Intelligent evaluation method and system suitable for analogue simulation training of underwater vehicle

Abstract

The invention relates to the technical field of intelligent evaluation suitable for analogue simulation training of an underwater vehicle, discloses an intelligent evaluation method and system suitable for analogue simulation training of the underwater vehicle, and provides the intelligent evaluation method suitable for analogue simulation training of the underwater vehicle. Aiming at the problems that an existing training evaluation process depends on manual scoring, a quantitative standard is lacked, and abnormal behaviors are difficult to recognize in real time, a full-process intelligent evaluation scheme from multi-source data acquisition, behavior modeling, normative scoring to feedback suggestion and path optimization is provided; high-frequency and synchronous acquisition of operation data and system state information is realized, accurate alignment of heterogeneous data on a time sequence is ensured by adopting a timestamp unified labeling mechanism, and a complete and continuous data support is provided for subsequent behavior modeling.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent evaluation applicable to underwater vehicle simulation training, and in particular to an intelligent evaluation method and system applicable to underwater vehicle simulation training. Background Art

[0002] In underwater vehicle training, simulation training has become a key means of improving operator skills and their ability to handle complex tasks. However, existing training effectiveness evaluation methods have shortcomings in many aspects, affecting the effectiveness and relevance of training.

[0003] Traditional underwater vehicle simulation training systems typically focus on the completion time of operational tasks to assess the operator's reaction speed and efficiency. However, this approach often overlooks details in the operational process, such as accuracy and consistency, leading to one-sided evaluation results.

[0004] Existing assessment methods typically use pre-set standard operating procedures (SOPs) and compare them with actual operator performance. However, the mission environment of underwater vehicles is complex and ever-changing, and SOPs may not cover all situations, resulting in a disconnect between assessment results and actual needs.

[0005] At the tactical level, evaluating the rationality of operator decisions typically relies on expert manual judgment. This approach is highly subjective, inefficient, and difficult to apply in large-scale training.

[0006] Traditional evaluation methods primarily focus on hardware system stability, such as equipment failure rates. However, as software becomes increasingly important in simulation training, the performance and stability of software systems have an increasingly significant impact on training effectiveness, yet this has not been adequately addressed.

[0007] Research shows that underwater vehicles face challenges in autonomous navigation, target positioning and identification in complex environments, and traditional evaluation methods are difficult to fully reflect the operator's capabilities in these areas.

[0008] Existing methods often evaluate indicators such as operational timeliness, process standardization, tactical rationality, and system stability separately, lacking comprehensive analysis of the correlations between these indicators and failing to fully reflect the operator's actual capabilities. Due to the complex and ever-changing mission environments of underwater vehicles, fixed evaluation criteria are difficult to adapt to diverse training scenarios, resulting in reduced applicability and reliability of evaluation results. Traditional evaluation methods overly rely on human judgment, are subject to subjective bias, and are difficult to effectively apply in large-scale training, limiting the real-time and objectivity of the evaluation. Summary of the Invention

[0009] In view of the above-mentioned problems, the present invention is proposed.

[0010] Therefore, the technical problems solved by the present invention are:

[0011] To solve the above technical problems, the present invention provides the following technical solutions: an intelligent evaluation method suitable for underwater vehicle simulation training, comprising: real-time collection of operation data and system status information during the training process.

[0012] Build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, and output the standardization and rationality data of the evaluation operation.

[0013] Provide feedback and improvement suggestions based on the normative and rationality data of the evaluation operations output by the model.

[0014] Building an intelligent evaluation model includes constructing an intelligent evaluation model through nonlinear interactive feature extraction, state stability modeling, behavioral deviation discriminant analysis, feature compression and dimensionality reduction via a historical behavior memory network. The model performs nonlinear feature mapping on the operation data within each time step, and characterizes the intensity and complexity of the control behavior at a moment by taking the logarithm of the norm square of the original input data.

[0015] Analyzing and identifying operator behavior patterns involves extracting nonlinear interaction features and evaluating state offsets of chronologically collected operation data and system state information.

[0016] As a preferred solution of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the operation data and system status information collected during the training process include:

[0017] Deploy multi-source data acquisition modules to collect operation data and system status information in real time.

[0018] The operation data includes the operator's mouse trajectory, key operations, joystick input, touch commands and voice commands on the training terminal.

[0019] As a preferred solution of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the system status information includes:

[0020] CPU usage, GPU load, memory usage, device temperature, power status, and internal running status of the training simulation.

[0021] A high-precision time synchronization mechanism is used to stamp each sampling operation with a unified timestamp, establishing a one-to-one correspondence between operation data and system status information.

[0022] As a preferred solution of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the construction of the intelligent evaluation model includes:

[0023] Organize the operation data collected in chronological order during the training process to form original behavior data;

[0024] The original behavioral data includes a combination of mouse tracks, key operations, joystick inputs, touch commands, and voice command interaction signals generated by the operator on the training terminal.

