A method and system for aircraft fuel consumption prediction based on quick access recorder data

Abstract

The application provides a kind of aircraft fuel consumption prediction method and system based on quick access recorder data, applied to data processing application field.Based on each parameter information recorded in historical quick access recorder data, the application extracts features from quick access recorder data using data processing algorithms to generate a multi-source heterogeneous dataset;process the multi-source heterogeneous dataset to generate a target directed multi-view dynamic graph;model the target directed multi-view dynamic graph to generate node embedding information;inject the node embedding information into each layer of the initial fuel consumption prediction model, adjust the model using an adaptive learning rate, and generate fusion graph data and multi-source data semantic information;process the quick access recorder data of the target aircraft based on the target machine learning model combined with the fusion graph data and multi-source data semantic information to generate aircraft fuel consumption prediction result information.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, and in particular to a method and system for predicting aircraft fuel consumption based on quick access recorder data. BACKGROUND

[0002] Accurate prediction of fuel consumption is crucial for the operation and management of airlines. It not only concerns cost control, but also involves environmental protection. In a highly competitive market environment, accurate prediction of fuel consumption can significantly reduce costs and improve economic efficiency. Reducing fuel consumption can also reduce carbon emissions and help slow down climate change. However, there are many challenges in predicting aircraft fuel consumption. Aviation fuel consumption is influenced by many factors, such as the choice of flight path, which directly affects the length of the journey and thus the fuel consumption. Different flight speeds can change air resistance, affecting engine work and fuel consumption. The frequency and manner of maneuvering operations also have an impact on fuel consumption. Weather conditions, such as temperature, pressure, wind speed and direction, have a significant impact on fuel consumption at different stages of flight. These factors interact with each other, making the prediction of fuel consumption extremely complex.

[0003] Traditional fuel consumption prediction methods mostly use a single algorithm, without fully considering factors such as engine power, attitude changes, and changes in the surrounding environment during flight. For example, during different stages of flight such as takeoff, cruising, and landing, engine power and attitude differ greatly. A single model cannot accurately describe the fuel consumption characteristics of each stage, resulting in limited prediction accuracy. At the same time, when applying existing prediction models to actual aviation operations, there are problems with adaptability, making it difficult to meet the diverse needs of airlines, such as fuel consumption prediction for different routes, different aircraft types, and different seasons.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore includes information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The present application aims to provide a method and system for predicting aircraft fuel consumption based on quick access recorder data, which at least partially overcomes the problems of the prior art. By obtaining QAR data and training sample sets of the target aircraft, a multi-source heterogeneous data set is generated through a data processing algorithm, and then processed into a target directed multi-view dynamic graph. Next, node embedding information is generated using a multi-view dynamic graph neural network, and an initial model is injected and adjusted to achieve semantic fusion of graph data and multi-source data. Finally, based on the target machine learning model, the fusion data is processed, analyzed and predicted, and the final prediction result is obtained after verification and adjustment. The fusion of multi-source data and multiple algorithm models effectively improves the prediction accuracy and helps airlines optimize fuel management and reduce operating costs.

[0006] Other features and advantages of the present application will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, which illustrate, by way of example, the principles of the application.

[0007] According to an aspect of the present application, there is provided a method for predicting aircraft fuel consumption based on quick access recorder data, comprising: obtaining quick access recorder data of a target aircraft and a training sample set; performing feature extraction on the quick access recorder data based on various parameter information recorded in historical quick access recorder data using a data processing algorithm to generate a multi-source heterogeneous data set; processing the multi-source heterogeneous data set based on a data mapping relationship and a data conversion method to generate a target directed multi-view dynamic graph; modeling and processing the target directed multi-view dynamic graph based on a multi-view dynamic graph neural network to generate node embedding information; injecting the node embedding information into each layer of an initial fuel consumption prediction model, adjusting the model using an adaptive learning rate, and generating fusion graph data and multi-source data semantic information; processing the quick access recorder data of the target aircraft based on a target machine learning model in combination with the fusion graph data and the multi-source data semantic information to generate aircraft fuel consumption prediction result information.

[0008] According to another aspect of the present application, there is provided an apparatus for predicting aircraft fuel consumption based on quick access recorder data, characterized in that it comprises: an obtaining module for obtaining quick access recorder data of a target aircraft and a training sample set; a processing module for performing feature extraction on the quick access recorder data based on various parameter information recorded in historical quick access recorder data using a data processing algorithm to generate a multi-source heterogeneous data set; processing the multi-source heterogeneous data set based on a data mapping relationship and a data conversion method to generate a target directed multi-view dynamic graph; modeling and processing the target directed multi-view dynamic graph based on a multi-view dynamic graph neural network to generate node embedding information; injecting the node embedding information into each layer of an initial fuel consumption prediction model, adjusting the model using an adaptive learning rate, and generating fusion graph data and multi-source data semantic information; processing the quick access recorder data of the target aircraft based on a target machine learning model in combination with the fusion graph data and the multi-source data semantic information to generate aircraft fuel consumption prediction result information.

[0009] According to yet another aspect of the present application, there is provided a computer-readable storage medium having stored thereon a computer program, which, when executed by a second processor, implements the method for predicting aircraft fuel consumption based on quick access recorder data as described above.

[0010] The application provides a kind of aircraft fuel consumption prediction method and system based on quick access recorder data, server obtains the QAR data of target aircraft and training sample set, generates multiple-source heterogeneous data set by data processing algorithm, then is processed into target directed multiview dynamic graph. Then, node embedding information is generated using multiview dynamic graph neural network, and the initial model is injected and adjusted to realize semantic fusion of graph data and multiple-source data. Finally, based on the target machine learning model, the fusion data is processed, analyzed and predicted, and the final prediction result is obtained after verification and adjustment. The technology integrates multiple-source data and various algorithm models, effectively improves the prediction accuracy, and helps airlines optimize fuel management and reduce operating costs.

[0011] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 A flow chart of an aircraft fuel consumption prediction method based on quick access recorder data provided by an embodiment of the application is shown.

[0013] Figure 2 A structure schematic diagram of an aircraft fuel consumption prediction device based on quick access recorder data provided by an embodiment of the application is shown. DETAILED DESCRIPTION

[0014] The preferred embodiments of the application are described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the application, and cannot limit the application.

