CN-121998206-A - Intelligent prediction method and system for aviation material demand based on dynamic weight self-adaptive adjustment

CN121998206ACN 121998206 ACN121998206 ACN 121998206ACN-121998206-A

Abstract

The invention discloses an intelligent prediction method and system for air material demands based on dynamic weight self-adaptive adjustment, and belongs to the field of intelligent scheduling of civil air material. The method comprises the steps of constructing a multi-source feature matrix integrating historical maintenance data features, model parameter features and environmental factor features, carrying out weighted integration on the multi-source features based on a dynamic weight function comprising seasonal adjustment components and gradient feedback components, inputting the weighted integrated features into a mixed prediction model comprising a time sequence feature extraction layer and a long-term associated modeling layer to carry out space material demand prediction, and training the mixed prediction model based on a weighted loss function embedded with dynamic weight vectors. According to the invention, the self-adaptive adjustment of the multisource characteristic weight is realized through the dynamic weight function, the seasonal fluctuation of the aircraft material demand can be accurately captured, the sudden change can be responded quickly, and the prediction precision and the inventory turnover efficiency are effectively improved.

Inventors

SUN ZIXUAN
XIE KUNMING
XIE FEI
LUO YUBO
GAO KAI
WANG YONGQIANG
PENG SEN
LI YONG
LIU YI
HU MIN
XIE YUN

Assignees

成都九洲电子信息系统股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. The intelligent prediction method for the aircraft material demand based on the dynamic weight self-adaptive adjustment is characterized by comprising the following steps: Constructing a multi-source feature matrix, wherein the multi-source feature matrix fuses historical maintenance data features, model parameter features and environmental factor features; Weighting and fusing various features in the multi-source feature matrix based on a dynamic weight function, wherein the dynamic weight function comprises a seasonal adjustment component and a gradient feedback component; Inputting the weighted and fused characteristics into a hybrid prediction model for predicting the demand of the aviation materials, wherein the hybrid prediction model comprises a time sequence characteristic extraction layer and a long-term association modeling layer; Training the hybrid prediction model based on a weighted loss function, and embedding a dynamic weight vector output by the dynamic weight function in the weighted loss function.
2. The intelligent prediction method for aircraft demand based on dynamic weight adaptive adjustment according to claim 1, wherein the historical maintenance data features comprise component failure frequency and maintenance cycle time sequence data, the model parameter features comprise flight hours, take-off and landing times and model code data, and the environmental factor features comprise temperature, humidity and airport altitude data.
3. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 1, wherein the dynamic weight function is expressed as: ; Wherein, the Is an initial weight base value; for seasonal adjustment component, T is the current time step, T is the periodic parameter, and lambda is the fluctuation amplitude coefficient; for the gradient feedback component, beta is the learning rate parameter, The gradient of loss is predicted for the previous cycle.
4. The intelligent prediction method of the aircraft demand based on the dynamic weight self-adaptive adjustment according to claim 1, wherein the time sequence feature extraction layer is a long-period memory network, the long-period associated modeling layer is a transducer network, the hybrid prediction model adopts an LSTM-transducer series architecture, and the output of the long-period memory network is used as the input of the transducer network.
5. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 4, wherein the transducer network adopts a multi-head attention mechanism, the number of attention heads is 8, and the feature dimension of each attention head is 64.
6. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 3, wherein the initial weight base value is allocated to various features according to a preset ratio, and the initial weight ratio of the historical maintenance data feature, the model parameter feature and the environmental factor feature is 0.45:0.30:0.25.
7. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 1, further comprising respectively performing data preprocessing on the historical maintenance data characteristic, the model parameter characteristic and the environmental factor characteristic before constructing the multi-source characteristic matrix, wherein the data preprocessing comprises data normalization processing and missing value filling processing.
8. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 1, further comprising applying a safety factor to the predicted demand output by the hybrid prediction model to obtain a safe redundancy demand when a predicted object is a flight key, and setting an emergency purchase threshold, and triggering an emergency purchase procedure when the inventory of the flight key is lower than the emergency purchase threshold.
9. The intelligent prediction method for air traffic demand based on dynamic weight adaptive adjustment according to claim 8, wherein the safety factor has a value ranging from 1.2 to 1.5, and the emergency purchase threshold is set to a predicted amount within a preset number of days of the flight key.
10. Intelligent prediction system of aviation material demand based on dynamic weight self-adaptation adjustment, its characterized in that includes: the multi-source data acquisition module is used for acquiring historical maintenance data characteristics, model parameter characteristics and environmental factor characteristics and constructing a multi-source characteristic matrix; The dynamic weight calculation module is used for calculating the dynamic weights of various features based on a dynamic weight function comprising seasonal adjustment components and gradient feedback components, and carrying out weighted fusion on the multi-source feature matrix; the mixed prediction model module comprises a time sequence feature extraction layer and a long-term association modeling layer and is used for receiving the weighted and fused features and predicting the requirements of the aircrafts, and the mixed prediction model module is trained based on a weighted loss function embedded with a dynamic weight vector; And the prediction output module is used for outputting the prediction result of the air material demand.

