AI-based unmanned combat evaluation method

Abstract

The invention relates to the technical field of combat evaluation, in particular to an AI-based unmanned combat evaluation method, which comprises the following steps of: based on an unmanned combat unit, analyzing a relationship between control signal triggering and attitude change, judging action lag characteristics, analyzing action errors under the influence of wind speed, electromagnetism and illumination, identifying a pairing relationship, and distinguishing abnormal actions. And determining a tactical behavior key path and a grading result. According to the method, layered structure judgment of the target task is realized through the proportion of the time interval to the action amplitude, the sensitivity of the unit action response and the dynamic corresponding relation between the target sorting and the response process, environment disturbance data and action error attribution are integrated, the recognition capability of the non-interference abnormal action is enhanced, and the recognition accuracy of the non-interference abnormal action is improved. Based on full-process distribution and path structure expression, high-value tactical nodes can be distinguished and graded, a quantitative basis is provided for credibility and structural hierarchy of unmanned combat behaviors in a complex scene, and finer and systematic evaluation and judgment are realized.

Description

Technical Field

[0001] The present invention relates to the field of combat assessment technology, and in particular to an AI-based unmanned combat assessment method. Background Art

[0002] The field of combat assessment involves the technical methods for quantitative and qualitative analysis of key elements such as enemy and friendly combat capabilities, tactical effects, equipment performance, and command efficiency during military operations. These methods include data acquisition and processing, battlefield situation analysis, force deployment assessment, tactical execution effectiveness assessment, and comparative assessment of operational plans. These methods are widely used in a variety of areas, including combat training feedback, actual combat effectiveness assessment, decision support, and unmanned system combat capability analysis. Traditional combat assessment involves evaluating the effectiveness of combat operations through the experience and judgment of combat commanders, manual recording and analysis of enemy and friendly situation data, and manual battlefield process deduction using diagrams. These methods typically involve manually comparing flight record images, manually summarizing target strike hits on-site, interpreting path deviations based on video playback, and manually matching execution processes using standard mission scripts to complete combat effectiveness assessments.

[0003] Existing technologies often face problems with incomplete information collection and subjective manual interpretation in the manual-dependent combat assessment process. Data processing is limited to manual analysis, making it difficult to achieve global synchronous comparison of task details and action performance. Field records and empirical interpretations make it difficult to make scientific responses to high-frequency coordinated actions or multi-factor disturbances. Action features are easily missed, target responses are ambiguous, and the attribution of environmental errors is unclear in the combat process, affecting the comprehensiveness and accuracy of the assessment and making it difficult to support hierarchical quantitative judgment of complex dynamic tasks and key nodes of the path. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose an AI-based unmanned combat assessment method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an AI-based unmanned combat assessment method, comprising the following steps:

[0006] S1: Based on the unmanned combat unit, the control signal triggering moment and the flight attitude change node are analyzed, the time interval between the signal and the attitude response is calculated, the attitude change amplitude and interval ratio at each time interval are determined, the action sequence is adjusted, and the hysteresis feature sequence is obtained;

[0007] S2: Based on the hysteresis feature sequence, the target ranking in the task is screened, the order of the target suppression response and the command trigger is compared, the action with the highest ranking and delayed response is determined, the distribution of each link is analyzed, and the priority response offset is obtained;

[0008] S3: Based on the priority response offset, analyze the wind speed change time, count the electromagnetic interference and light brightness change time, calculate the action error frequency corresponding to the disturbance parameter, identify the disturbance and error pairing relationship, determine the coverage situation, and obtain the interference attribution discrimination;

[0009] S4: Based on the interference attribution discrimination, determine the action features with low matching conditions, analyze the abnormal content of the hysteresis disturbance ratio in the action, classify the results as the abnormal action segment, and obtain the non-disturbance hysteresis amplitude.

[0010] The present invention has the following improvements: the hysteresis feature sequence includes action response hysteresis, response stability index, and timing consistency parameter; the priority response offset includes priority target recognition parameter, response offset grading item, and link connection status; the interference attribution discrimination includes environmental interference sensitivity, error attribution weight, and pairing validity feature; the non-disturbance hysteresis amplitude includes non-disturbance hysteresis amplitude, action abnormality distribution coefficient, and segment performance factor.

[0011] The present invention is improved in that the step of obtaining the hysteresis characteristic sequence is specifically as follows:

[0012] S111: Based on the unmanned combat unit, analyze the correspondence between the triggering moment of the control signal and the flight attitude change node, calculate the reaction time between any two sets of signals, and compare the changes in the roll angle, pitch angle, and yaw angle to obtain the attitude response time interval set;

[0013] S112: Based on the posture response time interval set, compare each reaction time segment with the variation range of the corresponding posture parameter, determine the correlation between the signal reaction speed and the movement range, and obtain a response proportional coefficient sequence;

[0014] S113: Based on the response proportional coefficient sequence, the execution order of the actions and the response synchronization are analyzed, the hysteresis of each action is calculated, and a hysteresis feature sequence is obtained.

[0015] The present invention is improved in that the step of obtaining the priority response offset is specifically as follows:

[0016] S211: Based on the hysteresis feature sequence, analyzing the action response hysteresis, response stability and timing consistency, comparing the control signal with the action occurrence sequence, determining the difference between the target ranking and the actual response sequence, screening the target numbers with delayed response sequence, and obtaining the action sequence hysteresis number;

[0017] S212: Based on the action sequence lag numbers, the correspondence between the response sequence and the command issuance sequence of each target in the identification, aiming, and strike phases is calculated, the response segments with key sequence differences are identified, and the sequence distribution of the response actions and command time points of each segment is optimized to obtain a target response distribution offset set;

[0018] S213: Based on the target response distribution offset set, determine the order changes of key nodes in the response process, calculate the difference between the response order of each node and the command order, analyze the process interval between adjacent actions in the task structure, obtain the degree of timing misalignment of each target key response action in the task, screen the rhythm offset, and obtain the priority response offset.

[0019] The present invention is improved in that the step of obtaining the interference attribution discrimination is specifically as follows:

[0020] S311: Based on the priority response offset, analyzing the time matching between wind speed changes and task nodes, determining the impact of wind speed changes on the state of the unmanned combat unit at each stage, and comparing the changing moments of the electromagnetic interference signal and the light illumination signal to obtain a disturbance feature sequence;

[0021] S312: Based on the disturbance feature sequence, calculating the posture change and motion offset performance of the unmanned combat unit corresponding to each type of disturbance event, determining the occurrence frequency of motion errors under different disturbance types, screening the pairing relationship between disturbance and motion error, and obtaining the disturbance error pairing strength value;

[0022] S313: Based on the disturbance error pairing strength, analyze the paired combinations with asynchronous disturbance responses, compare the rates of posture changes in each combination, calculate the degree of difference in the strength of the disturbance attribution, and obtain the disturbance attribution discrimination.

