Counterfactual reasoning based adversarial simulation experiment result root cause analysis method and system

Abstract

This invention relates to the field of adversarial simulation analysis, and discloses a method and system for root cause analysis of adversarial simulation experiment results based on counterfactual reasoning, in order to identify the causes of simulation results and provide experimental settings for improving the results. The method includes: collecting factor data, state data, response data, and log data from the adversarial simulation experiment; calculating the counterfactual sufficiency based on the Siamese network of the event causal graph; calculating the counterfactual necessity based on the Siamese network of the event causal graph; identifying the causal events that lead to the adversarial simulation results based on the counterfactual sufficiency and counterfactual necessity; constructing a counterfactual structural causal model of the causal events and the factors and states influencing them; and using random search-adaptive particle swarm optimization to find the optimal factor and state settings that improve the simulation results.

Description

Technical Field

[0001] This invention relates to the field of adversarial simulation analysis, and in particular to a method and system for root cause analysis of adversarial simulation experiment results based on counterfactual reasoning. Background Technology

[0002] Using adversarial simulation to describe combat elements and processes is the most effective way to study the complexity of war under dynamic conditions during peacetime. Adversarial simulation constructs a virtual battlefield environment that includes the weapons and equipment of opposing parties, simulating the information, firepower, mobility, protection, and support functions of combat forces. It achieves full-process, full-element simulation of command, action, and support under different combat styles, plans, times, and conditions, obtaining data on various combat actions and their results. This provides support for research on decision-making issues such as weapon system demonstration, combat concept and tactical research, combat plan verification, battlefield situation and outcome prediction, and troop combat capability analysis.

[0003] Current root cause analysis methods for adversarial simulation results primarily rely on correlation-based approaches. These methods include association analysis, knowledge graphs, event graphs, behavioral modeling, Bayesian networks, and fuzzy cognitive graphs. However, correlation-based methods are prone to misidentification of root causes, thus reducing the accuracy of root cause analysis results and making it difficult to accurately analyze the reasons behind the adversarial simulation results. Summary of the Invention

[0004] The purpose of this invention is to disclose a root cause analysis method and system for adversarial simulation experiment results based on counterfactual reasoning, so as to accurately analyze the causes of adversarial simulation results and provide factors and state variable settings to improve the simulation results.

[0005] To achieve the above objectives, in a first aspect, the present invention discloses a root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning, comprising:

[0006] S1: Develop adversarial simulation experiments for target adversarial simulation scenarios and collect experimental data from the adversarial simulation experiments. The experimental data includes factor data, state variable data, response data, and log data.

[0007] S2: Construct the event causal graph of the adversarial simulation experiment based on the experimental data, calculate the counterfactual sufficiency based on the Siamese network of the event causal graph, and calculate the counterfactual necessity based on the Siamese network of the event causal graph;

[0008] S3: Determine the causal events that lead to the adversarial simulation results based on the sufficiency and necessity of the counterfactual facts;

[0009] S4: Construct a counterfactual causal model of the causal event and the factors and states that influence the occurrence of the event;

[0010] S5: Based on the counterfactual causal model, determine the values ​​of each intervention parameter, determine the optimal intervention parameter and the optimal intervention value, and generate root cause analysis results.

[0011] Secondly, the present invention also discloses a root cause analysis system for adversarial simulation experiment results based on counterfactual reasoning, 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 implement the steps of the above method.

[0012] The present invention has the following beneficial effects:

[0013] This application provides a root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning. It proposes the concepts of counterfactual sufficiency and counterfactual necessity, and identifies the causal events leading to the adversarial simulation experiment results based on these concepts. Subsequently, a probabilistic model is constructed, relating the factors, state variables, and the probability of the event's occurrence to the event's occurrence. Then, a counterfactual causal model is constructed between the factors and state variables of the event's occurrence. Finally, a stochastic search-adaptive particle swarm optimization algorithm is used to find the factor and state variable settings that minimize the probability of the classification model, thereby improving the adversarial simulation experiment results. This method can accurately analyze the causes of adversarial simulation results and provide factor and state variable settings for improving the simulation results.

