Hierarchical aggregation combat simulation training evaluation method based on multiple dimensions

Abstract

The invention relates to the technical field of combat simulation training evaluation, and discloses a hierarchical aggregation combat simulation training evaluation method based on multiple dimensions. The method comprises the following steps: receiving original behavior data streams of combat units in real time by establishing a data connection channel, and extracting space coordinates, time sequences, equipment interaction and tactical action features of the original behavior data streams to construct a feature vector set; and performing multi-round iterative computation on the feature vectors by using a dynamic clustering engine, adaptively terminating iteration by monitoring a grouping stability index, and outputting a hierarchical aggregation unit of a final group. After the feature data of the aggregation units are stored in an assessment data warehouse, the feature data are analyzed and processed by a multi-stage assessment model to generate assessment index data for each aggregation unit, and finally a visual assessment report is formed and pushed to a command terminal. According to the method, automatic, multi-dimensional and systematic efficiency evaluation on the tactical unit dynamically formed in the simulation training is realized.

Description

Technical Field

[0001] This invention relates to the field of combat simulation training and evaluation technology, specifically to a multi-dimensional, hierarchical aggregation combat simulation training and evaluation method. Background Technology

[0002] Existing combat simulation training evaluation methods generally rely on post-hoc statistical analysis of behavioral data from fixed organizational units or individual combat platforms. The evaluation process is usually based on preset rules and indicators, summarizing discrete parameters such as the number of kills, survival time, and ammunition consumption of individual units. This evaluation model treats combat units in training as static and isolated individuals, and their interactions and synergies are often judged through simple data aggregation or the commander's subjective experience.

[0003] Existing technologies struggle to effectively handle the massive, high-dimensional, and dynamic streams of raw behavioral data generated during simulated training. Rule-based analysis methods cannot automatically identify the functional clusters dynamically formed between combat units due to tactical coordination. Evaluation perspectives are limited to individual equipment or fixed formations, resulting in assessments that fail to reflect the true nature of tactical coordination and self-organized collaboration among soldiers on the battlefield. The temporary tactical groups emerging during training and their overall combat effectiveness are overlooked.

[0004] Due to a lack of in-depth analysis of dynamic behavioral patterns, the assessment conclusions fail to reveal the efficiency and quality of coordination in system-of-systems warfare. Commanders are unable to ascertain which units spontaneously formed effective tactical clusters during specific operational phases, or whether these clusters exhibited deeper issues such as disjointed operations or inefficient resource allocation. The assessment results remain at a macro level or focus on individual performance, failing to provide a meso-level perspective on the dynamic composition and collaborative effectiveness of tactical units. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional hierarchical aggregation combat simulation training and evaluation method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a multi-dimensional hierarchical convergence combat simulation training and evaluation method, the method comprising:

[0007] Establish data connection channels with each combat unit in the simulation training system to receive raw behavioral data streams generated during the operation of each combat unit in real time.

[0008] The received raw behavioral data stream is subjected to feature extraction to identify spatial coordinate features, time series features, equipment interaction features, and tactical action features.

[0009] A feature vector set is constructed based on the extracted multi-class features, and the feature vector set is input into the dynamic clustering engine for initial grouping;

[0010] The feature vector set is subjected to multiple rounds of iterative calculations through a dynamic clustering engine. In each round of iteration, the grouping boundaries are adjusted and intermediate grouping results are generated.

[0011] The grouping stability index is calculated based on the intermediate grouping results. The iteration is terminated when the grouping stability index reaches the preset threshold, and the final grouping result is output as the hierarchical aggregation unit.

[0012] The feature data of each aggregation unit at each level are sent to the evaluation data warehouse for standardized storage;

[0013] Retrieve standardized stored feature data from the assessment data warehouse and input it into the multi-level assessment model for analysis and processing;

[0014] The evaluation index data for each level of aggregation unit is generated through a multi-level evaluation model;

[0015] A visual assessment report is generated based on the assessment indicator data, and then pushed to the command terminal display interface.

[0016] Preferably, the establishment of data connection channels with each combat unit in the simulation training system includes:

[0017] Configure network communication protocol parameters to establish a two-way data transmission link with each combat unit;

[0018] Set the data acquisition frequency parameters and receive the raw behavioral data streams uploaded by each combat unit at the set time intervals;

[0019] Add time tags and source identifiers to the raw behavioral data stream to form a raw data set with spatiotemporal tags.

[0020] Preferably, the feature extraction operation on the received raw behavioral data stream includes:

[0021] The position coordinate sequence is separated from the original dataset with spatiotemporal labels, and the variation pattern of the position coordinate sequence is extracted as a spatial motion feature;

[0022] The equipment status change records are parsed from the original dataset as equipment interaction features, and the pattern features in the tactical command execution records are identified as tactical action features.

[0023] Preferably, the initial grouping of the dynamic clustering engine includes:

[0024] Calculate the similarity matrix of each feature vector in the feature vector set, and construct an initial set of grouping center points based on the similarity matrix;

[0025] Based on the initial set of group center points, the feature vector set is divided into initial group sets.

[0026] Preferably, the multi-round iterative calculation includes:

[0027] In each iteration, the center point coordinates of each group are recalculated, and the group assignment of each feature vector is adjusted according to the recalculated center point coordinates. The changes in group assignment during each iteration are recorded, and a group adjustment log is generated.

[0028] Preferably, the calculation of the group stability index includes:

[0029] The magnitude of group assignment changes during multiple consecutive iterations is statistically analyzed. The moving average of the magnitude of group assignment changes is calculated as a stability reference value. The stability reference value is compared with a preset stability threshold to determine the iteration termination condition.

[0030] Preferably, the standardized storage of the evaluation data warehouse includes:

[0031] Perform format validation and integrity checks on the received feature data, and convert the validated feature data into a standard data format.

[0032] A multidimensional index structure is established based on the time dimension and the hierarchical aggregation unit dimension, and the standardized feature data is stored in the corresponding data storage partition.

[0033] Preferably, the analysis and processing of the multi-level evaluation model includes:

[0034] Configure an evaluation index system, including basic performance indicators, tactical coordination indicators, and combat effectiveness indicators;

[0035] Feature data corresponding to the evaluation index system is extracted from the evaluation data warehouse, and a weighted fusion algorithm is used to calculate the comprehensive evaluation value of each level of aggregation unit, generating an intermediate evaluation data table containing the evaluation results of each level of aggregation unit.

