Piloting training effect evaluation system and method based on multi-source data fusion

Abstract

The invention relates to the technical field of pilot training effect evaluation, in particular to a pilot training effect evaluation system and method based on multi-source data fusion. A pilot training effect evaluation system based on multi-source data fusion comprises a data acquisition module, a preprocessing module, a data fusion module, an evaluation model module and a feedback optimization module, the data acquisition module comprises a multi-modal sensor, the preprocessing module is in communication connection with the data acquisition module, and the data fusion module is connected with the preprocessing module; the evaluation model module is connected with the data fusion module and outputs a comprehensive evaluation score and a weak link diagnosis result; and the feedback optimization module is connected with the evaluation model module to generate a visual report and an optimization instruction. According to the method, high-precision processing and self-adaptive fusion of multi-source data are realized through multi-modal sensor synchronous acquisition, refined preprocessing, dynamic weight fusion and intelligent model evaluation, a closed loop is formed in combination with visual feedback, and training evaluation accuracy and optimization efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pilot training effect evaluation, and in particular to a pilot training effect evaluation system and method based on multi-source data fusion. Background Art

[0002] With the specialization and complexity of pilot training, higher requirements are placed on the scientific evaluation of training effects. Early evaluation methods mainly relied on single-dimensional data, such as operational behavior records or simple physiological indicators, which could not fully reflect the comprehensive performance of the trainees. With the development of sensor technology, the Internet of Things and data processing algorithms, multi-source data fusion technology has gradually been applied to the field of training evaluation. By integrating multimodal data such as physiological monitoring, operational behavior, and environmental perception, a more three-dimensional evaluation system is constructed. Traditional solutions have initially achieved the collection and simple fusion of multi-source data, such as integrating different data sources through fixed weights. However, there are still technical bottlenecks in the accuracy of data preprocessing, the dynamic adaptability of the fusion algorithm, the real-time update capability of the evaluation model, and the intuitiveness of the feedback mechanism.

[0003] The shortcomings of existing technologies are mainly reflected in the following aspects: First, the spatiotemporal alignment and noise reduction processing of multi-source data are not refined enough. The time synchronization errors and spatial scale differences of different sensors may cause data distortion, affecting the accuracy of subsequent analysis. Second, the weight allocation in the data fusion process is often based on empirical settings, and the dynamic changes in data quality and task relevance are not fully considered. This may cause low-quality or low-relevance data to interfere with the evaluation results. Third, the evaluation model often relies on offline training, which makes it difficult to adapt to the data distribution drift caused by changes in training scenarios. Long-term use is prone to a decrease in evaluation accuracy. Fourth, the feedback form of evaluation results is relatively simple, lacking visual positioning of weak links in training and feature importance analysis, making it difficult to effectively guide training optimization. These problems limit the objectivity, real-time nature, and practicality of existing evaluation systems. More efficient data processing and intelligent evaluation technologies are urgently needed to improve the scientific nature of training effect evaluation. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above problems and provide a pilot training effect evaluation system and method based on multi-source data fusion. To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] A pilot training effect evaluation system based on multi-source data fusion includes a data acquisition module, a preprocessing module, a data fusion module, an evaluation model module and a feedback optimization module. The data acquisition module includes a multimodal sensor and collects physiological monitoring data, operational behavior data, environmental perception data and task completion index data in real time; the preprocessing module is communicatively connected to the data acquisition module and performs noise reduction, normalization and time-space alignment processing on the multi-source data to generate a standardized time series data set; the data fusion module is connected to the preprocessing module and fuses the standardized data into a multidimensional feature vector based on a dynamic weight allocation algorithm; the evaluation model module is connected to the data fusion module and outputs a comprehensive evaluation score and weak link diagnosis results; the feedback optimization module is connected to the evaluation model module and generates a visualization report and optimization instructions.

[0006] Furthermore, the weight distribution algorithm of the data fusion module satisfies:

[0007]

[0008] Among them, W i is the weight value of the i-th category data, S i Score for signal-to-noise ratio, D i is the correlation coefficient with the task completion index, α and β are adjustment factors and satisfy α + β = 1, and n is the total number of data categories.

[0009] Furthermore, the spatiotemporal alignment of the preprocessing module includes timestamp calibration and spatial normalization.

[0010] The timestamp calibration is based on the timestamp of the operation behavior data and performs cubic spline interpolation on the physiological monitoring data. The interpolation formula is:

[0011]

[0012] Among them, t j is the time of adjacent sampling points, and the coefficients a, b, c, and d are determined by the interpolation point data;

[0013] The spatial normalization maps the spatial coordinates of the environmental perception data to a uniform range, and the calculation formula is:

[0014]

[0015] Among them, x min ,x max ,y min ,y max are the boundary coordinates of the environment space.

