network; E = (e1, e2, ..., e m ) represents the set of links in the network, e m It is the m-th link, where m is the number of links in the network; B = {b i |1≤i≤m},b i Represents the i-th link e i Maximum remaining bandwidth; T = {t} i |1≤i≤m},t i Represents the i-th link e i The current latency; d represents the amount of data transmitted over the network; f represents the data flow rate during network transmission; S represents the satellite set; E s Represents a set of satellite links;

[0020] Step 1.3, in the communication mechanism modeling stage, analyze the data format, packet loss rate and bit error rate of the low-orbit Internet, and construct the communication stage fault model shown in equation (2):

[0021] G={D,TD,PLR,BER,RC} (2)

[0022] In the formula, D represents the size of the received data packet; TD represents the transmission delay, i.e., the transmission time from the sender to the receiver; PLR represents the packet loss rate during data transmission; BER represents the bit error rate, i.e., the bit error rate that occurs during data transmission; and RC represents the number of times the data packet is retransmitted.

[0023] Step 1.4: Simulate the occurrence of a transient fault during the operation of the communication program, and construct a fault model of the communication program based on the collected system operation information before and after the transient fault occurs;

[0024] Based on the different stages of single-event effects in the communication system and the impact of faults on the output, a preliminary transient fault model C for a low-Earth orbit internet communication system is constructed. fault As shown in equation (3):

[0025] C fault ={F point ,F ins-block ,F pro ,R}|R={error,correct} (3)

[0026] In the formula, F point This indicates the location and corresponding stage of the fault point in the communication program affected by the single-event effect of radiation. F ins This indicates the characteristics of the instruction and its associated program block when a fault occurs in a register, buffer, or memory unit related to the communication system; F pro R represents the propagation characteristics of the fault in the target communication program; R represents the impact of the fault on the current stage when it occurs, and is divided into two categories: error and correct. Error indicates that a transient fault has occurred, and correct indicates that no transient fault has occurred.

[0027] Furthermore, steps 2 and 3 specifically include:

[0028] Step 3.1: Inject faults at different stages of the communication system, analyze the results of the dynamic execution of the communication program at each stage, and extract the characteristics of the fault points as shown in Equation (4):

[0029] F point ={v,S,num,IP s IP t ,P,TD,RC,f i ,f o} (4)

[0030] In the formula, v represents the communication transmission rate; S = max{s1,s2,…,s num} represents the size of the data packet, taking the maximum value of all data packets within the time window; num represents the number of data packets; IP s Indicates the source IP address; IPt represents the destination IP address; P represents the protocol type used for communication; TD represents the communication delay time; RC represents the number of retransmissions of the data packet, used to detect whether the data packet is lost or corrupted; f i Indicates the stage of fault injection, f o Indicates the stage at which the fault occurs;

[0031] Step 3.2: When a single-event effect acts on registers, buffers, and memory units, a transient fault is introduced at the instruction level using a fault injection method. The instruction sequence I of the communication program is then analyzed. ins The features of the j-th instruction contained therein are extracted, and the features of the instruction that causes the transient fault and the program block to which it belongs are extracted as shown in Equation (5):

[0032] F ins-block ={ins j ,ins type ,reg name ,bit i ,val op ,add start ,add end ,ins num} 5)

[0033] In the formula, ins j Ins represents the instruction number in the intermediate code of the communication program segment. type Indicates the command type; reg name Indicates the register name corresponding to the instruction operand, bit i This indicates that the bit flip occurred at the i-th bit in the corresponding memory cell; val op Indicates the operand stored in the data buffer in the instruction; add start Indicates the starting address of the program block, add end Indicates the end address of the program block, ins num Indicates the number of instructions in the program block;

[0034] Step 3.3: Construct the instruction type vector of the communication program fault point as shown in Equation (6).

[0035]

[0036] The communication program contains a set of all static instructions, including 11 instruction types: load, call, add, cmp, leave, mov, pop, store, xor, and getelementptr. Among them, the load and call instructions are input-related instructions; the getelementptr, call, and cmp instructions are output-related instructions; corresponding to the instruction fault point F in equation (5). ins-block Instagram type ;

[0037] Step 3.4: By tracing the instructions that cause transient failures in the target program, a fault propagation feature as shown in equation (7) is constructed:

[0038] F pro={BBL,FL,Ins pre Ins last Ins pre-idx} (7)

[0039] In the formula, BBL represents the size of the basic block containing the instruction that caused the transient fault; FL represents the size of the function containing it; Ins pre Instruction I indicates a transient fault occurring in the basic block. j Previous instruction count; Ins last Instruction I indicates a transient fault occurring in the basic block. j The number of instructions; Ins pre-idx Indicates the instruction number upon which the instruction that caused the transient fault depends.

[0040] Furthermore, the instantaneous fault model of the Internet communication system reconstructed in step 4 is shown in equation (8):

[0041]

[0042] In the formula, F pos The system indicates the possible location of the fault, including the data buffer, control register, and status register. R′ represents the result of a momentary fault at the fault point on the communication system. Based on the impact on the communication system, the fault types are divided into three categories: sdc, set, and crash. Among them, sdc (silent data corruption) indicates that the momentary fault did not cause an explicit fault to the communication system, but the running result was incorrect. set indicates that the momentary fault caused the communication program to fall into a blocked state or enter an infinite loop, and the program could never finish executing and exit. crash indicates that the communication system crashed after being subjected to a single-event effect.