[0015] The aircraft fuel consumption prediction method based on quick access recorder data according to the exemplary embodiments of the application will be described below in conjunction with Figure 1 An embodiment of the application also provides an aircraft fuel consumption prediction method and system based on quick access recorder data. As shown in the figure, the method is applied to a server, which includes: Figure 1

[0016] S101, obtaining the quick access recorder data of target aircraft and training sample set.

[0017] ​In one embodiment, an airline selects 50 Boeing 737-800 aircrafts that have flown both long and short routes in the past month as target aircrafts for optimizing fuel management. The flight data of these aircrafts are diverse and representative, providing rich information for subsequent fuel consumption prediction. After each flight, ground personnel use special equipment that meets the aviation data collection standard to connect the QAR through the data interface of the aircraft. Using the matching data collection software, the flight data stored in the QAR is exported in CSV format. For example, for a flight from Beijing to Shanghai, the collected data includes the engine speed every minute, the flight altitude, speed, heading recorded every 5 minutes, and other detailed parameters. These data record the real-time state of the flight process.

[0018] From the technical manual and maintenance records of the Boeing 737-800 aircraft, relevant performance parameters are obtained. For example, the fuel consumption rate of the engine under different working conditions, and the aerodynamic parameters such as the lift-drag ratio of the aircraft wing. Taking a specific engine model as an example, in the cruising state, when the engine speed stabilizes in a certain interval, the corresponding fuel consumption rate is X kg / hour. Cooperate with professional meteorological data suppliers to obtain meteorological data corresponding to the flight period and route. For example, for the Beijing-Shanghai flight mentioned above, the ground temperature in Beijing at takeoff is 25°C, the air pressure is 1010 hPa, and the wind speed and direction data at different altitude layers during the flight are obtained, such as at 9000 meters, the wind speed is 30 knots, and the wind direction is 270°. Extract route-related data from the airline's flight operation system. For the Beijing-Shanghai route, the actual distance of the route is 1088 kilometers, the height change range during the climb phase (from the takeoff height to the cruising height), the height change range during the descent phase, and the planned takeoff time, actual takeoff time, estimated arrival time, and actual arrival time of the flight are obtained.

[0019] The QAR data, performance parameter data, meteorological environment history data and route feature data of the same flight are integrated. Flight data every 10 minutes is grouped as a unit, and the actual fuel consumption in this time period is labeled. For example, in the 30-40 minute time period of the flight, according to the change of the fuel gauge reading in the QAR data, the actual fuel consumption is determined to be Y kilograms, and this group of data is taken as a training sample. The data verification algorithm is used to check the integrity and reasonableness of the QAR data. For example, check whether there is a parameter missing in a plurality of consecutive record points, if the engine speed is found to be missing in a certain 5 minutes, mark this part of the data for processing; at the same time, check the reasonableness of the data, if the speed data appears a value obviously exceeding the normal flight speed range of the aircraft type (such as the normal cruising speed range of Boeing 737-800 passenger aircraft is 800-900 km / h, if a speed value of 1500 km / h appears), it is judged as an abnormal value. For missing values, linear interpolation based on the data before and after is used for supplement; for abnormal values, according to the normal data range of other related parameters of the flight and the same type of flight, the abnormal values are corrected, if they cannot be corrected, they are rejected. For missing values in meteorological data, Kriging interpolation method is used to supplement according to the data of surrounding meteorological stations; for route feature data, check the logical consistency of flight time and route distance data, if the time and distance calculation are found to be inconsistent, correct them by comparing with the actual flight trajectory and flight plan.

[0020] In S102, based on the parameter information recorded in the historical quick access recorder data, a data processing algorithm is used to extract features from the quick access recorder data to generate a multi-source heterogeneous data set.

[0021] In an embodiment, parameter screening is performed on each parameter information recorded in the historical quick access recorder data, core parameters with high correlation to fuel consumption are identified, dimension influence is eliminated through normalization processing, key feature vectors are extracted through principal component analysis algorithm, and preliminary quick access recorder feature data is generated. For Boeing 737-800 passenger aircraft, historical quick access recorder (QAR) data contains numerous parameters, such as engine speed, flight altitude, speed, fuel flow, etc. Through research on the physical mechanism of fuel consumption and statistical analysis of a large amount of historical flight data, core parameters with high correlation to fuel consumption are determined. For example, the fuel flow of the engine directly reflects the fuel consumption, the engine speed affects the fuel combustion efficiency and then affects the fuel consumption, the flight altitude and speed are related to air resistance and indirectly affect the fuel consumption, which are determined as core parameters. The core parameters are normalized to eliminate the dimension influence. Taking engine speed (unit: revolutions / minute) and flight speed (unit: kilometers / hour) as examples, assuming that the engine speed value range is [4000, 12000] and the flight speed value range is [300, 900], the minimum-maximum normalization method is adopted. For engine speed X, after normalization, its value will be mapped to the [0, 1] interval; the flight speed is the same. Then, principal component analysis (PCA) algorithm is used to extract key feature vectors. The principle of PCA algorithm is to transform the original data to a new set of orthogonal basis through linear transformation, so that the variance of the data on the new basis is as large as possible, and these variance large directions are the principal component directions. The data matrix composed of the normalized core parameters is analyzed by PCA, for example, 3 principal components are extracted from the original 5 core parameters, which can retain most of the information of the original data, and compose the preliminary quick access recorder feature data.

[0022] The aircraft performance parameter data is processed by the performance index classification system, and the data is divided into dynamic performance category and aerodynamic performance category. The missing value completion processing is performed on the data of different categories to generate standardized aircraft performance data. For Boeing 737-800 aircraft, the aircraft performance parameter data is divided into dynamic performance category and aerodynamic performance category according to the performance index classification system. The dynamic performance category includes engine power, thrust and other parameters; the aerodynamic performance category covers wing lift coefficient, drag coefficient, aircraft angle of attack and other parameters. In actual data, there may be missing values in the parameters. For engine power data in the dynamic performance category, if the power value is missing for a certain period of time, the mean value interpolation based on the power data of the same type of engine under similar working conditions is used. Assuming that 100 sets of power data of the same type of engine under similar flight conditions are collected, the average value of these data is calculated, and the average value is used to fill in the missing value. For the wing lift coefficient data in the aerodynamic performance category, if the missing value appears, the lift coefficient is estimated by using the design parameters of the aircraft combined with the current flight state (such as known parameters such as flight height, speed, angle of attack, etc.) to fill in the missing value. The missing value completion processing is performed on the data of different categories to generate standardized aircraft performance data.