[0023] The present invention is improved in that the step of obtaining the non-disturbance hysteresis amplitude is specifically as follows:

[0024] S411: Based on the interference attribution discrimination, analyze the matching effectiveness and each action, select action features with weak pairing relationships, and compare the changing patterns between the features and the hysteresis disturbance ratio to determine the actions with response delays and obtain abnormal response combination items;

[0025] S412: Based on the abnormal response combination items, the sequence of each action during the interference triggering and the actual response is compared, the structural dislocation during the response process is analyzed, and the action segments with consecutive asynchronous responses are determined. The segment identification method is optimized to obtain the abnormal action segment identification result;

[0026] S413: Based on the abnormal motion segment identification result, the motion sequences that cannot be explained by interference attribution are screened, and their response sequence and motion amplitude distribution under the control signal are analyzed to determine the non-interference hysteresis performance in the sequence and obtain the non-interference hysteresis amplitude.

[0027] The present invention is improved in that the steps further include:

[0028] S5: Based on the non-disturbance hysteresis amplitude, optimize the distribution of the entire action process, determine the performance of each link in the path scoring, analyze the closed state of the high-scoring path in the high-frequency suppression task flow, determine the tactical behavior key path, and obtain the tactical key classification result;

[0029] The tactical key classification results include critical path structure parameters, path closure evaluation items, and mission penetration level factors.

[0030] The present invention is improved in that the steps for obtaining the tactical key classification results are specifically as follows:

[0031] S511: Based on the non-disturbance hysteresis amplitude, analyze the distribution of the action sequence as it changes with the task nodes, compare the action arrangement and the connection between the task nodes in each stage, determine the links where the action distribution is concentrated and the hysteresis performance is frequent, and obtain the action task node distribution density;

[0032] S512: Based on the distribution density of the action task nodes, optimize their performance in each combat path, identify paths where actions appear synergistically and are reasonably distributed, analyze the coordination between the action combination and the task structure, and obtain a path structure matching sequence;

[0033] S513: Based on the path structure matching sequence, the logical coherence of each task path is determined, and whether the starting and ending actions of the path form a closed loop is analyzed. Paths with uninterrupted task coverage, reasonable action sequence, and complete loop are selected to obtain tactical key classification results.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are:

[0035] In the present invention, when processing unmanned combat missions, a direct correlation is established between the action response lag and the amplitude of posture change. The sensitivity of the unit action response is refined through the ratio of time interval and action amplitude. The dynamic correspondence between target sorting and response process realizes the hierarchical structure judgment of the target task. The integrated environmental disturbance data and action error attribution enhance the recognition ability of non-interference abnormal actions. Based on the full-process distribution and path structure performance, high-value tactical nodes can be distinguished and graded, providing a quantitative basis for the credibility and structural hierarchy of unmanned combat behavior in complex scenarios, and realizing more refined and systematic evaluation and judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the main steps of the present invention;

[0037] Figure 2 This is a flow chart for obtaining the hysteresis characteristic sequence in the present invention;

[0038] Figure 3 This is a flow chart for obtaining the priority response offset in the present invention;

[0039] Figure 4 This is a flow chart for obtaining interference attribution discrimination in the present invention;

[0040] Figure 5 This is a flow chart for obtaining the non-disturbance hysteresis amplitude in the present invention;

[0041] Figure 6 This is a flowchart for obtaining tactical key classification results in the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0043] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0044] Example

[0045] See also Figure 1 The present invention provides a technical solution: an AI-based unmanned combat assessment method, comprising the following steps:

[0046] S1: Based on the unmanned combat unit, the triggering moment of the control signal and the corresponding nodes of the flight attitude change are analyzed, the attitude response time interval between each group of signals is calculated, the proportional relationship between the attitude change amplitude and the interval under each group of time intervals is determined, and the action sequence is arranged and adjusted to obtain the hysteresis feature sequence;

[0047] S2: Based on the hysteresis feature sequence, the target ranking in the unmanned combat mission is screened, the target suppression response process and the command trigger time are compared, the actions with higher target ranking and delayed response period are determined, the distribution order of each link in the response structure is analyzed, and the priority response offset is obtained;

[0048] S3: Based on the priority response offset, analyze the wind speed change moments during the combat process, count the electromagnetic interference occurrence periods and light intensity change points, calculate the frequency of motion errors corresponding to each disturbance parameter, identify the pairing relationship between disturbance and error, determine the coverage of the pairing results, and obtain the interference attribution discrimination;

[0049] S4: Based on the interference attribution discrimination, determine the action features with low matching conditions, analyze the abnormal content of the hysteresis disturbance ratio in the action, classify the results as the abnormal action segment, and obtain the non-disturbance hysteresis amplitude;

[0050] S5: Based on the non-disturbance hysteresis amplitude, optimize the distribution of actions in the entire combat process, judge the performance of each link in the path scoring, analyze the closed state of high-scoring paths in the high-frequency unmanned suppression mission flow, determine them as tactical behavior critical paths, and obtain tactical criticality classification results.

[0051] The lag feature sequence includes action response lag, response stability index, and timing consistency parameter; the priority response offset includes priority target identification parameter, response offset grading item, and link connection status; the interference attribution discrimination includes environmental interference sensitivity, error attribution weight, and pairing effectiveness characteristics; the non-disturbance lag amplitude includes non-disturbance lag amplitude, action abnormality distribution coefficient, and segment performance factor; the tactical key grading results include critical path structure parameters, path closure evaluation items, and task continuity level factor.

[0052] In S1, unmanned combat units refer to unmanned equipment platforms with autonomous or remote control capabilities, which can complete tactical tasks independently or collaboratively, such as drones, unmanned vehicles, and unmanned boats; control signals refer to various command signals issued by the system to unmanned combat units, including trigger commands such as navigation, attitude adjustment, and task execution; corresponding nodes refer to key action points or moments associated with specific control signals during flight or movement, such as the actual occurrence time of a turn, take-off, or dive; attitude response time interval refers to the time difference experienced by the combat unit in the actual response of attitude parameters (such as roll, pitch, yaw, etc.) after the control signal is issued; attitude change amplitude refers to the change in attitude parameters (angle or vector) of the unmanned combat unit within a certain period of time, which is used to reflect the intensity of the action response; proportional relationship refers to the ratio between the attitude response time interval and the corresponding attitude change amplitude, which is used to evaluate the sensitivity and efficiency of the action response; arrangement adjustment refers to the optimization and adjustment of the order of action occurrence, so that the analysis sequence conforms to the subsequent evaluation logic, which is convenient for the identification of abnormal behavior or lag phenomena.

