Method and system for measuring fuel quantity of fuel tank of aircraft and aircraft

Abstract

The invention discloses an aircraft fuel tank fuel quantity measuring method and system and an aircraft, and belongs to the technical field of aircraft fuel quantity measurement, and the method comprises the steps that target feature information is generated based on the current attitude angle of the aircraft and the current oil height of fuel in a fuel tank of the aircraft; selecting the sample feature information to form a sample set based on a sequence of distances between the target feature information and the sample feature information in a preset database from small to large; performing prediction processing based on the target feature information, the sample set and a sample oil volume corresponding to the sample set in a preset database by using a Gaussian process regression model to obtain a prediction result, the prediction result comprising an estimated value of the current oil volume; and determining the current oil liquid weight based on the estimated value and the current oil liquid density of the fuel oil in the oil tank. A complex nonlinear relation can be more accurately modeled based on a nearest neighbor local Gaussian process interpolation strategy, so that the prediction precision is remarkably improved, and the fuel quantity measurement error is reduced.

Description

Technical Field

[0001] This application relates to the field of aircraft fuel quantity measurement technology, specifically to aircraft fuel tank fuel quantity measurement methods, systems, and aircraft. Background Technology

[0002] The remaining fuel level in an aircraft's (such as an airplane) fuel tank is a crucial indicator for assessing its range and ensuring flight safety. An aircraft fuel tank level measurement system is designed to monitor the fuel level in the aircraft's fuel tank, providing fuel information to the flight crew. In measuring fuel level, the system needs to consider the fuel level in the fuel tank and the aircraft's attitude angle, and consult a fuel measurement database to determine the fuel volume.

[0003] The typical method for establishing a fuel measurement database is as follows: First, under a series of fixed attitude angles, the fuel tank model is sliced ​​into sections at specific step sizes from minimum to maximum fuel level, and the fuel volume corresponding to each section is calculated. Then, these volume data at specific attitude angles and fuel levels are stored in the fuel computer of the fuel tank fuel quantity measurement system to form a fuel measurement database. In actual use, when the aircraft is in a specific attitude and the sensor measures a certain fuel level, the system needs to interpolate to calculate the fuel volume under the current condition.

[0004] Because the aircraft's attitude angles vary widely, related methods are prone to introducing significant errors when using attitude angles and fuel level for multidimensional interpolation calculations. The main reason for these errors is that the relevant interpolation methods struggle to accurately predict the actual fuel quantity in high-dimensional space. Summary of the Invention

[0005] This application provides a method, system, and aircraft for measuring the fuel quantity in an aircraft's fuel tank, aiming to solve the problem of large measurement errors in fuel quantity.

[0006] Firstly, a method for measuring the fuel level in an aircraft's fuel tank is provided, including:

[0007] Obtain the current fuel level and density in the aircraft's fuel tank, as well as the aircraft's current attitude angle;

[0008] Based on the current attitude angle and current oil level, generate target feature information;

[0009] Based on the order of increasing distance between the target feature information and the sample feature information in the preset database, sample feature information is selected to form a sample set; each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information;

[0010] Using a Gaussian process regression model, prediction is performed based on target feature information, sample set, and the corresponding sample oil volume to obtain prediction results, including an estimate of the current oil volume.

[0011] Determine the current oil weight based on the estimated value and the current oil density.

[0012] In some of these design approaches, the prediction results also include posterior variance, which is used to quantify the uncertainty of the above estimates.

[0013] The above-mentioned method for measuring fuel level in fuel tanks also includes:

[0014] The confidence level of the above estimates is determined based on the posterior variance.

[0015] In some of these design approaches, the Gaussian process regression model includes a kernel function;

[0016] Using a Gaussian process regression model, prediction processing is performed based on target feature information, a sample set, and the corresponding sample oil volume, including:

[0017] Using kernel functions, the first kernel matrix between the target feature information and the sample set, and the second kernel matrix between the sample sets are determined;

[0018] The above estimated values ​​are determined based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set.

[0019] In some of these design approaches, the Gaussian process regression model includes the observation noise variance and the identity matrix;

[0020] Based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set, the above estimated values ​​are determined, including:

[0021] Based on the first kernel matrix, the second kernel matrix, the observation noise variance, the identity matrix, and the sample oil volume corresponding to the sample set, the posterior mean of the oil volume corresponding to the target feature information is determined, and the posterior mean is determined as the above estimated value.

[0022] In some of these design approaches, the prediction results also include posterior variance, which is used to quantify the uncertainty of the above estimates.

[0023] Using a Gaussian process regression model, prediction processing is performed based on target feature information, a sample set, and the corresponding sample oil volume. This also includes:

[0024] Using kernel functions, we determine the third kernel matrix between target feature information and target feature information, and the fourth kernel matrix between sample set and target feature information;

[0025] The posterior variance is determined based on the first, second, third, and fourth kernel matrices. In some of these designs, the Gaussian process regression model includes the observation noise variance and the identity matrix.

[0026] Based on the first, second, third, and fourth kernel matrices, the posterior variance is determined, including:

[0027] The posterior variance is determined based on the first kernel matrix, the second kernel matrix, the third kernel matrix, the fourth kernel matrix, the observation noise variance, and the identity matrix.

[0028] In some of these design approaches, the current fuel level and density in the aircraft's fuel tanks are obtained, including:

[0029] The current oil altitude and current oil density are obtained from the aircraft's fiber optic sensing measurement system;

[0030] Obtain the current attitude angles of the aircraft, including:

[0031] Obtain the current attitude angle from the aircraft's avionics system.

[0032] In some of these designs, the fiber optic sensing measurement system includes a fiber optic density sensor and multiple fiber optic level sensors;

[0033] The current oil altitude and current oil density are obtained from the aircraft's fiber optic sensing measurement system, including:

[0034] The current oil level is obtained from the aforementioned multiple fiber optic level sensors;

[0035] The current oil density is obtained from the fiber optic density sensor.