[0036] Preferably, the generation of the visualization evaluation report includes:

[0037] Extract key evaluation indicator data from intermediate evaluation data tables and design visualization chart templates, including situation distribution charts, performance comparison charts, and trend analysis charts;

[0038] Key evaluation indicator data are populated into the visualization chart template to generate chart elements, and the chart elements are combined to form a complete visualization evaluation report document.

[0039] Preferably, the push notifications on the command terminal display interface include:

[0040] The visual assessment report document is converted into a display data packet in a specified format, and the display data packet is sent to the command terminal through an encrypted transmission channel;

[0041] The command terminal parses and displays data packets and renders a visualization interface, updating the evaluation data content in the visualization interface in real time.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] By using a dynamic clustering engine to calculate group stability indices in each iteration and using them as termination criteria, the clustering process can autonomously perceive the evolution of the data's inherent structure. This mechanism overcomes the rigid grouping problem caused by traditional methods relying on preset parameters, achieving adaptive capture of dynamic relationships between combat units. The iteration automatically terminates when the indices show that the group boundaries are stabilizing, ensuring that the output hierarchical aggregated units truly reflect the tactical clusters emerging during training. The clustering results are no longer limited by fixed formations or subjective presets, providing an objective and combat-relevant basis for unit division, improving the automation and accuracy of behavioral data analysis.

[0044] By using dynamically generated hierarchical aggregation units as the core evaluation object, the evaluation dimension shifts from individual indicators to cluster collaborative effectiveness. The multi-level evaluation model quantifies system capability indicators such as overall task synchronization, decision-making response speed, and tactical coordination density by analyzing the timing coordination of actions, spatial situational awareness sharing, and resource interaction logic among members within the unit. This analysis can reveal micro-level collaborative phenomena that traditional evaluations cannot detect. This shift in evaluation perspective allows the analytical conclusions to directly point to the rationality of tactical formations and the effectiveness of collaborative rules.

[0045] This technical solution achieves a deep transformation from raw data to tactical insights through a closed-loop process of feature extraction, dynamic clustering, and multi-dimensional evaluation. Standardized storage of evaluation data supports historical comparison and trend analysis, while visualized reports transform complex cluster interaction situations into intuitive decision-making information. The entire process uncovers the tactical value inherent in simulated training data, providing data-driven analytical tools for optimizing combat formations and improving tactical doctrine. Attached Figure Description

[0046] Figure 1 This is a schematic diagram illustrating the working principle of the multi-dimensional hierarchical aggregation combat simulation training and evaluation method described in this invention.

[0047] Figure 2 A flowchart for establishing a data connection channel;

[0048] Figure 3 A flowchart for initializing groupings for the dynamic clustering engine;

[0049] Figure 4 This is a distribution map of collaborative feature clustering;

[0050] Figure 5 A heatmap showing the correlation between core indicators for combat training assessment. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 This invention provides a multi-dimensional hierarchical aggregation combat simulation training evaluation method. The method includes: establishing a data connection channel with each combat unit in the simulation training system to receive raw behavioral data streams generated during the operation of each combat unit in real time; establishing a bidirectional data transmission link with each combat unit through configuring network communication protocol parameters, and setting data acquisition frequency parameters to receive data at predetermined time intervals; simultaneously adding time tags and source identifiers to the raw behavioral data streams to form a raw data set with spatiotemporal markings. Feature extraction is performed on the received raw behavioral data streams to identify spatial coordinate features, time series features, equipment interaction features, and tactical action features from the spatiotemporally marked raw data set; the feature extraction operation includes separating position coordinate sequences to extract spatial motion features, and parsing equipment status change records and tactical command execution records to obtain equipment interaction features and tactical action features. A feature vector set is constructed based on the extracted multi-class features, and the feature vector set is input into a dynamic clustering engine for initial grouping; the dynamic clustering engine constructs an initial grouping center point set by calculating a feature vector similarity matrix, and uses this as the basis for dividing the initial grouping set. The dynamic clustering engine performs multiple rounds of iterative calculations on the feature vector set. In each round of iteration, the coordinates of the center point of each group are recalculated and the grouping of the feature vectors is adjusted. The grouping adjustment log is recorded to generate intermediate grouping results.

[0053] The moving average of the changes in group affiliation based on intermediate grouping results is used as a grouping stability index. The iteration terminates when this index reaches a preset threshold, and the final grouping result is output as the hierarchical aggregation unit. Feature data from each hierarchical aggregation unit is sent to an evaluation data warehouse for standardized storage. The evaluation data warehouse performs format validation and integrity checks on the feature data, converts it to a standard data format, and then establishes a multi-dimensional index structure for storage based on time and hierarchical aggregation unit dimensions. The standardized feature data is then retrieved from the evaluation data warehouse and input into a multi-level evaluation model for analysis and processing. The multi-level evaluation model is configured with an evaluation index system including basic performance indicators, tactical coordination indicators, and operational effectiveness indicators. A weighted fusion algorithm is used to calculate the comprehensive evaluation value of each hierarchical aggregation unit, generating an intermediate evaluation data table. Evaluation index data for each hierarchical aggregation unit is generated through the multi-level evaluation model. Based on the evaluation index data, key evaluation index data is extracted from the intermediate evaluation data table, and visualization chart templates are designed, including situational distribution charts, performance comparison charts, and trend analysis charts. Data is populated into the templates to generate chart elements, which are then combined to form a complete visualization evaluation report document. The visualization assessment report document is converted into a display data packet in a specified format and sent to the command terminal through an encrypted transmission channel. The command terminal parses the display data packet and renders the visualization interface, updating the assessment data content in real time.