[0053] In S2, target ranking refers to the ranking of various targets (such as enemy targets, mission points, areas, etc.) in unmanned combat missions based on tactical priority, threat level or mission importance; target suppression response process refers to the entire process of action execution when the unmanned combat unit suppresses or attacks the target, including identification, aiming, attack and other links; command trigger time refers to the time point when a specific execution instruction is issued to the target, which can be compared and analyzed with the actual time of suppression behavior; delayed response time refers to the situation where the actual action of the target with a high ranking lags significantly behind the ideal order or trigger time during the response or suppression process; the distribution order of each link refers to the actual order of occurrence of each action or step in the task response structure, which is used to discover timing logic anomalies or bottleneck links.

[0054] In S3, each disturbance parameter refers to various environmental disturbance quantities collected in real time during combat, such as wind speed, electromagnetic interference intensity, light brightness, etc., and each item is recorded separately; the frequency of motion errors refers to the number of times in the combat process that the action deviates from the predetermined trajectory, posture or rhythm due to environmental disturbances or other factors; the pairing relationship between disturbance and error refers to the correspondence between the occurrence time of each disturbance parameter and the time of occurrence of the motion error, which is used for attribution judgment; the coverage situation refers to the breadth or proportion of the scope of explanation of the motion error by the environmental disturbance event after pairing analysis, which can assist in identifying the error attribution type.

[0055] In S4, low-level action features refer to action performances with weak correlation and insufficient disturbance attribution in paired attribution analysis, which are usually characterized as abnormal or atypical reactions; abnormal content in the lag-disturbance ratio refers to the part of the action where the ratio of lag features to posture changes is significantly different from the normal range, which is used to identify potential abnormal nodes; abnormal action segments refer to continuous intervals or action sets in the action sequence that are attributed to abnormal responses and are not explained by the main disturbance.

[0056] In S5, the distribution of actions in the entire combat process refers to the positional distribution of all analyzed actions in the entire task process, time sequence or task nodes, which is used for overall feature evaluation; the performance of each link refers to the specific response effect, execution smoothness, rhythm and coordination of the path or action at each stage of the combat process; a high-scoring path refers to a path sequence that performs well in comprehensive analysis, responds to key nodes in a timely manner and has a reasonable structure, and usually has a high score or weight; a closed state refers to a high-scoring path that can form an effective cycle or closed loop in task logic and spatial structure, representing efficient coordination and target coverage; a tactical behavior key path refers to a set of actions or paths that have been screened through evaluation and have a decisive impact on the success or failure of the task and combat effectiveness.

[0057] See also Figure 2 , the specific steps for obtaining the lagged characteristic sequence are:

[0058] S111: Based on the unmanned combat unit, analyze the correspondence between the triggering moment of the control signal and the flight attitude change node, calculate the reaction time between any two sets of signals, and compare the changes in the roll angle, pitch angle, and yaw angle to obtain the attitude response time interval set;

[0059] To extract the issuing time of each control signal during the task execution, it is necessary to call the instruction trigger record table in the control system, extract the timestamp of the issuing time of each control instruction, and then sort them in chronological order. The sorted timestamps are bound to the corresponding control command types, such as navigation adjustment, attitude correction, etc., and then the aircraft state change data recorded by the flight attitude monitoring module are called. In this data, the node time closest to the issuing time of the control signal and the first sudden change of the roll angle, pitch angle or yaw angle is identified, and this moment is used as the actual response time point of the corresponding action. Then, the time interval difference between the issuing time of each control signal and the corresponding response moment is extracted to obtain the response time value of each control signal. Then, all control signals in the task execution process are paired in pairs, and the difference between their response times is calculated one by one to form a set of reaction time difference sequences. To ensure accurate classification of the action features of the response behavior, the reaction time for the three attitude change actions of roll, pitch and yaw is further extracted from the above time difference sequence, and classified into roll response time set, pitch response time set and yaw response time set according to the direction of angle change. For example, in an unmanned combat test, the first control command is issued to perform a left turn, and its issuance time is 0.50 seconds. The sensor identifies that the actual time of the left turn start action is 0.65 seconds, so the action response time is identified as 0.15 seconds; then the second command is issued for a rapid dive, and its command time is 1.20 seconds. The action starts at 1.35 seconds, and the identification response time is 0.15 seconds; the response behavior of the two commands is subtracted, and the reaction time interval is zero, indicating that the two responses are consistent. This processing method is repeated to finally organize a complete attitude response time set.

[0060] S112: Based on the posture response time interval set, compare each reaction time segment with the change range of the corresponding posture parameter, determine the correlation between the signal reaction speed and the movement range, and obtain a response proportional coefficient sequence;

[0061] Each group of response time is associated with the posture parameter changes that occur in the same time period. The reaction period and posture change data frame are compared through the data synchronization processing module to extract the angle change value in the posture action period. Then, the angle change amplitude value and the corresponding response time value are extracted according to the time segment. The two types of data are bound in sequence, and then grouped according to the pairing of signal response speed and angle change amplitude. Records with short reaction time but large angle change amplitude are marked as fast response high amplitude group, records with long reaction time but small angle change are marked as low response low amplitude group, and the remaining records are classified as middle segment. In order to accurately divide the data group, the judgment basis is set according to the experimental record. If the response time is within 0.1 second, it is considered fast response; if it is between 0.1 seconds and 0.3 seconds, it is considered a medium response; if it is greater than 0.3 seconds, it is considered a slow response. In terms of the amplitude of angle change, if the posture change angle exceeds 15 degrees, it is considered a high amplitude, between 5 and 15 degrees is a medium amplitude, and below 5 degrees is a low amplitude. For example, in an unmanned suppression mission, a response time of 0.09 seconds and an angle change of 17 degrees were recorded. According to the rules, it was judged to be a fast response high amplitude group; another group of response time was 0.35 seconds and an angle change of 3 degrees, which was judged to be a slow response low amplitude group. According to the above judgment method, all combinations of response time and angle change are labeled and arranged on the time series axis, summarized and organized into a response proportional coefficient sequence for subsequent posture stability analysis and control response evaluation.