[0014] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0016] Figure 1 This is a flowchart of a root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning, according to a preferred embodiment of this application.

[0017] Figure 2 This is a schematic diagram of a twin network for calculating the sufficiency of counterfactual facts according to a preferred embodiment of this application.

[0018] Figure 3 This is a schematic diagram of a twin network for calculating the counterfactual necessity of a preferred embodiment of this application. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0021] Please see Figure 1 This application provides a root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning, comprising:

[0022] S1: Develop adversarial simulation experiments for target adversarial simulation scenarios and collect experimental data from the adversarial simulation experiments. The experimental data includes factor data, state variable data, response data, and log data.

[0023] S2: Construct the event causal graph of the adversarial simulation experiment based on the experimental data, calculate the counterfactual sufficiency based on the Siamese network of the event causal graph, and calculate the counterfactual necessity based on the Siamese network of the event causal graph;

[0024] S3: Determine the causal events that lead to the adversarial simulation results based on the sufficiency and necessity of the counterfactual facts;

[0025] S4: Construct a counterfactual causal model of the causal event and the factors and states that influence the occurrence of the event;

[0026] S5: Based on the counterfactual causal model, determine the values ​​of each intervention parameter, determine the optimal intervention parameter and the optimal intervention value, and generate root cause analysis results.

[0027] It should be noted that during the simulation, numerous factor data reflecting information such as the firepower, troop deployment, environmental settings, and outcome of the opposing forces, as well as state data reflecting the combat process, were recorded. Log data recording the occurrence and timing of various events during the simulation, along with response data evaluating the combat process, were also included. These factor data, state data, response data, and log data are collectively referred to as experimental data. Analyzing this experimental data can identify the causes of the simulation results and provide constructive guidance for improving them.

[0028] The aforementioned root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning proposes the concepts of counterfactual sufficiency and counterfactual necessity. Based on these concepts, it identifies the causal events leading to the adversarial simulation experiment results. Subsequently, it constructs a classification model of the factors, state variables, and probability of this event, which influence its occurrence. Next, it constructs a counterfactual structural causal model among the factors and state variables of this event. Finally, it uses a stochastic search-adaptive particle swarm optimization algorithm to find the factor and state variable settings that minimize the probability of the classification model, thereby improving the adversarial simulation experiment results. This method can accurately analyze the causes of adversarial simulation results and provide factor and state variable settings for improving the simulation results.

[0029] Optionally, calculating the sufficiency of counterfactual facts based on the twin network of the event cause-effect graph includes the following steps:

[0030] S21: Obtain the event cause-effect graph for the i-th adversarial simulation based on expert knowledge. Q represents the set of events that occurred, Q represents the number of event types that occurred in this simulation, and the last event... This represents the result of the simulation. This represents the set of edges in the cause-effect graph of the i-th simulation event. Represents the real number field;

[0031] S22: Obtain a cause-and-effect diagram of events based on knowledge. Each event in The factors and state variables that occur;

[0032] S23: Will affect the event The factors and state variables that occur are abstracted as exogenous variables.

[0033] S24: Construct a structural causal model for an event cause-effect graph, which is defined as: a model with a set of node variables (node ​​variables of events). A set of latent variables A directed acyclic graph This is called the causal structure of the model, and its node variables are E. i ∪Ui and the set of functions F = {f1, ..., f} Q}, where f j yes arrive The mapping, E ij The value of the parent variable. Set F forms a sequence from U. i To E i The mapping can be represented as:

[0034]

[0035] When the set of functions F forms from the noise variable U i To node variable E i During mapping, the distribution on the noise variables causes the distribution on the nodal variables, as given by the following equation:

[0036]

[0037] In the formula, := represents assignment, the event node is a Bernoulli variable with only two states, 0 and 1. Any causal event can lead to the occurrence of the result event. Let its parent variable be {Y1, ..., Y}. L}, and its corresponding generating function f j As shown below:

[0038]

[0039] Indicates in Under the values ​​of exogenous variables, the event The probability that it will not happen spontaneously. As shown below:

[0040]

[0041] Indicates that the event Y has only a cause. k In the event of occurrence, The probability that it will not happen. The value is uniformly set to 0.001 (if the causative event of a certain event occurs, then the event itself is highly likely to occur).