[0043] Furthermore, step 5, based on the instantaneous fault model of the Internet communication system, constructs and trains a fault type prediction model based on CatBoost according to the collected fault data and feature information. This includes two stages: feature representation of the fault model and construction of the prediction model. The specific steps are as follows:

[0044] Step 5.1: Construct a feature vector of the target communication program's fault points based on the fault model; specifically including:

[0045] Step 5.1.1, based on the reconstructed instantaneous fault model C of the Internet communication system fault From the fault point information F point F ins-block and F pro Construct the fault point feature vector V in the target communication program as shown in equation (9). fault :

[0046]

[0047] In the formula, V ins-block This indicates the fault characteristics of instructions and their associated program blocks within the IR instruction layer; V pro This indicates the characteristics that may result from fault propagation;

[0048] Step 5.1.2, traverse the sample dataset; define the fault result generated by the communication program after the fault injection of the i-th sample in the sample dataset as Y. i Y represents the label value of the i-th sample, where i = 1, 2, ..., num, and num is the total number of samples in the dataset. i Represented as equation (10):

[0049]

[0050] Step 5.2: Construct and train a CatBoost-based fault type prediction model according to the model in Equation (8).

[0051] Furthermore, step 5.2 specifically includes:

[0052] Step 5.2.1: Based on the fault point feature vector constructed in Step 5.1, construct the sample dataset D required for training the CatBoost-based fault type prediction model. set As shown in equation (11):

[0053] D set ={(X1,Y1),(X2,Y2),…,(X n ,Y n (11)

[0054] In the formula, X i =(x i1 ,x i2 ,…,x im Let x be the feature vector of the i-th sample category. im Y represents the feature of the m-th sample category, where m is the number of features in the sample; i Let be the label value of the i-th sample; n is the total number of samples; the sample is the instruction.

[0055] Step 5.2.2, for the fault point feature vector V fault Encode the target feature values;

[0056] For the fault point feature vector V fault Based on the fault results, the classification features are encoded using OrderedTarget Encoding according to Equation (12):

[0057]

[0058] In the formula, Enc(X) i =x ik ) represents the feature of the i-th sample category as x. ik The encoding solution function; targe tmean This represents the average value of the target variable for each category value; N(*) represents the number of samples, such as N(X=x ik |Y=1) means that the sample category is x given that the sample label is 1. ik The number of; and N(X=X ik ) indicates that the sample category is x ik The number of samples; count represents the number of samples for that category; global target-mean This represents the average value of the target features across the entire training set; priori is a smoothing term used to avoid overfitting.

[0059] Step 5.2.3, for the fault point feature vector V fault Construct a decision tree based on classification features;

[0060] For the decision model T of the t-th tree t (X), as shown in equation (13):

[0061] T t (X)=θ t *h(X,θ t (13)

[0062] In the formula, h(X,θ) t ) is a decision tree structure, representing the decision tree's decision path for input feature X, θ t These are parameters of the decision tree, representing the threshold for splitting nodes and leaf nodes;

[0063] Decision Model T t The original output Z of (X) i (X) is represented by equation (14):

[0064] Z i (X)=θ t *g t (h(X,θ t (14)

[0065] Among them, g t It is the predicted value of the decision leaf node;

[0066] Step 5.2.4: Predict the type of fault caused by a transient fault at the fault point. The corresponding prediction function is Equation (15):

[0067]

[0068] In the formula, P(Y=(Y) i |X) represents the sample belonging to the category Y given input feature X. i The probability; exp(x) represents the natural exponential function e x , Z represents the original output for all categories j. j The summation is performed after indexation, where p is the total number of all categories.

[0069] Furthermore, the specific process of step 6 includes:

[0070] Step 6.1: Use the CatBoost-based fault prediction model to predict faults in the communication program;

[0071] Step 6.2: Based on the fault prediction results, classify the fault types of instantaneous faults occurring at the communication program fault points, and calculate the probability of instantaneous faults occurring in each instruction;

[0072] Step 6.3: Based on the fault prediction results and the probability of each instruction being prone to momentary faults, obtain the communication program segments that are prone to momentary faults in the communication program, and use the redundancy reinforcement method to reinforce the communication program segments.

[0073] Furthermore, step 6.2 specifically includes:

[0074] Step 6.2.1: Compare the fault conditions predicted by the CatBoost-based fault prediction model with the actual fault types, and classify the fault instances into different fault type categories according to Equation (10);

[0075] Step 6.2.2: Use the function in equation (16) to calculate the probability of each instruction being prone to transient failure:

[0076]

[0077] in, X is the probability that the i-th sample, i.e., the i-th instruction, is prone to a momentary failure. ik It is the k-th feature of the i-th sample, and m represents the number of features in the sample.

[0078] Furthermore, step 6.3 specifically includes:

[0079] Step 6.3.1, based on the prediction function P(Y=T) from step 5.2.4 i |X), calculate the probability of each type of fault when a transient fault occurs in each instruction of the target communication program, as shown in equation (17):

[0080]

[0081] In the formula, Kpoint It is the sum of the probabilities of all instructions in the communication program segment being prone to momentary failures;

[0082] Step 6.3.2, determine K point If the value is greater than the preset threshold, it indicates that the target communication program is a communication program segment that is prone to instantaneous failures.