[0036] In some of these design approaches, before performing prediction processing using a Gaussian process regression model based on target feature information, a sample set, and the corresponding sample oil volume, the following steps are also included:

[0037] A training set is created by combining the feature information of a subset of samples in the sample set.

[0038] By maximizing the marginal log-likelihood function, the parameter values ​​of the hyperparameters of the kernel function in the Gaussian process regression model are estimated based on the training set, the number of sample feature information in the training set, and the oil volume of the corresponding sample in the training set.

[0039] Update the hyperparameters of the kernel function in the Gaussian process regression model using the above parameter values.

[0040] Some of these design approaches also include:

[0041] The test set is composed of the feature information of samples other than the training set in the sample set.

[0042] After updating the hyperparameters of the kernel function in the Gaussian process regression model using the above parameter values, the following steps are also included:

[0043] Using a Gaussian process regression model, the predicted oil volume corresponding to the feature information of each sample in the test set is predicted.

[0044] Based on the predicted oil volume and sample oil volume corresponding to the feature information of each sample in the test set, the prediction error is determined.

[0045] Based on the prediction error, the fit of the Gaussian process regression model is determined.

[0046] Secondly, a fuel level measurement system for an aircraft's fuel tank is also provided, comprising:

[0047] The acquisition unit is configured to acquire the current fuel level and current fuel density in the fuel tank of the aircraft, as well as the current attitude angle of the aircraft;

[0048] The generation unit is configured to generate target feature information based on the current attitude angle and the current oil level.

[0049] The selection unit is configured to select sample feature information to form a sample set based on the order of increasing distance between the target feature information and the sample feature information in the preset database; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information;

[0050] The prediction unit is configured to use a Gaussian process regression model to perform prediction processing based on target feature information, sample set and sample oil volume corresponding to the sample set, and obtain prediction results, including the estimated value of the current oil volume.

[0051] The determining unit is configured to determine the current oil weight based on the above estimate and the current oil density.

[0052] Thirdly, an aircraft is also provided, including a fuel tank quantity measurement system as described in the second aspect.

[0053] Beneficial effects:

[0054] The solution provided in this application can obtain the current fuel level and density in the fuel tank of an aircraft, as well as the current attitude angle of the aircraft. Then, based on the current attitude angle and current fuel level, target feature information is generated. Next, sample feature information is selected to form a sample set based on the ascending distance between the target feature information and sample feature information in a preset database. Each sample feature information is generated based on the corresponding sample attitude angle and sample fuel level. The preset database also includes the sample fuel volume corresponding to the sample feature information. Then, a Gaussian process regression model is used to perform prediction processing based on the target feature information, the sample set, and the corresponding sample fuel volume to obtain a prediction result. The prediction result includes an estimated value of the current fuel volume. Finally, based on the estimated value and the current fuel density, the current fuel weight is determined. Since the sample feature information used to form the sample set is selected based on the ascending distance between the target feature information and sample feature information in the preset database, this sample set is the nearest neighbor sample set of the target feature information. The nearest neighbor-based local Gaussian process interpolation strategy can more accurately model complex nonlinear relationships, thereby significantly improving prediction accuracy and reducing fuel quantity measurement errors. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of a method for measuring the fuel level in an aircraft's fuel tank, provided in an embodiment of this application.

[0057] Figure 2 This is a schematic diagram of a fuel tank fuel quantity measurement system for an aircraft provided in an embodiment of this application;

[0058] Figure 3 This is a flowchart of the hyperparameter estimation method for the kernel function provided in the embodiments of this application;

[0059] Figure 4 This is another flowchart of the method for measuring the fuel level in the fuel tank of an aircraft provided in the embodiments of this application;

[0060] Figure 5 This is a schematic diagram of the absolute percentage error distribution provided in an embodiment of this application;

[0061] Figure 6 This is another schematic diagram of the absolute percentage error distribution provided in the embodiments of this application;

[0062] Figure 7 This is a schematic diagram showing the distribution of the absolute percentage error reduction at each prediction point using the scheme provided in this application;

[0063] Figure 8 This is another schematic diagram showing the distribution of the absolute percentage error reduction at each prediction point using the scheme provided in this application;

[0064] Figure 9 This is a schematic diagram of a fuel tank fuel quantity measurement system for an aircraft provided in an embodiment of this application. Detailed Implementation

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0067] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.

[0068] The use of "applies to" or "configured to" in this application implies open and inclusive language, which does not exclude the applicability to or configuration to devices performing additional tasks or steps. Additionally, the use of "based on" implies openness and inclusivity, because processes, steps, calculations, or other actions "based on" one or more conditions or values ​​may in practice be based on additional conditions or values ​​beyond those stated.

[0069] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0070] As mentioned earlier, due to the large range of attitude angle variations in aircraft, related methods are prone to introducing significant errors when performing multidimensional interpolation calculations using attitude angles and fuel level. The main reason for these errors is that the relevant interpolation methods struggle to accurately predict the actual fuel quantity in high-dimensional space.

[0071] One related technology involves a method for storing fuel model and calculating fuel volume in a fuel measurement system. This method involves traversing and cutting the fuel model of the fuel measurement unit under each fuel surface posture according to the required cutting height step, and simultaneously measuring the volume and center of gravity of the fuel model below the fuel surface. This yields the numerical relationship between the fuel volume, center of gravity, and fuel surface height for each posture combination. Then, using the least squares method and the corresponding Chebyshev polynomial function, the fuel model of the fuel measurement system for different fuel surface postures is obtained. Finally, the fuel volume and center of gravity coordinates can be obtained by solving the Chebyshev polynomial function. However, this related technology has accuracy limitations: in complex fuel tank shapes or regions with rapidly changing fuel levels, the Chebyshev polynomial needs to be of a higher order to ensure accuracy, increasing computational complexity.

[0072] This application provides a method, system, and aircraft for measuring the fuel level in an aircraft's fuel tank, which can significantly improve prediction accuracy and reduce fuel level measurement errors.

[0073] In some embodiments, the aircraft in this application includes, but is not limited to, airplanes.