[0016] Furthermore, the evaluation model module includes an offline training unit and an online updating unit.

[0017] The offline training unit uses expert scoring label data to train the gradient boosting decision tree model, and its objective function is:

[0018]

[0019] Among them, y k Rating labels for experts, is the model prediction value, λ is the regularization coefficient, and ||w||1 is the L1 norm of the model weight parameter;

[0020] In the online update unit, when the new input data distribution P new With historical data distribution P old The KL divergence of triggers the model update when it meets the following conditions:

[0021]

[0022] Wherein, θ is the preset threshold.

[0023] Furthermore, the data acquisition module includes a multi-protocol communication interface and a clock synchronization unit. The multi-protocol communication interface integrates Bluetooth 5.0, CAN bus and RS-485 interfaces, and synchronously collects data with periods of 10ms, 5ms and 20ms respectively; the clock synchronization unit adopts the IEEE 1588 protocol to achieve multi-source data synchronization, and the synchronization error is less than 5ms.

[0024] Furthermore, a method for evaluating the effectiveness of pilot training based on multi-source data fusion includes the following steps:

[0025] Step S1: synchronously collecting training data through multimodal sensors;

[0026] Step S2, performing preprocessing on the original data;

[0027] Step S3: fusing multi-source data based on a dynamic weight allocation algorithm to generate a multi-dimensional feature vector;

[0028] Step S4: Input the fusion features into the training effect evaluation model and output the comprehensive score and weak link diagnosis results;

[0029] Step S5: Generate a visualization report and feed it back to the training optimization system.

[0030] Furthermore, in step S2, the preprocessing includes:

[0031] Wavelet packet decomposition is used to reduce noise of physiological signals, and the noise threshold is calculated as:

[0032]

[0033] Where σ is the estimated value of the noise standard deviation and N is the signal length;

[0034] Sliding window mean filtering is used for the operation behavior data, and the window length L is dynamically adjusted to:

[0035]

[0036] Among them, f s is the sampling rate, f max The highest frequency of the effective signal.

[0037] Furthermore, in step S5, the visualization report generation includes: a spatiotemporal trajectory heat map, marking the sections where the path deviates by more than 0.5m; a time-series synchronization comparison curve of physiological indicators and operational behaviors; and a feature importance analysis map based on SHAP values, wherein the SHAP value calculation satisfies:

[0038]

[0039] Where N is the total number of features and f is the model prediction function.

[0040] The advantages of the present invention are:

[0041] 1. This invention achieves high-frequency synchronous acquisition of multi-source data, including physiological, operational, and environmental data, through the integrated design of multimodal sensors and multi-protocol communication interfaces, combined with the IEEE 1588 clock synchronization protocol. In the preprocessing stage, wavelet packet decomposition and noise reduction, dynamic sliding window filtering, and spatiotemporal alignment techniques are used to eliminate data noise, spatiotemporal deviations, and scale differences, generating a standardized time series dataset. This provides a high-precision data foundation for subsequent fusion analysis and resolves the evaluation errors caused by uneven data quality and spatiotemporal misalignment in traditional solutions.

[0042] 2. The present invention adopts a dynamic weight allocation algorithm based on signal-to-noise ratio and task relevance, adaptively adjusting the weight of each data source according to data quality and task relevance, thereby avoiding the interference of low-quality or irrelevant data on the evaluation results under fixed weights; the fused multi-dimensional feature vector is input into the gradient boosting decision tree model, combined with L1 regularization objective function optimization, to improve the model's nonlinear fitting ability for training effects, and achieve accurate output of comprehensive evaluation scores and weak link diagnosis. Compared with the traditional fixed weight fusion method, it significantly enhances the reliability of feature representation and the accuracy of the evaluation model.

[0043] 3. The present invention constructs an evaluation model architecture that includes offline training and online updating. In the offline stage, the initial model is trained using expert scoring data. In the online stage, the KL divergence is used to detect data distribution changes in real time and initiate incremental learning to ensure that the model dynamically adapts to changes in training scenarios. The feedback optimization module generates spatiotemporal trajectory heat maps, time series comparison curves, and SHAP value feature importance maps, converting the evaluation results into visual and interpretable decision-making information, realizing closed-loop feedback from data collection to training optimization, solving the problems of easy failure of traditional model offline training and the single feedback form, and improving the practicality and intelligence of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings constituting a part of this application are used to provide a further understanding of this application and make other features, purposes and advantages of this application more apparent. The drawings and descriptions of the exemplary embodiments of this application are used to explain this application and do not constitute an improper limitation on this application.