[0083] Step 6.3.3: For communication program segments that are prone to momentary failures, an instruction redundancy reinforcement method is used for reinforcement.

[0084] Furthermore, step 6 also includes:

[0085] Step 6.4 evaluates the reinforcement results from Step 6.3. The evaluation metrics include accuracy, recall, precision, and F1 score; specifically as follows:

[0086] The accuracy ACC is calculated according to formula (18):

[0087]

[0088] In the formula, TP represents the number of samples where both the predicted and actual results are 1, i.e., the predicted result is sdc or crash, and the actual result is also sdc or crash; TN represents the number of samples where both the predicted and actual results are 0, i.e., both the predicted and actual results are set; FP represents the number of samples where the predicted result is 1 and the actual result is 0, i.e., the predicted result is sdc or crash, but the actual result is set; FN represents the number of samples where the predicted result is 0 and the actual result is 1, i.e., the predicted result is set, but the actual result is sdc or crash.

[0089] Calculate the recall rate according to equation (19):

[0090]

[0091] Calculate the precision according to equation (20):

[0092]

[0093] The F1 score (F1score) is calculated according to equation (21). The F1 score is defined as the harmonic mean of precision and recall, and is an indicator used to measure the accuracy of a binary classification model.

[0094]

[0095] Compared with the prior art, the significant advantages of this invention are:

[0096] (1) This invention constructs and improves the transient fault model of the low-Earth orbit Internet communication system. Based on the transient fault model, relevant features at the instruction level of the communication routing program are extracted. The dataset constructed based on the features is used to train the CatBoost transient fault prediction model, and vulnerability prediction and redundancy reinforcement are performed on the instructions, thereby improving the reliability and availability of the low-Earth orbit Internet communication system.

[0097] (2) This paper uses the CatBoost instantaneous fault prediction model to predict the vulnerability of the communication program segment, and performs fine-grained partial redundancy reinforcement on the entire communication routing program based on the prediction results, effectively reducing the time and space overhead while ensuring the instantaneous fault detection rate.

[0098] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0099] Figure 1 This is a flowchart of the instantaneous fault detection and hardening method for low-orbit internet communication systems based on CatBoost, as described in this invention.

[0100] Figure 2 This is a performance comparison chart of the method of the present invention with other methods. Detailed Implementation

[0101] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0102] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0103] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0104] In one embodiment, addressing the problem of transient faults in low-Earth orbit (LEO) internet communication systems causing the entire communication system to fail due to single-event effects, this invention provides a CatBoost-based method for transient fault detection and hardening of communication systems. By constructing a fault model of the entire communication system and analyzing fault models when single-event effects occur on registers, buffers, and storage units, a CatBoost-based transient fault prediction model for LEO internet communication systems is established. Combined with... Figure 1 Specifically, it includes the following steps:

[0105] Step 1: Analyze different execution stages and existing failure cases of the low-Earth orbit (LEO) internet communication system to initially establish a transient failure model for the LEO internet communication system; specifically including:

[0106] Step 1.1: Based on the execution flow of the low-Earth orbit (LEO) Internet communication system, the LEO Internet communication system is divided into two stages: target network modeling and communication mechanism modeling. The first stage is network topology analysis, and the second stage is further divided into three parts: information reception, transmission, and data transmission.

[0107] Step 1.2, in the target network modeling stage, construct the fault model of the target network according to the six-tuple shown in equation (22):

[0108] W = {V,E,B,T,d,f,S,E} s} (twenty two)

[0109] In the formula, V = (v1, v2, ..., v n ) represents the set of nodes in the network, v n It is the nth node, where n is the number of nodes in the network; E = (e1, e2, ..., e m ) represents the set of links in the network, e m It is the m-th link, where m is the number of links in the network; B = {b i |1≤i≤m},b i Represents the i-th link e i Maximum remaining bandwidth; T = {t} i |1≤i≤m},t i Represents the i-th link e i The current latency; d represents the amount of data transmitted over the network; f represents the data flow rate during network transmission; S represents the satellite set; E s Represents a set of satellite links;

[0110] Step 1.3, in the communication mechanism modeling stage, analyze the data format, packet loss rate and bit error rate of the low-orbit Internet, and construct the communication stage fault model shown in equation (23):

[0111] G = {D, TD, PLR, BER, RC} (23)

[0112] In the formula, D represents the size of the received data packet; TD represents the transmission delay, i.e., the transmission time from the sender to the receiver; PLR represents the packet loss rate during data transmission; BER represents the bit error rate, i.e., the bit error rate that occurs during data transmission; and RC represents the number of times the data packet is retransmitted.

[0113] Step 1.4: Simulate the occurrence of a transient fault during the operation of the communication program, and construct a fault model of the communication program based on the collected system operation information before and after the transient fault occurs.

[0114] Based on the different stages of single-event effects in the communication system and the impact of faults on the output, a preliminary transient fault model C for a low-Earth orbit internet communication system is constructed. fault As shown in equation (24):

[0115] C fault ={F point ,F ins-block ,F pro ,R}|R={error,correct} (24)

[0116] In the formula, F point This indicates the location and corresponding stage of the fault point in the communication program affected by the single-event effect of radiation. F ins This indicates the characteristics of the instruction and its associated program block when a fault occurs in a register, buffer, or memory unit related to the communication system; F pro R represents the propagation characteristics of the fault in the target communication program; R represents the impact of the fault on the current stage when it occurs, and is divided into two categories: error and correct. Error indicates that a transient fault has occurred, and correct indicates that no transient fault has occurred.

