CN-121998517-A - Aircraft control quality evaluation method based on transducer model

Abstract

The invention discloses an aircraft control quality evaluation method based on a transducer model, and belongs to the field of aviation ergonomics. The method comprises the steps of designing flight task subject primitives, collecting multi-mode physiological signals and flight task performance data of pilots, extracting key features from the physiological signals through multi-level screening, fusing the key features with the results of task performance grading, constructing a transform prediction model combining convolution and multi-head self-attention mechanisms, effectively modeling local features and cross-mode global dependency relationships, realizing automatic prediction of manipulation quality grades, adopting a small sample fine tuning strategy based on dynamic weight freezing, enabling the model to be quickly adapted to different pilots, improving generalization capability, and finally obtaining final evaluation of an aggregate group prediction result. Objective and automatic evaluation of multi-source data driving is realized, the limitation of the traditional subjective evaluation method is overcome, and the consistency and efficiency of evaluation are improved.

Inventors

• ZHANG SHUGUANG
• LI YUHAN

Assignees

• 北京航空航天大学

Dates

Publication Date

20260508

Application Date

20260409

Claims (10)

1. 1. An aircraft manipulation quality assessment method based on a transducer model is characterized by comprising the following steps of: Designing a flight mission, collecting multi-mode physiological signals and flight mission performance of a pilot when the pilot executes the flight mission, and grading the flight mission performance; Extracting and screening the characteristics of the multi-mode physiological signals to obtain key physiological characteristics, and forming a control quality quantification key index by combining the flight task performance grading results; Taking a quantized key index of the manipulation quality as input, modeling local and global features through a convolution unit and a self-attention mechanism in a combined mode, introducing a global dependency relationship between position codes and modeling features of the multi-head self-attention mechanism, and constructing a prediction model based on a transducer; Performing cross-pilot adaptation and generalization processing on the trained prediction model based on the transducer to obtain an operation quality prediction model; And predicting the pilot data by using the control quality prediction model to obtain a prediction result of the pilot data, and obtaining a final aircraft control quality evaluation result according to the prediction result of the pilot data.
2. 2. The method for evaluating the maneuvering quality of the aircraft according to claim 1, wherein the designed flight mission comprises the pilot constructing a mission subject element MTE in a loop experimental platform for simulating a maneuvering mission scene, and the flight mission performance of the pilot when the mission subject element MTE is completed is converted into a plurality of grades by setting the flight mission completion time, the maneuvering accuracy and the track deviation.
3. 3. The method of claim 1, wherein the step of feature extracting and screening the multi-modal physiological signals comprises: Extracting time domain, frequency domain and nonlinear characteristics from the multi-mode physiological signals to construct a candidate characteristic set, evaluating the candidate characteristics by adopting at least two different characteristic importance evaluation methods, and reserving the characteristics which are ranked at the front in the at least two evaluation methods as the key physiological characteristics.
4. 4. The method for evaluating the maneuvering quality of an aircraft according to claim 3, wherein the feature importance evaluation method comprises a statistical analysis method based on Copula entropy, a machine learning method based on Shapley values and a dynamic analysis method based on channel attention weights.
5. 5. The method for evaluating the maneuvering quality of an aircraft according to claim 1, wherein the step of executing a prediction model based on a transducer is as follows: The method comprises the steps of carrying out local feature coding on key physiological features through one-dimensional convolution operation, adding the coded features and position codes to form sequence embedding, inputting the sequence embedding into at least one transducer coding layer, calculating global correlation among the features by the aid of a multi-head self-attention mechanism by the aid of the transducer coding layer, finally splicing output of the transducer coding layer with a result of flight task performance grading, and outputting prediction probability distribution of manipulation quality grades through a classification layer.
6. 6. The method for evaluating the quality of an aircraft maneuver of claim 1, wherein the step of fine-tuning the small samples is: The method comprises the steps of calculating the contribution degree of model parameters to output to determine the importance of the parameters on the basis of a pre-trained transducer-based prediction model, generating gradient gating coefficients based on the importance of the parameters to attenuate or mask gradients of parameters with high importance in back propagation, and then updating only the parameters which are not frozen or have low gradient attenuation degree by using new pilot data.
7. 7. The method according to claim 6, characterized in that the gradient gating coefficients are controlled by calculating the contribution of the model parameters by means of an integral gradient method and converting the importance of the parameters into freezing probabilities depending on the contribution.
8. 8. The method of claim 1, wherein the multi-modal physiological signals include at least three of electrocardiographic, dermatologic, respiratory, myoelectrical, and eye movement signals.
9. 9. The method of claim 6, wherein new pilot data is a result of a multimodal physiological signal sample and its corresponding mission performance classification acquired by the new pilot while performing the mission.
10. 10. The method for evaluating the maneuvering quality of the aircraft according to claim 1, wherein the final aircraft maneuvering quality evaluation result is a maneuvering quality perception score average value obtained by averaging the predicted scores of all the reference aircraft under the same flight mission.