[0042] S25: Constructing a causal graph of twin events like Figure 2 As shown, the counterfactual sufficiency of each event is calculated based on the structural causal model of the twin event causal graph and the event causal graph. The definition of counterfactual sufficiency is as follows: For event... For Y k All other causal events {Y1, ..., Y} L}\Y k The probability that the event will still occur even if intervention is made to prevent it from happening is shown in the following formula.

[0043]

[0044] ε represents the event state in a counterfactual situation, and ε represents evidence, excluding the event. The state of other events related to the cause event. do({Y1, ..., Y... L}\Y k =0) In counterfactual events, for each other except Y k All other causal events {Y1, ..., Y} L}\Y k Intervene to prevent it from happening.

[0045] The counterfactual necessity is calculated based on the twin network of the event causal graph, specifically including:

[0046] S26: Based on the twin event causal diagram Cause-and-effect diagram of events The structural causal model calculates the counterfactual necessity of each event, such as... Figure 3 As shown, the definition of counterfactual necessity is as follows: for an event For causal event Y k Intervene to prevent it from happening, the event The probability of it not occurring is shown in the following formula:

[0047]

[0048] ε represents the event state in a counterfactual situation, and ε represents evidence, excluding the event. The state of other events related to the cause event. do(Y) k =0) represents the pair of Y k Intervene to prevent it from happening.

[0049] Optionally, finding the causal events leading to the adversarial simulation results based on the sufficiency and necessity of counterfactual facts includes the following steps:

[0050] For each resulting event, calculate the counterfactual sufficiency and counterfactual necessity for each causal event, and the event with the largest sum is the causal event that caused the resulting event to occur.

[0051] Optionally, constructing a counterfactual causal model of the causal event and the factors and states that influence the occurrence of the event specifically includes the following steps:

[0052] S41: Using a class of support vector machines (OCSVM) to construct factor data X i，jState variable data S i，j With the event The input-output model between X and X, where the input is X. i，j S i，j The output is an event. The probability of occurrence. This model is a probabilistic model, using... Let h represent the probability model, and let h represent the probability model. If h(·) is greater than 0.5, the event will occur; if it is less than 0.5, the event will not occur.

[0053] S42: Construct a structural causal model of the factors and state variables that influence the occurrence of this event. The Gaussian process SCM (CP-SCM) refers to the following model:

[0054]

[0055] This includes the covariance function. This refers to a Gaussian distribution, where d represents the number of all variables, and U... r Exogenous variables, f r Refers to the mapping function, X pa(r) Refers to the set of parent variables for each parameter. Indicates a normal distribution. Let represent variance, r represent the variable, and ~ indicate conformity. For example: continuous X pa(r) The RBF core, in multiple simulations The data was fitted to obtain GP-SCM.

[0056] S43: Construct a counterfactual structural causal model of the factors and state variables influencing the occurrence of this event. The counterfactual structural causal model is built upon the structural causal model constructed in S42 and is defined as follows: Let... It comes from For the observed samples, for r∈[d] where |pa(r)|>0, X r The counterfactual structural causal model makes it possible for a single It is given by the following formula:

[0057]

[0058] in, Let K denote the Gram matrix. and The posterior mean and variance, This represents the parent variable of each parameter in the counterfactual world.

[0059] Set the counterfactual factor and the state variable x CF, to counterfactual x CF Input the OCSVM model to obtain the probability of the event occurring after resetting the factors and states.