[0117] Step 2: Conduct fault injection experiments at different stages of the communication system to identify vulnerable nodes and components of the target low-Earth orbit Internet that are prone to transient faults, and collect and analyze the corresponding fault information.

[0118] Step 3 involves extracting features from vulnerable nodes prone to transient failures at different stages, tracking the failure propagation trajectory when a transient failure occurs, and analyzing the correlation between different operational stages of the target LEO Internet after the transient failure occurs; specifically including:

[0119] Step 3.1: Inject faults at different stages of the communication system, analyze the results of the dynamic execution of the communication program at each stage, and extract the characteristics of the fault points as shown in Equation (25):

[0120] Fpoint ={v,S,num,IP s IP t ,P,TD,RC,f i ,f o} (25)

[0121] In the formula, v represents the communication transmission rate; S = max{s1,s2,…,s num} represents the size of the data packet, taking the maximum value of all data packets within the time window; num represents the number of data packets; IP s Indicates the source IP address; IP t Indicates the destination IP address; P indicates the protocol type used for communication; TD indicates the communication delay time; RC indicates the number of retransmissions of the data packet, used to detect whether the data packet is lost or corrupted; f i Indicates the stage of fault injection, f o Indicates the stage at which the fault occurs;

[0122] Step 3.2: When a single-event effect acts on registers, buffers, and memory units, a transient fault is introduced at the instruction level using a fault injection method. The instruction sequence I of the communication program is then analyzed. ins The features of the j-th instruction contained therein are extracted, and the features of the instruction that causes the transient fault and the program block to which it belongs are extracted as shown in Equation (26):

[0123] F ins-block ={ins j ,ins type ,reg name ,bit i ,val op ,add start ,add end ,ins num} (26)

[0124] In the formula, ins j Ins represents the instruction number in the intermediate code of the communication program segment. type Indicates the command type; reg name Indicates the register name corresponding to the instruction operand, bit i This indicates that the bit flip occurred at the i-th bit in the corresponding memory cell; val op Indicates the operand stored in the data buffer in the instruction; add start Indicates the starting address of the program block, add end Indicates the end address of the program block, ins num Indicates the number of instructions in the program block;

[0125] Step 3.3: Construct the instruction type vector of the communication program fault point as shown in Equation (27).

[0126]

[0127] The communication program contains a set of all static instructions, including 11 instruction types such as load, call, add, cmp, leave, mov, pop, store, xor, and getelementptr. Among them, the load and call instructions are input-related instructions; the getelementptr, call, and cmp instructions are output-related instructions; corresponding to the instruction fault point F in equation (26). ins-block Instagram type ;

[0128] Step 3.4: By tracing the instructions that cause transient failures in the target program, a fault propagation feature as shown in equation (28) is constructed:

[0129] F pro ={BBL,FL,Ins pre Ins last Ins pre-isx} (28)

[0130] In the formula, BBL represents the size of the basic block containing the instruction that caused the transient fault; FL represents the size of the function containing it; Ins pre Instruction I indicates a transient fault occurring in the basic block. j Previous instruction count; Ins last Instruction I indicates a transient fault occurring in the basic block. j The number of instructions; Ins pre-idx Indicates the instruction number upon which the instruction that caused the transient fault depends.

[0131] Step 4: Based on Steps 2 and 3, reconstruct the instantaneous fault model of the Internet communication system;

[0132] The fault model reconstructed after modification based on the characteristic information of the target network fault points is shown in Equation (29):

[0133]

[0134] In the formula, F posThis indicates the possible location of the fault, including the data buffer, control register, and status register. R′ represents the result of a transient fault at the fault point on the communication system. Based on the impact on the communication system, fault types are divided into three categories: sdc (Silent Data Corruption), set (Single Event Transient), and crash. Sdc indicates that the transient fault did not cause an explicit fault in the communication system, but the execution result was incorrect. Set indicates that the transient fault caused the communication program to enter a blocked state or an infinite loop, and the program could never complete and exit. Crash indicates that the communication system collapsed after experiencing a single event.

[0135] Step 5: Based on the instantaneous fault model of the Internet communication system, construct and train a fault type prediction model based on CatBoost according to the collected fault data and feature information; this includes two stages: feature representation of the fault model and construction of the prediction model, with the specific steps as follows:

[0136] Step 5.1: Construct a feature vector of the target communication program's fault points based on the fault model; specifically including:

[0137] Step 5.1.1, based on the reconstructed instantaneous fault model C of the Internet communication system fault From the fault point information F point F ins-block and F pro Construct the fault point feature vector V in the target communication program as shown in equation (30). fault :

[0138]

[0139] In the formula, V ins-block This indicates the fault characteristics of instructions and their associated program blocks within the IR instruction layer; V pro This indicates the characteristics that may result from fault propagation;

[0140] Step 5.1.2, traverse the sample dataset; define the fault result generated by the communication program after the fault injection of the i-th sample in the sample dataset as Y. i Y represents the label value of the i-th sample, where i = 1, 2, ..., num, and num is the total number of samples in the dataset. i Represented as equation (31):

[0141]

[0142] Step 5.2, construct and train a CatBoost-based fault type prediction model according to the model in equation (29), specifically including:

[0143] Specifically, it includes:

[0144] Step 5.2.1: Based on the fault point feature vector constructed in Step 5.1, construct the sample dataset D required for training the CatBoost-based fault type prediction model. set As shown in equation (32):

[0145] D set ={(X1,Y1),(X2,Y2),…,(X n ,Y n )} (32)

[0146] In the formula, X i =(x i1 ,x i2 ,…,x im Let x be the feature vector of the i-th sample category. im Y represents the feature of the m-th sample category, where m is the number of features in the sample; i Let be the label value of the i-th sample; n is the total number of samples; the sample is the instruction.