[0025] An intelligent evaluation model is constructed through nonlinear interactive feature extraction, state stability modeling, behavioral deviation discriminant analysis, feature compression and dimensionality reduction, and historical behavior memory network. The model performs nonlinear feature mapping on the operation data within each time step, and generates single-step behavioral features representing the intensity and complexity of the control behavior by taking the logarithm of the norm square of the original input data.

[0026] The single-step behavior features of multiple time steps are constructed into a historical behavior sequence in chronological order, and the historical behavior sequence is encoded based on the gated recurrent network to extract the time-correlation memory vector.

[0027] The system state information is used to construct a state deviation response function through an exponential function to measure the deviation between the current state and the reference stable state, reflecting the instantaneous impact of the operation behavior on the system.

[0028] The distance calculation between the learning behavior output and the standard sample data is introduced to characterize the degree of deviation between the current behavior and the standard paradigm, and the error is sent to the compression function for nonlinear normalization processing.

[0029] The principal component analysis method is used to reduce the dimensionality of the original behavioral data and extract the representative operation pattern feature vectors.

[0030] As a preferred embodiment of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the analysis and identification of the operator's behavior pattern includes:

[0031] The operation data and system status information collected in chronological order are used for nonlinear interaction feature extraction and state offset evaluation respectively.

[0032] The nonlinear interaction feature extraction includes obtaining the operation amplitude feature by squaring the norm of the operation vector at each time step and then performing logarithmic processing.

[0033] The state offset assessment includes calculating the exponential deviation between the system state sequence and the historical mean to form a state offset factor, and feeding the operation amplitude characteristics and the state offset factor into the first behavior scoring fraction function to calculate the initial behavior normative score of the current time step.

[0034] As a preferred solution of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the standardization and rationality of the evaluation operation include:

[0035] The operation data at the same time step are standardized and then high-order features are extracted. The weighted sum expression of the activation function is constructed through high-order feature extraction. The weighted nonlinear combination of the operation signals of each dimension is performed to obtain the operation activity features. The partial derivative response value of the operation data with respect to the system state data is calculated. The behavior sensitivity index is constructed through the cubic square root calculation method. The operation activity features and the behavior sensitivity index are used as input for normative scoring to form a behavior-system joint normativeness evaluation factor.

[0036] As a preferred embodiment of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the forming of the intelligent scoring function includes:

[0037] The absolute value of the difference between the behavioral feature vector predicted by the deep learning model at the corresponding time step in training and the standard behavioral sample of the same type is calculated, the abnormal deviation is amplified by two-thirds power, and the output is sent to the S-type compression function for normalization to obtain the behavioral deviation.

[0038] The operation data is processed by principal component dimensionality reduction to obtain a compressed expression vector, and a historical behavior memory representation is constructed based on a gated recurrent network. The compressed expression vector is vector-point multiplied with the historical behavior memory representation and then normalized with the norm squared difference to obtain a behavior consistency index.

[0039] As a preferred embodiment of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, forming the intelligent scoring function further comprises:

[0040] The behavior-system joint normativeness evaluation factor, behavior deviation and behavior consistency index are nested and fused to form an intelligent scoring function, and the output value of the scoring function is distributed in the range of [-1,1].

[0041] As a preferred embodiment of the intelligent evaluation method for underwater vehicle simulation training according to the present invention, the providing of improvement suggestions includes:

[0042] Improvement suggestions are generated based on the sources of operation deviations identified during the feedback suggestion generation process and the corresponding abnormal score intervals. Based on the three-dimensional difference analysis results between the current behavior expression vector, the standard operation template vector, and the historical behavior deviation vector, a behavior adjustment space is constructed, and a list of operation items to be improved is generated based on the operation type label.

[0043] The operation item list includes dimension fields related to action instructions, system interaction rhythm, control strength, or status response.

[0044] The vector cosine similarity and average response error analysis are performed on the multi-round performance of the corresponding operation item in the current behavior expression vector in the simulated training records and the optimal performance in the standard sample. If the error exceeds the set threshold in three consecutive operations, it is identified as a dimension that needs to be improved, and a cross-validation evaluation is performed on the operation item to be improved.

[0045] The similarity calculation and error analysis results include improving path selection based on the similarity calculation and error analysis results, calling multiple rounds of candidate optimization solutions for improved path selection, and building a parallel list of multiple solutions including behavior rhythm reshaping suggestions, interaction rhythm buffering solutions, and operation sequence rescheduling suggestions.

[0046] The parallel suggestion list is intelligently screened, and dynamic priority weights are assigned to each type of suggestion based on the current training phase goals, the operator's historical behavioral style characteristics, and the system status tolerance level. The suggestions are then compared and matched with proven effective improvement solutions in the retrospective database to select the optimal set of improvement paths.

[0047] Convert the optimal improvement path into a structured improvement suggestion package.