Description

Aircraft control quality evaluation method based on transducer model Technical Field The invention belongs to the technical field of aviation ergonomics and control quality evaluation, and particularly relates to an aircraft control quality evaluation method based on a transformer model, which is used for uniformly modeling characteristics fused with flight task performance and multi-mode physiological signals so as to output an aircraft control quality score evaluation method. Background The quality assessment of the maneuvering of an aircraft has long been mainly based on the traditional cooper Hamper scale, which requires a grading judgment by a pilot with a great deal of experience according to the completion of the mission and subjective feeling. This evaluation method is widely used in civil aviation field, and although forming a unified standard in practice, the essence of the evaluation method still belongs to a subjective evaluation method based on experience. With the rapid development of urban air traffic and electric vertical take-off and landing (ELECTRIC VERTICAL TAKE-off AND LANDING, eVTOL) aircrafts, the configuration of the aircrafts is increasingly complex, pilot groups are gradually diversified, and the requirements of a new generation of aircrafts in rapid iterative design, performance verification and airworthiness authentication are difficult to meet only by relying on subjective evaluation of a few expert pilot test aircrafts. There are significant limitations to the prior art, the cooper Hamper method. First, subjective scores are affected by pilot experience levels and individual differences, and the results are easily disturbed by bias, and it is difficult to ensure consistency and repeatability of the assessment. Secondly, the flight test of relying on expert pilot not only has high cost, but also has long period, and causes remarkable restriction on the development efficiency of the aircraft. Finally, as the flight modes of new aircraft are diversified and the operating environment is complicated, the traditional scale is difficult to comprehensively reflect the pilot's maneuvering load and compensation behavior, resulting in limited guidance of the evaluation results on design optimization and maneuvering law improvement. Furthermore, in some studies, performance metrics have been used as an objective measure of maneuver quality, such as by measuring mission completion through flight trajectory, flight path angle, and their deviation from performance limits, and classifying performance into ideal, adequate, controllable, and uncontrollable levels. The method overcomes the limitation of pure subjective evaluation to a certain extent, but mainly focuses on the task result itself, and ignores the artificial load and compensation behavior generated by the pilot in the task completion process. In recent years, the development of wearable physiological sensors has provided a new technological approach to monitor pilot status. The multi-mode physiological signals such as electrocardio, skin electricity, respiration, myoelectricity, eye movement and the like can objectively reflect the cognitive load and the compensation mode of the pilot. However, the existing researches are mostly to split the flight performance and the physiological signals, and are respectively used for researching the task completion quality or the pilot state, and a method for fusing the two through a model and forming a unified control quality prediction framework is lacked. Meanwhile, a transducer model based on a self-attention mechanism is widely applied to the fields of natural language processing, time sequence prediction, multi-modal learning and the like. Compared with the traditional cyclic neural network (RNN, LSTM and the like), the transducer establishes the dependency relationship between any features in the same layer through a self-attention mechanism, and is more suitable for modeling complex correlations between multi-modal features. The prior art begins to try to introduce a transducer into physiological signal analysis such as electrocardio and electroencephalogram and human-computer interaction state identification, but the proposal is mostly aimed at a single mode or a common scene, and a transducer network structure and a training flow for evaluating the control quality of an aircraft are not formed yet. In view of the above problems, the present invention provides an aircraft maneuvering quality evaluation method based on a transducer model. The method comprehensively utilizes flight performance indexes and pilot physiological signals on the basis of task subject primitives (Mission TASK ELEMENT, MTE), combines feature screening and deep learning models, and achieves objective and automatic prediction of control quality grades. The method not only overcomes the subjectivity of the traditional Cooper Hamper method, but also overcomes the defect that the human factor is ignored only depending