[0060] Optionally, S5 specifically includes the following steps:

[0061] S51: Define the best course of action. The best course of action must meet the following three conditions:

[0062] 1. The best intervention needs to be able to successfully prevent the causal event from occurring;

[0063] 2. When selecting intervention parameters, it is necessary to ensure that the selected parameters are easy to intervene in;

[0064] 3. The values ​​of each intervention parameter must be within the normal range of the parameter, and the closer they are to the original value of the parameter, the better.

[0065] For events that have an impact Parameters that occur Let array I = <0, ..., 1, ..., 1> represent whether to intervene in these parameters, where 0 indicates intervention and 1 indicates no intervention; let array do = {do1, ..., do...} K+P The value} represents the difficulty of intervening in these operations; the smaller the value, the simpler and easier it is to intervene in this variable. V = < 0, ..., a i , ..., a K+P > indicates the value of each intervention parameter, parameter a i The value of needs to be within the normal range of parameter i;

[0066] S52: Let the objective function of the optimization algorithm be Tar, which consists of three sub-objective functions: Tar1, Tar2, and Tar3, each corresponding to one of the three conditions for the best preventive measures.

[0067] After intervening in the parameters, the counterfactual structural causal model is used to obtain the counterfactual values ​​of each parameter, assuming they are... Will Input the OCSVM model to obtain the probability of the event occurring. The smaller the better. Tar1 is shown in the following formula:

[0068]

[0069] The selected intervention parameters should be as easy to operate as possible, as shown in the following formula (Tar2):

[0070] Tar2 = min|I T *do| (10);

[0071] The symbol T represents the transpose of the matrix.

[0072] set up This represents the original value of the variable being intervened upon. Let... The intervention value should be as close as possible to the original value of the variable, as shown in the Tar3 formula below:

[0073]

[0074] In summary, the objective function of the optimization algorithm is:

[0075] Tar=w1Tar1+w2Tar2+w3Tar3 (12);

[0076] w1, w2, and w3 represent the weights of each sub-objective function.

[0077] S53: Generate a large number of initial intervention variables I and intervention values ​​I based on the constraints. V , generate I, I V The constraints are as follows:

[0078] The intervention variable I is a binary array with a value of 0 or 1, where 0 represents no intervention and 1 represents intervention.

[0079] Intervention value I V Given a real number array, each parameter has a normal range. Intervention values ​​must be taken within the normal range of the corresponding parameter. The intervention value I... V This is the part that the random search-adaptive particle swarm optimization algorithm ultimately aims to optimize.

[0080] The optimization algorithm based on random search-adaptive particle swarm optimization requires setting up a large initial population, i.e., candidate interventions. Let the initial population size be *size*, and the population include intervention variables and intervention values. Assume that the number of intervention variables *I* randomly generated according to the above rules is *s_size*, that is, assume the set of randomly generated intervention variables is {*I1*, ..., *I2*}. s_size The initial population size, s_size, and the intervention value I. V The variable I was obtained through optimization algorithm, and the possible worlds can be calculated using counterfactual calculation (SCM). The optimal intervention measure I is obtained by optimizing the algorithm based on random search-adaptive particle swarm optimization and determining whether the convergence condition is met. best , These represent the optimal intervention parameters and the optimal intervention values, respectively, to find the best values ​​to avoid the occurrence of the causal event.

[0081] The steps of the above-described root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning are described below with a complete example:

[0082] The following adversarial simulation scenario is designed: After the Blue Team's unmanned vehicle swarm destroys a key Red Team target, it retreats to a mountainous and jungle area. The Red Team quickly dispatches reconnaissance drones and attack unmanned vehicle swarms to coordinate an encirclement and capture operation in the mountainous and jungle area, attempting to destroy or block the Blue Team's unmanned vehicles. This adversarial simulation experiment is repeated 100 times, collecting factor data, state data, response data, and log data for each experiment.

[0083] In the 10th simulation, the Blue Team failed to retreat to a safe zone within the specified time. It is necessary to find the causal events that caused this failure. First, a causal graph of the events in this simulation is constructed based on expert knowledge. The events that occurred include the Red Team's drones detecting the Blue Team's drone swarm. Blue team encounters obstacles War Interaction The Blue Team failed to retreat to a safe area within the stipulated time. and All are events Causes of the occurrence.