[0048] The improvement suggestion package includes the suggestion item number, improvement dimension, original operation reference vector, suggested alternative vector, suggested adjustment cycle and system status preset range, and is dynamically displayed through the simulated interactive terminal in the form of pop-up prompts, path demonstration and virtual guidance.

[0049] The present invention also proposes an intelligent evaluation system suitable for underwater vehicle simulation training, which is characterized by comprising a multi-source data acquisition module, a model building and evaluation module, and a feedback improvement module;

[0050] The multi-source data acquisition module is used to collect operation data and system status information during the training process in real time. The multi-source data acquisition module includes a sensor control interface submodule, a data aggregation cache submodule, and a timestamp marking and synchronization submodule;

[0051] The model building and evaluation module is used to build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, and output the standardization and rationality data of the evaluation operation;

[0052] The feedback improvement module is used to provide feedback and improvement suggestions based on the normative and rationality data of the evaluation operation output by the model.

[0053] Beneficial effects of the present invention: The present invention provides an intelligent evaluation method suitable for underwater submersible simulation training. In response to the problems that the existing training evaluation process relies on manual scoring, lacks quantitative standards, and is difficult to identify abnormal behavior in real time, the present invention proposes a full-process intelligent evaluation solution from multi-source data collection, behavior modeling, normative scoring, to feedback suggestions and path optimization. By constructing a multi-level data collection architecture, high-frequency and synchronous collection of operation data and system status information is achieved. The use of a unified timestamp annotation mechanism ensures the precise alignment of heterogeneous data in time series, providing complete and continuous data support for subsequent behavior modeling.

[0054] The constructed intelligent assessment model introduces nonlinear measurement of operation intensity, state offset response function, behavioral paradigm deviation compression mechanism and historical behavior memory structure. The model structure is complex and precise, and can realize multi-dimensional modeling and refined scoring of operator behavior patterns; among them, the behavioral scoring function adopts nested fraction expression, nonlinear scaling and vector space projection mechanism in structure to ensure that the model has strong discrimination and discrimination stability.

[0055] A feedback strategy and deviation tracing mechanism based on score value range segmentation is proposed, which can not only output structured feedback suggestions in real time, but also accurately locate abnormal behavior types and triggering conditions based on the difference vector analysis results, significantly improving the response speed and interpretability of the training process.

[0056] By constructing an improved module that includes behavioral expression difference analysis, candidate path optimization, and recommended parameterized generation mechanism, the present invention can generate multi-round adjustable optimization plans in a targeted manner, and dynamically screen and guide them in combination with the operator's behavioral style characteristics, so that the simulation training system has adaptive personalization capabilities, effectively improving training efficiency and behavioral norms. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0058] Figure 1 This is an overall flow chart of an intelligent evaluation method and system for underwater vehicle simulation training provided by the first embodiment of the present invention; DETAILED DESCRIPTION

[0059] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0060] Example 1, with reference to Figure 1 , which is an embodiment of the present invention, provides an intelligent evaluation method suitable for underwater vehicle simulation training, comprising:

[0061] S1: Collect operation data and system status information during training in real time.

[0062] Deploy multi-source data acquisition modules to collect operation data and system status information in real time.

[0063] The multi-source data acquisition module includes: a sensor control interface submodule, which interfaces with the device driver and reads low-level operation signals; a data aggregation cache submodule, which caches data according to time windows; and a timestamp labeling and synchronization submodule, which timestamps each data item and performs multi-source synchronization.

[0064] The operation data includes the operator's mouse trajectory, key operations, joystick input, touch commands and voice commands on the training terminal.

[0065] System status information includes CPU usage, GPU load, memory usage, device temperature, power status, and the internal operating status of the training simulation system.

[0066] A high-precision time synchronization mechanism is used to stamp each sampling operation with a unified timestamp, establishing a one-to-one correspondence between operation data and system status information.

[0067] Furthermore, a fixed sampling interval is set to periodically collect the above information, and the operation data and system status data at each moment are organized into a unified data structure, which is cached in the edge cache area in chronological order. The cache structure adopts a double buffering mechanism to ensure continuous data collection while realizing data batch packaging.

[0068] Further set the trigger conditions, reach the set time window, the upper limit of the number of data items, and detect abnormal system status, trigger the data packaging and uploading, and transmit the current batch of data to the back-end intelligent evaluation module.

[0069] During the data collection process, to ensure data integrity and suitability for subsequent evaluation model training, a preliminary abnormal state identification mechanism was introduced. This mechanism determines whether the system state is abnormal based on preset thresholds and includes abnormality identification tags in the data for supervised learning. This allows for the simultaneous multi-source collection of key behavioral information and operating environment status during training, creating a structured and traceable training data set.

[0070] S2: Build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, and output the standardization and rationality data of the evaluation operation.

[0071] Organize the operation data collected in chronological order during the training process to form original behavior data;

[0072] The original behavioral data includes a combination of mouse tracks, key operations, joystick inputs, touch commands, and voice command interaction signals generated by the operator on the training terminal.