[0147] Step 5.2.2, for the fault point feature vector V fault Encode the target feature values;

[0148] For the fault point feature vector V fauklt Based on the fault results, the classification features are encoded using OrderedTarget Encoding according to Equation (33):

[0149]

[0150] In the formula, target mean The global variable represents the average value of the target variable for each category value, and the count represents the number of samples for that category value. target-mean This represents the average value of the target features across the entire training set; priori is a smoothing term used to avoid overfitting.

[0151] Step 5.2.3, for the fault point feature vector V fault Construct a decision tree based on classification features;

[0152] For the decision model T of the t-th tree t (X), as shown in equation (34):

[0153] T t (X)=θ t *h(X,θ t (34)

[0154] In the formula, h(X,θ)t ) is a decision tree structure, representing the decision tree's decision path for input feature X, θ t These are parameters of the decision tree, representing the threshold for splitting nodes and leaf nodes;

[0155] Decision Model T t The original output Z of (X) i (X) is represented by equation (35):

[0156] Z i (X)=θ t *g t (h(X,θ t (35)

[0157] Among them, g t It is the predicted value of the decision leaf node;

[0158] Step 5.2.4, predict the type of fault caused by a transient fault at the fault point. The corresponding prediction function is Equation (36):

[0159]

[0160] In the formula, P(Y=Y i |X) represents the sample belonging to the category Y given input feature X. i The probability; exp(x) represents the natural exponential function e x , Z represents the original output for all categories j. j The summation is performed after indexation, where p is the total number of all categories.

[0161] Step 6: Utilize the CatBoost-based fault prediction model to predict communication program segments prone to transient faults during the target low-Earth orbit internet communication process, and then strengthen these predicted segments to enable autonomous transient fault detection. This step specifically includes:

[0162] Step 6.1: Use the CatBoost-based fault prediction model to predict faults in the communication program.

[0163] Step 6.2: Based on the fault prediction results, classify the fault types of instantaneous faults occurring at the communication program fault points, and calculate the probability of instantaneous faults occurring in each instruction;

[0164] Step 6.2.1: Compare the fault conditions predicted by the CatBoost-based fault prediction model with the actual fault types, and classify the fault instances into different fault type categories according to Equation (31);

[0165] Step 6.2.2, use the function of equation (37) to calculate the probability of each instruction being prone to transient failure:

[0166]

[0167] in, X is the probability that the i-th sample, i.e., the i-th instruction, is prone to a momentary failure. ik It is the k-th feature of the i-th sample, and m represents the number of features in the sample.

[0168] Step 6.3: Based on the fault prediction results and the probability of transient faults occurring in each instruction, obtain the communication program segments prone to transient faults in the communication program, and strengthen these communication program segments using redundancy reinforcement methods; specifically including:

[0169] Step 6.3.1, based on the prediction function P(Y=Y) from step 5.2.4 i |X), calculate the probability of each type of fault when a transient fault occurs in each instruction of the target communication program, as shown in equation (38):

[0170]

[0171] In the formula, K point It is the sum of the probabilities of all instructions in the communication program segment being prone to momentary failures;

[0172] Step 6.3.2, determine K point If the value is greater than the preset threshold, it indicates that the target communication program is a communication program segment that is prone to instantaneous failures.

[0173] Step 6.3.3: For communication program segments that are prone to transient failures, an instruction redundancy hardening method is used for hardening.

[0174] Step 6.4 evaluates the reinforcement results from Step 6.3. The evaluation metrics include accuracy, recall, precision, and F1 score; specifically as follows:

[0175] The accuracy ACC is calculated according to equation (39):

[0176]

[0177] In the formula, TP represents the number of samples where both the predicted and actual results are 1, i.e., the predicted result is sdc or crash, and the actual result is also sdc or crash; TN represents the number of samples where both the predicted and actual results are 0, i.e., both the predicted and actual results are set; FP represents the number of samples where the predicted result is 1 and the actual result is 0, i.e., the predicted result is sdc or crash, but the actual result is set; FN represents the number of samples where the predicted result is 0 and the actual result is 1, i.e., the predicted result is set, but the actual result is sdc or crash.

[0178] Calculate the recall rate (REcall) according to equation (40):

[0179]

[0180] Calculate the precision according to equation (41):

[0181]

[0182] The F1 score (F1score) is calculated according to equation (42). The F1 score is defined as the harmonic mean of precision and recall, and is an indicator used to measure the accuracy of a binary classification model.

[0183]

[0184] In one embodiment, a CatBoost-based instantaneous fault detection and hardening system for communication systems is provided, characterized in that the system comprises:

[0185] The first module is used to analyze different execution stages and existing failure cases of the low-orbit internet communication system, and to initially establish a transient failure model of the low-orbit internet communication system.