[0084] Based on expert knowledge, the factors and state variables that influence the occurrence of each event are identified, thus affecting the event. The factors that cause this event include the red team's reconnaissance ability x1 and the blue team's counter-reconnaissance ability x2; the influencing events... The factors that occurred included the blue square's environmental adaptability x3; the influence The factors affecting E3 include the blue team's cluster reliability (X4) and cluster resilience (X5). The state variables influencing E3 are the blue team's troop strength (S1) and the red team's troop strength (S2). The factor that occurs is the blue team's forward speed x6.

[0085] Will affect the event The factors and state variables that occur are abstracted as exogenous variables. Will affect the event The factors and state variables that occur are abstracted as exogenous variables. Will affect the event The factors and state variables that occur are abstracted as exogenous variables. Will affect the event The factors and state variables that occur are abstracted as exogenous variables.

[0086] structure The structural causal model between them.

[0087] Based on the structural causal model of the event graph and its twin network, calculate respectively The sufficiency of counterfactual facts.

[0088] Based on the structural causal model of the event graph and its twin network, calculate respectively The counterfactual necessity.

[0089] right The counterfactual sufficiency and counterfactual necessity of the four events are added together, and their values ​​are compared. The largest value is found in the interaction of war. Therefore, war interaction is the causal event leading to the simulation results.

[0090] Impact on war interaction The factors that occur include the blue team's cluster reliability (X4) and the blue team's cluster resilience (X5), which have an impact. The state variables involved are the blue team's troop strength S1 and the red team's troop strength S2. Record whether these 100 simulated war interactions occurred, and the corresponding values ​​of X4, X5, S1, and S2. Use an Optical Support Vector Machine (OCSVM) to construct the relationships between X4, X5, S1, and S2. The probability model takes the values ​​of X4, X5, S1, and S2 as inputs and outputs as... The probability of occurrence;

[0091] Construct a structural causal model among X4, X5, S1, and S2;

[0092] Based on the structural causal model among X4, X5, S1, and S2, construct a counterfactual structural causal model among X4, X5, S1, and S2.

[0093] The process of assigning values ​​to X4, X5, S1, and S2 based on the counterfactual structural causal model of X4, X5, S1, and S2 is called counterfactual value assignment. This refers to intervention measures.

[0094] Set the objective function of the optimization algorithm as follows: [The function is to...] Input the OCSVM model, and the output should be as small as possible; the operational difficulty of X4 and X5 interventions is 0.1, and the operational difficulty of S1 and S2 is 0.5; The closer the value is to the original value, the better. The weights of the three sub-objective functions are all 1.

[0095] A large number of initial intervention variables and intervention values ​​are generated. The optimal values ​​of X4, X5, S1, and S2 are obtained using a random search-adaptive particle swarm optimization algorithm. These are the measures to improve the simulation results.

[0096] This application also provides a root cause analysis system for adversarial simulation experiment results based on counterfactual reasoning, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the corresponding methods in the above two embodiments.

[0097] The counterfactual reasoning-based root cause analysis system for adversarial simulation experiment results can implement all the embodiments of the counterfactual reasoning-based root cause analysis method for adversarial simulation experiment results described above, and can achieve the same beneficial effects. Here, it will not be elaborated further.