[0073] An intelligent evaluation model is constructed through nonlinear interactive feature extraction, state stability modeling, behavioral deviation discriminant analysis, feature compression and dimensionality reduction, and historical behavior memory network. The model performs nonlinear feature mapping on the operation data within each time step, and generates single-step behavioral features representing the intensity and complexity of the control behavior by taking the logarithm of the norm square of the original input data.

[0074] The single-step behavior features of multiple time steps are constructed into a historical behavior sequence in chronological order, and the historical behavior sequence is encoded based on the gated recurrent network to extract the time-correlation memory vector.

[0075] The system state information is used to construct a state deviation response function through an exponential function to measure the deviation between the current state and the reference stable state, reflecting the instantaneous impact of the operation behavior on the system.

[0076] The distance calculation between the learning behavior output and the standard sample data is introduced to characterize the degree of deviation between the current behavior and the standard paradigm, and the error is sent to the compression function for nonlinear normalization processing.

[0077] The principal component analysis method is used to reduce the dimensionality of the original behavioral data and extract the representative operation pattern feature vectors.

[0078] Furthermore, the operation data and system status information collected in chronological order are subjected to nonlinear interaction feature extraction and state offset evaluation respectively.

[0079] The nonlinear interaction feature extraction includes obtaining the operation amplitude feature by squaring the norm of the operation vector at each time step and then performing logarithmic processing.

[0080] A preferred solution for obtaining the operating amplitude characteristics is:

[0081] A t =log(1+||x t || 2 )

[0082] Among them, A t represents the operation amplitude characteristic at time t, x t Represents the operation data vector at time t.

[0083] The first behavior score, joint normativeness evaluation factor, behavior deviation and behavior consistency index are nested and integrated to form an intelligent scoring function. The output value of the scoring function is set to be distributed in the range of [-1,1] based on the experience of the experimenters.

[0084] State offset assessment includes calculating the exponential deviation between the system state sequence and the historical mean to form a state offset factor.

[0085] A preferred solution for forming the state shift factor is:

[0086]

[0087] Among them, R t Indicates the system state response deviation index at time t, s t,i represents the value of the i-th system parameter at time t, represents the historical mean of the i-th system parameter, and n represents the number of system parameters.

[0088] The partial derivative response value of the operation data relative to the system state data is calculated, and the behavioral sensitivity index is constructed by cubic square root calculation.

[0089] A preferred solution for constructing behavioral sensitivity indicators is:

[0090]

[0091] Among them, t represents the behavioral sensitivity index at time t, x t Represents the operation data vector at time t, s k represents the kth system state parameter, and m represents the number of system state dimensions involved in the sensitivity analysis.

[0092] According to the operation amplitude characteristics, state offset factors and sensitivity indicators, the initial behavior normativeness score of the current time step is calculated.

[0093] A preferred method for calculating the initial behavior normativeness score at the current time step is:

[0094]

[0095] in, represents the first behavior norm score at time t, A t Represents the operating amplitude characteristics, R t represents the system state response deviation index at time t, ζ t represents the behavioral sensitivity index at time t, and η represents a minimum constant that prevents the denominator from being zero.

[0096] The operation data at the same time step are standardized and then high-order features are extracted. The weighted sum expression of the activation function is constructed through high-order feature extraction, and the weighted nonlinear combination of the operation signals of each dimension is performed to obtain the operation activity features.

[0097] A preferred solution for obtaining the operation activity feature is:

[0098]

[0099] Among them, H t represents the operation activity score at time t, x t,j represents the j-th dimension operation eigenvalue at time t, ω j represents the behavioral importance weight of the dimension, σ(·) represents the number of operation dimensions, and d represents the number of operation feature dimensions, which refers to the total number of feature dimensions of various operation signals in the original behavior data vector at the same time step. The number of operation feature dimensions determines the representation accuracy of the operation behavior and directly affects the input dimension structure of the behavior modeling and scoring function.

[0100] The operational activity characteristics and behavioral sensitivity indicators are used as input for normative scoring to form a behavior-system joint normativeness evaluation factor.

[0101] The absolute value of the difference between the behavioral feature vector predicted by the deep learning model at the corresponding time step in training and the standard behavioral sample of the same type is calculated, the abnormal deviation is amplified by two-thirds power, and the output is sent to the S-type compression function for normalization to obtain the behavioral deviation.

[0102] A preferred method for obtaining the behavior deviation is:

[0103]

[0104] Among them, B t represents the behavioral deviation at time t, The j-th eigenvalue of the operation behavior at the t-th time step, represents the feature of the standard behavior sample corresponding to the t-th time step in the j-th dimension, F(·) represents the sigmoid function, d represents the number of dimensions of the behavior feature vector, j represents the index of the current feature dimension, ()2 / 3 Represents the two-thirds power function.