[0054] Example 1: See Figure 2 In practical implementation, establishing data connection channels with each combat unit in the simulation training system is a fundamental step, and configuring network communication protocol parameters is the primary operation. These parameters include specific settings for the Transmission Control Protocol (TCP) or User Datagram Protocol (UDP), such as setting the port number to 8080, the timeout to 5000 milliseconds, and the data packet size to 1024 bytes. These parameter configurations aim to establish a reliable bidirectional data transmission link with each combat unit distributed in the simulation training environment. Setting the data acquisition frequency parameters is a synchronous task. The data acquisition frequency parameters are set according to the real-time requirements of the simulation training scenario. For example, the acquisition frequency is set to 100 Hz for a highly dynamic battlefield environment and 1 Hz for the command and decision-making process. The system actively polls or passively receives the raw behavioral data streams uploaded by each combat unit at set time intervals. Adding time stamps and source identifiers to the raw behavioral data stream is a key step in data preprocessing. The time stamps use Coordinated Universal Time (UTC) format and are accurate to the millisecond level. The source identifiers use the unique identification code of the combat unit, such as a MAC address or a preset equipment serial number. This results in a raw data set with spatiotemporal markers, which is temporarily cached in a circular buffer to prevent data overflow.

[0055] In practical implementation, feature extraction from the received raw behavioral data stream is the core processing stage. Separating the position coordinate sequence from the spatiotemporally labeled raw data set is the starting point for spatial feature extraction. The position coordinate sequence originates from the latitude, longitude, and altitude data stream output by the GPS receiver or inertial navigation system. Extracting the variation patterns of the position coordinate sequence as spatial motion features involves continuous data processing. Spatial motion features include instantaneous velocity values, average acceleration vectors, radius of curvature of the motion trajectory, and displacement direction angles. These features are calculated from the raw coordinate sequence using a sliding window difference algorithm and a least-squares fitting method. Parsing equipment status change records from the raw data set as equipment interaction features is a parallel process. Equipment status change records include discrete event logs such as radar power on / off status, weapon and ammunition inventory, and communication module operating modes. The parsing process uses a finite state machine model to identify key nodes of state transitions and record timestamps. Identifying pattern features in tactical command execution records as tactical action features requires combining domain knowledge. Tactical command execution records originate from the interaction logs of the command and control system. Pattern features include command type distribution, command execution delay, and command sequence coherence index. These features are extracted using named entity recognition and sequence labeling techniques in natural language processing.

[0056] In some embodiments, the overall architecture of the feature extraction operation adopts a pipeline design. The data parsing module first performs format parsing and field splitting on the original dataset with spatiotemporal markers, separating the mixed data stream into a positioning data substream, an equipment status substream, and a tactical command substream. The spatial motion feature extractor specifically processes the positioning data substream, smoothing and denoising the original coordinate data using a Kalman filter algorithm, then calculating the moving average velocity and first-order differential acceleration, while the trajectory curvature is estimated using the three-point circular method. The equipment interaction feature extractor monitors the equipment status substream, establishes a state transition matrix based on a time window, and statistically analyzes the frequency and duration of equipment mode switching within a specific time period, such as calculating the average duration of continuous radar operation in detection mode. The tactical action feature extractor analyzes the tactical command substream, constructs a command dictionary and grammar rule base, models the command sequence using a hidden Markov model, and identifies the command feature vectors corresponding to typical tactical patterns such as "alternating cover" or "concentrated fire."

[0057] Understandably, the computational complexity of feature extraction needs optimization. For the high-frequency sampled raw behavioral data stream, the system employs a multi-threaded parallel processing mechanism, dividing the data stream into time slices and distributing them to different feature extractors for concurrent execution. Spatial motion feature extraction uses a graphics processing unit to accelerate matrix operations, equipment interaction feature extraction adopts an event-driven architecture to reduce computational overhead, and tactical action feature extraction utilizes pre-compiled regular expression patterns for fast matching. All extracted feature values ​​are normalized and then assembled into feature vectors. These feature vectors are arranged in chronological order and labeled with group identifiers, providing standardized input for the subsequent dynamic clustering engine.

[0058] It is understandable that the stability and reliability of the data connection channel need to be ensured through redundant design. The network communication protocol parameter configuration includes a heartbeat detection mechanism and disconnection reconnection logic. The heartbeat detection sends a status query packet every 5 seconds, and the disconnection reconnection automatically triggers a reconnection sequence within 3 seconds after a link interruption is detected. The data acquisition frequency parameter supports dynamic adjustment. When the system detects insufficient network bandwidth, it automatically reduces the acquisition frequency of non-critical data, for example, reducing the acquisition frequency of equipment status data from 100 Hz to 10 Hz. The raw data set with spatiotemporal stamps adopts a multi-level caching strategy. The raw data retains the most recent 300 seconds of content in memory, and data exceeding the capacity limit is automatically archived to the solid-state drive temporary storage area to ensure that the feature extraction module can continuously access historical data.

[0059] Optionally, the configurability of feature extraction operations is achieved through parameter templates. Spatial motion feature extraction supports configuring different sliding window sizes, which can be adjusted between 10 and 100 data points to adapt to the recognition needs of different motion patterns. Equipment interaction feature extraction allows users to customize a list of key states, such as marking "ammunition depleted" and "radar lock" as high-priority states for focused monitoring. Tactical action feature extraction provides a rule editor, allowing training directors to manually add new tactical pattern recognition rules, such as defining the command sequence template corresponding to the "flanking maneuver" tactic. All these configuration parameters are centrally managed through a graphical interface and support configuration file import and export functions.

[0060] Optionally, the real-time performance of the feature extraction process is further enhanced through pipelined parallel technology. The three stages—data parsing, feature computation, and vector assembly—form a three-stage pipeline, with each stage processing data blocks from different time slices. The data parsing stage is specifically responsible for parsing the data packet at second N; the feature computation stage simultaneously processes the data parsed at second N-1; and the vector assembly stage packages the features calculated at second N-2. This overlapping processing method increases system throughput by three times. The feature extraction module also integrates a quality monitoring subsystem, which calculates the signal-to-noise ratio and completeness indicators of the feature data in real time. When the indicators fall below a threshold, it automatically triggers data retransmission or an alarm mechanism to ensure that the data quality input to the dynamic clustering engine meets the requirements.