[0062] S113: Based on the response ratio coefficient sequence, analyze the order of action execution and response synchronization using the formula:

[0063] ;

[0064] Calculate the hysteresis of each action and obtain the hysteresis characteristic sequence, where Indicates action The hysteresis, Indicates action The reaction time, Indicates action The correlation coefficient between the reaction amplitude and reaction time, Indicates action and action The timing interval between Indicates action The trigger moment, Indicates action The change of posture parameters, Indicates the total number of actions.

[0065] Lag refers to the measurement of the overall delay and inconsistency between the actual time of action, change in action amplitude and the ideal sequence of an unmanned combat unit after receiving the control signal. This indicator combines the synchronization of the reaction delay and action amplitude of a single action, as well as the distribution difference in timing with other actions, comprehensively reflecting the response lag and coordinated performance of each action in the entire process, and is used to characterize the consistency of action execution and the timeliness of response.

[0066] The time interval between the control signal and the corresponding action response, in seconds, is the ratio of response amplitude to reaction time, expressed in degrees per second, For action and action The trigger time interval between For action The control trigger time point, For action The attitude angle change amplitude, in degrees, The number of actions involved in the summation is denoted by , taking action 1 as an example, setting:

[0067] , the original unit is seconds, and the normalized value is 0.60;

[0068] , the original unit is degrees per second, and the normalized value is 0.72;

[0069] , in seconds, normalized to 0.40, , after normalization it is 0.40, , unit degree, after normalization it is 0.65;

[0070] , after normalization it is 0.70, , after normalization it is 0.70, , after normalization, it is 0.18;

[0071] , after normalization it is 0.90, , after normalization it is 0.85, , after normalization, it is 0.92;

[0072] Square root of the first product:

[0073] ;

[0074] The second sum:

[0075] ;

[0076] ;

[0077] ;

[0078] Merge result:

[0079] ;

[0080] This result shows that the hysteresis corresponding to action 1 In the preset hysteresis range The formula is within the medium lag level, indicating that the action has a certain deviation in response time and posture adjustment amplitude but does not exceed the control allowable range. It is a response node that requires attention but does not require immediate intervention, and can be used as a basis for optimizing subsequent action sequencing. The formula obtains the single-point lag trend by multiplying the posture response time and the reaction ratio. At the same time, it combines the execution time and posture change ratio between multiple actions to comprehensively reflect the response deviation degree and system coordination consistency of the action in the action sequence.

[0081] See also Figure 3 , the steps for obtaining the priority response offset are as follows:

[0082] S211: Based on the hysteresis feature sequence, analyze the action response hysteresis, response stability and timing consistency, compare the control signal with the action occurrence sequence, determine the difference between the target ranking and the actual response sequence, filter the target numbers with hysteresis in the response sequence, and obtain the action sequence hysteresis number;

[0083] First, retrieve the action response lag record data, calculate the difference between each unmanned combat control command and the actual action occurrence time after it is triggered, and obtain the action response time difference. Then arrange all the differences according to the time axis to form a lag time set. Then evaluate the response stability of each action, extract more than three consecutive execution records in each type of action, calculate the maximum and minimum response time, and use the difference between the two as the response fluctuation value. If the fluctuation value is less than 0.1 second, it is defined as a stable response area, and the fluctuation value between 0.1 second and 0.3 second is defined as a general response area. If it is greater than 0.3 second, it is considered an unstable response area. Then, further extract the sequence of occurrence time points of each group of commands and actual actions, compare the control signal time with the action time one by one, and confirm their order on the time axis. Whether they are consistent, if the execution time point of a certain action appears earlier than its corresponding control command, it is determined to be a reverse response. If the action occurs later than the command but exceeds the average response time difference by more than 0.25 seconds, it is determined to be a delayed response. The above-identified reverse and delayed actions are numbered and marked. Combined with the target number list, the order of their actions is compared according to the initial ranking of the targets, and the numbers of the targets with higher rankings but responses occurring after the subsequent targets are identified. The numbers are extracted as the target numbers corresponding to the delayed actions. In the example, if target numbers 01, 02, and 03 trigger the strike command in sequence, and the actual response order is targets 03, 01, and 02, then target 03 is not delayed, target 01 lags behind 03, and target 02 lags behind 01. Targets 01 and 02 are respectively marked as delayed numbers in the action sequence.

[0084] S212: Based on the action sequence lag number, the correspondence between the response sequence and the command issuance sequence of each target in the identification, aiming, and attack phases is calculated, the response segments with key sequence differences are identified, and the sequence distribution of the response actions and command time points of each segment is optimized to obtain the target response distribution offset set;

[0085] The time series of the response behavior of each lagging target in the three stages of identification, aiming and attack are extracted, and the actual start time is recorded in stages and matched one by one with the time when the command is issued. The response time is compared stage by stage, and the time difference between the actual action of the target before or after the command in each stage is extracted. The sequence error is accumulated for each item in the three stages, and the stage with a cumulative error of more than 0.5 seconds is set as the key response segment. The action in the response segment is marked as the sequence difference key response segment, and the action triggering moment in the marked segment is fine-tuned. The adjustment plan is to adjust the mean of the action response time point to the reference time with the shortest response time among all lagging targets while keeping the command trigger time unchanged. If the reference time is less than 0.2 seconds of the current response time average, it is directly overwritten and updated. If it exceeds, it is linearly shortened to make the response process close to the command time distribution. For example, if the target recognition response time is 1.20 seconds, the command time is 1.00 seconds, and the average response difference is 0.20 seconds, if the optimal response time in the current group is 1.10 seconds, the response time is adjusted to 1.10 seconds. After the optimization is completed, the response sequence is recorded and rearranged according to the target number to form a corrected response distribution sequence. Finally, all optimized and adjusted target response behaviors are remapped on the time axis, and the average position difference of each target response time period relative to the command time is counted to form a target response distribution offset set.

[0086] S213: Based on the target response distribution offset set, determine the order change of key nodes in the response process, calculate the difference between the response order of each node and the command order, and analyze the process interval between adjacent actions in the task structure. The formula is:

[0087] ;

[0088] Obtain the degree of timing misalignment of each target's key response actions in the task, filter the rhythm offset, and obtain the priority response offset ,in, Indicates the total number of key nodes identified, Indicates the The response sequence number of the key nodes in the task structure, Indicates the The command trigger sequence number corresponding to each key node, Indicates the The process interval between a key node and its adjacent nodes, Indicates the The density of response action distribution within the structural segment where the key node is located.