[0074] In some embodiments, the fuel quantity measurement method for an aircraft's fuel tank provided in this application can be executed by the aircraft's fuel quantity measurement system, for example, by the fuel computer in the fuel quantity measurement system.

[0075] Figure 1 This is a flowchart of a method for measuring the fuel level in an aircraft's fuel tank, provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0076] S101: Obtain the current fuel level and density in the aircraft's fuel tank, as well as the current attitude angle of the aircraft;

[0077] S103: Generate target feature information based on the current attitude angle and current oil level;

[0078] S105: Based on the order of increasing distance between the target feature information and the sample feature information in the preset database, select the sample feature information to form a sample set; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information;

[0079] S107: Using a Gaussian process regression model, prediction is performed based on target feature information, sample set, and the sample oil volume corresponding to the sample set to obtain prediction results, including an estimate of the current oil volume.

[0080] S109: Determine the current oil weight based on the estimated value and the current oil density.

[0081] Figure 1 The corresponding embodiment provides a solution that can obtain the current fuel level and density in the aircraft's fuel tank, as well as the aircraft's current attitude angle. Then, based on the current attitude angle and current fuel level, target feature information is generated. Next, sample feature information is selected to form a sample set based on the ascending distance between the target feature information and sample feature information in a preset database. Each sample feature information is generated based on its corresponding sample attitude angle and sample fuel level. The preset database also includes the sample fuel volume corresponding to the sample feature information. Then, a Gaussian process regression model is used to perform prediction processing based on the target feature information, the sample set, and the corresponding sample fuel volume to obtain a prediction result. The prediction result includes an estimated value of the current fuel volume. Finally, based on the estimated value and the current fuel density, the current fuel weight is determined. Since the sample feature information used to form the sample set is selected based on the ascending distance between the target feature information and sample feature information in the preset database, this sample set is the nearest neighbor sample set of the target feature information. The nearest neighbor-based local Gaussian process interpolation strategy can more accurately model complex nonlinear relationships, thereby significantly improving prediction accuracy and reducing fuel quantity measurement errors.

[0082] Steps S101 to S109 will be explained below.

[0083] In step S101, the current fuel level and current fuel density in the fuel tank of the aircraft are obtained, as well as the current attitude angle of the aircraft are obtained.

[0084] Specifically, the fuel quantity measurement system for an aircraft's fuel tank can include a fiber optic sensing measurement system, a measurement cable, and a computer device, which can be called a fuel computer. The fiber optic sensing measurement system can be used to measure the fuel level and density in the aircraft's fuel tank in real time; for example, the fiber optic sensing measurement system senses the static pressure of the fuel, converts it into fuel level and density information, and transmits this information to the fuel computer. The measurement cable is used for signal transmission, for example, for signal transmission between the fiber optic sensing measurement system and the fuel computer. The fuel computer is used for fuel quantity calculation. Its embedded software uses the fuel level, combined with the attitude angles collected by the aircraft's avionics system, to query a preset database (which can be called a fuel measurement database) to determine the fuel volume. The preset database includes the correspondence between sample feature information and sample fuel volume. The sample feature information is determined based on the corresponding sample attitude angle and sample fuel level.

[0085] In practice, the fuel computer can obtain the current fuel level and density in the aircraft's fuel tanks from a fiber optic sensing measurement system. Furthermore, the fiber optic sensing measurement system includes, for example... Figure 2 The fiber optic density sensor and multiple fiber optic liquid level sensors shown (such as...) Figure 2 The fiber optic level sensor 1, fiber optic level sensor 2, and fiber optic level sensor M shown in the diagram allow the fuel computer to obtain the current fuel level from these multiple fiber optic level sensors and the current fuel density from the fiber optic density sensor. Figure 2 This is a schematic diagram of a fuel tank fuel quantity measurement system for an aircraft provided in an embodiment of this application. Figure 2 In this context, M is an integer greater than 1. Additionally, as... Figure 2 As shown, the fuel computer can also communicate with the aircraft's avionics system to obtain the current attitude angles. These current attitude angles include pitch and roll angles. The units for pitch and roll angles can be, for example, degrees. In one example, the measurement cable described above can also be used for signal transmission between the avionics system and the fuel computer.

[0086] The calculation principle of the fiber optic liquid level sensor is as follows: When the fiber optic liquid level sensor is placed in oil, the static pressure it experiences is: P = ρgh. Where ρ represents the oil density, g represents the acceleration due to gravity, and h represents the oil height. h is the parameter to be calculated by the fiber optic liquid level sensor, with units in meters (m). The unit of ρ can be kg / m³. 3 kg represents kilogram. g can be measured in m / s². 2 s represents seconds. Under static pressure P, the diaphragm of the fiber optic liquid level sensor deforms, and its deformation can be expressed as: c = k1ρgh. Here, k1 is a calibration function related to pressure. Based on the diaphragm deformation c, the fuel level in the tank can be deduced.

[0087] Fiber optic density sensors measure oil density based on the pressure difference method. The sensor uses two diaphragms to measure the pressure difference between them: ΔP = ρgΔh, where Δh represents the oil height difference. The pressure difference ΔP can be quantitatively expressed as the difference in reflected wavelengths between the two diaphragms, Δc: Δc = k1ρgΔh. The oil density can then be calculated.

[0088] In step S103, target feature information is generated based on the current attitude angle and current oil level. To improve computational efficiency, all features in the target feature information and those in the previously described sample feature information are normalized features, and the same normalization method is used. Furthermore, the target feature information and sample feature information have the same data structure; for example, each feature in the target feature information and its corresponding feature in the sample feature information belong to the same field, which can be oil level, pitch angle, or roll angle. Additionally, both the target feature information and the sample feature information can be feature vectors.

[0089] Let θ represent the pitch angle in the current attitude angle, Φ represent the roll angle in the current attitude angle, and the current oil height be M oil heights, denoted by h1,…,h M Taking the M oil heights as an example, when generating target feature information, initial feature information x = [θ, Φ, h1, ..., h] can be generated first. M ] T Then, each feature in the initial feature information x is normalized to obtain the target feature information x. ′ .