[0061] Example 2: See Figure 3 In its implementation, the dynamic clustering engine's initial grouping process begins with receiving a set of feature vectors. This set originates from the standardized data stream output by the feature extraction operation. Each feature vector contains a multi-dimensional numerical representation of spatial coordinate features, time-series features, equipment interaction features, and tactical action features. Calculating the similarity matrix of each feature vector in the set is the first step in initial grouping. The similarity matrix is ​​a symmetric square matrix with the number of rows and columns equal to the total number of vectors in the set. Matrix elements store the similarity value between each pair of feature vectors. The similarity value is calculated using the cosine similarity algorithm, expressed by the formula:

[0062] ;

[0063] in: This represents the similarity value between feature vector a and feature vector b. This represents the eigenvalue of eigenvector a in the k-th dimension. This represents the eigenvalue of eigenvector b in the k-th dimension. The total dimension of the feature vectors is represented. After the similarity matrix is ​​calculated, it is stored in memory for subsequent grouping. The initial set of group centroids is constructed based on the similarity matrix using the minimum-maximum distance method. First, a feature vector is randomly selected as the first centroid. Then, the feature vector with the lowest similarity to the selected centroid is iteratively selected as the new centroid, until the number of centroids reaches the preset number of groups K. The initial set of group centroids is saved as a coordinate list. Based on the initial set of group centroids, the feature vector set is divided into initial group sets. The division rule is based on the nearest neighbor principle; each feature vector is assigned to the group of the centroid with the highest similarity, forming a member assignment mapping table for the initial group sets.

[0064] In practice, the multi-round iterative calculation process is automatically triggered after group initialization. The iterative calculation is executed in a loop until the termination condition is met. During each iteration, the center point coordinates of each group are recalculated. The center point coordinates are calculated using the mean of all feature vectors within the group. For each group, the new center point coordinates are obtained by calculating the arithmetic mean of all feature vectors in each dimension. The mean calculation uses a weighted average method, with weights adjusted based on the freshness of the feature vector timestamps. The group assignment of each feature vector is adjusted based on the recalculated center point coordinates. This adjustment operation involves traversing the entire feature vector set. For each feature vector, its similarity value with all new center points is calculated, and the feature vector is reassigned to the group corresponding to the center point with the highest similarity. Parallel computing technology is used to accelerate the group assignment adjustment, dividing the feature vector set into multiple subsets for concurrent processing. The changes in group assignment during each iteration are recorded, generating a group adjustment log. The group adjustment log includes fields such as iteration number, feature vector identifier, old group identifier, new group identifier, and similarity difference. The log is stored in a circular buffer structure to prevent memory overflow. The group adjustment log is used for subsequent stability analysis.

[0065] In some embodiments, the similarity matrix calculation optimization employs dimensionality reduction techniques to handle high-dimensional feature vectors. When the feature vector dimension D exceeds a threshold, principal component analysis is used to reduce the dimension to the principal components, reducing computational complexity while retaining most of the variance information. The construction of the initial grouping centroid set supports various heuristic algorithms. In addition to the maximum-minimum distance method, k-means++ or density peak-based methods can also be used. These algorithms improve the representativeness of centroid selection by pre-compiling a similarity distribution histogram. The grouping adjustment process integrates an anomaly detection mechanism. When a feature vector is detected to have a similarity to all centroids below a threshold, the feature vector is marked as an outlier and temporarily stored in an independent buffer. Outliers are processed separately or reported to the monitoring system after the iteration ends.

[0066] In some embodiments, the execution framework for multi-round iterative computation employs a fault-tolerant design. Before each iteration, the system resource status is checked, and if memory is insufficient, data paging or compression operations are automatically triggered. A smoothing factor is added when recalculating the center point coordinates. The new center point coordinates are determined by a weighted combination of the current iteration's calculated values ​​and historical center point coordinates. The smoothing factor is dynamically adjusted based on the number of iterations to balance convergence speed and stability. Grouping adjustment log generation supports configurable levels of verbosity. The verbosity level controls the granularity of log content recording; for example, recording the complete similarity matrix in debug mode and only recording a change summary in production mode. Log data is periodically archived to persistent storage for auditing purposes.

[0067] It is understandable that the initial grouping and multi-round iterative computation of the dynamic clustering engine constitute a closed-loop feedback system. The accuracy of the similarity matrix calculation directly affects the grouping quality; therefore, the normalization preprocessing of feature vectors is a necessary step to ensure that all feature dimensions are of the same magnitude. The construction of the initial group centroid set depends on the global characteristics of the similarity matrix. Matrix computation uses a distributed algorithm; when the data scale is large, the feature vectors are distributed across multiple computing nodes to compute similarity values ​​in parallel. The termination conditions for multi-round iterative computation, in addition to stability indicators, also include a maximum iteration limit to prevent infinite loops. During the iteration process, resource usage is monitored in real time and performance metrics are recorded.

[0068] It is understandable that the accuracy of group assignment adjustments depends on the update strategy of the centroid coordinates. The mean-based centroid calculation is sensitive to outliers; therefore, a robust method using the median instead of the mean is more suitable in certain scenarios. The median calculation uses a fast selection algorithm to improve efficiency. The group assignment log analysis function supports offline replay, and log data can be exported to analysis tools for visual inspection, helping to debug group boundary change trends. The overall iterative calculation process is designed to be interruptible and recoverable. If the system unexpectedly stops, it can be restarted from the most recent iteration checkpoint, reducing wasted computational resources.

[0069] Optionally, the similarity matrix calculation supports multiple similarity measurement algorithms. Besides cosine similarity, the inverse of Euclidean distance or Pearson correlation coefficient can also be configured. Algorithm selection is controlled by configuration file parameters, and the system dynamically loads the corresponding calculation module at runtime. The construction of the initial group centroid set allows for manual intervention. Training instructors can manually specify the initial centroids through a graphical interface. Manually specifying centroids involves selecting representative combat unit feature vectors based on domain knowledge. The group assignment adjustment process integrates an adaptive learning mechanism. The adjustment strategy is optimized based on historical iteration data; for example, reinforcement learning is used to dynamically adjust the similarity threshold to improve group consistency.

[0070] Optionally, the parallelization of multi-round iterative computation adopts a multi-threaded model, dividing the feature vector set into multiple blocks. Each thread processes the grouping and assignment adjustment of one block, and the threads synchronize the centroid coordinates through shared memory. The step of recalculating the centroid coordinates supports incremental updates. When new feature vectors flow in, the centroid coordinates are only locally updated based on the changed groupings, reducing computational overhead. The storage format of the grouping adjustment log supports compression and encryption. Log files are written using standard formats such as AVRO or Parquet to ensure data security and query efficiency. The log analysis interface provides an SQL-like query language for custom report generation.