[0089] The priority response offset refers to the overall deviation between the actual response sequence and the command trigger sequence for each key target node during an unmanned combat mission. By quantifying the degree of misalignment in the timing of action responses, it comprehensively reflects the coordination and smoothness of the response rhythm of each target in the task execution link. It not only reflects the lag of individual actions, but also comprehensively considers the sequential misalignment of all key targets in the entire process. It can provide direct quantitative support for subsequent process optimization, anomaly detection, and task node adjustment.

[0090] The response sequence number of the calling node and command sequence number Perform difference calculations and take into account the process intervals of each node and action distribution density As a normalized interference factor, the rhythm offset calculation is performed using the formula, where, 、 、 、 Representing the The response sequence number, command sequence number, process interval and structure segment density of each key node, In order to determine the number of key nodes involved in the analysis, four key nodes were selected for analysis, and their original data are as follows:

[0091] Node 1: 、 、 、 (After normalization, they are: 、 、 、 );

[0092] Node 2: 、 、 、 (After normalization, they are: 、 、 、 );

[0093] Node 3: 、 、 、 (After normalization, they are: 、 、 、 );

[0094] Node 4: 、 、 、 (After normalization, they are: 、 、 、 );

[0095] Substitute the above normalized parameters into the formula for calculation:

[0096] Node 1:

[0097] ;

[0098] Node 2:

[0099] ;

[0100] Node 3:

[0101] ;

[0102] Node 4:

[0103] ;

[0104] Finally, summing up the offsets of the above nodes, we get:

[0105] ;

[0106] This result shows that when the preferential response offset When the calculated result of is 0.399, it is already in the middle section of the preset rhythm deviation judgment interval [0.3, 0.6]. According to the task response sequence coordination grading standard, this interval corresponds to a moderate level of rhythm deviation, which means that there are a certain number of response sequence dislocation phenomena in the current combat process, and the response rhythm of the key target nodes is not completely synchronized with the command trigger sequence. The closer the value is to the upper limit of the interval, the more serious the rhythm deviation is, and the rhythm structure of the key nodes needs to be corrected and the path nodes need to be rearranged; on the contrary, if it is lower than the lower limit of the interval, it means that the response path is relatively stable. Therefore, by By comparing with the set rhythm offset grading interval, it can be further sorted out that the lagging behavior of key nodes is concentrated in the phased task segments.

[0107] See also Figure 4 ,The steps to obtain the interference attribution discrimination are as follows:

[0108] S311: Based on the priority response offset, analyze the time matching between wind speed changes and mission nodes, determine the impact of wind speed changes on the status of the unmanned combat unit at each stage, compare the changing moments of the electromagnetic interference signal and the light illumination signal, and obtain the disturbance feature sequence;

[0109] The timestamp information of all task nodes in the task process is extracted, and the wind speed data of the unmanned combat unit within five seconds before and after the execution of each task node is extracted. The wind speed change values ​​in this section are arranged in chronological order to form a local wind speed change sequence. Then, the time difference between each node and its corresponding wind speed mutation moment is calculated. If a wind speed change occurs within the range of 1 second before and 1 second after the execution of the task node, it is judged to match the node in time. If the change occurs outside this range, it is judged to be mismatched. All matched wind speed changes are bound to the node number, the intensity and amplitude of the wind speed change at each node are identified, and the attitude data of the unmanned combat unit at the corresponding moment of each node are retrieved for comparison to identify whether there is angle drift, attitude disturbance or trajectory deviation in the stability of the unmanned combat unit before and after the wind speed change. If the angle change before and after the wind speed change exceeds 3 degrees, or the trajectory offset distance exceeds 1.5 meters, it is defined as the stage where the state is affected by wind speed. Next, the start and end times of the electromagnetic interference signals recorded during the entire task are analyzed. All disturbance events are sorted in chronological order and their duration is marked. At the same time, the light intensity signal is extracted synchronously, and the moment when the brightness fluctuation value exceeds 20% per unit time is recorded, and the start and end time points of the fluctuation are marked. All disturbance signals are merged according to the time axis, and the disturbance time node set is formed at the moment when each disturbance event occurs. Then, the disturbance events are grouped according to their types to generate three subsequences: wind speed disturbance sequence, electromagnetic interference sequence, and light change sequence. The full process linear interpolation is performed according to the time of event occurrence to unify the time axis and then combine to form a disturbance feature sequence. In this example, if the time of the sixth node of the task is 40 seconds, the wind speed detected before and after increases by 2.6 m / s at 38.5 seconds and decreases by 1.2 m / s at 40.8 seconds, it is recorded as a wind speed disturbance matching node. If the light change rate increases by 25% at the same time, this time point is included in the light disturbance sequence and matched with the interference signal intensity fluctuation from 35 seconds to 45 seconds in the electromagnetic interference period. The three disturbances together constitute the disturbance feature sequence of this section.

[0110] S312: Based on the disturbance feature sequence, calculate the posture change and motion deviation performance of the unmanned combat unit corresponding to each type of disturbance event, determine the frequency of motion errors under different disturbance types, screen the pairing relationship between disturbance and motion error, and obtain the disturbance error pairing strength;

[0111] Extract the unmanned combat unit posture change record data and action execution trajectory data within 5 seconds before and after the corresponding moment of each type of disturbance event, identify the change trend of the unmanned combat unit posture angle, spatial position, speed data, etc. during the start and end period of the disturbance event, extract the three-segment angle change difference of roll, pitch, and yaw angles before, during, and after the disturbance in the wind speed disturbance event, identify the control response delay, abnormal attitude control response, action interruption or loss record in the electromagnetic interference event, extract the number of image recognition misjudgments, image alignment offset changes and other behavioral performances in the illumination change event, correspond one-to-one with each disturbance event and its action response difference amplitude, and record it as a posture change event or action offset event, classify the events according to the disturbance type, and then count the frequency of action errors in each type of event by type, and count the events with a frequency of action errors higher than 20% of the total response times. The disturbance type is marked as a high-frequency error disturbance source, less than 10% is a low-frequency error source, and 10%-20% is a medium-frequency disturbance. In the example, the motion offset event under wind speed disturbance occurred 12 times in the total task execution segment, accounting for 22.6% of the total response time, marked as a high-frequency error disturbance source, and the illumination change event was associated with motion error 5 times, accounting for 9%, marked as a low-frequency error disturbance source. Each motion error was then paired with the corresponding disturbance event according to the absolute value of the time difference. The paired time difference of no more than 1.0 second was considered a valid pairing, and the time difference exceeding 1.0 second was considered an unpaired record. The ratio of the number of valid pairings to the total number of errors was counted according to the disturbance type. The disturbance type with a pairing ratio higher than 0.6 was identified as a high-intensity pairing source, the pairing ratio between 0.3 and 0.6 was medium-intensity, and the pairing ratio below 0.3 was a weak-pairing disturbance type. The corresponding disturbance error pairing intensity was output according to the disturbance type.