[0023] In processing the route feature data of Boeing 737-800 aircraft, the route topology analysis algorithm used to construct the route network model belongs to a graph model. The route network model is a graph structure-based model, which is composed of nodes and edges. In this scenario, each navigation point is taken as a node, and the route connecting these nodes is taken as an edge to represent the topology structure of the route for analyzing the route-related features. The model includes a node layer and an edge layer. Specifically, the node layer: each navigation point is a node, and the node contains the geographic location information (such as latitude and longitude) of the navigation point, which is the basis for calculating the distance and analyzing the route direction. Different navigation points have different roles in the route, such as the starting point, the intermediate key turning point and the end point, etc. The edge layer: the edge connecting each navigation point represents the flight path of the aircraft between two navigation points. The edge not only represents the connection relationship between the nodes, but also contains the direction information, i.e. the flight direction of the aircraft. At the same time, the attributes of the edge can include distance, flight time and other route-related data.

[0024] By the combination of nodes and edges, a directed graph structure is constructed, which fully presents the flight route network from Guangzhou to Chengdu. This graph structure intuitively shows the layout of the route, the connection relationship between the navigation points, and the flight direction, providing a basic framework for subsequent calculation of route distance and analysis of segment characteristics. The latitude and longitude coordinates of each navigation point are the key parameters of the model. Assuming that the coordinates of a departure navigation point in Guangzhou are (latitude1, longitude1), the coordinates of an arrival navigation point in Chengdu are (latitude2, longitude2), and the coordinates of the intermediate navigation points are (latitude_n, longitude_n). These coordinates are used to calculate the distance between navigation points, for example, using the spherical distance formula d = r x arccos(sin(latitude1) x sin(latitude2) + cos(latitude1) x cos(latitude2) x cos(longitude2 - longitude1)) (where r is the radius of the Earth) to calculate the distance between two navigation points. After calculating the distance between each navigation point using the above spherical distance formula, the total route distance is obtained by accumulating these distances, which is an important quantitative feature of the route. For example, the route from Guangzhou to Chengdu passes through 5 navigation points, and the distances d1, d2, d3, d4 between adjacent navigation points are calculated in turn, and the total route distance D = d1 + d2 + d3 + d4.

[0025] The starting and ending altitudes of the climb and descent phases are parameters used to calculate the climb and descent altitude characteristics of the segment. For example, in the take-off climb phase, the altitude is raised from the ground height (assuming 0 meters) to the cruising altitude of 10000 meters, and the climb altitude is 10000 meters. In the landing descent phase, the altitude is lowered from the cruising altitude of 10000 meters to the ground height, and the descent altitude is 10000 meters. A day is divided into morning (0:00-6:00), afternoon (6:00-12:00), evening (12:00-18:00), and night (18:00-24:00) time periods, and the departure time and arrival time of the flight are used to determine the flight period feature. If the departure time of the flight is 10:00 am and the arrival time is 13:00 pm, the flight period feature is marked as morning-afternoon.

[0026] Based on the preliminary QAR feature data, standardized aircraft performance data, dynamic meteorological environment data, and structured route feature data, a correlation rule mining algorithm (such as the Apriori algorithm) is used to establish data correlation. The Apriori algorithm discovers correlation rules by finding frequently occurring item sets in the data set. For example, in a large amount of flight data, it is found that when the engine speed is in a certain interval, the flight altitude is within a certain range, and the meteorological condition is a specific combination of temperature and wind speed, the fuel consumption rate will exhibit a certain pattern. After establishing the correlation rules, abnormal value detection and cleaning are performed. Taking fuel flow data as an example, if the fuel flow value at a certain time deviates greatly from the expected value calculated according to the correlation rules and exceeds the set threshold (such as ±20% of the expected value), it is determined that the value is an abnormal value. For abnormal values, if they can be corrected according to other related data, they are corrected; if they cannot be corrected, they are excluded. After abnormal value detection and cleaning, the various types of data are integrated according to a certain format to generate a multi-source heterogeneous data set for aircraft fuel consumption prediction, providing high-quality data support for subsequent fuel consumption prediction model training.

[0027] In S103, based on the data mapping relationship and the data conversion method, the multi-source heterogeneous data set is processed to generate a target directed multi-view dynamic graph.

[0028] In one implementation, for a Boeing 737-800 passenger plane, the multi-source heterogeneous data set that has been generated includes preliminary QAR feature data, standardized aircraft performance data, dynamic meteorological environment data, and structured route feature data. The preliminary QAR feature data covers engine speed, flight altitude, speed, and other features related to fuel consumption. According to the degree of direct correlation with fuel consumption and the trend of data change, key features are selected to generate a subset. Engine fuel flow, engine speed, and flight altitude are selected as the three features that have a greater impact on fuel consumption and obvious change rules to form a QAR feature data subset. In this way, the core data can be focused on in subsequent processing, improving processing efficiency and prediction accuracy. The standardized aircraft performance data is divided into power performance category and aerodynamic performance category. From the power performance category, engine thrust and engine power, which directly affect fuel consumption, are selected; from the aerodynamic performance category, wing lift coefficient and drag coefficient, which are related to air resistance when the aircraft is flying and thus affect fuel consumption, are selected. These parameters are combined into an aircraft performance data subset.

[0029] The dynamic meteorological environment data includes information such as air temperature, air pressure, wind speed and wind direction. Considering the meteorological factors that have a greater impact on fuel consumption, the air temperature and wind speed are selected to generate a subset of meteorological environment data. For example, during high-altitude flight, air temperature and wind speed have a significant impact on the aerodynamic force of the aircraft and the performance of the engine, thereby affecting fuel consumption. The structured route feature data includes route distance, segment climb and descent altitude features, flight period features, etc. The segment climb altitude and flight period are selected to generate a subset of route feature data. The change of the climb altitude is directly related to the power demand and fuel consumption of the aircraft, and the meteorological conditions and flight regulation requirements are different in different flight periods, which also indirectly affects the fuel consumption.