[0045] In the attached figure:

[0046] Figure 1 This is a system architecture diagram of a navigation training effect evaluation system based on multi-source data fusion in Example 1.

[0047] Figure 2 This is a flow chart of a method for evaluating the effectiveness of pilot training based on multi-source data fusion in Example 1. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0049] The present invention is described in detail and specifically below through specific examples to provide a better understanding of the present invention. However, the following examples do not limit the scope of protection of the present invention.

[0050] Example 1

[0051] like Figure 1-2As shown, a pilot training effect evaluation system based on multi-source data fusion includes a data acquisition module, a preprocessing module, a data fusion module, an evaluation model module and a feedback optimization module. The data acquisition module includes a multimodal sensor, which collects physiological monitoring data, operational behavior data, environmental perception data and task completion index data in real time; the preprocessing module is communicated with the data acquisition module, and performs noise reduction, normalization and time-space alignment processing on the multi-source data to generate a standardized time series data set; the data fusion module is connected to the preprocessing module, and fuses the standardized data into a multi-dimensional feature vector based on a dynamic weight allocation algorithm; the evaluation model module is connected to the data fusion module, and outputs a comprehensive evaluation score and weak link diagnosis results; the feedback optimization module is connected to the evaluation model module, and generates a visualization report and optimization instructions.

[0052] The system consists of five modules: data acquisition, preprocessing, data fusion, evaluation model, and feedback optimization. The data acquisition module uses multimodal sensors to acquire physiological monitoring data, operational behavior data, environmental perception data, and task completion indicator data in real time to ensure the comprehensiveness of data sources. The preprocessing module performs noise reduction, normalization, and spatiotemporal alignment on the raw data to eliminate noise interference and unify the data scale, generating a standardized time series data set. The data fusion module uses a dynamic weight allocation algorithm to integrate multi-source data to form a multi-dimensional feature vector, enhancing data representation capabilities. The evaluation model module outputs a comprehensive score and weak link diagnosis based on the fused features to accurately locate training problems. The feedback optimization module generates visual reports and optimization instructions to form a closed-loop feedback link. The modules are connected in series to achieve full-process automation from data acquisition to result feedback, significantly improving the objectivity and systematicity of training effect evaluation.

[0053] Furthermore, the weight distribution algorithm of the data fusion module satisfies:

[0054]

[0055] Among them, W i is the weight value of the i-th category data, S i Score for signal-to-noise ratio, D i is the correlation coefficient with the task completion index, α and β are adjustment factors and satisfy α + β = 1, and n is the total number of data categories.

[0056] By weighting the signal-to-noise ratio score and task relevance coefficient, the weight of each data source is dynamically adjusted. The α and β in the formula are adjustment factors, ensuring that data with high signal-to-noise ratios or high relevance dominates the fusion process. This algorithm effectively addresses the problem of uneven quality of multi-source data, preventing low-quality or low-relevance data from interfering with the evaluation results, and improving the reliability of the fused feature vector and the accuracy of the evaluation model.

[0057] Furthermore, the spatiotemporal alignment of the preprocessing module includes timestamp calibration and spatial normalization.

[0058] The timestamp calibration is based on the timestamp of the operation behavior data and performs cubic spline interpolation on the physiological monitoring data. The interpolation formula is:

[0059]

[0060] Among them, t j is the time of adjacent sampling points, and the coefficients a, b, c, and d are determined by the interpolation point data;

[0061] The spatial normalization maps the spatial coordinates of the environmental perception data to a uniform range, and the calculation formula is:

[0062]

[0063] Among them, x min ,x max ,y min ,y max are the boundary coordinates of the environment space.

[0064] Timestamp calibration uses cubic spline interpolation to align physiological data based on operational data, ensuring temporal consistency across data with varying sampling frequencies. Spatial normalization maps environmental perception data to a unified coordinate system, eliminating differences in spatial scale. Through mathematical constraints imposed by interpolation and normalization formulas, precise alignment of multi-source data in both temporal and spatial dimensions is ensured, providing standardized input for subsequent data fusion and model analysis, avoiding evaluation errors caused by temporal and spatial deviations.

[0065] Furthermore, the evaluation model module includes an offline training unit and an online updating unit.