[0071] See Figure 4This figure represents the core output of the dynamic clustering engine: using spatial and collaborative feature dimensions as coordinate systems, it displays the clustering distribution of feature vectors from 50 combat units. Different colored dots represent four levels of aggregation units, with black pentagrams serving as cluster centers, calculated iteratively from the feature vector similarity matrix. The core value of this figure lies in its ability, through multiple iterations of the dynamic clustering engine, to aggregate discrete combat units from training into spontaneously formed tactical clusters based on feature correlations. Each center point visually reflects the core characteristics of its corresponding unit. This result overcomes the limitations of traditional assessments that use fixed organizational structures as units, providing analytical units that align with actual training for subsequent multi-level assessment models, and more accurately reflecting the ad-hoc collaborative relationships between combat units.

[0072] Example 3: In specific implementation, the grouping stability index is calculated based on the intermediate grouping results generated by the dynamic clustering engine through multiple iterations. These intermediate grouping results are stored in an in-memory database as a grouping attribution mapping table, which records the grouping identifier corresponding to each feature vector in each iteration. The first step in the calculation process is to statistically analyze the magnitude of grouping attribution changes over multiple iterations. The magnitude of grouping attribution change is defined as the proportion of feature vectors whose grouping changes between two adjacent iterations relative to the total number of feature vectors. Specifically, the system maintains a fixed-length iteration window, for example, set to the size of the last five iterations. After each new iteration, the system scans the grouping attribution mapping table between the current iteration and previous iterations, comparing each feature vector's grouping identifier row by row to see if it has changed. The moving average of the grouping attribution change magnitude is used as a stability reference value using a weighted average algorithm. The moving average calculation assigns higher weight to recent iteration data, and the calculation formula is expressed as:

[0073] ;

[0074] in: This represents a stability reference value. This represents the weight coefficient for the i-th iteration, which increases as the iteration index increases. This represents the magnitude of the group assignment change detected in the i-th iteration. This represents the size of the iteration window. Comparing the stability reference value with the preset stability threshold is the core operation for determining the iteration termination condition. The stability threshold is preset by the system administrator according to the accuracy requirements of the training scenario, for example, set to 0.05. When the calculated stability reference value is less than or equal to the stability threshold, the dynamic clustering engine triggers the iteration termination process and outputs the current grouping result as the final grouping result.

[0075] In practical implementation, the standardized storage process for evaluating the data warehouse begins with receiving feature data sent by the dynamic clustering engine. This feature data includes identifiers of hierarchical aggregation units, feature vector values, and timestamp metadata. Format validation and integrity checks on the received feature data are necessary steps before storage. Format validation checks whether the data field type, length, and value range conform to a predefined pattern; for example, it verifies whether the feature vector dimension matches the data pattern definition and whether numerical feature values ​​are within a reasonable range. Integrity checks verify whether data blocks have been corrupted during transmission by calculating hash values ​​or checksums, and also check for null values ​​in required fields. The validated feature data is then converted to a standard data format using a format converter component. The standard data format is defined as a binary format of Avro or Protocol Buffers with a fixed pattern. The conversion process includes field renaming, data type conversion, and encoding compression to ensure that all incoming data has a unified structured representation. Building a multidimensional index structure based on time and hierarchical aggregation unit dimensions is key to optimizing query performance. The time-dimensional index uses millisecond-level timestamps as the primary key, while the hierarchical aggregation unit-dimensional index uses grouping identifiers as the primary key. The multidimensional index structure employs a hybrid implementation of B+ trees and bitmap indexes, supporting fast data retrieval based on time range and organizational structure. Standardized feature data is stored in corresponding data storage partitions based on a partitioning strategy. Data storage partitions are divided according to time intervals and hierarchical aggregation unit hash values, with each partition mapped to an independent physical storage file. The file format uses columnar storage to optimize analysis and query efficiency.

[0076] In some embodiments, the calculation process of the grouping stability index is integrated with a real-time monitoring dashboard. The dashboard visually displays the change in group affiliation values ​​and the trend of stability reference values ​​for each iteration, helping training instructors intuitively understand the clustering convergence state. The calculation of the moving average supports dynamic adjustment of the window size. The system automatically expands or shrinks the iteration window based on the volatility of historical iteration data. When data fluctuations are large, the window size is increased to smooth noise; when the data is stable, the window size is decreased to improve response speed. The comparison logic of the stability threshold includes a hysteresis mechanism to prevent frequent oscillations near the threshold from causing premature termination. Only when the stability reference value is below the threshold for three consecutive iterations is the iteration termination condition confirmed.

[0077] In some embodiments, the standardized storage process for evaluating the data warehouse includes data version management functionality. Each data update operation generates a new data version and retains a snapshot of the old version. The version number is stored in association with the data checksum. The format validation and integrity check module is implemented as a pluggable validation rule engine, allowing administrators to dynamically add custom validation rules, such as adding business rules to check the logical consistency between feature vectors. The construction process of the multidimensional index structure is executed in parallel with the data loading pipeline. Index construction uses an incremental update algorithm to avoid the performance overhead of rebuilding the index every time. The index file is periodically optimized to eliminate fragmentation. It can be understood that the calculation accuracy of the grouping stability index depends on the reasonable selection of the iteration window size. Too small a window may lead to oversensitivity to temporary fluctuations, while too large a window will delay the detection of the convergence state. The allocation strategy of the weight coefficients in the moving average calculation affects the sensitivity of the stability reference value. Linearly increasing weights are suitable for most scenarios, while exponentially increasing weights can emphasize recent trends. The setting of the stability threshold needs to balance clustering quality and computational efficiency. A lower threshold will produce more stable grouping results but increase the number of iterations, while a higher threshold can complete clustering quickly but may sacrifice grouping consistency.

[0078] It is understandable that standardized storage in an evaluation data warehouse is fundamental to ensuring the reliability of subsequent evaluation and analysis. Format validation effectively intercepts erroneous data from contaminating the warehouse content, while integrity checks prevent analytical biases caused by partial data loss. The use of standard data formats eliminates the heterogeneity of multi-source data, providing a consistent data interface for upper-level evaluation models. The multidimensional index structure reduces the complexity of linear scans to logarithmic levels, greatly improving the query efficiency of large-scale historical data. The data storage partitioning design supports horizontal scaling; as the data volume increases, capacity can be expanded by adding partition nodes.