Description

Intelligent prediction method and system for aviation material demand based on dynamic weight self-adaptive adjustment Technical Field The invention belongs to the field of intelligent scheduling of civil aviation materials, and particularly relates to an intelligent prediction method and an intelligent prediction system for aviation material demands based on dynamic weight self-adaptive adjustment, which are suitable for accurate inventory management and demand prediction of scenes such as airport aviation material libraries, airline company maintenance centers and the like. Background The prediction of the aviation material demand is a key link of a civil aviation maintenance support system, and the accuracy of the prediction directly influences the aviation material inventory level and the maintenance support efficiency. With the continuous expansion of civil aviation transportation scale and the increasing complexity of fleet structures, the technical challenges facing the prediction of the demand for aircrafts are also increasing. The existing aviation material demand prediction technology mainly has the following defects: First, in terms of feature weight distribution, the existing method generally adopts a static weight strategy, that is, after the weights of all features are determined through an optimization algorithm in a model training stage, the weights are kept constant in actual operation. For example, in the prior art, an improved particle swarm algorithm is adopted to optimize the initial weight and the threshold value of the BP neural network, while the prediction accuracy is improved to a certain extent, the weight of the BP neural network is fixed after training is completed, and seasonal fluctuation characteristics of the aircraft material requirements cannot be responded. The field of civil aviation maintenance has remarkable seasonal rules, namely, flight density increase during spring transportation causes accelerated consumption of aviation materials, high-temperature environments in summer aggravate failure rate of specific components, low-temperature influences on reliability of precise components such as a hydraulic system in winter, and the like, and the seasonal factors require that characteristic weights can be dynamically adjusted along with time, and a static weight method is difficult to meet the requirement. Second, in terms of predictive model architecture, existing approaches mostly employ a single type of neural network model. The traditional BP neural network belongs to a shallow feedforward network, and has certain nonlinear fitting capability, but has limited network depth, so that the time sequence dependency relationship in the historical maintenance data is difficult to effectively capture. A single long and short term memory network, while good at processing time series data, is not capable of long-distance correlation between modeling type parameters, environmental factors and other multidimensional features and aircraft demands. The single model architecture has difficulty in simultaneously satisfying the dual requirements of short-term time sequence feature extraction and long-term correlation feature modeling, so that prediction accuracy is limited. Thirdly, in terms of safety guarantee, the existing aircraft demand prediction method generally regards all aircraft as being equally important, and does not implement a differential prediction strategy for the aircraft with different safety levels. However, the guarantee requirements of the civil aviation field on flight critical parts (such as engines, landing gear and other core parts which directly affect the flight safety) are far higher than those of general aviation materials, and a safety redundancy mechanism needs to be introduced on the basis of a prediction result so as to ensure that the supply of the flight critical parts is not interrupted under any condition. Disclosure of Invention The invention aims to provide an intelligent prediction method and system for the demand of the aviation material based on the self-adaptive adjustment of the dynamic weight, which realize the dynamic self-adaptive adjustment of the multi-source characteristic weight in the demand prediction of the aviation material, improve the prediction precision and the inventory turnover efficiency in a complex scene and simultaneously meet the safety redundancy requirement of civil aviation. In order to achieve the above object, the present invention provides a technical solution comprising: an intelligent prediction method for the demand of the aviation material based on the self-adaptive adjustment of the dynamic weight comprises the following steps: Constructing a multi-source feature matrix, wherein the multi-source feature matrix fuses historical maintenance data features, model parameter features and environmental factor features; Weighting and fusing various features in the multi-source feature matrix based on a dynamic weigh