[0030] Each factor that affects the fuel consumption of the Boeing 737-800 passenger aircraft is taken as a node. The factors that directly promote fuel consumption are taken as source nodes. For example, engine thrust is the direct power to propel the aircraft forward, and the greater the engine thrust, the fuel consumption will generally also increase, so the engine thrust is taken as a source node. In addition, engine speed affects the combustion efficiency of fuel and is also an important source node. The fuel consumption related indicators affected by the source nodes are taken as target nodes. For example, fuel flow directly reflects the fuel consumption, which is affected by factors such as engine thrust and speed, so fuel flow is taken as a target node. In addition, the fuel consumption rate comprehensively reflects the efficiency of fuel consumption of the aircraft under different working conditions, and is also taken as a target node. The node information is generated according to the previously generated data subsets. For each source node and target node, the corresponding data value and related metadata (such as data collection time, corresponding flight number, etc.) are combined to form node information. For example, for the engine thrust source node, the node information may include the engine thrust value at a certain time in a flight (such as 10000 pounds), as well as information such as the flight number (such as CA1234) and the collection time (such as October 10, 2024 10:00:00).

[0031] According to each data subset, the edge information is generated based on the influence relationship of fuel consumption. The edge information represents the association relationship and influence degree between nodes. For example, from the relationship between engine thrust and fuel flow, according to the working principle of the engine, the increase of engine thrust will cause the increase of fuel flow. Therefore, a directed edge is established between the two nodes, from the engine thrust node to the fuel flow node, indicating the positive influence of engine thrust on fuel flow.

[0032] The weight of an edge is determined by statistical analysis of a large amount of flight data to quantify the degree of influence. Suppose that it is found through analysis that, for every 1000 pounds of engine thrust increase, the fuel flow rate increases by 50 kg / h on average, given that other conditions remain unchanged. Then the weight of the edge from engine thrust to fuel flow rate can be set to 0.05 (i.e. 50 ÷ 1000). Similarly, for the edge between engine speed and fuel consumption rate, it is found through data analysis that, for every 100 revolutions per minute increase in engine speed, the fuel consumption rate increases by 0.1 kg / (h·km), so the weight of this edge can be set to 0.001 (i.e. 0.1 ÷ 100). In this way, each edge contains information about the start node, the end node, and the weight representing the degree of influence, constituting complete edge information.

[0033] Based on the node information and edge information generated above, a directed graph is constructed to generate the preliminary structure of the directed multi-view dynamic graph. According to the source and nature of the data, the directed graph is divided into different views. For example, the nodes and edges related to QAR data are divided into a QAR view, including engine speed, fuel flow rate nodes and their connecting edges; the nodes and edges related to aircraft performance data are divided into an aircraft performance view, such as engine thrust, wing lift coefficient nodes and their edges; the nodes and edges related to meteorological environmental data are divided into a meteorological view, including air temperature, wind speed nodes and their edges; the nodes and edges related to route feature data are divided into a route view, such as flight segment climb altitude, flight period nodes and their edges. In each view, the nodes and edges are connected according to the node information and edge information. Taking the QAR view as an example, the engine speed nodes, fuel flow rate nodes, etc. are connected according to the edge information between them to form a local directed graph structure. Similarly, corresponding construction is also carried out in other views. Then the directed graph structures of these different views are integrated together to form the preliminary structure of the directed multi-view dynamic graph. This preliminary structure shows the basic relationship between different factors, but there may be some unreasonable or redundant parts.

[0034] The preliminary directed multi-view dynamic graph is integrated and optimized using a graph optimization algorithm. Here, a graph optimization algorithm based on the principle of minimum spanning tree (MST) can be used, such as Prim's algorithm. In the preliminary structure, there may be some redundant edges that have little effect on the overall structure and fuel consumption prediction. For example, some edges represent relationships that have very little effect in reality, or there are edges that can be indirectly represented by the combination of other edges. Through the graph optimization algorithm, the contribution of each edge to the overall graph structure and the target (fuel consumption prediction) is calculated. Suppose that it is found through calculation that a certain edge connecting two meteorological nodes (such as air temperature and air pressure) has negligible effect on fuel consumption prediction when considering other edges, then this edge is removed to simplify the graph structure and improve the efficiency of subsequent processing.

[0035] In order to make the graph structure more clearly reflect the degree of influence of each factor on fuel consumption and the mutual relationship, the node positions are adjusted according to the weights of the edges and the closeness of the association between nodes. For example, nodes closely related to fuel consumption (such as fuel flow, engine thrust, etc.) are placed in the central position of the graph, while nodes with relatively small influence are placed at the edge position. The weight of the edge is further optimized. When considering the combined action of multiple factors, it is found that the originally set edge weight of engine speed on fuel consumption rate is not very accurate. By reanalyzing the data and calculation, the weight of this edge is adjusted to make it more consistent with the actual situation.

[0036] During data acquisition and processing, there may be some abnormal data corresponding to the nodes. These abnormal nodes may interfere with the accurate analysis and prediction of fuel consumption. For example, the data of a certain engine thrust node is obviously beyond the normal range, and it is found through inspection that it is caused by data acquisition error. By setting reasonable threshold and data verification rules, these abnormal nodes are identified and removed. After the above integration and optimization processing, the target directed multiview dynamic graph is generated, which can more accurately and clearly show the influence of each factor on the fuel consumption of Boeing 737-800 passenger aircraft, and provide a more reliable basis for subsequent fuel consumption prediction.

[0037] In one embodiment, the target directed multiview dynamic graph is processed by a multiview dynamic graph neural network to generate node embedding information.

[0038] In one embodiment, the target directed multiview dynamic graph is processed by a multiview dynamic graph neural network to generate node embedding information.

[0039] In each view, nodes are connected by edges. In the QAR data view, engine speed changes affect fuel flow, so there is a directed edge from the engine speed node to the fuel flow node. In the aircraft performance view, the engine thrust node and the engine power node may have a connection edge because of the working principle of the engine. Detailed analysis of the direction, weight, and connection method of the connection edge between these nodes constitutes node connection relationship information. This information reveals the direct or indirect influence path between different factors. Since the data is recorded based on the flight process, it has a time dimension. Taking a flight data of a Boeing 737-800 aircraft as an example, in different stages of flight, such as take-off, cruising, and landing, engine speed, flight altitude, and other parameters change over time. By analyzing the trend of these parameters on the time axis, such as the rapid rise of engine speed during take-off and the relative stability during cruising, time series feature information can be obtained. This information is crucial for understanding the change pattern of fuel consumption in different flight stages. Integrating the above hierarchical structure information, node connection relationship information, and time series feature information forms comprehensive graph structure information, which fully presents all key features of the target directed multiview dynamic graph and provides detailed and accurate basic data for subsequent neural network processing.