[0090] Furthermore, for each feature in the initial feature information x, the min-max normalization method can be used to map that feature to the interval [0, 1]. Let x... i This characteristic, x max x represents the maximum value corresponding to this feature. min Taking the minimum value corresponding to this feature as an example, the value of this feature after normalization is... Where i can be an integer greater than or equal to 1 and less than or equal to M+2.

[0091] It should be noted that the normalization process described above ensures that all input features are on the same scale, which helps improve the numerical stability of the nearest neighbor metric and Gaussian process regression (GPR) models. Further data processing can be performed as needed to ensure data quality, thereby improving the model's accuracy and stability.

[0092] In this embodiment, a technical approach based on a nearest neighbor sample selection strategy and a Gaussian process is adopted. The nearest neighbor sample selection strategy is used to construct a nearest neighbor sample set. The Gaussian process regression model is a non-parametric probabilistic model used to represent the distribution of a function. Based on Bayesian learning theory, the core idea of ​​the Gaussian process is to predict the function value at any point given a set of observation data and a corresponding kernel function. Furthermore, it can also predict the uncertainty of the function value. In this embodiment, the input features of the Gaussian process regression model include pitch angle, roll angle, and oil height, and the output target variable includes oil volume.

[0093] In step S105, sample feature information is selected to form a sample set based on the ascending distance between the target feature information and the sample feature information in the preset database. Each sample feature information is generated based on its corresponding sample attitude angle and sample oil height. The preset database also includes the sample oil volume corresponding to the sample feature information. Therefore, in the actual calculation process, a sample set can be formed by selecting a portion of the sample feature information based on the ascending distance between the target feature information and the sample feature information in the preset database, instead of directly using all the sample feature information in the preset database. This reduces the computational burden and improves interpolation accuracy.

[0094] When determining the distance between target feature information and the feature information of each sample in a preset database, an Euclidean distance algorithm or other distance calculation algorithms can be used. When using the Euclidean distance algorithm, the Euclidean distance between the target feature information and the feature information of each sample in the preset database is determined.

[0095] Take x′=[θ′, Φ′, h1′,…,h M ′] T Represents target feature information, Q′ j =[θ′ j ,Φ′ j h 1,j ′,…,h M,j ′] T Taking any sample feature information from the preset database as an example, x′ and Q′ can be calculated using the following formulas. j Euclidean distance (x′, Q′) j ):

[0096]

[0097] Where j can be an integer greater than or equal to 1 and less than or equal to the number of sample feature information in the preset database. θ′, Φ′, h′1, h M θ′ can sequentially represent the pitch angle, roll angle, first oil height, and Mth oil height in the target feature information x′. j,Φ′ j h′ 1,j h M,j ' can be represented sequentially as sample feature information Q' j The sample pitch angle, sample roll angle, oil height of the first sample, and oil height of the Mth sample are included.

[0098] When selecting sample features to form a sample set based on the ascending distance between the target feature information and the sample feature information in a preset database, for example, a preset number of sample features with the smallest distance to the target feature information can be selected. Let n represent this preset number; n can be an integer greater than 1. Furthermore, the value of n can be set according to actual needs and is not specifically limited here. It should be understood that this sample set is the nearest neighbor sample set of the target feature information. By using the nearest neighbor sample set, the use of all sample feature information can be avoided, thereby effectively reducing the computational load. In addition, using the nearest neighbor sample set for subsequent oil volume interpolation prediction in a Gaussian process regression model can improve computational efficiency and increase local prediction accuracy.

[0099] In practice, Gaussian process regression models include a kernel function, which serves as a covariance function and has hyperparameters. In one example, this kernel function includes, but is not limited to, the radial basis function (RBF) kernel. The RBF kernel can be defined as follows:

[0100]

[0101] Where k() is the radial basis function kernel, a and b are the input parameters of the radial basis function kernel, and a and b can both be passed in a single feature or a set of multiple feature information, σ f σ is the signal standard deviation, and l is the length scale. f and l are hyperparameters of the radial basis kernel function.

[0102] In one implementation, if the hyperparameter optimization of the kernel function of the Gaussian process regression model is complete, step S107 can be executed after step S105. If the hyperparameter optimization of the kernel function of the Gaussian process regression model is not complete, step S107 can be executed as follows: Figure 3 The hyperparameter estimation method for the kernel function is shown. Among them, Figure 3 This is a flowchart of a hyperparameter estimation method for a kernel function provided in an embodiment of this application. The hyperparameter estimation method includes the following steps:

[0103] S301: Combine the feature information of some samples in the sample set to form a training set;

[0104] S303: By maximizing the marginal log-likelihood function, the parameter values ​​of the hyperparameters of the kernel function in the Gaussian process regression model are estimated based on the training set, the number of sample feature information in the training set, and the oil volume of the corresponding sample in the training set.

[0105] S305: Update the hyperparameters of the kernel function in the Gaussian process regression model using the above parameter values.

[0106] As one implementation method, maximizing the marginal log-likelihood function can be expressed as:

[0107]

[0108] Where logp() represents maximizing the marginal log-likelihood function, X train Let y represent the training set. train N represents the volume of oil in the samples corresponding to the training set. train This indicates the amount of feature information in the samples in the training set. K(X) represents the variance of the observation noise, I represents the identity matrix, and K(X) represents the variance of the observation noise. train ,X train ) represents X determined based on the kernel function. train With X train The kernel matrix is ​​obtained by maximizing the marginal log-likelihood function with respect to the hyperparameters of the kernel function (such as the signal standard deviation σ). f By differentiating the hyperparameter with the length scale (l) and iterating using an optimization algorithm, the optimal parameter value can be obtained. This parameter value is then used to update the hyperparameter of the kernel function in the Gaussian process regression model, thereby completing the training of the Gaussian process regression model.