[0112] S313: Based on the disturbance error pairing strength, analyze the paired combinations with asynchronous disturbance responses, compare the rates of posture change in each combination, and calculate the degree of difference in the strength of the disturbance attribution using the formula:

[0113] ;

[0114] Obtain interference attribution discrimination ,in, represents the number of perturbation error pair combinations, Indicates the The time difference between the occurrence of the group disturbance event and the occurrence of the action error, Indicates the The frequency of group action errors under corresponding disturbance events, Indicates the The change in attitude disturbance angle of the unmanned combat unit corresponding to the group, It represents the average frequency of all disturbance error pair combination action errors, Indicates the minimum data of the attitude disturbance angle change among all disturbance error pairing combinations.

[0115] The Disturbance Attribution Discrimination metric quantifies the ability to attribute the effects of different types of environmental disturbances (such as wind speed changes, electromagnetic interference, and sudden changes in illumination) on motion errors during unmanned combat missions. It comprehensively considers the pairing of various disturbance events with motion errors in terms of timing, response frequency, and posture changes, reflecting the differences in the causal relationships between disturbances and errors and the ability to distinguish them. Larger values ​​indicate a stronger ability to distinguish between different disturbance sources and motion errors, meaning it is easier to clearly identify the correspondence between specific disturbances and errors. Smaller values ​​indicate a high degree of overlap in the effects of disturbance types on errors or ambiguous attribution, making it difficult to distinguish the primary causes of the errors. This discriminability provides a quantitative basis for subsequent combat anomaly analysis, disturbance tracing, and interference optimization.

[0116] Count the intervals between the disturbance time point and the action error occurrence time in all disturbance error pairs. If the time difference in the combination If the time response interval exceeds the task setting, it is marked as a response asynchronous combination, and then the attitude disturbance angle change of the unmanned combat unit in the corresponding combination is extracted. , and simultaneously record the frequency of action errors For example, for the five pairing combinations numbered from o1 to o5, their original parameters are as follows: (Second), (Second-rate), (degrees), normalized, the parameters after normalization are:

[0117] ;

[0118] ;

[0119] ;

[0120] Calculation is done using the formula:

[0121] in, ;

[0122] ;

[0123] ;

[0124] The numerator part is calculated as follows:

[0125] Item 1:

[0126] ;

[0127] Item 2:

[0128] ;

[0129] Item 3:

[0130] ;

[0131] Item 4:

[0132] ;

[0133] Item 5:

[0134] ;

[0135] Adding the above results together, the sum of the numerators is:

[0136] ;

[0137] Then calculate the denominator:

[0138] ;

[0139] Substituting into the calculation, we get:

[0140] ;

[0141] The results show that the interference attribution discrimination calculated by the formula , in the preset judgment interval, it belongs to the "sufficient discrimination ability" level. According to the defined attribution differentiation judgment criteria, the interval is set as follows:

[0142] : Weak attribution ability and insufficient correlation between disturbance-error matching;

[0143] : The attribution ability is moderate and the matching logic is identifiable;

[0144] : Strong attribution ability, clear and concentrated disturbance impact;

[0145] Therefore, the current discrimination result is near the upper limit of the moderate interval, indicating that there is a relatively stable response mapping relationship between different types of disturbance events and the motion errors of unmanned combat units, and the interference attribution structure has recognition accuracy. However, some disturbance types still show fuzzy boundaries in the matching process. The numerical value represents the discrimination strength of the attribution structure, which can be used as an important measurement factor for motion anomaly identification and attribution structure screening in subsequent processing. After further organization, it can be used to define disturbance traceability confidence marks or construct anomaly classification labels to support tactical key path adjustments or control strategy reconstruction.

[0146] See also Figure 5, the specific steps for obtaining the non-disturbance hysteresis amplitude are:

[0147] S411: Based on the interference attribution discrimination, analyze the matching effectiveness and each action, screen the action features with weak pairing relationships, and compare the changing patterns between the features and the hysteresis disturbance ratio to determine the actions with response delays and obtain abnormal response combination items;

[0148] Extract the disturbance error pairing strength value corresponding to each action record, and extract the action identifier and the number of the paired disturbance event, and count the ratio of the number of successful pairings to the total number of pairing attempts. If the pairing success rate is lower than 0.3, mark the action record as a weak pairing item. Extract three types of data indicators, namely response time, posture change, and action offset, from all the weak pairing action features and form an action response parameter set. Then extract the hysteresis disturbance ratio record value during the interference period, that is, the ratio sequence of the delayed response time of each action when the disturbance event occurs and the posture offset value. According to the one-to-one correspondence of the action identifier, compare and calculate the action response parameter set with the hysteresis disturbance ratio. For each action record, if If the hysteresis disturbance ratio differs from the average ratio of other actions in the paired weak items by more than 0.2, it is determined to be a deviation from the sample, forming a characteristic change difference set. Then, the actions with values ​​greater than the preset error limit in the difference set are marked as delayed abnormal candidate actions. Subsequently, the recognition order of each candidate action is extracted, and they are sorted according to the action trigger time. The records that appear in the early stage of the task but have a significantly delayed response time are separately classified to form a head-to-tail misalignment combination. In the example, if action A has a preset trigger order of 3rd, but its response record time appears at the 8th second of the task and the first two actions are triggered at the 2nd and 4th seconds respectively, then action A is classified into the abnormal response combination item.

[0149] S412: Based on the abnormal response combination items, the sequence of each action during the interference triggering and the actual response is compared, the structural misalignment during the response process is analyzed, and the action segments with consecutive asynchronous responses are determined. The segment identification method is optimized to obtain the abnormal action segment identification result;

[0150] Obtain the time record of each action in each combination item when the disturbance is triggered and the time point when the response actually occurs, and judge the two time values ​​in sequence. If the response time is more than 0.5 seconds later than the trigger time, it is counted as a response misalignment record. All misalignment record actions are arranged according to the time axis and grouped by continuity. If three or more actions are misaligned continuously and the time interval between adjacent responses is less than 1.2 seconds, it is classified as a continuous response asynchronous segment. Such segments are marked as preliminary abnormal segment candidates. Then calculate the functional node position of the internal action of each candidate item in the task execution process to determine whether there is duplication If there are more than three repeated types in the segment and the timing is disordered, it is considered that the structure is seriously misaligned and the segment is classified into a high misalignment segment. Then the response records in the high misalignment segment are refitted according to the average time interval, and the difference between the standard interval average and the original interval average is compared. If the difference exceeds 0.4 seconds, it is considered that the response time distribution is unbalanced. Its feedback is used as the basis for optimizing the discrimination rule. After the update, all candidates are re-traversed for correction and discrimination. The time segment of the action set judged to be abnormal is output, and its task number and action identifier are recorded as the action abnormal segment recognition result.