[0066] The offline training unit uses expert scoring label data to train the gradient boosting decision tree model, and its objective function is:

[0067]

[0068] Among them, y k Rating labels for experts, is the model prediction value, λ is the regularization coefficient, and ||w||1 is the L1 norm of the model weight parameter;

[0069] In the online update unit, when the new input data distribution P new With historical data distribution P old The KL divergence of triggers the model update when it meets the following conditions:

[0070]

[0071] Wherein, θ is the preset threshold.

[0072] The offline training unit uses a gradient boosting decision tree model, combined with L1 regularization to prevent overfitting and ensure model generalization. The online update unit detects changes in data distribution using KL divergence. When the new data distribution deviates significantly from historical data, it triggers incremental model learning, dynamically adapting to changes in the training environment. This design ensures that the evaluation model combines initial training accuracy with continuous optimization capabilities, ensuring the timeliness and adaptability of evaluation results.

[0073] Furthermore, the data acquisition module includes a multi-protocol communication interface and a clock synchronization unit. The multi-protocol communication interface integrates Bluetooth 5.0, CAN bus and RS-485 interfaces, and synchronously collects data with periods of 10ms, 5ms and 20ms respectively; the clock synchronization unit adopts the IEEE 1588 protocol to achieve multi-source data synchronization, and the synchronization error is less than 5ms.

[0074] The multi-protocol communication interface integrates Bluetooth, CAN bus, and RS-485, synchronously collecting physiological, operational, and environmental data at 10ms, 5ms, and 20ms cycles, respectively, meeting the real-time requirements of multimodal sensors. The clock synchronization unit utilizes the IEEE 1588 protocol, limiting multi-source data synchronization errors to less than 5ms and ensuring data temporal consistency. The collaborative design of hardware protocols and synchronization algorithms resolves the issue of timing misalignment in multi-device data collection, providing a high-precision data foundation for subsequent processing.

[0075] Furthermore, a method for evaluating the effectiveness of pilot training based on multi-source data fusion includes the following steps:

[0076] Step S1: synchronously collecting training data through multimodal sensors;

[0077] Step S2, performing preprocessing on the original data;

[0078] Step S3: fusing multi-source data based on a dynamic weight allocation algorithm to generate a multi-dimensional feature vector;

[0079] Step S4: Input the fusion features into the training effect evaluation model and output the comprehensive score and weak link diagnosis results;

[0080] Step S5: Generate a visualization report and feed it back to the training optimization system.

[0081] This method encompasses five steps: data acquisition, preprocessing, fusion, evaluation, and feedback. Training data is acquired synchronously through multimodal sensors. Preprocessing eliminates noise and unifies data scales. Dynamic weight fusion enhances feature expression. Model evaluation outputs comprehensive scores and diagnostic results, ultimately generating a visual report to guide training optimization. This method achieves end-to-end processing from raw data to decision recommendations, significantly improving the efficiency and scientific nature of training effect analysis.

[0082] Furthermore, in step S2, the preprocessing includes:

[0083] Wavelet packet decomposition is used to reduce noise of physiological signals, and the noise threshold is calculated as:

[0084]

[0085] Where σ is the estimated value of the noise standard deviation and N is the signal length;

[0086] Sliding window mean filtering is used for the operation behavior data, and the window length L is dynamically adjusted to:

[0087]

[0088] Among them, f s is the sampling rate, f max The highest frequency of the effective signal.

[0089] Wavelet packet decomposition is used to reduce noise in physiological signals, and a noise threshold formula is used to filter out high-frequency interference, preserving valid physiological features. Sliding window mean filtering is applied to behavioral data, with the window length dynamically adjusted based on the signal spectrum, balancing denoising effectiveness with preservation of signal detail. These two methods, each designed for different data characteristics, balance noise reduction accuracy and computational efficiency, providing high-quality data input for subsequent analysis.

[0090] Furthermore, in step S5, the visualization report generation includes: a spatiotemporal trajectory heat map, marking the sections where the path deviates by more than 0.5m; a time-series synchronization comparison curve of physiological indicators and operational behaviors; and a feature importance analysis map based on SHAP values, wherein the SHAP value calculation satisfies:

[0091]

[0092] Where N is the total number of features and f is the model prediction function.

[0093] The spatiotemporal trajectory heat map marks areas of path deviation, visually demonstrating the differences between the operation trajectory and the standard path. A time-series comparison curve of physiological indicators and operation behavior reveals the correlation between the two. A feature importance map based on SHAP values quantifies the impact of various factors on the evaluation results. This multi-dimensional visualization helps users quickly identify weak links in training, provides data support for optimization strategies, and enhances the interpretability and practicality of evaluation results.