[0109] Furthermore, after updating the hyperparameters of the kernel function in the Gaussian process regression model using this parameter value, the fit of the Gaussian process regression model can be evaluated. For example, the feature information of samples other than the training set in the sample set is used to form a test set. The Gaussian process regression model is then used to predict the predicted oil volume corresponding to each sample feature information in the test set. Based on the predicted oil volume and the sample oil volume corresponding to each sample feature information in the test set, the prediction error is determined, and based on the prediction error, the fit of the Gaussian process regression model is determined. This prediction error can be, for example, the mean squared error (MSE) or the mean absolute error (MAE). The fit can be one of normal fit, underfit, or overfit. If the fit is normal, step S107 can be executed. If the fit is underfit or overfit, steps S301 to S305 can be executed to further optimize the hyperparameters of the kernel function in the Gaussian process regression model. This ensures that the Gaussian process regression model does not exhibit underfit or overfit.

[0110] It should be noted that by making all features in the target feature information and all features in the sample feature information as described above normalized features, the stability of model training can be improved.

[0111] Continue reading Figure 1 In step S107, a Gaussian process regression model is used to perform prediction processing based on target feature information, sample set and sample oil volume corresponding to the sample set, and the prediction result is obtained. The prediction result includes the estimated value of the current oil volume.

[0112] Specifically, using the kernel function in the Gaussian process regression model, the first kernel matrix of the target feature information and the sample set, as well as the second kernel matrix of the sample set, are determined. Based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set, the above estimated values ​​are determined.

[0113] Furthermore, the Gaussian process regression model includes observation noise variance and identity matrix. Based on the first kernel matrix, the second kernel matrix, observation noise variance, identity matrix, and the sample oil volume corresponding to the sample set, the posterior mean of the oil volume corresponding to the target feature information can be determined, and the posterior mean can be determined as the above-mentioned estimated value.

[0114] For example, the formula used to calculate the posterior mean in a Gaussian process regression model can be:

[0115]

[0116] Where x′ represents the target feature information, X represents the sample set, y represents the sample oil volume corresponding to the sample set, μ(x′) represents the posterior mean of the oil volume corresponding to the target feature information, K(x′,X) represents the first kernel matrix of x′ and X, and K(X,X) represents the second kernel matrix of X and X. Let I represent the variance of the observation noise, and let I represent the identity matrix.

[0117] In step S109, the current oil weight is determined based on the above estimated value and the current oil density. Specifically, the product of the above estimated value and the current oil density is determined as the current oil weight.

[0118] It should be noted that the related technologies described above, which involve the storage of oil model and the calculation of oil quantity in a fuel measurement system, still suffer from the problem of lacking uncertainty assessment. These technologies only provide deterministic predictions and cannot provide prediction results with confidence.

[0119] In addition, some related technologies involve measuring the remaining fuel level in aircraft fuel tanks. For example, one related technology involves a method for measuring the remaining fuel level in aircraft fuel tanks based on convolutional neural networks, including the following steps: standardizing flight data and obtaining a training set based on the standardized flight data; establishing a convolutional neural network model and training the model using the training set to obtain a trained convolutional neural network model; detecting the aircraft's flight data in real time during flight and feeding it into the trained convolutional neural network model to obtain the remaining fuel level in the aircraft fuel tank. This related technology uses an infrared camera to acquire images, angular velocity, and angular acceleration, and uses a convolutional neural network to calculate the remaining fuel level in the aircraft fuel tank.

[0120] Another related technology involves a learning model-based system and method for measuring the remaining fuel level in an aircraft fuel tank. The system includes a processor connected to an angle sensor, an angle acceleration sensor, and at least two level sensors. The angle sensor detects the aircraft's flight angle, the angle acceleration sensor detects the acceleration of the aircraft's flight angle, and the level sensors are positioned at different locations on the aircraft fuel tank to detect the fuel level. The processor processes the data detected by each sensor according to a learning model to obtain the remaining fuel level in the aircraft fuel tank. This related technology detects the aircraft's flight angle, flight angle acceleration, and fuel tank level, and uses an LSTM (long short-term memory network) model to determine the remaining fuel level in the aircraft fuel tank based on the data detected by each sensor.

[0121] Both of the above-mentioned related technologies use neural networks for calculation. However, neural networks have some inherent drawbacks in fuel quantity measurement applications, such as:

[0122] 1. "Black box" characteristics: The decision-making process inside neural networks is difficult to interpret, and lacks the necessary interpretability in aviation safety-critical systems;

[0123] 2. High data dependency: A large amount of training data is required to achieve good performance;

[0124] 3. Weak uncertainty assessment capability: Standard neural networks cannot provide confident prediction results;

[0125] 4. High computational resource requirements: Training and optimizing complex neural networks requires high computational resources, making them unsuitable for airborne systems with limited computing power.

[0126] To quantify uncertainty and enhance interpretability, in one embodiment, the prediction result further includes a posterior variance, which is used to quantify the uncertainty of the estimated value; the fuel tank quantity measurement method further includes determining the confidence level of the estimated value based on the posterior variance. For example, assuming μ(x′) represents the estimated value and Σ(x′) represents the posterior variance, the 95% confidence interval could be μ(x′) ± 1.96*Σ(x′).

[0127] The solution provided in this application uses a Gaussian process regression model for oil volume interpolation prediction. Compared to "black box" models such as neural networks, a Gaussian process is a stochastic process and can be viewed as an infinite-dimensional Gaussian distribution. In a Gaussian process, the model's output (such as the predicted value of the regression function) is inferred from the correlation of given input points (determined by the covariance function) and the distribution of known data. Therefore, each step of the prediction in the Gaussian process regression model can be traced back to the choice of the covariance function, the influence of the training data, and how the Gaussian process regression model handles uncertainty. This traceability and transparency make the Gaussian process highly interpretable.

[0128] Furthermore, in the process of predicting based on target feature information, sample set, and corresponding sample oil volume using a Gaussian process regression model, while determining the first and second kernel matrices as described above using kernel functions, a third kernel matrix relating the target feature information and the sample set, as well as a fourth kernel matrix relating the target feature information and the sample set, can also be determined. Then, based on the first, second, third, and fourth kernel matrices, the posterior variance is determined. Further, the posterior variance can be determined based on the first, second, third, and fourth kernel matrices, the observation noise variance, and the identity matrix.