[0105] The operation data is processed by principal component dimensionality reduction to obtain a compressed expression vector, and a historical behavior memory representation is constructed based on a gated recurrent network. The compressed expression vector is vector-point multiplied with the historical behavior memory and then normalized with the norm squared difference to obtain a behavior consistency index.

[0106] A preferred solution for obtaining the behavioral consistency index is:

[0107]

[0108] Among them, C t represents the behavioral consistency score at time t, u t represents the compressed expression vector, v t Represents the historical behavior memory vector, r represents the vector dimension, and ∈ represents the smoothing term.

[0109] The behavior-system joint normativeness evaluation factor, behavior deviation and behavior consistency index are nested and fused through the hyperbolic tangent function to form an intelligent scoring function, and the output value of the scoring function is distributed in the range of [-1,1].

[0110] A preferred solution for forming an intelligent scoring function is:

[0111]

[0112] Among them, E t Represents the final behavior score value, and tanh() represents the hyperbolic tangent function.

[0113] S3: Provide feedback and improvement suggestions based on the normative and rational data of the evaluation operation output by the model.

[0114] According to the behavioral normativeness score output by the intelligent scoring function at each time step, the score value is compared with the preset multi-level threshold interval, which includes a highly normative interval, an acceptable deviation interval, a slightly abnormal interval, and a severely abnormal interval.

[0115] When the behavioral normativity score is in the highly normative range, the feedback module generates affirmative prompts and records the current operating mode as a reference paradigm.

[0116] If the score is within the acceptable deviation range, suggestive suggestions will be generated.

[0117] Suggestive suggestions include suggestions for adjusting operation details, rhythm control suggestions, and status monitoring suggestions. Suggestive suggestions are dynamically generated based on the difference vector calculation results between historical behavior memory and current operations.

[0118] Suggestions for adjusting operation details include optimizing the operation sequence, correcting the button click position, optimizing the command switching logic, and fine-tuning the touch operation range.

[0119] Rhythm control suggestions include: input frequency is too high, operation interval is too short, and behavior rhythm fluctuations exceed the expected range.

[0120] Status monitoring suggestions include: current operations are not timely linked to system status judgment, important feedback windows are ignored, and equipment temperature fluctuations are not paid attention to.

[0121] If the score is in the mild or severe abnormal range, the behavioral deviation tracing process will begin.

[0122] The behavioral deviation tracing process includes extracting the most relevant behavioral segment at the current moment from the historical behavioral memory vector, calculating the characteristic distance between the current behavioral expression and the historical memory expression, and combining the behavioral deviation factor with the state response deviation to locate and analyze the specific operation type, time period, and system status that caused the deviation.

[0123] Based on the typical error types and risk rule tables preset in the behavior pattern library, the category labels and recommended templates of the current deviation pattern are matched.

[0124] After parameterizing the suggestion template, structured feedback suggestions are generated. The structured feedback suggestions include the suggestion level, suggestion content, positioning operation category, corresponding time step and system status reference value. These suggestions are pushed to the operator in real time through a graphical interactive interface or voice feedback, and the feedback suggestions and their corresponding behavioral characteristics are recorded in the training backtracking database.

[0125] Improvement suggestions are generated based on the sources of operation deviations identified during the feedback suggestion generation process and the corresponding abnormal score intervals. Based on the three-dimensional difference analysis results between the current behavior expression vector, the standard operation template vector, and the historical behavior deviation vector, a behavior adjustment space is constructed, and a list of operation items to be improved is generated based on the operation type label.

[0126] The operation item list includes dimension fields related to action instructions, system interaction rhythm, control strength, or status response.

[0127] The vector cosine similarity and average response error analysis are performed on the multi-round performance of the corresponding operation item in the current behavior expression vector in the simulated training records and the optimal performance in the standard sample. If the error exceeds the set threshold in three consecutive operations, it is identified as a dimension that needs to be improved, and a cross-validation evaluation is performed on the operation item to be improved.

[0128] The similarity calculation and error analysis results include: similarity calculation and error analysis results are used to drive improved path selection; path selection calls multiple rounds of candidate optimization solutions to build a parallel list of multiple solutions including behavior rhythm reshaping suggestions, interaction rhythm buffering solutions, and operation sequence rescheduling suggestions.

[0129] Behavioral rhythm reshaping suggestions include: Behavioral rhythm reshaping suggestions refer to structural reconstruction suggestions made after analyzing rhythm parameters such as input frequency, action switching interval, and operation maintenance duration in the operator's operating behavior. The purpose is to make the operating behavior closer to the standard rhythm model and improve the controllability and consistency of the task process.

[0130] The interactive rhythm buffering scheme includes: The interactive rhythm buffering scheme refers to a rhythm coordination mechanism that is proactively proposed when the system detects a mismatch or interactive conflict between the user behavior rhythm and the system response rhythm. It makes human-computer interaction smoother by inserting buffer instructions, rhythm transition prompts, interface waiting guidance, etc.