[0094] While the specific embodiments of the present invention have been described in detail above, they are merely exemplary, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions of the present invention are also within the scope of the present invention. Therefore, equivalent changes and modifications made without departing from the spirit and scope of the present invention are intended to be encompassed within the scope of the present invention.

Claims

1. A pilot training effectiveness evaluation system based on multi-source data fusion, characterized by: It includes a data acquisition module, a preprocessing module, a data fusion module, an evaluation model module and a feedback optimization module. The data acquisition module contains a multimodal sensor, which collects physiological monitoring data, operational behavior data, environmental perception data and task completion indicator data in real time; the preprocessing module is communicated with the data acquisition module, performs noise reduction, normalization and time-space alignment processing on multi-source data, and generates a standardized time series data set; the data fusion module is connected to the preprocessing module, and fuses the standardized data into a multi-dimensional feature vector based on a dynamic weight allocation algorithm; the evaluation model module is connected to the data fusion module, and outputs a comprehensive evaluation score and weak link diagnosis results; the feedback optimization module is connected to the evaluation model module, and generates a visualization report and optimization instructions.

2. The pilot training effect evaluation system based on multi-source data fusion according to claim 1, characterized in that: The weight distribution algorithm of the data fusion module satisfies: Among them, W i is the weight value of the i-th category data, S i Score for signal-to-noise ratio, D i is the correlation coefficient with the task completion index, α and β are adjustment factors and satisfy α + β = 1, and n is the total number of data categories.

3. The pilot training effect evaluation system based on multi-source data fusion according to claim 2, characterized in that: The spatiotemporal alignment of the preprocessing module includes timestamp calibration and spatial normalization. The timestamp calibration is based on the timestamp of the operation behavior data and performs cubic spline interpolation on the physiological monitoring data. The interpolation formula is: Among them, t j is the time of adjacent sampling points, and the coefficients a, b, c, and d are determined by the interpolation point data; The spatial normalization maps the spatial coordinates of the environmental perception data to a uniform range, and the calculation formula is: Among them, x min ,x max ,y min ,y max are the boundary coordinates of the environment space.

4. The pilot training effect evaluation system based on multi-source data fusion according to claim 3 is characterized by: The evaluation model module includes an offline training unit and an online updating unit. The offline training unit uses expert scoring label data to train the gradient boosting decision tree model, and its objective function is: Among them, y k Rating labels for experts, is the model prediction value, λ is the regularization coefficient, and ||w||1 is the L1 norm of the model weight parameter; In the online update unit, when the new input data distribution P new With historical data distribution P old The KL divergence of triggers the model update when it meets the following conditions: Wherein, θ is the preset threshold.

5. The pilot training effect evaluation system based on multi-source data fusion according to claim 4 is characterized in that: The data acquisition module includes a multi-protocol communication interface and a clock synchronization unit. The multi-protocol communication interface integrates Bluetooth 5.0, CAN bus and RS-485 interfaces, and synchronously collects data with periods of 10ms, 5ms and 20ms respectively; the clock synchronization unit uses the IEEE 1588 protocol to achieve multi-source data synchronization, and the synchronization error is less than 5ms.

6. A method for evaluating the effect of pilot training based on multi-source data fusion, used to implement the system for evaluating the effect of pilot training based on multi-source data fusion according to claims 1-5, characterized in that: The following steps are involved: Step S1: synchronously collecting training data through multimodal sensors; Step S2, performing preprocessing on the original data; Step S3: fusing multi-source data based on a dynamic weight allocation algorithm to generate a multi-dimensional feature vector; Step S4: Input the fusion features into the training effect evaluation model and output the comprehensive score and weak link diagnosis results; Step S5: Generate a visualization report and feed it back to the training optimization system.

7. The method for evaluating the effectiveness of pilot training based on multi-source data fusion according to claim 6, characterized in that: In step S2, the preprocessing includes: Wavelet packet decomposition is used to reduce noise of physiological signals, and the noise threshold is calculated as: Where σ is the estimated value of the noise standard deviation and N is the signal length; Sliding window mean filtering is used for the operation behavior data, and the window length L is dynamically adjusted to: Among them, f s is the sampling rate, f max The highest frequency of the effective signal.

8. The method for evaluating the effectiveness of pilot training based on multi-source data fusion according to claim 6, characterized in that: In step S5, the visualization report generated includes: a spatiotemporal trajectory heat map, marking the sections where the path deviates by more than 0.5m; a time-series synchronization comparison curve of physiological indicators and operational behaviors; and a feature importance analysis map based on SHAP values, where the SHAP value calculation satisfies: Where N is the total number of features and f is the model prediction function.

Images