[0129] For example, the formula used to calculate the posterior variance in a Gaussian process regression model can be:

[0130]

[0131] Where x′ represents the target feature information, X represents the sample set, Σ(x′) represents the posterior variance, K(x′,X) represents the first kernel matrix of x′ and X, X(X,X) represents the second kernel matrix of X and X, K(x′,x′) represents the third kernel matrix of x′ and x′, and K(X,x′) represents the fourth kernel matrix of X and x′. Let I represent the variance of the observation noise, and let I represent the identity matrix.

[0132] Figure 4 This is another flowchart of the method for measuring the fuel level in an aircraft's fuel tank provided in this application embodiment. (See also...) Figure 4 As shown, the method includes the following steps:

[0133] S401: Obtain the current fuel level and density in the aircraft's fuel tank, as well as the current attitude angle of the aircraft;

[0134] S403: Generate target feature information based on the current attitude angle and current oil level;

[0135] S405: Based on the order of increasing distance between the target feature information and the sample feature information in the preset database, select sample feature information to form a sample set; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information;

[0136] S407: Using a Gaussian process regression model, prediction is performed based on target feature information, sample set, and the sample oil volume corresponding to the sample set to obtain prediction results. The prediction results include the estimated value of the current oil volume and the posterior variance. The posterior variance is used to quantify the uncertainty of the estimated value.

[0137] S409: Determine the current oil weight based on the estimated value and the current oil density;

[0138] S411: Determine the confidence level of the above estimates based on the posterior variance.

[0139] For an explanation of steps S401 to S411, please refer to the relevant explanations above, which will not be repeated here.

[0140] exist Figure 4 In the corresponding embodiment, the sample set is the nearest neighbor sample set of the target feature information. This scheme, based on the nearest neighbor local Gaussian process interpolation strategy, can more accurately model complex nonlinear relationships, thereby significantly improving prediction accuracy and reducing fuel quantity measurement errors. Furthermore, by using a Gaussian process regression model to predict the estimated value of the current fuel volume, the posterior variance corresponding to this estimate is quantified, and the confidence level of the estimate is determined based on this posterior variance. Thus, the Gaussian process regression model not only provides a point estimate of the fuel volume but also an estimate of the uncertainty of the prediction result, providing a prediction result with confidence. Moreover, compared to "black box" models such as neural networks, using a Gaussian process regression model for fuel volume interpolation prediction enhances interpretability.

[0141] This application and related multidimensional linear interpolation calculations were tested on a fuel measurement database, and the absolute percentage error was calculated as: |(actual value - predicted value) / predicted value|, with the following results:

[0142] The mean absolute percentage error of this application is approximately 0.0145%, and the mean absolute percentage error of the related multidimensional linear interpolation calculation is approximately 0.4150%.

[0143] The absolute percentage median of this application is approximately 0.004146%, and the absolute percentage median calculated by relevant multidimensional linear interpolation is approximately 0.04%.

[0144] The absolute percentage error distribution of this application and related multidimensional linear interpolation calculations is as follows: Figure 5 and Figure 6 As shown. Figure 5 This is a schematic diagram of the absolute percentage error distribution provided in the embodiments of this application. Figure 6 This is another schematic diagram of the absolute percentage error distribution provided in the embodiments of this application. Figure 6 and Figure 5 Correspondingly, Figure 6 The vertical axis displays a logarithmic scale to provide a clearer view of the absolute percentage error distribution. Figure 5 and Figure 6 In this context, "this scheme" refers to this application, and "linear interpolation" refers to related multidimensional linear interpolation calculations.

[0145] Furthermore, the distribution of the absolute percentage error reduction at each prediction point using the scheme provided in this application is as follows: Figure 7 and Figure 8 As shown. Figure 7 This is a schematic diagram illustrating the distribution of the absolute percentage error reduction at each prediction point using the scheme provided in this application. Figure 8 This is another schematic diagram illustrating the distribution of the absolute percentage error reduction at each prediction point using the scheme provided in this application. Figure 8 and Figure 7 Correspondingly, Figure 8 The vertical axis is displayed on a logarithmic scale to more clearly show the distribution of the absolute percentage error reduction. Figure 7 and Figure 8 In this context, "linear interpolation - this scheme" represents this application.

[0146] Figure 9 This is a schematic diagram of a fuel tank level measurement system for an aircraft provided in an embodiment of this application. The system includes:

[0147] The acquisition unit 901 is configured to acquire the current fuel level and current fuel density in the fuel tank of the aircraft, as well as the current attitude angle of the aircraft.

[0148] The generation unit 902 is configured to generate target feature information based on the current attitude angle and the current oil level.

[0149] The selection unit 903 is configured to select sample feature information to form a sample set based on the distance between the target feature information and the sample feature information in the preset database in ascending order; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information;

[0150] Prediction unit 904 is configured to use a Gaussian process regression model to perform prediction processing based on target feature information, sample set and sample oil volume corresponding to the sample set, and obtain prediction results, including an estimated value of the current oil volume.

[0151] The determining unit 905 is configured to determine the current oil weight based on the above estimate and the current oil density.

[0152] In some implementations, the prediction results also include posterior variance, which is used to quantify the uncertainty of the estimated values.

[0153] Unit 905 is also configured as follows:

[0154] The confidence level of the above estimates is determined based on the posterior variance.

[0155] In some implementations, the Gaussian process regression model includes a kernel function;

[0156] Prediction unit 904 is configured to use a Gaussian process regression model to perform prediction processing based on target feature information, a sample set, and the corresponding sample oil volume, including:

[0157] Using kernel functions, the first kernel matrix between the target feature information and the sample set, and the second kernel matrix between the sample sets are determined;

[0158] The above estimated values ​​are determined based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set.