[0151] S413: Based on the abnormal motion segment identification results, the motion sequences that cannot be explained by interference attribution are screened, and their response sequence and motion amplitude distribution under the control signal are analyzed to determine the non-interference hysteresis performance in the sequence and obtain the non-interference hysteresis amplitude;

[0152] Extract the action sequence identifiers recorded in all sections, and call the corresponding interference attribution matching records. Count whether each record has a paired disturbance event number. If the action does not find a valid match in all disturbance sequences, or its matching strength is lower than 0.2, it is considered to have failed the interference attribution action, and its action sequence is separately sorted as an unexplained sequence. For each action in the sequence, extract the time when the control signal is issued and the actual response time of the action, sort and classify them according to the time difference, and then extract the change amplitude of the posture parameters in the response period by angle, and judge the corresponding relationship between the angle change and the response time difference. If the response time is between 0.2 If the response time is between 0.1 and 0.4 seconds but the angle change exceeds 15 degrees, it is judged as an abnormal delayed behavior. If the response time is less than 0.1 second but the angle change is less than 2 degrees, it is judged as an atypical reaction. All action records whose relationship between response time and angle amplitude deviates from the average sequence trend are classified into the delayed non-disturbance group. Then, the delay amplitude of each non-disturbance action is counted and grouped according to the segment number. In this example, if the response time of action B is 0.35 seconds, the angle change is 17 degrees, and its segment does not match any interference items, then this record is counted into the non-disturbance delayed action set. The response amplitude values ​​corresponding to all such records are extracted to form the non-disturbance hysteresis amplitude.

[0153] See also Figure 6 ,The specific steps for obtaining tactical key classification results are:

[0154] S511: Based on the non-perturbation hysteresis amplitude, analyze the distribution of the action sequence with the change of task nodes, compare the action arrangement and the connection between the task nodes in each stage, determine the links where the action distribution is concentrated and the hysteresis performance is frequent, and obtain the distribution density of the action task nodes;

[0155] Bind each hysteresis amplitude record with its corresponding action number and task node number, extract the appearance position of all action sequences on the task timeline, and divide the time period into task nodes. All action records are respectively classified into corresponding node segments for statistics. Then, the number of hysteresis actions and their average hysteresis amplitude values ​​under each task node are counted, and the distribution of all action nodes is extracted. The number of actions in each task node time period is compared. If the number of actions in a node segment is greater than 1.5 times the average action density of the whole process, it is judged to be an action distribution concentrated segment. Combined with the hysteresis amplitude value in the node, the distribution of all action nodes is analyzed. If the average lag amplitude is greater than 1.3 times the average value of the entire process, it is marked as a frequent lag segment. If both conditions are met at the same time, it is determined to be a high-density lag segment. For example, if the task process contains a total of 10 nodes, each node has an average of 8 action records, if the fourth node records 14 actions and the average lag action amplitude is 0.35 seconds, while the overall average is 0.26 seconds, then the fourth node is judged to be an action-intensive and high-frequency lag node. Repeat this judgment process, re-aggregate all action sequences by node number, and finally summarize them into an action task node distribution density set.

[0156] S512: Based on the distribution density of action task nodes, optimize their performance in each combat path, identify paths where actions appear synergistically and are reasonably distributed, analyze the coordination between action combinations and task structures, and obtain path structure matching sequences;

[0157] Obtain the frequency of occurrence of each node in each combat path and record its position in the path. Identify path segments where action numbers appear continuously and are distributed in the same logical path. Determine whether there is a coordinated action in the path segment, that is, multiple actions occur together with different control types in the same time period. If there are three or more action combinations with an action trigger time difference of less than 0.2 seconds, it is determined to be a coordinated action segment. The action node numbers in such segments are mapped to combat path identifiers. A mapping table of paths and action densities is established. Then, compare whether the action nodes in each path are evenly distributed. If the difference in the number of actions in two consecutive nodes in the path is less than 2 and the lag amplitude does not change by more than 0.1 seconds, the distribution is considered reasonable. , mark the path as a reasonable path, and then match the action combination type in the reasonable path with the target requirements set in the task structure to determine whether there is a structural consistency relationship. If the three functional modules of target identification, positioning, and attack appear in the action combination and the order corresponds, it is identified as a structural matching path. For example, in path P1, actions A, B, and C appear in nodes 3, 4, and 5 respectively, and the corresponding functional modules of the three actions are identification, positioning, and attack in order. At the same time, the number of actions per node is 5, 6, and 5, and the lag amplitude is 0.28, 0.30, and 0.29 seconds. Then, the path meets the action coordination and structural sequence requirements. Finally, the path segment that meets the above conditions is output as the path structure matching sequence.

[0158] S513: Based on the path structure matching sequence, the logical coherence of each task path is determined. The starting and ending actions of the path are analyzed to see if they form a closed loop. Paths with uninterrupted task coverage, reasonable action sequences, and complete loops are selected to obtain tactical key classification results.

[0159] The action sequence in each path is extracted section by section, and the functional type and logical position of the starting action and the ending action are recorded. It is judged whether the starting action belongs to the pre-action type of the task start, such as start, search or navigation, and whether the ending action belongs to the closed action type, such as return, lock or shutdown. If the functional types at both ends form a task logic closed loop, it is determined to be a start-end closed path. Then, the trigger time of each action in the path is compared with the internal event response logic of the path. If the response sequence of adjacent actions meets the sequence relationship and the response time difference is between 0.1 seconds and 0.8 seconds, the sequence is considered reasonable. Then, combined with whether there are continuous task nodes in the path, the path is considered to be a closed path. Coverage: If the actions in the entire path cover three or more consecutive task nodes, it is marked as a task-uninterrupted path. For example, if actions D, E, F, and G appear in path P2 in sequence, located at nodes 2, 3, 4, and 5 respectively, and the time is 10.2 seconds, 10.9 seconds, 11.4 seconds, and 11.8 seconds respectively, and their corresponding functions are search, identification, positioning, and attack, and the termination action is attack and no subsequent commands are extended, then the path is identified as a closed path. Combined with its response sequence and the action coverage node situation, the paths that meet the logical coherence, closed action loop, and continuous node conditions are finally summarized and output as the tactical key classification results.