[0040] The adaptive parameter initialization algorithm automatically adjusts the initial parameters of the neural network according to the characteristics of the input data and the features of the graph structure. For a multiview dynamic graph neural network, considering the differences in data distribution and node connection relationship between different views, this algorithm will initialize the parameters of different levels and different types of neurons in a targeted manner. When processing the data of the Boeing 737-800 aircraft, the node data in the QAR data view changes frequently, while the temperature and wind speed data in the weather environment view are relatively stable. The adaptive parameter initialization algorithm will set a larger initial learning rate for neurons processing the QAR data view to capture data changes faster, and set a smaller initial learning rate for neurons processing the weather environment view to ensure the stability of the model. The generated model initialization information enables the neural network to better adapt to the data characteristics of different views in the early stage of training, improving training efficiency and model performance.

[0041] The graph convolution attention mechanism combines graph convolution and attention mechanism. Graph convolution is used to extract local features in graph structure, and attention mechanism is used to automatically learn the importance weight of different nodes and edges. When processing the graph structure information of Boeing 737-800 passenger aircraft, for the nodes in the QAR data view, the graph convolution operation will aggregate the information of the node and its neighbor nodes. For example, when calculating the features of the fuel flow node, the information of the nodes connected to it, such as engine speed and flight altitude, will be considered. At the same time, the attention mechanism will assign weights according to the influence of each node on the fuel flow node. If the engine speed has a more critical impact on the fuel flow under certain flight conditions, the weight of the engine speed node will be relatively high when calculating. Through this graph convolution attention mechanism, the initial feature representation of each node containing rich spatio-temporal features can be generated by processing the graph structure information. For example, the initial features of the fuel flow node not only contain its own data information, but also fuse the information of the surrounding related nodes, and are weighted according to the importance of different nodes, which provides more valuable information for subsequent analysis.

[0042] The initial node feature information is processed by the attention mechanism that fuses spatio-temporal dependencies to generate time series fusion feature information. This mechanism considers the dependency relationship in both time and space dimensions. In the time dimension, it captures the trend of node feature change over time; in the space dimension, it focuses on the association between different nodes at the same time. For Boeing 737-800 passenger aircraft, the change of engine speed at different time points will affect fuel consumption, and different engine speeds will interact with other spatial dimension factors such as flight altitude and wing lift coefficient. Taking the engine speed node as an example, this mechanism analyzes its feature changes in different flight stages (takeoff, cruise, landing), and considers the association with other related nodes (such as fuel flow and flight altitude) at each time point, fuses these spatio-temporal information, and generates time series fusion feature information that can reflect the comprehensive influence of engine speed in the entire flight process. Such feature information more comprehensively reflects the influence of various factors on fuel consumption in the time and space dimensions.

[0043] The time series fusion feature information is processed by the multi-view collaborative attention mechanism to generate node embedding information reflecting the influence of various factors on fuel consumption. This mechanism aims to integrate the time series fusion feature information of different views, learn the importance of different view features, and realize the collaborative effect between multiple views. When processing the data of Boeing 737-800 passenger aircraft, the QAR data view reflects real-time flight parameters, the aircraft performance view reflects the performance of the aircraft itself, the meteorological environment view displays external meteorological conditions, and the route feature view contains route-related information, which affect fuel consumption from different angles.

[0044] The multi-view collaborative attention mechanism automatically learns the importance weight of the time-series fusion features of each view on the final fuel consumption prediction. In the high-altitude cruise phase, the wind speed in the meteorological environment view has a greater impact on fuel consumption, and the weight of this view will be increased accordingly; in the take-off and landing phase, the parameters in the aircraft performance view and the QAR data view are more critical, and their weights will increase. By weighting and fusing the time-series fusion features of different views, node embedding information reflecting the influence of each factor on fuel consumption is generated. These node embedding information integrates the key information of multiple views and can represent the role of each factor in the fuel consumption process, providing high-quality input data for the subsequent fuel consumption prediction model.

[0045] In S105, the node embedding information is injected into each layer of the initial fuel consumption prediction model, and an adaptive learning rate is used to adjust the model to generate fused graph data and multi-source data semantic information.

[0046] In an embodiment, the node embedding information is subjected to dimension adaptation and format conversion processing to generate adapted node embedding information. The node embedding information generated by the multi-view dynamic graph neural network cannot be directly accepted by the initial fuel consumption prediction model in terms of dimension and format. Different models have different requirements for the dimension and format of input data, and if adaptation and conversion are not performed, the model cannot effectively use this information for training and prediction. Assuming that the node embedding information generated by the multi-view dynamic graph neural network is a tensor with a shape of (128,), and the input layer of the initial fuel consumption prediction model (assuming it is a multi-layer perceptron MLP) expects a vector with a shape of (64,) and a data type of numpy array. At this time, the principal component analysis (PCA) algorithm is used to reduce the dimension of the node embedding information. The PCA algorithm projects high-dimensional data into a low-dimensional space through linear transformation while preserving the main features of the data as much as possible. In this example, the 128-dimensional node embedding information is projected into a 64-dimensional space. Then, the processed tensor is converted to a numpy array using the corresponding function, thereby generating the adapted node embedding information.

[0047] The initial fuel consumption prediction model is loaded and the pre-trained weight is frozen, and the key parameters of the adaptive learning rate adjustment component are initialized to generate the initial state information of the model. A pre-trained multi-layer perceptron (MLP) is selected as the initial fuel consumption prediction model. The MLP model has an input layer, multiple hidden layers, and an output layer, and has been trained on a large amount of aircraft fuel consumption related data, learning some general fuel consumption patterns and rules. Through a specific loading function, the trained model parameters are loaded into the current prediction system.

[0048] To prevent over-modifying the general knowledge learned by the model in subsequent training, the pre-training weights of the model are frozen. For example, in the Keras framework, the weights of the layers in the MLP model can be frozen by setting layer.trainable = False, so that during the initial training, the weights of these layers will not change due to the input of new data, and the model will only optimize part of the parameters according to the new data in the subsequent fine-tuning process. The model is trained using the adaptive learning rate algorithm Adam. Adam algorithm combines the advantages of momentum method and RMSProp algorithm, and can adaptively adjust the learning rate. When initialized, the learning rate learning_rate = 0.001, the exponential decay rate of the matrix estimate beta_1 = 0.9, beta_2 = 0.999, and a small constant epsilon = 1e-8 are set to prevent division by zero operations. These parameters determine how the Adam algorithm adjusts the learning rate during training based on gradient information, and appropriate initialization can make the model converge to the optimal solution faster. After the above operations are completed, the initial state information of the model is generated, which is prepared for subsequent training.