[0159] In some implementations, the Gaussian process regression model includes observation noise variance and an identity matrix;

[0160] Prediction unit 904 is configured to determine the above estimated value based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set, including:

[0161] Based on the first kernel matrix, the second kernel matrix, the observation noise variance, the identity matrix, and the sample oil volume corresponding to the sample set, the posterior mean of the oil volume corresponding to the target feature information is determined, and the posterior mean is determined as the above estimated value.

[0162] In some implementations, the prediction results also include posterior variance, which is used to quantify the uncertainty of the estimated values.

[0163] Prediction unit 904 is configured to use a Gaussian process regression model to perform prediction processing based on target feature information, a sample set, and the corresponding sample oil volume, and also includes:

[0164] Using kernel functions, we determine the third kernel matrix between target feature information and target feature information, and the fourth kernel matrix between sample set and target feature information;

[0165] The posterior variance is determined based on the first, second, third, and fourth kernel matrices. In some implementations, the Gaussian process regression model includes the observation noise variance and the identity matrix;

[0166] Prediction unit 904 is configured to determine the posterior variance based on a first kernel matrix, a second kernel matrix, a third kernel matrix, and a fourth kernel matrix, including:

[0167] The posterior variance is determined based on the first kernel matrix, the second kernel matrix, the third kernel matrix, the fourth kernel matrix, the observation noise variance, and the identity matrix.

[0168] In some embodiments, the acquisition unit 901 is configured to acquire the current fuel level and current fuel density in the aircraft's fuel tank, and to acquire the aircraft's current attitude angle, including:

[0169] The current oil altitude and current oil density are obtained from the aircraft's fiber optic sensing measurement system;

[0170] Obtain the current attitude angle from the aircraft's avionics system.

[0171] In some embodiments, the fiber optic sensing measurement system includes a fiber optic density sensor and multiple fiber optic liquid level sensors;

[0172] The acquisition unit 901 is configured to acquire the current oil altitude and current oil density from the aircraft's fiber optic sensing measurement system, including:

[0173] The current oil level is obtained from the aforementioned multiple fiber optic level sensors;

[0174] The current oil density is obtained from the fiber optic density sensor.

[0175] In some implementations, the features in the target feature information and the features in each sample feature information are normalized features.

[0176] In some embodiments, the selection unit 903 is configured to select sample feature information to form a sample set based on the order of increasing distance between the target feature information and the sample feature information in a preset database, including:

[0177] A sample set is formed by selecting a preset number of sample feature information that are closest to the target feature information.

[0178] In some implementations, the Gaussian process regression model includes a kernel function, and the fuel tank fuel quantity measurement system also includes a training unit (not shown in the figure), which is configured as follows:

[0179] A training set is created by combining the feature information of a subset of samples in the sample set.

[0180] By maximizing the marginal log-likelihood function, the parameter values ​​of the hyperparameters of the kernel function in the Gaussian process regression model are estimated based on the training set, the number of sample feature information in the training set, and the oil volume of the corresponding sample in the training set.

[0181] Update the hyperparameters of the kernel function in the Gaussian process regression model using the above parameter values.

[0182] In some implementations, the training unit is further configured to:

[0183] The test set is composed of the feature information of samples other than the training set in the sample set.

[0184] After updating the hyperparameters of the kernel function in the Gaussian process regression model using the above parameter values, the Gaussian process regression model is used to predict the predicted oil volume corresponding to the feature information of each sample in the test set.

[0185] Based on the predicted oil volume and sample oil volume corresponding to the feature information of each sample in the test set, the prediction error is determined.

[0186] Based on the prediction error, the fit of the Gaussian process regression model is determined.

[0187] In some embodiments, the fuel tank fuel quantity measurement system includes a fuel computer as described above, with the units described above specifically located within the fuel computer.

[0188] It should be noted that other aspects and implementation details of the fuel tank quantity measurement system provided in some embodiments of this application are the same as or similar to the fuel tank quantity measurement method described above, and will not be repeated here.

[0189] This application also provides an aircraft, including, as described in the embodiments. Figure 9 The described fuel tank fuel level measurement system.

[0190] This application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, it implements the method of any of the above embodiments.

[0191] This application also provides a computer-readable storage medium storing a computer program thereon, which is loaded by a processor to execute the steps of any of the methods described in the above embodiments. In this application, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0192] Based on the foregoing description, the solution provided in this application addresses the following key issues in the related technologies by employing a Gaussian process-based calculation method:

[0193] For the relevant interpolation methods:

[0194] 1. Accuracy limitations: In areas with complex tank shapes or rapidly changing oil levels, the accuracy of relevant interpolation methods is insufficient;

[0195] 2. Prediction uncertainty problem: Related interpolation methods cannot provide prediction results with confidence.

[0196] For neural network methods:

[0197] 1. "Black box" characteristics: The decision-making process inside neural networks is difficult to interpret, and lacks the necessary interpretability in aviation safety-critical systems;

[0198] 2. High data dependency: A large amount of training data is required to achieve good performance;

[0199] 3. Weak uncertainty assessment capability: Standard neural networks cannot provide confident prediction results;

[0200] 4. High computational resource requirements: Training and optimizing complex neural networks requires high computational resources, making them unsuitable for airborne systems with limited computing power.

[0201] The solution provided in this application has the following significant beneficial effects:

[0202] 1. Significantly improved prediction accuracy: The nearest-neighbor local Gaussian process interpolation strategy can more accurately model complex nonlinear relationships, thereby improving prediction accuracy;

[0203] 2. Provides uncertainty quantification: The Gaussian process regression model not only provides a point estimate of the oil volume, but also provides an uncertainty estimate of the prediction results, giving a prediction result with confidence.

[0204] 3. Enhanced interpretability: Compared with "black box" models such as neural networks, Gaussian processes are stochastic processes and can be viewed as an infinite-dimensional Gaussian distribution. In a Gaussian process, the model's output (such as the predicted value of the regression function) is inferred from the correlation of given input points (determined by the covariance function) and the distribution of known data. Therefore, each step of the model's prediction can be traced back to the choice of the covariance function, the influence of the training data, and how the model handles uncertainty. This traceability and transparency make Gaussian processes highly interpretable.