[0098] In summary, the root cause analysis method and system for adversarial simulation experiment results based on counterfactual reasoning disclosed in the above embodiments of the present invention have at least the following beneficial effects:

[0099] This invention first proposes the concepts of counterfactual sufficiency and counterfactual necessity based on counterfactual reasoning. Based on these concepts, it identifies the causal events leading to the adversarial simulation experiment results. Then, it constructs a classification model of the factors, state variables, and probability of the event influencing its occurrence. Next, it constructs a counterfactual structural causal model among the factors and state variables of the event. Finally, it uses a stochastic search-adaptive particle swarm optimization algorithm to find the factor and state variable settings that minimize the probability of the classification model, thereby improving the adversarial simulation experiment results.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning, characterized in that, include: S1: Develop adversarial simulation experiments for target adversarial simulation scenarios and collect experimental data from the adversarial simulation experiments. The experimental data includes factor data, state variable data, response data, and log data. S2: Construct the event causal graph of the adversarial simulation experiment based on the experimental data, calculate the counterfactual sufficiency based on the Siamese network of the event causal graph, and calculate the counterfactual necessity based on the Siamese network of the event causal graph; S3: Determine the causal events that lead to the adversarial simulation results based on the sufficiency and necessity of the counterfactual facts; S4: Construct a counterfactual causal model of the causal event and the factors and states that influence the occurrence of the event; S5: Based on the aforementioned counterfactual causal model, determine the values ​​of each intervention parameter, identify the optimal intervention parameter and optimal intervention value, and generate root cause analysis results; S4 includes: S41: Constructing factor data using a class of support vector machines, OCSVM State variable data With the event The input-output model between them, the input is The output is an event. The probability of occurrence; this model is a probabilistic model, using... express, Let the probability model be represented by the following: ,if A value greater than 0.5 indicates that the event will occur; a value less than 0.5 indicates that the event will not occur. S42: Construct a structural causal model of the factor data and state variable data that influence the occurrence of this event. The Gaussian process SCM refers to the following model: Official (7) Among them, the covariance function , Gaussian distribution This refers to the total number of all variables. Exogenous variables Refers to the mapping function. Refers to the set of parent variables for each parameter. Indicates a normal distribution. Represents variance. Represents variables, To show obedience; S43: Construct a counterfactual structural causal model of the factor data and state variable data that influence the occurrence of this event. The counterfactual structural causal model is built on the structural causal model constructed in S42 and is defined as follows: set up It comes from The observed samples, for of , The counterfactual structural causal model makes it possible for a single It is given by the following formula: （8）； in, , and Represents the Gram matrix. and Indicates the posterior mean and variance. Represents the parent variable of each parameter in the counterfactual world; Set the counterfactual factor data and state variable data as , will be counterfactual Input the OCSVM model to obtain the probability of the event occurring after resetting the factors and states.

2. The root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning according to claim 1, characterized in that, The calculation of counterfactual sufficiency based on the twin network of the event causal graph includes: S21: Based on expert knowledge, the first... Cause-and-effect graph of the impact of sub-adversarial simulation , Represents a set of events that have occurred. This represents the number of event types that occurred in this simulation. This represents the result of the simulation. Representing the The set of edges in the causal graph of the simulated events. Represents the real number field; S22: Obtain a cause-and-effect diagram of the events based on expert knowledge. Each impact event in The factors and state variables that occur; S23: Will affect the event The factors and state variables that occur are abstracted as exogenous variables. ; S24: Construct a structural causal model of the causal graph of the events, wherein the structural causal model includes a set of node variables. A set of potential noise variables A directed acyclic graph Its node variables are function set ,in yes arrive The mapping, represent The value of the parent variable, the set of functions Formed from arrive The mapping is represented as follows: （1）； When the function set Formation from noise variables To node variables During the mapping, the distribution on the noise variables causes the distribution on the node variables, as shown below: （2）； In the formula, Event nodes are Bernoulli variables, with only two states: 0 and 1. Any causal event can lead to the occurrence of a result event. Let its parent variable be Its corresponding generating function As shown below: （3）； In the formula, Indicates in Under the values ​​of exogenous variables, the event The probability that it will not occur spontaneously. Indicates only the cause of the event. In the event of occurrence, The probability of it not occurring is as follows: （4）； S25: Constructing a causal graph of twin events Based on the structural causal model of the twin event causal graph and the influencing event causal graph, the counterfactual sufficiency of each event is calculated, where counterfactual sufficiency represents: for event , except All other cause events The probability that the event will still occur even if intervention is taken to prevent it from happening. It satisfies the following relationship: （5）； In the formula, Represents the state of events in a counterfactual context. Representative except for events The state of other events related to the causal event, Indicating counterfactual events, except for All other cause events Intervene to prevent it from happening.