[0186] The second module is used to conduct fault injection experiments at different stages of the communication system, to obtain the vulnerable nodes and components of the target low-orbit Internet that are prone to instantaneous faults, and to collect and analyze the corresponding fault information.

[0187] The third module is used to extract features of vulnerable nodes that are prone to transient failures at different stages, track the failure propagation trajectory when a transient failure occurs, and analyze the correlation between different working stages of the target low-orbit Internet after the transient failure occurs.

[0188] The fourth module is used to reconstruct the transient fault model of the Internet communication system;

[0189] The fifth module is used to construct and train a CatBoost-based fault type prediction model based on the instantaneous fault model of the Internet communication system and the collected fault data and feature information.

[0190] The sixth module is used to predict communication program segments that are prone to transient faults during the target low-orbit Internet communication process using the CatBoost-based fault prediction model, and to strengthen the predicted communication program segments that are prone to transient faults so that they have the function of autonomous detection of transient faults.

[0191] Specific limitations regarding the CatBoost-based instantaneous fault detection and hardening system for communication systems can be found in the limitations of the CatBoost-based instantaneous fault detection and hardening method described above, and will not be repeated here. Each module in the aforementioned CatBoost-based instantaneous fault detection and hardening system for communication systems can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independent of the processor in the computer device, or stored in software in the memory of the computer device, so that the processor can call and execute the corresponding operations of each module.

[0192] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps:

[0193] Step 1: Analyze the different execution stages and existing failure cases of the low-Earth orbit (LEO) Internet communication system, and initially establish a transient failure model for the LEO Internet communication system.

[0194] Step 2: Conduct fault injection experiments at different stages of the communication system to identify vulnerable nodes and components of the target low-Earth orbit Internet that are prone to transient faults, and collect and analyze the corresponding fault information.

[0195] Step 3: Extract features from vulnerable nodes that are prone to transient failures at different stages, track the failure propagation trajectory when a transient failure occurs, and analyze the correlation between different working stages of the target LEO Internet after the transient failure occurs.

[0196] Step 4: Based on Steps 2 and 3, reconstruct the instantaneous fault model of the Internet communication system;

[0197] Step 5: Based on the instantaneous fault model of the Internet communication system, construct and train a fault type prediction model based on CatBoost according to the collected fault data and feature information.

[0198] Step 6: Use the CatBoost-based fault prediction model to predict the communication program segments that are prone to transient faults during the target low-orbit Internet communication process, and strengthen the predicted communication program segments that are prone to transient faults to enable them to have the function of autonomous detection of transient faults.

[0199] For specific limitations on each step, please refer to the limitations on the instantaneous fault detection and hardening method for CatBoost-based communication systems mentioned above, which will not be repeated here.

[0200] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0201] Step 1: Analyze the different execution stages and existing failure cases of the low-Earth orbit (LEO) Internet communication system, and initially establish a transient failure model for the LEO Internet communication system.

[0202] Step 2: Conduct fault injection experiments at different stages of the communication system to identify vulnerable nodes and components of the target low-Earth orbit Internet that are prone to transient faults, and collect and analyze the corresponding fault information.

[0203] Step 3: Extract features from vulnerable nodes that are prone to transient failures at different stages, track the failure propagation trajectory when a transient failure occurs, and analyze the correlation between different working stages of the target LEO Internet after the transient failure occurs.

[0204] Step 4: Based on Steps 2 and 3, reconstruct the instantaneous fault model of the Internet communication system;

[0205] Step 5: Based on the instantaneous fault model of the Internet communication system, construct and train a fault type prediction model based on CatBoost according to the collected fault data and feature information.

[0206] Step 6: Use the CatBoost-based fault prediction model to predict the communication program segments that are prone to transient faults during the target low-orbit Internet communication process, and strengthen the predicted communication program segments that are prone to transient faults to enable them to have the function of autonomous detection of transient faults.

[0207] For specific limitations on each step, please refer to the limitations on the instantaneous fault detection and hardening method for CatBoost-based communication systems mentioned above, which will not be repeated here.

[0208] The invented method, after training the CatBoost transient fault prediction model, can automatically predict communication program segments in the communication routing program that are prone to transient faults, thereby guiding instruction redundancy hardening and eliminating the need for extensive fault injection. Regarding the prediction results, this paper uses prediction accuracy (ACC), recall, computational precision (Precession), and F1-Score to quantify the performance metrics of the transient fault prediction model, such as... Figure 2As shown, comparing the two prediction models, Random Forest and XGBoost, it can be seen that the CatBoost instantaneous fault prediction model proposed in this invention has better performance.