[0205] Through the above improvements, this application achieves high-precision fuel quantity prediction, while enhancing the reliability of the results and ensuring the interpretability of the model, thereby effectively improving the performance and reliability of the aircraft's fuel tank fuel quantity measurement system and providing more stable and reliable technical support for flight safety.

[0206] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0207] The above provides a detailed description of the fuel tank fuel quantity measurement method, system, and aircraft of the aircraft provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for measuring the fuel quantity in an aircraft's fuel tank, characterized in that, include: Obtain the current fuel level and density in the aircraft's fuel tank, and obtain the current attitude angle of the aircraft; Based on the current attitude angle and the current oil level, target feature information is generated; Based on the order of increasing distance between the target feature information and the sample feature information in the preset database, the sample feature information is selected to form a sample set; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information; Using a Gaussian process regression model, a prediction process is performed based on the target feature information, the sample set, and the sample oil volume corresponding to the sample set to obtain a prediction result, which includes an estimated value of the current oil volume. Based on the estimated value and the current oil density, the current oil weight is determined.

2. The fuel tank quantity measurement method according to claim 1, characterized in that, The prediction results also include posterior variance, which is used to quantify the uncertainty of the estimated value; The fuel tank quantity measurement method also includes: The confidence level of the estimated value is determined based on the posterior variance.

3. The fuel tank quantity measurement method according to claim 1, characterized in that, The Gaussian process regression model includes a kernel function; The method of using a Gaussian process regression model to perform prediction processing based on the target feature information, the sample set, and the corresponding sample oil volume includes: Using the kernel function, determine the first kernel matrix between the target feature information and the sample set, and the second kernel matrix between the sample set and the sample set; The estimated value is determined based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set.

4. The fuel tank quantity measurement method according to claim 3, characterized in that, The Gaussian process regression model includes the observation noise variance and the identity matrix; Based on the first kernel matrix, the second kernel matrix, and the sample oil volume corresponding to the sample set, the estimated value is determined, including: Based on the first kernel matrix, the second kernel matrix, the observation noise variance, the identity matrix, and the sample oil volume corresponding to the sample set, the posterior mean of the oil volume corresponding to the target feature information is determined, and the posterior mean is determined as the estimated value.

5. The fuel tank quantity measurement method according to claim 3, characterized in that, The prediction results also include posterior variance, which is used to quantify the uncertainty of the estimated value; The method of using a Gaussian process regression model to perform prediction processing based on the target feature information, the sample set, and the sample oil volume corresponding to the sample set further includes: Using the kernel function, determine the target feature information and the third kernel matrix of the target feature information, and the sample set and the fourth kernel matrix of the target feature information; The posterior variance is determined based on the first kernel matrix, the second kernel matrix, the third kernel matrix, and the fourth kernel matrix.

6. The fuel tank quantity measurement method according to claim 5, characterized in that, The Gaussian process regression model includes the observation noise variance and the identity matrix; Determining the posterior variance based on the first kernel matrix, the second kernel matrix, the third kernel matrix, and the fourth kernel matrix includes: The posterior variance is determined based on the first kernel matrix, the second kernel matrix, the third kernel matrix, the fourth kernel matrix, the observation noise variance, and the identity matrix.

7. The fuel tank quantity measurement method according to claim 1, characterized in that, The acquisition of the current fuel level and current fuel density in the aircraft's fuel tank includes: The current oil altitude and current oil density are obtained from the aircraft's fiber optic sensing measurement system; Obtaining the current attitude angle of the aircraft includes: The current attitude angle is obtained from the avionics system of the aircraft.

8. The fuel tank quantity measurement method according to claim 7, characterized in that, The fiber optic sensing and measurement system includes a fiber optic density sensor and multiple fiber optic liquid level sensors. Obtaining the current oil altitude and current oil density from the aircraft's fiber optic sensing measurement system includes: The current oil level is obtained from the plurality of fiber optic level sensors; The current oil density is obtained from the fiber optic density sensor.

9. The method for measuring fuel quantity in a fuel tank according to any one of claims 1-8, characterized in that, Before performing prediction processing using a Gaussian process regression model based on the target feature information, the sample set, and the corresponding sample oil volume, the method further includes: A training set is formed by combining some of the sample feature information from the sample set. By maximizing the marginal log-likelihood function, the parameter values ​​of the hyperparameters of the kernel function in the Gaussian process regression model are estimated based on the training set, the number of sample feature information in the training set, and the oil volume of the sample corresponding to the training set. The hyperparameters of the kernel function in the Gaussian process regression model are updated using the parameter values.

10. The fuel tank quantity measurement method according to claim 9, characterized in that, Also includes: The sample feature information in the sample set other than the training set is used to form a test set; After updating the hyperparameters of the kernel function in the Gaussian process regression model using the parameter values, the method further includes: Using the Gaussian process regression model, the predicted oil volume corresponding to the feature information of each sample in the test set is predicted; Based on the predicted oil volume and the sample oil volume corresponding to the feature information of each sample in the test set, the prediction error is determined; Based on the prediction error, the fit of the Gaussian process regression model is determined.

11. A fuel level measurement system for an aircraft's fuel tank, characterized in that, include: The acquisition unit is configured to acquire the current fuel level and current fuel density in the fuel tank of the aircraft, and to acquire the current attitude angle of the aircraft; The generation unit is configured to generate target feature information based on the current attitude angle and the current oil level; The selection unit is configured to select sample feature information to form a sample set based on the distance between the target feature information and each sample feature information in the preset database in ascending order; wherein, each sample feature information is generated based on the corresponding sample attitude angle and sample oil height, and the preset database also includes the sample oil volume corresponding to the sample feature information; The prediction unit is configured to use a Gaussian process regression model to perform prediction processing based on the target feature information, the sample set, and the sample oil volume corresponding to the sample set, and obtain a prediction result, the prediction result including an estimated value of the current oil volume; The determining unit is configured to determine the current oil weight based on the estimated value and the current oil density.

12. An aircraft, characterized in that, Includes the fuel tank fuel quantity measurement system as described in claim 11.

Images