[0160] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An AI-based unmanned combat assessment method, characterized in that: The following steps are involved: S1: Based on the unmanned combat unit, the control signal triggering moment and the flight attitude change node are analyzed, the time interval between the signal and the attitude response is calculated, the attitude change amplitude and interval ratio at each time interval are determined, the action sequence is adjusted, and the hysteresis feature sequence is obtained; S2: Based on the hysteresis feature sequence, the target ranking in the task is screened, the order of the target suppression response and the command trigger is compared, the action with the highest ranking and delayed response is determined, the distribution of each link is analyzed, and the priority response offset is obtained; S3: Based on the priority response offset, analyze the wind speed change time, count the electromagnetic interference and light brightness change time, calculate the action error frequency corresponding to the disturbance parameter, identify the disturbance and error pairing relationship, determine the coverage situation, and obtain the interference attribution discrimination; S4: Based on the interference attribution discrimination, determine the action features with low matching conditions, analyze the abnormal content of the hysteresis disturbance ratio in the action, classify the results as the abnormal action segment, and obtain the non-disturbance hysteresis amplitude.

2. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The hysteresis feature sequence includes action response hysteresis, response stability index, and timing consistency parameter; the priority response offset includes priority target recognition parameter, response offset classification item, and link connection status; the interference attribution discrimination includes environmental interference sensitivity, error attribution weight, and pairing validity feature; the non-disturbance hysteresis amplitude includes non-disturbance hysteresis amplitude, action abnormality distribution coefficient, and segment performance factor.

3. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The steps for obtaining the hysteresis characteristic sequence are specifically as follows: S111: Based on the unmanned combat unit, analyze the correspondence between the triggering moment of the control signal and the flight attitude change node, calculate the reaction time between any two sets of signals, and compare the changes in the roll angle, pitch angle, and yaw angle to obtain the attitude response time interval set; S112: Based on the posture response time interval set, compare each reaction time segment with the variation range of the corresponding posture parameter, determine the correlation between the signal reaction speed and the movement range, and obtain a response proportional coefficient sequence; S113: Based on the response proportional coefficient sequence, the execution order of the actions and the response synchronization are analyzed, the hysteresis of each action is calculated, and a hysteresis feature sequence is obtained.

4. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The steps for obtaining the priority response offset are specifically as follows: S211: Based on the hysteresis feature sequence, analyzing the action response hysteresis, response stability and timing consistency, comparing the control signal with the action occurrence sequence, determining the difference between the target ranking and the actual response sequence, screening the target numbers with delayed response sequence, and obtaining the action sequence hysteresis number; S212: Based on the action sequence lag numbers, the correspondence between the response sequence and the command issuance sequence of each target in the identification, aiming, and strike phases is calculated, the response segments with key sequence differences are identified, and the sequence distribution of the response actions and command time points of each segment is optimized to obtain a target response distribution offset set; S213: Based on the target response distribution offset set, determine the order changes of key nodes in the response process, calculate the difference between the response order of each node and the command order, analyze the process interval between adjacent actions in the task structure, obtain the degree of timing misalignment of each target key response action in the task, screen the rhythm offset, and obtain the priority response offset.

5. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The steps for obtaining the interference attribution discrimination are specifically as follows: S311: Based on the priority response offset, analyzing the time matching between wind speed changes and task nodes, determining the impact of wind speed changes on the state of the unmanned combat unit at each stage, and comparing the changing moments of the electromagnetic interference signal and the light illumination signal to obtain a disturbance feature sequence; S312: Based on the disturbance feature sequence, calculating the posture change and motion offset performance of the unmanned combat unit corresponding to each type of disturbance event, determining the occurrence frequency of motion errors under different disturbance types, screening the pairing relationship between disturbance and motion error, and obtaining the disturbance error pairing strength value; S313: Based on the disturbance error pairing strength, analyze the paired combinations with asynchronous disturbance responses, compare the rates of posture changes in each combination, calculate the degree of difference in the strength of the disturbance attribution, and obtain the disturbance attribution discrimination.

6. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The steps for obtaining the non-disturbance hysteresis amplitude are specifically as follows: S411: Based on the interference attribution discrimination, analyze the matching effectiveness and each action, select action features with weak pairing relationships, and compare the changing patterns between the features and the hysteresis disturbance ratio to determine the actions with response delays and obtain abnormal response combination items; S412: Based on the abnormal response combination items, the sequence of each action during the interference triggering and the actual response is compared, the structural dislocation during the response process is analyzed, and the action segments with consecutive asynchronous responses are determined. The segment identification method is optimized to obtain the abnormal action segment identification result; S413: Based on the abnormal motion segment identification result, the motion sequences that cannot be explained by interference attribution are screened, and their response sequence and motion amplitude distribution under the control signal are analyzed to determine the non-interference hysteresis performance in the sequence and obtain the non-interference hysteresis amplitude.

7. The AI-based unmanned combat assessment method according to claim 1, characterized in that: The steps also include: S5: Based on the non-disturbance hysteresis amplitude, optimize the distribution of the entire action process, determine the performance of each link in the path scoring, analyze the closed state of the high-scoring path in the high-frequency suppression task flow, determine the tactical behavior key path, and obtain the tactical key classification result; The tactical key classification results include critical path structure parameters, path closure evaluation items, and mission penetration level factors.

8. The AI-based unmanned combat assessment method according to claim 7, characterized in that: The steps for obtaining the tactical key classification results are as follows: S511: Based on the non-disturbance hysteresis amplitude, analyze the distribution of the action sequence as it changes with the task nodes, compare the action arrangement and the connection between the task nodes in each stage, determine the links where the action distribution is concentrated and the hysteresis performance is frequent, and obtain the action task node distribution density; S512: Based on the distribution density of the action task nodes, optimize their performance in each combat path, identify paths where actions appear synergistically and are reasonably distributed, analyze the coordination between the action combination and the task structure, and obtain a path structure matching sequence; S513: Based on the path structure matching sequence, the logical coherence of each task path is determined, and whether the starting and ending actions of the path form a closed loop is analyzed. Paths with uninterrupted task coverage, reasonable action sequence, and complete loop are selected to obtain tactical key classification results.

Images