[0079] Optionally, the group stability index calculation module supports switching between multiple moving average algorithms. Besides weighted moving average, simple moving average or exponential moving average algorithms can be selected, with algorithm selection controlled by configuration parameters. The stability threshold comparison operation allows setting different comparison operators; for example, it can be configured to terminate when the stability reference value is less than the threshold, or when the rate of change of the stability reference value is less than the threshold, adapting to different convergence criteria. The results of the iteration termination condition judgment are recorded in the audit log, including the termination time, the final number of iterations, and the stability index value, for subsequent clustering quality analysis.

[0080] Optionally, the standardized storage process of the evaluation data warehouse supports encrypted data storage. Sensitive data is encrypted using the AES-256 algorithm after being converted to a standard format, and the encryption keys are managed uniformly by a key management system. The data storage partitioning strategy supports custom partition keys; in addition to time and hierarchical aggregation unit dimensions, other dimensions such as data source and confidentiality level can be added as partitioning criteria. The storage system's metadata management maintains data lineage information, recording the source, transformation process, and quality indicators of each data block, achieving end-to-end data traceability. The warehouse system also provides data lifecycle management functions, automatically archiving historical data exceeding the retention period to low-cost storage media or securely deleting it.

[0081] Example 4: In practical implementation, the analysis and processing flow of the multi-level evaluation model begins with configuring the evaluation indicator system. This system comprises three levels: basic performance indicators, tactical coordination indicators, and combat effectiveness indicators. Basic performance indicators cover individual capability parameters such as mobility speed, strike accuracy, and reaction time for single equipment or service branches. Tactical coordination indicators measure collaboration parameters such as information sharing rate, action synchronization, and resource allocation efficiency between units. Combat effectiveness indicators comprehensively reflect overall effectiveness parameters such as mission completion rate, casualty exchange ratio, and battlefield control. The configuration process is completed through a graphical configuration interface. Administrators can drag and drop predefined indicator templates or define new indicators. Each indicator requires setting a name, data type, unit of measurement, and threshold range. The configuration results are saved in JSON format to the evaluation model configuration library. Feature data corresponding to the evaluation indicator system is extracted from the evaluation data warehouse using a distributed query engine. The query engine parses the evaluation indicator system to generate SQL query statements, which include time range filtering conditions, hierarchical aggregation unit grouping conditions, and feature field selection lists. The query results are returned to the multi-level evaluation model in the form of a data stream. The core processing step is to calculate the comprehensive evaluation value of each level of aggregation unit using a weighted fusion algorithm. The algorithm assigns weight coefficients to each evaluation indicator, which are determined using the Analytic Hierarchy Process (AHP). AHP calculates the relative importance between indicators by constructing a judgment matrix. The final weight values ​​take effect after a consistency check. The formula for calculating the comprehensive evaluation value is expressed as follows:

[0082] ;

[0083] in: The comprehensive evaluation value representing the hierarchical aggregation unit. The weight coefficient representing the i-th evaluation indicator. Represents the normalized value of the i-th evaluation indicator. This represents the total number of evaluation indicators. Generating an intermediate evaluation data table containing the evaluation results of each level of aggregation unit is the output step. The intermediate evaluation data table is stored using a relational database table structure (see Table 1). The table structure includes fields such as the hierarchical aggregation unit identifier, evaluation timestamp, original value of each indicator, normalized value of each indicator, and comprehensive evaluation value. A composite index based on the hierarchical aggregation unit identifier and timestamp is established in the data table to support fast querying.

[0084] Table 1: Intermediate Evaluation Data Table

[0085] In practical implementation, the evaluation indicator system supports a dynamic update mechanism. During simulation training, administrators can adjust indicator weights or add / delete indicator items based on real-time situations. Adjustments are implemented through version control, with each configuration version bearing an effective timestamp and operator identifier. Data quality checks are performed when extracting feature data from the evaluation data warehouse, including data integrity verification, outlier detection, and timeliness verification. Missing data is handled using interpolation or default value strategies, and outlier data is marked and stored separately for subsequent analysis. The weighted fusion algorithm supports multiple preset weighting schemes, allowing the system to quickly switch between them based on different training scenarios. For example, the offensive scenario weighting scheme emphasizes strike accuracy and maneuver speed, while the defensive scenario weighting scheme focuses on information sharing and reaction time. Intermediate evaluation data tables are generated using a batch processing mode, triggering batch calculations every five minutes or when data accumulates to a certain scale. The calculation process is accelerated using a memory computing engine, and the data tables are periodically archived to a historical evaluation database for trend analysis.

[0086] In some embodiments, the analysis and processing of the multi-level evaluation model integrates a real-time feedback loop. The evaluation results are transmitted back to the dynamic clustering engine as optimization parameters. When the evaluation finds abnormal performance of certain hierarchical aggregation units, the similarity calculation weights of the clustering algorithm are automatically adjusted. The configuration interface of the evaluation index system provides an index correlation analysis function. The system automatically calculates the correlation coefficients between indicators and displays them visually, helping administrators optimize index selection and avoid duplicate evaluations. The calculation process of the weighted fusion algorithm supports multi-dimensional verification, including weight and sum verification, result range verification, and sensitivity analysis, to ensure the scientificity and stability of the evaluation results. The intermediate evaluation data table is stored in a columnar storage format to optimize query performance, and an evaluation result snapshot mechanism is established to support the retrospective and comparative analysis of evaluation results at any point in time.

[0087] It is understandable that the analytical processing effectiveness of a multi-level evaluation model directly depends on the scientific nature of the evaluation indicator system. Basic performance indicators need to cover the basic tactical and technical performance of the equipment, tactical coordination indicators should reflect the system-of-systems combat capability under informationized conditions, and combat effectiveness indicators should reflect the comprehensive effect of realistic training. The determination of weight coefficients in the weighted fusion algorithm needs to combine expert experience and historical data, and the judgment matrix of the analytic hierarchy process (AHP) should be independently filled out by experts from multiple fields and then comprehensively processed. The structural design of the intermediate evaluation data table needs to consider the needs of subsequent visualization, and field settings should balance machine readability and human readability; numerical fields should retain appropriate precision.