[0209] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A method for instantaneous fault detection and hardening of a low-Earth orbit Internet communication system based on CatBoost, characterized in that, The method includes the following steps: Step 1: Analyze the different execution stages and existing failure cases of the low-Earth orbit (LEO) Internet communication system, and initially establish a transient failure model for the LEO Internet communication system. Step 2: Conduct fault injection experiments at different stages of the communication system to identify vulnerable nodes and components of the target low-Earth orbit Internet that are prone to transient faults, and collect and analyze the corresponding fault information. Step 3: Extract features from vulnerable nodes that are prone to transient failures at different stages, track the failure propagation trajectory when a transient failure occurs, and analyze the correlation between different working stages of the target LEO Internet after the transient failure occurs. Step 4: Based on Steps 2 and 3, reconstruct the instantaneous fault model of the Internet communication system; Step 5: Based on the instantaneous fault model of the Internet communication system, construct and train a fault type prediction model based on CatBoost according to the collected fault data and feature information. Step 6: Use the CatBoost-based fault prediction model to predict the communication program segments that are prone to transient faults during the target low-orbit Internet communication process, and strengthen the predicted communication program segments that are prone to transient faults to enable them to have the function of autonomous detection of transient faults. Steps 2 and 3 specifically include: Step 3.1: Inject faults at different stages of the communication system, analyze the results of the dynamic execution of the communication program at each stage, and extract the characteristics of the fault points as shown in Equation (1): In the formula, Indicates the rate of communication transmission; This indicates the size of the data packet, taking the maximum value of all data packets within the time window; Indicates the number of data packets; Indicates the source IP address; Indicates the destination IP address; Indicates the type of protocol used for communication; Indicates the communication delay time; This indicates the number of times a data packet has been retransmitted, used to detect whether a data packet has been lost or corrupted. This indicates the stage of fault injection. Indicates the stage at which the fault occurs; Step 3.2: When a single-event effect occurs in registers, buffers, and memory units, a transient fault is introduced at the instruction level using a fault injection method. The instruction sequence of the communication program is then analyzed. The features of the j-th instruction contained therein are extracted, and the features of the instruction that causes the transient fault and its corresponding program block are extracted as shown in Equation (2): In the formula, Instruction number indicating the intermediate code of the communication program segment. Indicates the instruction type; Indicates the register name corresponding to the instruction operand. This indicates that a bit flip occurs in the corresponding memory cell at the 1st... Bit; This indicates the operand stored in the data buffer in the instruction; Indicates the starting address of the program block. Indicates the end address of the program block. Indicates the number of instructions in the program block; Step 3.3: Construct the instruction type vector of the communication program fault point as shown in equation (3). ; The communication program contains a set of all static instructions, including 11 instruction types: load, call, add, cmp, leave, mov, pop, store, xor, and getelementptr. Among them, the load and call instructions are input-related instructions; the getelementptr, call, and cmp instructions are output-related instructions; corresponding to the instruction fault points in equation (3). In ; Step 3.4: By tracking the instructions that cause transient failures in the target program, a fault propagation feature as shown in equation (4) is constructed: In the formula, Indicates the size of the basic block containing the instruction that caused the transient fault; Indicates the size of the function it belongs to; Instructions that indicate a transient fault in a basic block Previous instruction count; Instructions that indicate a transient fault in a basic block The number of instructions; Indicates the instruction number upon which the instruction that caused the transient fault depends.

2. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 1, characterized in that, The construction process of the transient fault model of the low-orbit internet communication system described in step 1 is as follows: Step 1.1: Based on the execution flow of the low-Earth orbit (LEO) Internet communication system, the LEO Internet communication system is divided into two stages: target network modeling and communication mechanism modeling. The first stage is network topology analysis, and the second stage is further divided into three parts: information reception, transmission, and data transmission. Step 1.2, in the target network modeling stage, construct the fault model of the target network according to the six-tuple shown in equation (5): In the formula, Represents the set of nodes in the network. It is the number of nodes in the network; Represents the set of links in the network. It is the number of links in the network; , Indicates the first Link The maximum remaining bandwidth; , Indicates the first Link The current latency; This indicates the amount of data transmitted over the network. This represents the data flow during network transmission. Represents a collection of satellites; Represents a set of satellite links; Step 1.3, in the communication mechanism modeling stage, analyze the data format, packet loss rate and bit error rate of the low-orbit Internet, and construct the communication stage fault model shown in equation (6): In the formula, Indicates the size of the received data packet; This indicates the transmission delay, which is the transmission time from the sender to the receiver; This indicates the packet loss rate during data transmission. This represents the bit error rate, which is the rate of bit errors that occur during data transmission. Indicates the number of times the data packet has been retransmitted; Step 1.4: Simulate the occurrence of a transient fault during the operation of the communication program, and construct a fault model of the communication program based on the collected system operation information before and after the transient fault occurs; Based on the different stages of single-event effects in the communication system and the impact of faults on the output, a preliminary transient fault model for a low-Earth orbit internet communication system is constructed. As shown in equation (7): In the formula, This indicates the location and corresponding stage of the fault point in the communication process affected by the single-event radiation effect. This indicates the characteristics of the instruction and its associated program block when a fault occurs in a register, buffer, or memory unit related to the communication system. This indicates the propagation characteristics of a fault in the target communication program; This indicates the impact of a fault on the current stage when it occurs, and is divided into two categories: error and correct. Error indicates that a momentary fault has occurred, while correct indicates that no momentary fault has occurred.

3. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 2, characterized in that, The instantaneous fault model of the Internet communication system reconstructed in step 4 is shown in equation (8): In the formula, This indicates the possible locations of a fault, including the data buffer, control register, and status register; It is the result of a momentary failure at the fault point on the communication system. Based on the impact on the communication system, the fault type is divided into three categories: SDC, SET, and crash. Among them, sdc silent data corruption means that the transient fault did not cause an explicit fault in the communication system, but the running result was incorrect; set means that the transient fault caused the communication program to fall into a blocked state or enter an infinite loop, and the program could never finish executing and exit; crash means that the communication system crashed after being subjected to a single-event effect.

4. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 3, characterized in that, Step 5 involves constructing and training a CatBoost-based fault type prediction model based on the instantaneous fault model of the Internet communication system, using collected fault data and feature information. This includes two stages: feature representation of the fault model and construction of the prediction model. The specific steps are as follows: Step 5.1: Construct a feature vector of the target communication program's fault points based on the fault model; specifically including: Step 5.1.1, based on the reconstructed instantaneous fault model of the Internet communication system From fault point information , and Construct the fault point feature vector in the target communication program as shown in equation (9). : In the formula, This indicates the fault characteristics of instructions and their associated program blocks within the IR instruction layer; This indicates the characteristics that may result from fault propagation; Step 5.1.2, traverse the sample dataset; define the fault result generated by the communication program after the fault injection of the i-th sample in the sample dataset as... ,in The total number of samples in the dataset. Represented as equation (10): Step 5.2: Construct and train a CatBoost-based fault type prediction model according to the model in Equation (8).

5. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 4, characterized in that, Step 5.2 specifically includes: Step 5.2.1: Based on the fault point feature vector constructed in Step 5.1, construct the sample dataset required for training the CatBoost-based fault type prediction model. As shown in equation (11): In the formula, For the first Feature vectors of sample categories, For the first Features of each sample category The number of features in the sample; For the first The label values ​​of each sample; The total number of samples; the samples are the instructions. Step 5.2.2, analyze the feature vector of the fault point. Encode the target feature values; For the feature vector of the fault point Based on the fault results, the classification features are encoded using Ordered TargetEncoding according to Equation (12): In the formula, Indicates the first The characteristics of each sample category are The encoding and solving function; This represents the average value of the target variable corresponding to each category value; Indicates the number of samples, such as This indicates that the sample category is given by the condition that the sample label is 1. The number; and Indicates the sample category is The number of samples; This indicates the number of samples for that category value; This represents the average value of the target features across the entire training set. It is a smoothing term used to avoid overfitting; Step 5.2.3, for the fault point feature vector Construct a decision tree based on classification features; For the Tree-based decision model As shown in equation (13): In the formula, It is a decision tree structure, representing the decision tree's response to input features. The decision path, These are parameters of the decision tree, representing the threshold for splitting nodes and leaf nodes; Decision Model raw output Represented as equation (14): in, It is the predicted value of the decision leaf node; Step 5.2.4: Predict the type of fault caused by a momentary fault at the fault point. The corresponding prediction function is Equation (15): In the formula, Indicates the given input features of the sample The sample belongs to the category The probability of; Represents the natural exponential function , Indicates all categories raw output The summation is performed after indexation, where p is the total number of all categories.

6. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 5, characterized in that, Step 6 includes the following specific steps: Step 6.1: Use the CatBoost-based fault prediction model to predict faults in the communication program; Step 6.2: Based on the fault prediction results, classify the fault types of instantaneous faults occurring at the communication program fault points, and calculate the probability of instantaneous faults occurring in each instruction; Step 6.3: Based on the fault prediction results and the probability of transient faults occurring in each instruction, obtain the communication program segments that are prone to transient faults in the communication program, and reinforce these communication program segments using a redundancy reinforcement method.

7. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 6, characterized in that, Step 6.2 specifically includes: Step 6.2.1: Compare the fault conditions predicted by the CatBoost-based fault prediction model with the actual fault types, and classify the fault instances into different fault type categories according to Equation (10); Step 6.2.2, use the function of equation (16) to calculate the probability of transient failure for each instruction: in, It is the first It is the first The k-th feature of a sample, where m represents the number of features in the sample.

8. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 7, characterized in that, Step 6.3 specifically includes: Step 6.3.1, based on the prediction function in step 5.2.4 Calculate the probability of various faults when each instruction of the target communication program experiences a transient fault, as shown in equation (17): In the formula, It is the sum of the probabilities of all instructions in the communication program segment being prone to momentary failures; Step 6.3.2, Determine If the value is greater than the preset threshold, it indicates that the target communication program is a communication program segment that is prone to instantaneous failures. Step 6.3.3: For communication program segments that are prone to momentary failures, an instruction redundancy reinforcement method is used for reinforcement.

9. The method for instantaneous fault detection and hardening of a low-orbit internet communication system based on CatBoost according to claim 6, characterized in that, Step 6 also includes: Step 6.4 evaluates the reinforcement results from Step 6.

3. The evaluation metrics include accuracy, recall, precision, and F1 score; specifically as follows: Calculate the accuracy rate according to formula (18) : In the formula, This represents the number of samples where both the predicted and actual results are 1, meaning that the predicted result is sdc or crash, and the actual result is also sdc or crash. This represents the number of samples where both the predicted and actual results are 0, i.e., both the predicted and actual results are set. This represents the number of samples where the predicted result is 1 but the actual result is 0, i.e., the predicted result is sdc or crash, but the actual result is set; This represents the number of samples where the predicted result is 0 but the actual result is 1; that is, the predicted result is set, but the actual result is sdc or crash. Calculate the recall rate according to formula (19) : Calculate the accuracy according to formula (20) : Calculate the F1-fraction according to equation (21). The F1 score, defined as the harmonic mean of precision and recall, is a metric used to measure the accuracy of binary classification models. 。

Images