[0131] Operation reordering suggestions include: Operation reordering suggestions are optimization recommendations for reconstructing the logical order of operations when the system detects that the current operation sequence is inconsistent with the standard task process or a proven effective strategy. These suggestions adjust the operation sequence by referencing the standard operation path or historically effective path to improve task efficiency and accuracy.

[0132] The parallel suggestion list is intelligently screened, and dynamic priority weights are assigned to each type of suggestion based on the current training phase goals, the operator's historical behavioral style characteristics, and the system status tolerance level. The suggestions are then compared and matched with proven effective improvement solutions in the retrospective database to select the optimal set of improvement paths.

[0133] Convert the optimal improvement path into a structured improvement suggestion package.

[0134] The improvement suggestion package includes the suggestion item number, improvement dimension, original operation reference vector, suggested alternative vector, suggested adjustment cycle and system status preset range, and is dynamically displayed through the simulated interactive terminal in the form of pop-up prompts, path demonstration and virtual guidance.

[0135] Improvement dimensions include: each improvement dimension corresponds to a specific adjustable aspect of operational behavior, and has the characteristics of being quantifiable, monitorable, and verifiable.

[0136] The above embodiments also include an intelligent evaluation system suitable for underwater vehicle simulation training, specifically comprising: a multi-source data acquisition module, a model building and evaluation module, and a feedback improvement module.

[0137] The multi-source data acquisition module is used to collect operational data and system status information during the training process in real time. The multi-source data acquisition module includes a sensor control interface submodule, a data aggregation cache submodule, and a timestamp labeling and synchronization submodule.

[0138] The model building and evaluation module is used to build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, and output the standardization and rationality data of the evaluation operation.

[0139] The feedback improvement module is used to provide feedback and improvement suggestions based on the normative and rationality data of the evaluation operation output by the model.

[0140] Example 2, referring to the figure, is applicable to the intelligent evaluation of underwater submersible simulation training. It is an embodiment of the present invention, which provides an intelligent evaluation method and system applicable to underwater submersible simulation training. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0141] To verify the practicality and advantages of the intelligent assessment model-based behavior analysis and feedback recommendation system in underwater vehicle simulation training, a set of simulation experimental scenarios was constructed. Seven operators who received a standard training process were selected for the experiment and performed standard operational tasks in a unified simulation chamber environment, including submersible path control, power adjustment, signal response, and task execution. During each training mission cycle, a multi-source data acquisition module was deployed to synchronously collect operational data and system status information, including operational behavior data such as mouse trajectory offset, keystroke frequency, joystick angle, touch command response, and voice recognition instructions, as well as multi-dimensional system status parameters such as CPU, GPU, memory, temperature, power load, and mission status during training.

[0142] The data is collected with a fixed sampling period of 500ms, and a double-buffered structure is used to cache data in batches, with each batch lasting 10 seconds. When the data reaches 128 items or an abnormal CPU state exceeding 85% or a temperature exceeding 75°C is detected, the upload mechanism is immediately triggered, and the structured data is uploaded to the intelligent evaluation model. The evaluation model extracts the operation amplitude at each moment through nonlinear interactive mapping and establishes a primary behavior score based on the state offset factor. The standardized behavior vector is fed into a high-order activation combination, and the sensitivity of the operation to the system response is calculated to form a joint scoring factor.

[0143] Furthermore, PCA and GRU are used to build a memory of historical operational behaviors and form a difference vector with current behaviors. The final behavioral norm score output by the model is controlled between [-1, 1] and automatically compared and classified with the four-level scoring range set by the system. The feedback improvement module delivers structured prompts in real time during training. When deviation scores repeatedly enter the abnormal range, the behavioral deviation tracing mechanism is automatically activated to analyze the cause of the behavior and generate personalized recommendation templates.

[0144] The experiment lasted one hour, generating approximately 7,200 data points per operator. All scores and recommendations were written to the training backtracking database for further optimization of the personalized behavior assessment model. The experimental data is shown in Table 1.

[0145] Table 1 Experimental data table

[0146]

[0147] During the simulation training of seven operators, the operating standard scores output by the intelligent assessment system ranged from 0.55 to 0.95, indicating that the model has good behavior recognition accuracy. Operators with scores above 0.80 have a clear advantage in the mean state offset and the number of abnormal labels, indicating that their operating behavior is relatively standardized and has less impact on system stability. In contrast, operators E and G have significantly lower scores and more abnormal label records. The feedback suggestion generation rate and improvement suggestion adaptation success rate are both below 0.7, indicating that the model of this invention can accurately locate non-standard operations and trigger an effective feedback mechanism.

[0148] For individuals with a high stability index and high behavioral scores, the model's recommendations focused on pacing and fine-tuning system responses, providing more detailed and specific feedback. For individuals with lower scores, however, the model triggered recommendations for path reconstruction improvements, providing more interventionist feedback. This dynamic feedback and recommendation mechanism demonstrates the advantages of this invention over existing static threshold judgment methods—it not only provides real-time identification of abnormal behavior, but also automatically generates graded recommendations based on behavioral vector history, paradigm references, and system state differences.