[0088] Optionally, the evaluation index system configuration supports template import and export functions, allowing mature index systems to be saved as template files for quick reuse in different training scenarios. When extracting feature data from the evaluation data warehouse, data sampling functionality is supported; when the data volume is too large, random sampling or stratified sampling methods can be used to improve processing efficiency. The weighted fusion algorithm can be configured with different normalization methods, including min-max normalization, Z-score standardization, etc., to adapt to the data distribution characteristics of different indicators.

[0089] See Figure 5 This chart serves as an auxiliary analytical tool for the evaluation indicator system. Using six core evaluation indicators as dimensions, it employs color depth to represent the correlation coefficients between indicators: darker colors indicate stronger correlations, while lighter colors indicate weaker correlations. The chart clarifies the inherent logical connections between the evaluation indicators. The high correlation between maneuver speed and mission completion, and between strike accuracy and mission completion, reflects the direct impact of basic performance on operational effectiveness. Meanwhile, the strong correlation between information sharing rate and operational synchronization reflects the synergistic nature of collaborative indicators. This analysis provides data support for the weighting of indicators in multi-level evaluation models, avoiding indicator duplication or imbalanced weight allocation, and enhancing the scientific rigor and rationality of the evaluation system.

[0090] Example 5: In specific implementation, the generation process of the visualization assessment report begins with extracting key assessment indicator data from the intermediate assessment data table output by the multi-level assessment model. This key assessment indicator data includes the comprehensive assessment value of the hierarchical aggregation unit, basic performance indicator values, tactical coordination indicator values, and combat effectiveness indicator values, as well as the trend data of these values ​​over time. The extraction operation is performed using a structured query language, with query conditions including time range limitations and hierarchical aggregation unit identifier filtering. The extracted data is temporarily stored in an in-memory data structure for preprocessing. Designing visualization chart templates is the core step in report generation. These templates include situational distribution maps, performance comparison charts, and trend analysis. Figure 3The system comprises three basic types: a situational distribution map, which uses a two-dimensional grid coordinate system overlaid with geographic information layers to display the position and status of different levels of aggregated units in the simulated battlefield space; a performance comparison map, which uses a rectangular tree diagram or parallel coordinate diagram to achieve intuitive comparison of multi-dimensional indicators; and a trend analysis map, which uses a stacked area diagram or line chart to reflect the changing patterns of evaluation indicators over time. Chart templates are defined using vector graphics and include elements such as axes, legends, color mapping rules, and interactive controls. Populating key evaluation indicator data into the visualization chart template to generate chart elements is an automated process. The data filling engine maps data fields to graphic attributes according to the binding rules defined in the template; for example, it maps the comprehensive evaluation value to the radius of the bubble chart and the tactical coordination indicator value to color depth. During the filling process, data scaling and labeling are performed to adapt to the visualization space. Combining the various chart elements to form a complete visualization evaluation report document uses a report compositor component. The report compositor arranges the generated situational distribution map, performance comparison map, and trend analysis map on the report page according to a preset layout, and adds text elements such as titles, summaries, explanatory text, and generation timestamps. The final output visualization evaluation report document is in PDF / A standard format to ensure long-term readability.

[0091] In practical implementation, the push process of the command terminal display interface begins with the format conversion of the visualization assessment report document. The visualization assessment report document is converted into a display data package of a specified format using a document converter module. The specified format is a lightweight data exchange format based on JSON. The conversion process includes extracting vector graphic path data, text content, and style information from the PDF document and re-encoding it into a layered display data package containing a metadata layer, a graphic element layer, and an interaction logic layer. The display data package is sent to the command terminal via an encrypted transmission channel using a secure communication protocol with two-way authentication. The encrypted transmission channel is established based on the TLS 1.3 protocol. Before transmission, the display data package is encrypted in AES-256-GCM mode and a digital signature is attached to the header. Upon receiving the data, the command terminal verifies the integrity of the signature before decrypting the data. The command terminal parses the display data package and renders the visualization interface using its built-in rendering engine. The parsing process includes decompressing the data stream, verifying the integrity of the data structure, and loading resource files. The rendering engine draws the chart coordinate system and graphic primitives on the terminal screen based on the description information in the display data package and fills them with color textures. Simultaneously, it initializes interactive functions such as zooming, panning, and tooltips. The evaluation data in the visualization interface is updated in real time through an incremental update mechanism. When a new display data packet is received, the rendering engine compares the differences between the old and new data packets and only redraws the parts that have changed. The data update frequency is synchronized with the evaluation cycle of the simulation training system to ensure the consistency between the interface content and the background data.

[0092] In some embodiments, the generation of visualization assessment reports supports multi-version management. The system can generate report versions with different levels of detail, such as detailed version, summary version, and thematic version, for the same assessment period. The version differences lie in the number and level of detail of the included charts, and users can choose to view them as needed. The style rules of the visualization chart templates are defined through CSS-like style sheets, allowing users to customize color themes, font sizes, and layouts. The separation of style sheets from data logic facilitates maintenance and updates. The mapping rules during the data population process support formula definition; for example, the radius ratio of a bubble chart can be calculated using the following formula:

[0093] ;

[0094] in: This represents the final rendered bubble radius. Represents the reference radius constant. This represents the comprehensive evaluation value of the current level aggregation unit. This represents the maximum overall evaluation value across all units. The report synthesizer supports pagination logic, automatically paginating when the number of charts exceeds the single-page capacity and generating a page navigation table of contents.

[0095] In some embodiments, the command terminal display interface pushes data using an asynchronous communication mode. The sending priority of display data packets is configurable; high-priority data packets can interrupt the transmission of low-priority packets to ensure the timely delivery of critical information. The encrypted transmission channel supports adaptive bandwidth adjustment, automatically reducing data transmission resolution or employing lossy compression algorithms when network bandwidth is limited. The terminal rendering engine is implemented as a cross-platform component, running on multiple operating systems such as Windows, Linux, and Android, and is optimized for touchscreen operation. The real-time update mechanism supports a data subscription mode, allowing the command terminal to specify receiving only evaluation data updates from specific types or levels of aggregation units, reducing unnecessary network transmission and rendering overhead.