[0049] Based on the initial state information of the model, the adapted node embedding information is injected layer by layer based on the layer structure of the initial fuel consumption prediction model to generate model input information with node embedding. According to the layer structure of the initial fuel consumption prediction model (MLP), the adapted node embedding information is injected layer by layer starting from the input layer. The MLP model has 3 hidden layers, and the input layer receives the preprocessed QAR data feature vector with a dimension of (50,). The adapted node embedding information (dimension (64,)) is spliced with the QAR data feature vector of the input layer in dimension to obtain a new input vector with a dimension of (50+64,) = (114,). This new vector is used as the actual input of the input layer, and then passes through the 3 hidden layers and the output layer in turn. In each layer, the node embedding information is operated with the weights of the layer to participate in the model calculation process, so that the model can use the fuel consumption influencing factor information carried by the node embedding information in each layer, thereby generating model input information with node embedding.

[0050] An adaptive learning rate strategy is adopted to train and adjust the model input information with node embedding. During the training process, the parameters of the adaptive learning rate adjustment component are dynamically updated to generate adjusted model parameter information. The model input information with node embedding is used as input, and the model is trained using a training sample set containing Boeing 737-800 aircraft flight data. During the training process, the Adam algorithm is used for optimization. In the first few iterations of training, since the model parameters are far from the optimal solution, the gradient is large, and the Adam algorithm updates the model parameters with a relatively large learning rate (such as 0.001) according to the initialized parameters, accelerating the learning speed of the model. As the training progresses, the gradient gradually becomes smaller, and the Adam algorithm automatically reduces the learning rate according to the first-order moment estimate and the second-order moment estimate of the gradient, avoiding the model skipping the optimal solution when approaching the optimal solution.

[0051] In each training iteration, the Adam algorithm calculates the exponentially weighted moving average of the gradient (the first-order moment estimate m t and the second-order moment estimate v t ) according to the formula: m t =β1m t-1 +(1-β1)g t , where g t is the gradient of the current iteration, m t-1 and v t-1 are the first-order moment estimate and the second-order moment estimate of the last iteration. Then, the learning rate is adjusted according to these estimates, and the model parameters are updated. Through continuous training and parameter updating, adjusted model parameter information is generated, enabling the model to better fit the data and improving the prediction accuracy of Boeing 737-800 aircraft fuel consumption.

[0052] Using the adjusted model parameters, the input graph data (i.e., the data information contained in the target directed multiview dynamic graph) and multi-source data (such as QAR data, aircraft performance data, meteorological environment data, and route feature data) are jointly semantically modeled. The model learns the semantic associations between different data and uncovers the complex relationships between various factors and fuel consumption. In this process, the layers of the model (input layer, hidden layer, and output layer) work together to extract and combine features from the input data. For example, the input layer receives multi-source data such as QAR data and meteorological data, the hidden layer transforms the data through a nonlinear activation function (such as ReLU) to extract potential features from the data, and the output layer analyzes these features to obtain semantic representations related to fuel consumption.

[0053] After joint semantic modeling, the model generates semantic information that integrates the features and correlations of multiple data types, providing a more comprehensive and in-depth reflection of the various factors affecting the fuel consumption of the Boeing 737-800 passenger aircraft and their interrelationships. For example, the model may find that, under low-temperature and strong headwind weather conditions, when the aircraft is in the cruise phase and the engine speed is maintained within a certain range, the fuel consumption will exhibit a specific pattern. These integrated semantic information provides strong support for subsequent accurate prediction of fuel consumption.

[0054] In one embodiment, based on the target machine learning model, the fusion graph data and the semantic information of the multi-source data are processed to generate the fuel consumption prediction result information of the target aircraft.

[0055] In one embodiment, based on the target machine learning model, the fusion graph data and the semantic information of the multi-source data are processed to generate the fuel consumption prediction result information of the target aircraft.

[0056] The fusion data is input to the input layer of the CNN in the form of a two-dimensional matrix. In the convolution layer, different size and weight convolution kernels are designed to perform convolution operations on the data. For example, a 3x3 convolution kernel is used to slide across different positions of the fusion data to extract local features. The weights of the convolution kernel are continuously optimized during the training process to capture key features related to fuel consumption. For the part of the fusion graph data representing the relationship between engine speed and fuel flow, the convolution kernel can learn the change pattern between the two. After alternating processing by multiple convolution layers and pooling layers, the features of the data are gradually extracted and compressed. The pooling layer (such as max pooling) is used to reduce the data dimension and retain the main features. Finally, the extracted features are integrated through the fully connected layer to generate fusion data feature information. These feature information contains key information after the fusion of multiple data, such as the comprehensive feature representation of the influence of factors such as flight state, weather conditions, and aircraft performance on fuel consumption.

[0057] The flight parameter sequence in the quick access recorder data of the target aircraft and the related multi-source data sequence are parsed to generate target event feature information. For a Boeing 737-800 passenger aircraft, the flight parameter sequence in the quick access recorder (QAR) data includes data such as engine speed, flight altitude, and speed changing over time. The related multi-source data sequence includes aircraft performance parameters (such as engine power and wing lift coefficient), meteorological environmental data (air temperature and wind speed), and route feature data (route distance and flight period). Taking a flight from Beijing to Shanghai as an example, the engine speed sequence recorded every 5 minutes in the QAR data is parsed, and statistical features such as its change rate and average value are calculated. At the same time, the flight altitude and speed data in the same period are combined to analyze their mutual relationship. For meteorological environmental data, the air temperature and wind speed sequences at different altitude layers during the flight are parsed to determine whether there are sudden temperature changes or strong wind areas.

[0058] According to the parsing results, target event feature information is generated. For example, if the engine speed continues to rise during a certain flight time, the flight altitude increases, the speed remains stable, and the meteorological data shows that it is in a headwind environment at that time, a target event feature representing the "headwind climb phase" is generated based on these information. This feature not only includes the changes of flight parameters, but also integrates the influences of multi-source data such as meteorological and route data, and can more accurately describe the potential association between specific events in the flight process and fuel consumption.