[0149] The present invention achieves an average success rate of over 78% in the adaptation of improvement suggestions, which is significantly better than the average response rate of traditional training guidance methods based on manual experience feedback. It effectively shortens the closed-loop cycle from problem identification to behavior correction, and improves training efficiency and behavior consistency.

[0150] Example 3, referring to the figure, is applicable to the intelligent evaluation of underwater submersible simulation training. It is an embodiment of the present invention and provides an intelligent evaluation system applicable to underwater submersible simulation training, including: a multi-source data acquisition module 100, a model construction and evaluation module 200, and a feedback improvement module 300.

[0151] Among them, S4: multi-source data acquisition module 100 is used to collect operation data and system status information during the training process in real time. The multi-source data acquisition module includes a sensor control interface layer, a data aggregation cache layer, and a timestamp labeling and synchronization layer.

[0152] It should also be noted that the upload trigger mechanism in the multi-source data acquisition module 100 is triggered based on the time window, data batch capacity or status abnormality condition, and transmits the cached data to the back-end model construction and evaluation module 200 in real time.

[0153] S5: The model building and evaluation module 200 is used to build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, output the standardization and rationality data of the evaluation operation and input it into the feedback improvement module 300.

[0154] It should also be noted that the model building and evaluation module 200 supports the coordinated output of behavior scores and score explanations, providing both quantitative scores and generating behavior-sensitive dimension explanation vectors for reference by the feedback module.

[0155] S6: The feedback improvement module 300 is used to provide feedback and improvement suggestions based on the normativeness and rationality data of the evaluation operation output by the model.

[0156] It should also be noted that the feedback improvement module 300 has the ability to interact with the backtracking database, and the generated feedback results and behavior scoring data are automatically archived into the training records, supporting subsequent behavior analysis and model retraining iterative use.

[0157] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0158] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0159] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0160] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having logic gate circuits for implementing logical functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc. It should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications should be encompassed by the claims of the present invention.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent evaluation method suitable for underwater vehicle simulation training, characterized in that: include: Collect operational data and system status information during training in real time; Build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify operator behavior patterns, and output evaluation operation standardization and rationality data; Provide feedback and improvement suggestions based on the normative and rational data of the evaluation operations output by the model; The construction of an intelligent evaluation model involves extracting nonlinear interactive features, modeling state stability, analyzing behavioral deviations, and compressing and reducing features through a historical behavior memory network. The model performs nonlinear feature mapping on the operational data within each time step and characterizes the intensity and complexity of the control behavior within a moment by taking the logarithm of the squared norm of the original input data. Analyzing and identifying operator behavior patterns involves extracting nonlinear interaction features and evaluating state offsets of chronologically collected operation data and system state information.

2. The intelligent evaluation method for underwater vehicle simulation training according to claim 1, characterized in that: The operation data and system status information collected during the training process include: Deploy multi-source data acquisition modules to collect operational data and system status information in real time; The operation data includes the operator's mouse trajectory, key operations, joystick input, touch commands and voice commands on the training terminal.

3. The intelligent evaluation method for underwater vehicle simulation training according to claim 1 or 2, characterized in that: The system status information includes: CPU usage, GPU load, memory usage, device temperature, power status, and internal running status of the training simulation; A high-precision time synchronization mechanism is used to stamp each sampling operation with a unified timestamp, establishing a one-to-one correspondence between operation data and system status information.

4. The intelligent evaluation method for underwater vehicle simulation training according to claim 3, characterized in that: The constructing of the intelligent evaluation model includes: Organize the operation data collected in chronological order during the training process to form original behavior data; The raw behavioral data includes a combination of mouse tracks, key operations, joystick inputs, touch commands, and voice command interaction signals generated by the operator on the training terminal; An intelligent evaluation model is constructed through nonlinear interaction feature extraction, state stability modeling, behavioral deviation discriminant analysis, feature compression and dimensionality reduction, and a historical behavior memory network. The model performs nonlinear feature mapping on the operation data within each time step and generates single-step behavioral features representing the intensity and complexity of the control behavior by taking the logarithm of the square of the norm of the original input data. The single-step behavior features of multiple time steps are constructed into a historical behavior sequence in chronological order, and the historical behavior sequence is encoded based on a gated recurrent network to extract the time-correlation memory vector; The system state information is used to construct a state deviation response function through an exponential function to measure the deviation between the current state and the reference stable state, reflecting the instantaneous impact of the operation behavior on the system. The distance calculation between the learning behavior output and the standard sample data is introduced to characterize the degree of deviation between the current behavior and the standard paradigm. The error is sent to the compression function for nonlinear normalization processing. The principal component analysis method is used to reduce the dimensionality of the original behavioral data and extract the representative operation pattern feature vectors.