[0096] Understandably, the quality of the generated visualization assessment report depends on the scientific nature of the chart template design. Situational distribution maps need to accurately reflect battlefield spatial relationships, performance comparison charts should highlight key differences, and trend analysis charts must clearly demonstrate patterns of change. The structural design of the displayed data packets affects parsing efficiency; a reasonable layered structure can accelerate the rendering engine's parsing process. The security of the encrypted transmission channel is crucial, requiring regular updates to the encryption algorithm and keys to address potential security threats. The performance of the real-time update function depends on the efficiency of the incremental calculation algorithm; an efficient difference comparison algorithm can minimize the area requiring interface redrawing.

[0097] Optionally, the visualization assessment report generation system provides a template editor tool, allowing users to customize chart layouts and styles through drag-and-drop. The template editor has a built-in preview function that displays the design effect in real time. The data package conversion process supports a caching mechanism, allowing unmodified chart elements to directly reuse previous conversion results to improve processing speed. The command terminal rendering engine supports offline mode, enabling the interface to continue displaying based on locally cached data when the network is interrupted, and automatically synchronizing the latest data after the network is restored. The real-time update function can be configured to be manually triggered or automatically pushed to meet the needs of different scenarios.

[0098] It is understandable that the generation of the visual assessment report and its delivery to the command terminal display interface constitute a complete information delivery chain. The report generation side needs to fully consider the display capabilities and interactive characteristics of the terminal devices, while the terminal side needs to have efficient parsing and rendering capabilities to ensure user experience. The entire implementation process requires a rigorous quality control mechanism, including verification of chart accuracy, verification of data transmission integrity, and testing of interface rendering effects. The system should have good scalability to adapt to future additions of assessment indicators and visualization needs.

[0099] Optionally, the visualization assessment report can be output as an interactive HTML5 format, supporting direct viewing in a web browser and providing dynamic filtering and drill-down functions. The transmission of display data packets supports multicast technology, reducing network load when multiple command terminals need to receive the same data. The command terminal rendering engine integrates a performance monitoring module, monitoring the interface rendering frame rate and response latency in real time, and automatically simplifying visualization effects when performance degrades. The system provides complete logging capabilities, tracking the entire process from report generation to terminal display, facilitating troubleshooting and performance optimization.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-dimensional based hierarchical aggregation campaign simulation training evaluation method, characterized in that, The method comprises: establishing a data connection channel with each combat unit in the simulation training system, and receiving raw behavior data streams generated during the operation of each combat unit in real time; performing feature extraction on the received raw behavior data streams, and identifying spatial coordinate features, time sequence features, equipment interaction features and tactical action features contained therein; constructing a feature vector set according to the extracted multi-class features, and inputting the feature vector set into a dynamic clustering engine for initialization grouping; performing multi-round iteration calculation on the feature vector set by the dynamic clustering engine, adjusting the grouping boundary and generating an intermediate grouping result in each iteration process; calculating a grouping stability index based on the intermediate grouping result, terminating the iteration when the grouping stability index reaches a preset threshold, and outputting the final grouping result as a hierarchical aggregation unit; sending the feature data of each hierarchical aggregation unit to an evaluation data warehouse for standardized storage; retrieving the standardized stored feature data from the evaluation data warehouse and inputting it into a multi-level evaluation model for analysis and processing; generating evaluation index data for each hierarchical aggregation unit through the multi-level evaluation model; generating a visual evaluation report according to the evaluation index data, and pushing the visual evaluation report to the command terminal display interface.

2. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method according to claim 1, wherein, The establishment of the data connection channel with each combat unit in the simulation training system comprises: configuring network communication protocol parameters, and establishing a bidirectional data transmission link with each combat unit; setting data collection frequency parameters, and receiving raw behavior data streams uploaded by each combat unit at a set time interval; adding time labels and source identifiers to the raw behavior data streams to form a set of raw data with space-time markers.

3. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 2, wherein, The feature extraction operation on the received raw behavior data streams comprises: separating the position coordinate sequence from the set of raw data with space-time markers, and extracting the change rule of the position coordinate sequence as the spatial motion feature; parsing the equipment state change record from the raw data set as the equipment interaction feature, and identifying the pattern feature in the tactical instruction execution record as the tactical action feature.

4. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method according to claim 3, wherein, The initialization grouping of the dynamic clustering engine comprises: calculating the similarity matrix of each feature vector in the feature vector set, and constructing an initial grouping center point set according to the similarity matrix; dividing the feature vector set into an initial grouping set based on the initial grouping center point set.

5. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 4, wherein, The multi-round iteration calculation comprises: recomputing the center point coordinates of each grouping in each iteration process, adjusting the grouping attribution of each feature vector according to the recomputed center point coordinates, recording the grouping attribution change in each iteration process, and generating a grouping adjustment log.

6. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method according to claim 5, wherein, The calculation of the grouping stability index comprises: statistically calculating the amplitude value of the grouping attribution change in continuous multiple iteration processes, calculating the moving average value of the grouping attribution change amplitude value as a stability reference value, comparing the stability reference value with a preset stability threshold, and judging the iteration termination condition.

7. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 1, wherein, The standardized storage of the evaluation data warehouse comprises: performing format verification and integrity check on the received feature data, and converting the feature data that passes the verification into a standard data format; A multi-dimensional index structure is established according to a time dimension and a hierarchical aggregation unit dimension, and the standardized feature data is stored in a corresponding data storage partition.

8. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 7, wherein, The analysis processing of the multi-level evaluation model comprises: An evaluation index system is configured, including basic performance indexes, tactical coordination indexes and combat effectiveness indexes; Feature data corresponding to the evaluation index system is extracted from an evaluation data warehouse, a weighted fusion algorithm is used to calculate comprehensive evaluation values of hierarchical aggregation units, and an intermediate evaluation data table containing evaluation results of the hierarchical aggregation units is generated.

9. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 8, wherein, The generation of the visual evaluation report comprises: Key evaluation index data is extracted from the intermediate evaluation data table, a visual chart template is designed, including a situation distribution chart, a performance comparison chart and a trend analysis chart; Chart elements are generated by filling the key evaluation index data into the visual chart template, and complete visual evaluation report documents are formed by combining the chart elements.

10. The multi-dimensional based hierarchical aggregated campaign simulation training evaluation method of claim 9, wherein, The pushing of the command terminal display interface comprises: The visual evaluation report document is converted into a display data packet in a specified format, the display data packet is sent to the command terminal through an encrypted transmission channel; The display data packet is parsed at the command terminal and the visual interface is rendered, and the evaluation data content in the visual interface is updated in real time.

Images