3. The root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning according to claim 2, characterized in that, The calculation of counterfactual necessity based on the twin network of the event causal graph includes: S26: Based on the twin event causal diagram Cause-and-effect diagram of events The structural causal model calculates the counterfactual necessity of each event. Counterfactual necessity is expressed as: for event , for causal events Intervene to prevent it from happening, the event The probability of it not happening It satisfies the following relationship: （6）； In the formula, In a counterfactual event, the causal event is indicated. Intervene to prevent it from happening.

4. The root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning according to claim 1, characterized in that, S3 includes: For each causal event corresponding to the result event, the counterfactual sufficiency and counterfactual necessity are calculated, and the cause that results in the result event is the one with the largest sum of counterfactual sufficiency and counterfactual necessity.

5. The method for root cause analysis of adversarial simulation experiment results based on counterfactual reasoning according to claim 1, characterized in that, S5 includes: S51: Define the best course of action; for events affecting [the situation]. Parameters that occur Using arrays This indicates whether to intervene in these parameters, where 0 indicates intervention and 1 indicates no intervention; an array is used. This indicates the difficulty of intervening in these operations; the smaller the value, the simpler and easier it is to intervene in that variable. This indicates the values ​​of each intervention parameter. The value needs to be determined by the parameter. Within the normal range; S52: Let the objective function of the optimization algorithm be... , Depend on , , It consists of three sub-objective functions, each corresponding to one of the three conditions for optimal preventive measures; After intervening in the parameters, the counterfactual structural causal model is used to obtain the counterfactual values ​​of each parameter, assuming they are... ,Will Input the OCSVM model to obtain the probability of the event occurring. As shown in the following formula: （9）； The selected intervention parameters should be as easy to operate as possible. As shown in the following formula: （10）； Among the symbols Represents the transpose of a matrix; set up , representing the original value of the variable being intervened upon, let . , As shown in the following formula: （11）； The objective function of the optimization algorithm is set as follows: （12）； , , This represents the weight of each sub-objective function; S53: Generate a large number of initial intervention variables based on the constraints. Intervention values ,generate , Constraints; The optimization algorithm based on random search-adaptive particle swarm optimization requires setting a large initial population as candidate intervention measures. Let the initial population size be... The population includes the intervention variable and the intervention value. Assume that the intervention variable is randomly generated according to the above rules. The number of That is, assuming the set of randomly generated intervention variables is The size of the initial population And intervention value The variables were optimized using an optimization algorithm, and thus obtained... Calculate possible worlds based on counterfactual SCM The optimal intervention measure is obtained by optimizing the algorithm based on random search-adaptive particle swarm optimization and determining whether the convergence condition is met. , , representing the optimal intervention parameter and the optimal intervention value, respectively, are used to find the best values ​​to avoid the occurrence of the causal event.

6. The root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning according to claim 5, characterized in that, The optimal course of action must meet the following three conditions: The best intervention needs to be able to successfully prevent the causal event from occurring; When selecting intervention parameters, it is necessary to ensure that the selected parameters are easy to intervene in; The values ​​of each intervention parameter must be within the normal range of the parameter and should be as close as possible to the original value of the parameter.

7. The root cause analysis method for adversarial simulation experiment results based on counterfactual reasoning according to claim 5, characterized in that, The constraints include: Intervention variables It is a binary array, with values ​​of 0 or 1, where 0 represents no intervention and 1 represents intervention; Intervention value Given a real number array, each parameter has a normal range. Intervention values ​​are taken within the normal range of the corresponding parameter. This is the part that the random search-adaptive particle swarm optimization algorithm ultimately aims to optimize.

8. A system for root cause analysis of adversarial simulation experiment results based on counterfactual reasoning, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

Images