[0059] Based on the fusion data feature information, the target event feature information is associated and analyzed to generate association analysis result information. Pearson correlation coefficient method is used for association analysis. Pearson correlation coefficient is used to measure the degree of linear correlation between two variables, and its value ranges from -1 to 1. In this scenario, Pearson correlation coefficients are calculated between each feature in the fusion data feature information and each feature in the target event feature information. For example, the correlation coefficient between the feature representing the relationship between engine performance and fuel consumption in the fusion data feature and the "headwind climb phase" feature in the target event feature is calculated. A series of correlation coefficients are obtained by calculation, forming the association analysis result information. If it is found that the correlation coefficient between the fusion feature representing the relationship between engine power and fuel consumption and the "headwind climb phase" target event feature is high (such as 0.8), it indicates that the influence of engine power on fuel consumption is more significant during the headwind climb phase. These association analysis result information reveals the correlation degree and direction between different features, providing an important basis for subsequent fuel consumption prediction.

[0060] The correlation analysis result information is input into a prediction module of the target machine learning model, and processed using the prediction capability of the model to generate preliminary aircraft fuel consumption prediction information. In the selected CNN model, the prediction module is composed of a last fully connected layer and an output layer. The fully connected layer further integrates and transforms the correlation analysis result information, and the output layer calculates the preliminary aircraft fuel consumption prediction value according to the output of the fully connected layer through a specific activation function (such as a linear activation function). The prediction module learns the mapping relationship between the correlation analysis result information and the actual fuel consumption during the training process.

[0061] The correlation analysis result information is input into the prediction module. For example, the vector containing the correlation degree of various features is input into the fully connected layer, and after weight calculation and nonlinear transformation, the preliminary aircraft fuel consumption prediction information is output by the output layer. This prediction information can be a specific fuel consumption value, or a trend of fuel consumption change (such as increase, decrease or stability). Assuming that the model calculation shows that the fuel consumption of the Boeing 737-800 aircraft in the next 10 minutes is expected to be 500 kg under the current flight conditions, this is the preliminary aircraft fuel consumption prediction information.

[0062] The preliminary aircraft fuel consumption prediction information is subjected to result verification and adjustment processing to generate aircraft fuel consumption prediction result information. The historical data comparison verification method is adopted. A large amount of actual fuel consumption data of the Boeing 737-800 aircraft under similar flight conditions is collected as historical reference. The preliminary prediction information is compared with the historical data to calculate the prediction error. For example, the difference between the predicted 500 kg of fuel consumption and the actual fuel consumption under the same flight phase, similar weather conditions and route conditions in history is calculated, and the relative error is calculated. If the prediction error exceeds the preset threshold (such as the relative error exceeds 10%), the preliminary prediction information is adjusted. Based on the direction and size of the error, a rule-based adjustment method or fine-tuning using a machine learning model can be used. For example, if the prediction value is always high and the error is relatively stable, the prediction value can be reduced by a certain percentage. After adjustment, the final aircraft fuel consumption prediction result information is obtained. Assuming that after verification and adjustment, the final fuel consumption in the next 10 minutes is determined to be 480 kg, which is the generated aircraft fuel consumption prediction result information, which can be used for fuel management and flight plan optimization of the airline.

[0063] The aircraft fuel consumption prediction method and system based on quick access recorder (QAR) data provided in the application mainly includes three core links of data processing, model construction and prediction. In the data processing stage, the QAR data and the training sample set of the target aircraft are first obtained, the historical QAR data is processed through parameter screening, normalization and principal component analysis, and combined with the aircraft performance, meteorological environment and route characteristic data, a multi-source heterogeneous data set is generated. Then, the data set is classified and screened to construct a directed multi-view dynamic graph, providing a basis for subsequent modeling.

[0064] In the model construction aspect, the multi-view dynamic graph neural network is used to model the graph, and the node embedding information is generated through layered disassembly, adaptive parameter initialization, graph convolution attention mechanism and other operations. Then the node embedding information is injected into the initial fuel consumption prediction model, and the adaptive learning rate adjustment is adopted to realize the semantic fusion of the graph data and the multi-source data. In the prediction stage, based on the target machine learning model, the fusion data is processed for feature extraction, the flight parameters in the QAR data and the multi-source data sequence analysis results are combined for correlation analysis, the analysis results are input into the model prediction module to obtain the preliminary prediction information, and after verification and adjustment, the final aircraft fuel consumption prediction result is obtained. By comprehensively utilizing multi-source data and using multiple algorithms and models, the accuracy of aircraft fuel consumption prediction is effectively improved, which provides strong support for airlines to optimize fuel management and reduce costs.

[0065] In one embodiment, as shown in Figure 2 The application also provides an aircraft fuel consumption prediction device based on quick access recorder data, which includes:

[0066] The acquisition module 201 is configured to acquire the quick access recorder data and the training sample set of the target aircraft.

[0067] The processing module 202 is configured to extract features from the quick access recorder data based on the parameter information recorded in the historical quick access recorder data, generate a multi-source heterogeneous data set by using a data processing algorithm, process the multi-source heterogeneous data set based on a data mapping relationship and a data conversion method, generate a target directed multi-view dynamic graph, model the target directed multi-view dynamic graph based on a multi-view dynamic graph neural network, generate node embedding information, inject the node embedding information into each layer of an initial fuel consumption prediction model, adjust the model by using an adaptive learning rate, generate semantic information of fused graph data and multi-source data, and process the quick access recorder data of the target aircraft based on a target machine learning model in combination with the fused graph data and the multi-source data semantic information, to generate aircraft fuel consumption prediction result information.

[0068] The various embodiments in this application are described in a related manner, and the same or similar parts among the various embodiments can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the aircraft fuel consumption prediction method, electronic device, electronic equipment, and readable storage medium embodiments based on quick access recorder data, since they are basically similar to the aircraft fuel consumption prediction method embodiments based on quick access recorder data described above, the description is relatively simple, and the relevant parts can be referred to the part of the description of the aircraft fuel consumption prediction method embodiments based on quick access recorder data described above.

Claims

1. A method for aircraft fuel consumption prediction based on quick access recorder data, characterized in that, The method comprises the following steps: obtaining quick access recorder data of a target aircraft, a training sample set; based on various parameter information recorded in historical quick access recorder data, using a data processing algorithm to extract features from the quick access recorder data, and generating a multi-source heterogeneous data set; based on data mapping relationships and data conversion methods, processing the multi-source heterogeneous data set to generate a target directed multi-view dynamic graph; based on a multi-view dynamic graph neural network, modeling and processing the target directed multi-view dynamic graph to generate node embedding information; injecting the node embedding information into each layer of an initial fuel consumption prediction model, adjusting the model using an adaptive learning rate, and generating fusion graph data and multi-source data semantic information; based on a target machine learning model, processing the quick access recorder data of the target aircraft by combining the fusion graph data and the multi-source data semantic information to generate aircraft fuel consumption prediction result information.