5. The intelligent evaluation method for underwater vehicle simulation training according to claim 1, 2 or 4, characterized in that: The analysis and identification of operator behavior patterns include: The operation data and system status information collected in chronological order are subjected to nonlinear interaction feature extraction and state offset evaluation respectively; Nonlinear interaction feature extraction includes obtaining the operation amplitude feature by squaring the norm of the operation vector at each time step and then performing logarithmic processing; The state offset assessment includes calculating the exponential deviation between the system state sequence and the historical mean to form a state offset factor, and feeding the operation amplitude characteristics and the state offset factor into the first behavior scoring fraction function to calculate the initial behavior normative score of the current time step.

6. The intelligent evaluation method for underwater vehicle simulation training according to claim 5, characterized in that: The standardization and rationality of the assessment operations include: The operation data at the same time step are standardized and then high-order features are extracted. The weighted sum expression of the activation function is constructed through high-order feature extraction. The weighted nonlinear combination of the operation signals of each dimension is performed to obtain the operation activity features. The partial derivative response value of the operation data with respect to the system state data is calculated. The behavior sensitivity index is constructed through the cubic square root calculation method. The operation activity features and the behavior sensitivity index are used as input for normative scoring to form a behavior-system joint normativeness evaluation factor.

7. The intelligent evaluation method for underwater vehicle simulation training according to claim 6, characterized in that: The forming of the intelligent scoring function comprises: The absolute value of the difference between the behavior feature vector predicted by the deep learning model at the corresponding time step in training and the standard behavior sample of the same type is calculated, the abnormal deviation is amplified by two-thirds power, and the output is sent to the S-type compression function for normalization to obtain the behavior deviation; The operation data is processed by principal component dimensionality reduction to obtain a compressed expression vector, and a historical behavior memory representation is constructed based on a gated recurrent network. The compressed expression vector is vector-point multiplied with the historical behavior memory representation and then normalized with the norm squared difference to obtain a behavior consistency index.

8. The intelligent evaluation method for underwater vehicle simulation training according to claim 7, characterized in that: The forming of the intelligent scoring function further comprises: The behavior-system joint normativeness evaluation factor, behavior deviation and behavior consistency index are nested and fused to form an intelligent scoring function, and the output value of the scoring function is distributed in the range of [-1,1].

9. The intelligent evaluation method for underwater vehicle simulation training according to claim 1, 2, 4 or 8, characterized in that: The improvement suggestions provided include: Improvement suggestions are generated based on the sources of operation deviations identified during the feedback suggestion generation process and the corresponding abnormal score intervals. Based on the three-dimensional difference analysis results between the current behavior expression vector, the standard operation template vector, and the historical behavior deviation vector, a behavior adjustment space is constructed, and a list of operation items to be improved is generated based on the operation type label. The operation item list includes dimension fields related to action instructions, system interaction rhythm, control strength, or status response; The vector cosine similarity and average response error analysis are performed on the multi-round performance of the corresponding operation item in the current behavior expression vector in the simulated training records and the best performance in the standard sample. If the error exceeds the set threshold in three consecutive operations, it is identified as a dimension that needs improvement, and a cross-validation evaluation is performed on the operation item to be improved. The similarity calculation and error analysis results include: improving path selection based on the similarity calculation and error analysis results; improving path selection calls multiple rounds of candidate optimization solutions; and building a parallel list of multiple solutions including behavior rhythm reshaping suggestions, interaction rhythm buffering solutions, and operation sequence rescheduling suggestions; Intelligently screen the parallel suggestion list, assign dynamic priority weights to each suggestion based on the current training phase objectives, operator historical behavioral style characteristics, and system status tolerance level, and compare and match proven improvement solutions in the retrospective database to select the optimal set of improvement paths; Convert the optimal improvement path into a structured improvement suggestion package; The improvement suggestion package includes the suggestion item number, improvement dimension, original operation reference vector, suggested alternative vector, suggested adjustment cycle and system status preset range, and is dynamically displayed through the simulated interactive terminal in the form of pop-up prompts, path demonstration and virtual guidance.

10. An intelligent evaluation system suitable for underwater vehicle simulation training, characterized by: It includes a multi-source data acquisition module (100), a model building and evaluation module (200), and a feedback improvement module (300); The multi-source data acquisition module (100) is used to collect operation data and system status information in the training process in real time, and the multi-source data acquisition module includes a sensor control interface submodule, a data aggregation cache submodule, and a timestamp marking and synchronization submodule; The model building and evaluation module (200) is used to build an intelligent evaluation model, input operation data and system status information into the intelligent evaluation model, analyze and identify the operator's behavior pattern, and output the standardization and rationality data of the evaluation operation; The feedback improvement module (300) is used to provide feedback and improvement suggestions based on the normativeness and rationality data of the evaluation operation output by the model.

Images