2. The method of claim 1, wherein, Based on various parameter information recorded in historical quick access recorder data, using a data processing algorithm to extract features from the quick access recorder data, and generating a multi-source heterogeneous data set, comprising: performing parameter filtering processing on various parameter information recorded in historical quick access recorder data, identifying core parameters with high correlation with fuel consumption, performing normalization processing to eliminate dimensional influence, combining principal component analysis algorithm to extract key feature vectors, and generating preliminary quick access recorder feature data; processing aircraft performance parameter data through a performance index classification system, and dividing the data into dynamic performance category and aerodynamic performance category; performing missing value completion processing on different category data to generate standardized aircraft performance data; using a spatiotemporal interpolation algorithm to supplement and process missing data of meteorological environment historical data, and generating dynamic meteorological environment data; for route feature data, constructing a route network model by means of a route topology analysis algorithm, calculating route distance and flight segment climb and descent height features, extracting flight period features in combination with flight schedule data, and generating structured route feature data; based on preliminary QAR feature data, standardized aircraft performance data, dynamic meteorological environment data and structured route feature data, establishing data association through an association rule mining algorithm, and generating a multi-source heterogeneous data set for aircraft fuel consumption prediction after abnormal value detection and cleaning.

3. The method of claim 1, wherein, Based on data mapping relationships and data conversion methods, processing the multi-source heterogeneous data set to generate a target directed multi-view dynamic graph, comprising: performing classification and filtering processing on preliminary QAR feature data, standardized aircraft performance data, dynamic meteorological environment data and structured route feature data in the multi-source heterogeneous data set to generate QAR feature data subset, aircraft performance data subset, meteorological environment data subset and route feature data subset; taking various factors affecting fuel consumption as nodes, taking factors directly promoting fuel consumption as source nodes, and taking fuel consumption related indicators affected by the source nodes as target nodes, generating node information according to each data subset; processing the action relationship affecting fuel consumption based on each data subset to generate edge information; Based on the generated node information and edge information, a directed graph is constructed to generate a preliminary structure of the directed multi-view dynamic graph; the preliminary constructed directed multi-view dynamic graph is integrated and optimized, a graph optimization algorithm is used to adjust the node position and the weight of the edge, so that the graph structure reflects the influence degree and the mutual relationship of each factor on the fuel consumption, and the redundant edges and the abnormal nodes are removed to generate a target directed multi-view dynamic graph.

4. The method of claim 3, wherein, Based on the multi-view dynamic graph neural network, the target directed multi-view dynamic graph is modeled and processed to generate node embedding information, including: The target directed multi-view dynamic graph is processed by hierarchical decomposition to generate hierarchical structure information, node connection relationship information and time sequence feature information, which together constitute comprehensive graph structure information; For the multi-view dynamic graph neural network, the network parameters are initialized by using an adaptive parameter initialization algorithm according to the graph structure characteristics and data characteristics to generate model initialization information; Based on the model initialization information, the graph structure information is processed by using an innovative graph convolution attention mechanism to generate initial node feature information containing rich spatio-temporal features; The initial node feature information is processed by using a spatio-temporal dependent attention mechanism to generate time sequence fusion feature information; The time sequence fusion feature information is processed by using a multi-view collaborative attention mechanism to generate node embedding information reflecting the influence of each factor on fuel consumption.

5. The method of claim 1, wherein, The node embedding information is injected into each layer of the initial fuel consumption prediction model, and the model is adjusted by using an adaptive learning rate to generate fusion graph data and multi-source data semantic information, including: The node embedding information is processed by dimension adaptation and format conversion to generate adapted node embedding information; The initial fuel consumption prediction model is loaded and the pre-training weight is frozen, and the key parameters of the adaptive learning rate adjustment component are initialized to generate model initial state information; Based on the model initial state information, the adapted node embedding information is injected layer by layer based on the layer structure of the initial fuel consumption prediction model to generate model input information with node embedding; The model input information with node embedding is trained and adjusted by using an adaptive learning rate strategy, and the parameters of the adaptive learning rate adjustment component are dynamically updated during the training process to generate adjusted model parameter information; Based on the adjusted model parameter information, the input graph data and multi-source data are jointly modeled to generate fusion graph data and multi-source data semantic information.

6. The method of claim 1, wherein, Based on the target machine learning model, the fusion graph data and multi-source data semantic information are processed to generate aircraft fuel consumption prediction result information, including: Based on the target machine learning model, the fusion graph data and multi-source data semantic information are processed to generate fusion data feature information: The flight parameter sequence and related multi-source data sequence in the rapid access recorder data of the target aircraft are parsed to generate target event feature information: Based on the fusion data feature information, the target event feature information is associated and analyzed to generate association analysis result information: The correlation analysis result information is input into a prediction module of a target machine learning model, and the model prediction capability is used for processing to generate preliminary aircraft fuel consumption prediction information: The preliminary aircraft fuel consumption prediction information is subjected to result checking and adjustment processing to generate aircraft fuel consumption prediction result information.

7. An aircraft fuel consumption prediction apparatus based on quick access recorder data, characterised in that, The device comprises: An acquisition module is configured to acquire quick access recorder data of a target aircraft and a training sample set; A processing module is configured to perform feature extraction on the quick access recorder data based on various parameter information recorded in historical quick access recorder data by using a data processing algorithm, to generate a multi-source heterogeneous data set; to process the multi-source heterogeneous data set based on a data mapping relationship and a data conversion method, to generate a target directed multi-view dynamic graph; to model the target directed multi-view dynamic graph based on a multi-view dynamic graph neural network, to generate node embedding information; to inject the node embedding information into each layer of an initial fuel consumption prediction model, to adjust the model by using an adaptive learning rate, and to generate fused graph data and multi-source data semantic information; and to process the quick access recorder data of the target aircraft based on a target machine learning model in combination with the fused graph data and the multi-source data semantic information, to generate aircraft fuel consumption prediction result information.

8. An electronic device, comprising: It comprises: A first processor; and a memory for storing executable instructions of the first processor; The first processor is configured to execute the executable instructions to perform the aircraft fuel consumption prediction method based on quick access recorder data according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the second processor to implement the aircraft fuel consumption prediction method based on quick access recorder data according to any one of claims 1-6.

Images